Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities [1st ed.] 978-981-13-7947-5;978-981-13-7948-2

Table of contents :
Front Matter ....Pages i-xiv
Introduction (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 1-22
VCE Overpressure Prediction Using Empirical Methods (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 23-44
VCE Overpressure Prediction by CFD Modelling (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 45-79
CFD-Based Overpressure Prediction for Single Modules-Extended GAME Correlation (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 81-109
CFD-Based Overpressure Prediction for Double Modules—Data-Dump Technique (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 111-127
CFD-Based Overpressure Prediction for Congested Multi-Modules—Safety Gap Effect (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 129-151
Risk Analysis Methods for Gas Explosion (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 153-172
Event Tree Analysis of Offshore Hydrocarbon Release Events (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 173-189
Bayesian Network Analysis of Explosion Events at Petrol Stations (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 191-217
Grid-Based Risk Screening for Explosion Accidents at Large Onshore Facilities (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 219-237
Multi-Level Explosion Risk Analysis for VCEs in Super-Large FLNG Facilities (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 239-266
CFD-Based Explosion Risk Analysis of Blast Wall Effects on FLNG Platforms (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 267-288
Standard-Based Lifecycle Risk Management of Explosion Events (Guowei Ma, Yimiao Huang, Jingde Li)....Pages 289-310

Citation preview

Guowei Ma · Yimiao Huang · Jingde Li

Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities

Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities

Guowei Ma Yimiao Huang Jingde Li •

•

Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities

123

Guowei Ma School of Civil and Transportation Engineering Hebei University of Technology Tianjin, China

Yimiao Huang Department of Civil, Environmental and Mining Engineering, School of Engineering University of Western Australia Perth, WA, Australia

Jingde Li Centre for Infrastructural Monitoring and Protection Curtin University Perth, WA, Australia

ISBN 978-981-13-7947-5 ISBN 978-981-13-7948-2 https://doi.org/10.1007/978-981-13-7948-2

(eBook)

© Springer Nature Singapore Pte Ltd. 2019 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Preface

In the oil and gas industry, a vapour cloud explosion (VCE) induced by oil and gas leaks is one of the major threats to process safety that may lead to catastrophic consequences. Risk evaluation of such gas explosions is complicated as a large number of factors and sequential consequences, including gas leak, cloud formation, ignition location and strength, structure congestion and confinement, humanity and environment conditions, etc., should be considered. Further complexity arises from the specific site characteristics of different process facilities, i.e. petrol stations, process factories, underground pipelines, offshore platforms, which have strong influences on the explosion risk. The main purpose of this book is to introduce new risk analysis and load prediction methods proposed and developed by the authors. It provides brief descriptions of VCE mechanisms and the currently prevailing risk analysis models and tools, wherein no attempt is intended to cover all aspects of explosion risk analysis. Readers are expected to explore in-depth information according to the literature referred in the context. This book is divided into 13 chapters. Chapter 1 is a brief introduction and review of the current state-of-the-art for the risk analysis of the oil and gas explosion. Chapters 2–13 elaborate and exemplify mainly load predictions and risk analyses. Focus is biased on the load prediction part because blast effect is the most significant factor in consequence prediction, and it is the base for evaluation of structural damages, human losses, environmental issues, etc. The contents of each chapter are briefed below for reader’s quick reference to identify relevant contexts. Specifically, Chap. 1 is an introductory section, which gives the background for the researches in this book and the basic information of VCE mechanisms, such as cloud formation, ignition and structure condition. Followed are Chaps. 2–6 discussing load prediction methods of both empirical and computational fluid dynamics (CFD) in details, wherein new approaches for gas explosion prediction and accuracy improvement by using CFD model proposed by the authors are included. Figure 1 shows an example of gas explosion simulation by the authors using one of the most advanced CFD analysis software called “FLACS”.

v

vi

Preface

Fig. 1 FLACS simulation example of gas explosion

Chapter 2 illustrates three most prevailing empirical methods for load prediction, including the TNT equivalent model, the TNO Multi-Energy model and the Baker– Strehlow–Tang model. Case studies of all models by using DNV PHAST are given. Chapter 3 presents the principles of the CFD-based overpressure calculation approach. The gas explosion mechanism regarding fluid flow equations, thermodynamic relationships, and turbulence and combustion modelling are described, and the numerical simulation procedure is introduced. Chapter 4 demonstrates explosion load predictions for a single module based on the CFD simulations. A newly derived correlation, confinement specific correlation (CSC) and performance validation of CSC are presented and discussed. Chapter 5 introduces explosion overpressure calculations of two separate modules with consideration of the gap effects. A data-dump guideline is then provided to improve the overpressure prediction accuracy prior to further investigation of safety gaps. Chapter 6 evaluates blast effects for a super-large cylindrical floating liquefied natural gas (FLNG) vessel. Gas explosion simulations of different settings for safety gaps are covered. The second half of this book, Chaps. 7–13, presents the most popular and generally applied explosion risk analysis methods from theoretical to standardized approaches. Trendy developments in explosion risk analysis methods, such as a confidence-level-based event tree analysis, a multi-level evaluation and a grid-based model, are introduced. Figure 2 depicts an example of a grid-based risk screening map of a living area with a VCE at the centre.

Preface

vii

Fig. 2 Example of 3D grid-based risk screening method

Chapter 7 gives a broad literature review on the state-of-the-art explosion risk analysis methods including both qualitative and quantitative approaches, such as risk checklist, HAZOP, HAZIP, event tree, fault tree, and Bayesian network. Chapter 8 describes an event-tree-based risk analysis of hydrocarbon release accidents, which is the source for the VCE formation. A fuzzy-theory-based confidence level method for reducing uncertainties is also described. Chapter 9 is the description of a Bayesian-network-based quantitative risk analysis method for VCE accidents occurring at small oil and gas facilities, such as petrol stations. A case study using the proposed method to model the complete explosion process is presented. Chapter 10 demonstrates a grid-based risk mapping method of explosion risk screening for large oil and gas facilities combined with the surrounding living areas. Exemplification is demonstrated with a refinery factory. Chapter 11 illustrates a multi-level explosion risk analysis method for super-large oil and gas facilities, so as of the FLNG platform. The CFD method is applied for detailed risk quantification, and an as low as reasonably practical (ALARP) method is described as a calibration tool. Chapter 12 demonstrates a detailed quantitative risk assessment method using CFD modelling and exceedance curves. The effect of blast wall on the cylindrical FLNG to reduce explosion risks is also investigated. Chapter 13 outlines the prevailing explosion risk management methods based on worldwide industrial standards, wherein a lifecycle risk management process for explosion accidents based on structural integrity management (SIM) is introduced. Tianjin, China Perth, Australia Perth, Australia

Guowei Ma Yimiao Huang Jingde Li

Contents

1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Formation of VCEs . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Gas Release . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Dispersion and Ventilation . . . . . . . . . . . . . 1.3 Ignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Combustion Properties of Fuel–Air Mixtures . . . . . . 1.5 Deflagrations and Detonations . . . . . . . . . . . . . . . . . 1.5.1 Deflagrations . . . . . . . . . . . . . . . . . . . . . . . 1.5.2 Detonation . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.3 Deflagration to Detonation Transition (DDT) 1.6 Facility Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 1.7 Overpressure Prediction . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

1 1 5 5 6 8 9 11 11 13 15 15 18 20

2

VCE 2.1 2.2 2.3 2.4

Overpressure Prediction Using Empirical Methods TNT Equivalency Model . . . . . . . . . . . . . . . . . . . . . TNO Multi-Energy Model . . . . . . . . . . . . . . . . . . . . Baker–Strehlow–Tang Model . . . . . . . . . . . . . . . . . . Applications of Empirical Models Using PHAST . . . 2.4.1 Introduction to DNV PHAST . . . . . . . . . . . 2.4.2 Case Study . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

23 23 25 30 33 33 35 41 43

3

VCE Overpressure Prediction by CFD Modelling . 3.1 Methodology of CFD Simulation . . . . . . . . . . 3.1.1 Fluid Flow Equations . . . . . . . . . . . . 3.1.2 Stoichiometry . . . . . . . . . . . . . . . . . . 3.1.3 Thermodynamic Relationships . . . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

45 45 45 46 47

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

ix

x

Contents

3.1.4 Ignition Process . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.5 Geometry Counting and Porosity Calculations . . . . 3.1.6 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . 3.2 Modelling Turbulence and Combustion . . . . . . . . . . . . . . . 3.2.1 Turbulence Model . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Combustion Model . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Numerical Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Conservation Equations in Finite Domain . . . . . . . 3.3.2 Finite Domain Calculations . . . . . . . . . . . . . . . . . . 3.3.3 The Continuity and Momentum Equations in Finite Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.4 Calculation Procedure . . . . . . . . . . . . . . . . . . . . . . 3.4 Simulation Flow Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Simulation Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.2 Definition of Grid and Calculation of Porosities . . . 3.5.3 Gas Dispersion Simulation . . . . . . . . . . . . . . . . . . 3.5.4 Gas Explosion Simulation . . . . . . . . . . . . . . . . . . . 3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

CFD-Based Overpressure Prediction for Single ModulesExtended GAME Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 The GAME Correlations and CFD Case Studies . . . . . . . . . 4.2.1 GAME Correlations . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Modules Tested in CFD Simulations . . . . . . . . . . . 4.2.3 Verification of GAME Correlation by Case Study . 4.3 Parametric Studies and Development of a New Correlation . 4.3.1 Conceptual Definition of Confinement and Congestion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Definition of Parameters . . . . . . . . . . . . . . . . . . . . 4.3.3 Proposition of New Correlation . . . . . . . . . . . . . . . 4.3.4 Verification of the New Correlation . . . . . . . . . . . . 4.3.5 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Evaluation of Explosion Overpressures in Irregularly Structured Configurations Using New Correlation . . . . . . . . 4.4.1 Definition of Regularity or Irregularity of Confinement and Congestion . . . . . . . . . . . . . . . . . 4.4.2 Application of CSC to Irregularly Arranged Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.3 Rapid Prediction of Structural Damage . . . . . . . . . 4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

48 50 51 53 53 53 56 56 56

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

57 58 58 60 60 60 61 65 75 78

. . . . . . .

. . . . . . .

. . . . . . .

81 81 82 82 83 85 88

. . . . .

. . . . .

. . . . .

88 89 92 93 94

...

98

...

98

. . . .

. . . .

. 99 . 103 . 107 . 108

Contents

5

6

7

xi

CFD-Based Overpressure Prediction for Double Modules—Data-Dump Technique . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Numerical Models for Separated Congestions in Double Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Experimental Set-up . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 CFD Modelling Using FLACS . . . . . . . . . . . . . . . . 5.3 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Gas Explosion Overpressures with a Small Separation Distance in the Double Modules . . . . . . . . . . . . . . . 5.3.2 Gas Explosion Overpressures with a Large Separation Distance in the Double Modules . . . . . . . . . . . . . . . 5.3.3 Data-Dump Technique . . . . . . . . . . . . . . . . . . . . . . 5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . 111 . . 111 . . . .

. . . .

112 112 113 116

. . 116 . . . .

. . . .

119 121 125 127

. . . . . .

. . . . . .

129 129 130 131 131 134

CFD-Based Overpressure Prediction for Congested Multi-Modules—Safety Gap Effect . . . . . . . . . . . . . . . . . . . . . . 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Numerical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.1 Ship-Shaped FLNG . . . . . . . . . . . . . . . . . . . . . . . 6.2.2 Cylindrical FLNG . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Safety Gap Effect on Gas Explosion . . . . . . . . . . . . . . . . . 6.3.1 Near-Field Gas Explosion Simulation of Cylindrical FLNG Platform . . . . . . . . . . . . . . . . 6.3.2 Far-Field Gas Explosion Simulation of Cylindrical FLNG Platform . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . 138 . . . 148 . . . 149

Risk Analysis Methods for Gas Explosion . . . . . . . . . . . . 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Qualitative Risk Analysis . . . . . . . . . . . . . . . . . . . . 7.2.1 Hazard Checklists for Offshore Installations . 7.2.2 HAZOP . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.3 HAZIDs . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.4 Qualitative Methods for Explosion Risks . . . 7.3 Quantitative Risk Analysis . . . . . . . . . . . . . . . . . . . 7.3.1 Event Tree and Fault Tree Analysis . . . . . . . 7.3.2 Bayesian Network Modelling . . . . . . . . . . . 7.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . .

. . . 134

. . . . . . . . . . . .

. . . . . . . . . . . .

153 153 154 155 156 158 159 162 162 166 170 171

xii

8

9

Contents

. . 173 . . 173 . . 175

Event Tree Analysis of Offshore Hydrocarbon Release Events . . 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 BORA-Release Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Application of the Confidence Level Method to BORA Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.1 Analysis Using L-R Bell-Shaped Fuzzy Number . . . 8.3.2 Fuzzy Calculations . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.3 Defining Confidence Level of RIF-Scoring Process . 8.3.4 Deciding Degree of Optimism and Defuzzifying Final Fuzzy Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.5 Assumptions for Practical Implementation of Proposed Method . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4 Case Study—Application of Proposed Method to BORA Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . 183 . . 187 . . 188

Bayesian Network Analysis of Explosion Events at Petrol Stations . . . . . . . . . . . . . . . . . . . . . . . . . 9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 9.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . 9.2.1 Bayesian Network Modelling . . . . . 9.2.2 Quantification of BN . . . . . . . . . . . 9.2.3 Calculation of BN . . . . . . . . . . . . . 9.3 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . 9.3.1 Quantification of BN . . . . . . . . . . . 9.3.2 Calculation and Discussion . . . . . . . 9.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . .

. . . .

177 177 179 180

. . 181 . . 182

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

191 191 192 193 193 198 199 200 212 215 215

10 Grid-Based Risk Screening for Explosion Accidents at Large Onshore Facilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2 Grid-Based Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.3 Bayesian Network Modelling . . . . . . . . . . . . . . . . . . . . . . 10.3.1 Quantification of Bayesian Network . . . . . . . . . . 10.3.2 Calculation of Bayesian Network . . . . . . . . . . . . 10.3.3 Matrix Calculation and Result Display . . . . . . . . . 10.4 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.4.1 Quantification of Bayesian Network . . . . . . . . . . 10.4.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . 10.4.3 Mesh Convergence . . . . . . . . . . . . . . . . . . . . . . . 10.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

219 219 220 221 222 223 225 226 227 230 233 236 236

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

Contents

xiii

11 Multi-Level Explosion Risk Analysis for VCEs in Super-Large FLNG Facilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Multi-Level Explosion Risk Analysis . . . . . . . . . . . . . . . . . 11.2.1 First Level of Qualitative Risk Screening . . . . . . . . 11.2.2 Second Level of Semi-Quantitative Risk Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.3 Third Level of Quantitative Risk Assessment . . . . . 11.3 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.3.1 Qualitative Risk Screening of Cylindrical FLNG . . 11.3.2 Semi-Quantitative Risk Classification . . . . . . . . . . 11.3.3 Detailed Quantitative Risk Assessment . . . . . . . . . 11.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . .

. . . .

. . . .

239 239 241 242

. . . . . . . .

. . . . . . . .

. . . . . . . .

246 250 252 253 254 255 264 265

12 CFD-Based Explosion Risk Analysis of Blast Wall Effects on FLNG Platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.2 Numerical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3 Inputs for Gas Dispersion and Explosion Simulations . . . . . 12.4 Design of Blast Walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.5 Dispersion and Explosion Analyses . . . . . . . . . . . . . . . . . . 12.5.1 Dispersion Analyses . . . . . . . . . . . . . . . . . . . . . . . 12.5.2 Dispersion Simulation Results and Discussions . . . 12.5.3 Explosion Simulations . . . . . . . . . . . . . . . . . . . . . 12.5.4 Results and Discussions for Explosion Simulations 12.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

267 267 268 268 271 272 272 274 277 278 286 287

13 Standard-Based Lifecycle Risk Management of Explosion Events . . . . . . . . . . . . . . . . . . . . . . . . . 13.1 Introduction to Explosion Risk Management . . 13.1.1 Risk Management . . . . . . . . . . . . . . . . 13.1.2 Structural Integrity Management . . . . . 13.1.3 Lifecycle Explosion Risk Management 13.2 Data Collection . . . . . . . . . . . . . . . . . . . . . . . . 13.2.1 Facility Characteristic Data . . . . . . . . . 13.2.2 Structural Condition Data . . . . . . . . . . 13.2.3 Explosion-Related Data . . . . . . . . . . . 13.2.4 Data Management . . . . . . . . . . . . . . . 13.3 Risk Assessment . . . . . . . . . . . . . . . . . . . . . . . 13.3.1 Risk Assessment Process . . . . . . . . . . 13.3.2 Detailed Assessment . . . . . . . . . . . . . . 13.3.3 Risk Acceptance Criteria . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

289 289 289 291 293 293 293 294 294 295 295 295 297 303

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

xiv

Contents

13.4 Risk Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . 13.4.1 Explosion Risk Reduction Measures . . . . 13.4.2 Structural Strengthening . . . . . . . . . . . . . 13.5 Risk Management Implementation and Monitoring 13.5.1 Implementation . . . . . . . . . . . . . . . . . . . 13.5.2 Monitoring . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

304 304 304 308 308 308 309

Chapter 1

Introduction

Abstract This chapter is an introductory section which gives the background for the researches in this book. Several disastrous explosion events from various types of oil and gas facilities are discussed. Vapour cloud explosion (VCE) is defined and the basic VCE mechanisms, such as cloud formation, ignition, deformation and detonation are described. In addition, facility conditions and several overpressure predication methods are briefly introduced.

1.1 Background Historically, oil had been used for lighting purposes for thousands of years till the nineteenth century, whence it replaced most other fuels. For example, the oil was adopted as power fuel in the automobile industry and the aircraft was designed and practically put into operation due to the invention of gasoline engines. By replacing the coal-powered engines, the ships driven by oil move up to twice the speed as they used to. Therefore, exploration and investment in natural sources of energy, such as oil and gas, show a constant boom worldwide. According to the statistic review report of BP (2017), oil consumption has increased from around 0.6 million barrels daily (Mb/d) to approximately 1.6 Mb/d over the last twenty years. However, in concomitant with the oil and gas boom, one of the largest dangers to human life, assets and environment arises from the potential explosive risks associated with the petroleum-related operations and products, which vary in nature from reservoir fluids to methane. The gas explosion, which is also termed as the vapour cloud explosion (VCE) is one of the most potential severe hazards on oil and gas facilities. Many onshore and offshore VCE accidents have been reported with catastrophic consequences in view of human being casualty, economic loss and irreversible environmental contamination. For example, Fig. 1.1 shows some of the most disastrous explosion accidents in the industry. On 6 July 1988, an explosion destroyed the Piper Alpha oil platform in North Sea and led to a large oil fire as shown in Fig. 1.1a (O’Byrne, 2011). The accident was triggered by massive leakage of gas condensate, which was later ignited and caused the gas explosion with extreme heat. The gas explosion and fire lasted only 22 min. © Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_1

1

2

1 Introduction

(a) Piper Alpha platform explosion

(b) Ghana petrol station explosion (Bella

(Broadcast Greenlight, 2013)

Naija, 2015)

(c) Kaohsiung pipeline explosion (Mega,

(d) Texas BP refinery factory explosion

2014)

(The Guardian, 2010)

(e) Guadalajara sewer explosion (Vertigo

(f) BP Deepwater Horizon explosion (The

Politico, 2015)

Telegraph, 2015)

Fig. 1.1 Catastrophic explosion events (online resources and links are provided)

1.1 Background

3

However, 167 men died, while only 61 survived (Hendershot, 2013). As one of the worst offshore oil disasters in history, this accident destroyed approximately 10% of North Sea oil and gas supply during 1980s and the economic loss was estimated in an excess of US$3 billion (Patecornell, 1993). Another catastrophic fire and explosion event occurred at a petrol station as a result of a flood in Accra, Ghana, on 4 June 2015 and caused over 152 fatalities (Asumadu-Sarkodie, Owusu, & Rufangura, 2015). The fuel tanks in the station were believed to be the source of the explosion. Around 100 people sheltered inside the petrol station from heavy rains and flooding were killed by fire. More people were implicated because water from the flood expelled the fire to the nearby buildings. Figure 1.1b shows the destroyed petrol station after the accident. On 31 July 2014, a series of gas explosions from an underground pipeline took place in Kaohsiung, Taiwan, which caused 32 fatalities and 321 injuries. More than four main roads with a total length of approximately 6 km were damaged and traffic was blocked for several months (Liaw, 2016). Figure 1.1c demonstrates the road damage after explosions. The explosion was induced by the leakage of propylene and led to around NT$1.9 billion economic expense for rebuilding the damaged roads and drainage system. On 23 March 2005, a hydrocarbon vapour cloud was ignited and exploded at the isomerization (ISOM) process unit at BP’s Texas City refinery. It caused fatal blunt force trauma to 15 people in and around the trailers and injured over 180 others (CSB, 2005). Figure 1.1d shows the accident site after the explosion. According to BP’s accident investigation report, heavier-than-air hydrocarbon vapours were combusted and exploded when reached by an ignition source, probably an engine of a running vehicle. The explosion incurred approximately US$1.5 billion economic loss. Figure 1.1e shows the damage after the Guadalajara explosion that occurred on 22 April 1992. It was claimed with 206 people casualties, nearly 500–600 missing and 1800 injured during the accident. However, more than 1000 deaths were estimated by the catastrophe as the day was a market day and the nearby roads were crowded with people (ARIA, 2007). The explosion led to the destruction of 13 km street roads, 1500 houses, 1224 buildings and 637 vehicles. The total estimated potential financial loss ranges from $300 million to US$1 billion. Another explosion accident happened on 20 April 2010 in the Gulf of Mexico. The Deepwater Horizon semi-submersible mobile offshore drilling unit (MODU) was undermined by the gas explosion and the subsequent fire (Reader & O’Connor, 2014). According to the BP report (CSB, 2014), there was an initial hydrocarbon release being ingested into the air intakes of the diesel generator. The emitted hot gas from the exhaust outlets drove the vapour cloud to blow out, which killed 11 people (Dadashzadeh, Abbassi, Khan, & Hawboldt, 2013). After burning for more than a day, Deepwater Horizon sank on 22 April 2010 followed by a massive offshore oil spill in the Gulf of Mexico. It was estimated that BP could be required to pay penalties up to US$90 billion. The aforementioned accidents are just the most sensational gas explosions tagged as the most catastrophic records either in people casualties, economic loss or environmental contamination. In an analysis of those historical accidents, it is found that

4

1 Introduction

the very cause as well the initial consequence in the chain of catastrophic events is the failure of structural components, which is due to the extreme loads generated from the accidental gas explosions. However, it is not feasible for the engineers to effectively consider the accidental explosion risks in designing structures because such extreme events normally have a very low probability to occur. If a conservative design is adopted to accommodate the possible extreme explosion risk, construction cost may not be acceptable, and the facilities may not fit for its intended functions because of the overbulky structures. Since the large overpressures from the gas explosions are very unpredictable in nature and not proficient to be calculated with the current state-of-the-art computational technology and facilities, it is of practical engineering significance to research the gas explosion mechanism and risks to achieve the trade-off balance between the safety and the cost. Thorough understanding of VCE risks can help to reduce the human loss and environmental damage by application of appropriate gas explosion mitigating approaches. VCE is defined as ‘an explosion resulting from an ignition of a premixed cloud of flammable vapor, gas or spray with air, in which flames accelerate to sufficiently high velocities to produce significant overpressure’ (Mercx & van den Berg, 2005). Figure 1.2 illustrates the overall process from gas releases to VCE and the subsequent consequences. Both release and ignition should present to ensue fire or explosion accidents (Bjerketvedt, Bakke, & vanWingerden, 1997). To evaluate the risks of vapour cloud explosions, it is essential to consider a large array of random variables, such as gas property, facility conditions and weather situations in comprehension of multiple consequences, i.e. structural damage, human loss, environmental effects and so on. For different types of process facilities, environments and structures are variable, whereby appropriate respective methods should be applied in consideration of the specific site characteristics to achieve the reliability and efficiency of explosion risk assessments. In view of that, some of the most important characteristics of VCE, such as the formation of vapour clouds, ignition properties, combustion properties and structural conditions are elaborated in the following sections. The last section of this chapter briefly introduces the overpressure

Gas Leak

Not Ignited

Ignited Immediately

Fuel-Air Cloud Formatted and Ignited

Fire

Explosion

Damage to Personnel Or Structures

Fig. 1.2 Typical process of gas explosion

No Further Damage

Consequent Fire or Explosion

1.2 Formation of VCEs

5

prediction methods. The explosion risk analysis by combining risk analysis methods and the detailed load prediction approaches is discussed.

1.2 Formation of VCEs 1.2.1 Gas Release Various types of flammable gases, which may be used for energy purpose or as a reactant in the chemical process, are daily items to be dealt with in the process industry. Safety measures with regard to those gases have been given much attention, particularly in offshore petrochemical industry, where large amounts of flammable gas and oil are processed in a limited, usually confined and highly congested areas. There are various potential sources for the accidental gas releases and most of them can be categorized into one or more of the following types: • failure of pipes or vessels containing gas due to external physical impact, • failure of valve or gas filled pipe/vessel due to ageing, fatigue and lack of appropriate maintenance, • pressurized system overloaded with operating conditions exceeding the design specifications to blowout, • artefacts or human errors. It is significant to understand the features of the different types of the gas leak to estimate the potential consequences. Generally, the leak rate depends on the gas flow rate in the pipe, the volume and the pressure of the inventory or pressurized vessel, as well as the leakage hole size and shape. Whereas the leak momentum depends also on whether the release is blocked by equipment or walls. If effectively ventilated, the momentum is reduced quickly and may restrain the mixing of gases and the gas cloud size from expanding. The duration of the leak or the decay of the leak rate depends on various factors. Precautious measures of point detectors, line detectors or room detectors based on video pictures and image analysis may identify the gas leak in its early stage. Alternatively, falling pressure in an inventory may be detected in the control room by appropriately deployed sonic detectors or the operation staffs through the abnormal performance of the facilities. Time-dependent leak rate input for a numerical simulation is usually estimated based on the hole size/shape, mixture composition, pressure/temperature, volume of inventory and the risk reduction measure applied for gas detection. Release frequencies can be calculated according to the historical recorded frequencies, setting of the potential leak exposed equipment/piping and special considerations in human factors, pressure level, temperature, etc.

6

1 Introduction

1.2.2 Dispersion and Ventilation Gas dispersion refers to the phenomenon of that gas leaks mix with the surrounding air and dilute. In the dispersion process, turbulent mixing, density differences due to the different molecular weight or temperature and diffusion govern the mixing. Leaks with a high velocity and momentum will disperse much faster than those with a lower momentum do. Wherefore, the gas concentration in the cloud depends on the ventilation pattern and strength. The amount of potentially flammable gas within a region will increase until the amount of gas equivalent to the leaked amount is transported away due to ventilation. Therefore, a lower ventilation rate will cause a denser gas cloud. The dispersion pattern varies with different release sources. A typical jet release of gas, either two phases or evaporating liquid flows, or a diffuse release for evaporating pool is given in Fig. 1.3. The jet release has a high momentum and forms a strong flow field due to the additional air entrainment. For a medium and large jet release, the dispersion pattern in a module is dominated by the leak characteristics, while the practical concentration level depends on the natural and/or designed ventilation. For a small jet release, both the dispersion pattern and the concentration level are dominated by the ventilation. For the evaporating pool, the dispersion process is dominated by wind force and buoyancy. Compared to the jet releases, the flow velocity of the diffuse releases is usually much lower. If the evaporating liquid forms a dense gas cloud, a layer of combustible gas may be formed at the ground level or in the low compartment. Alternatively, a dense gas cloud in open area may disperse into confined spaces such as a building, which could lead to more severe accidents. Simple models for gas dispersion usually assume constant wind velocity inside the module for the ventilation rate prediction. The amount of gas in kg accumulating in a module may thus be calculated by the following equation: Mg =

L vent Rleak × × 3600 L module Rventilation

Fig. 1.3 Jet release and pool evaporation

(1.1)

1.2 Formation of VCEs

7

wherein M g is the mass of gas in the process area (kg); L vent is the distance of the module vent to the end of congested area (m); L module is the length of the module (m); Rleak is the gas leak rate (kg/s) and R ventilation is the change rate of ventilation in air. This model usually gives effective estimates with regard to the amount of gas inside a module if the ventilation flow field is approximately uniform. On the other hand, uncertainties of the model may become intractable for complicated ventilation patterns. Another improficiency of this model is that the degree of mixing is not inferred, wherein the information about the degree of mixing and concentration are essential inputs for a non-conservative evaluation of the explosion hazard. It is because the combustible feature of the natural gas occurs only in a limited concentration range. It is indispensable for the detailed explosion risk assessment to engage a dispersion model to identify the gas concentration level and distribution. Since the geometry details exert a strong influence on the turbulent mixing with air, a 3D computational fluid dynamic (3D-CFD) model should be adopted to approach the arbitrary geometry layouts. Such a model is promising to simulate effectively the complicated ventilation patterns. Advanced CFD models using distributed porosity concept, e.g. FLACS, or a high degree of unstructured grid refinement, so as of FLUENT, can be applied to simulate fuel–air mixing due to flow gradients and interaction with equipment. Those 3D gas dispersion simulations predict sufficiently reliable gas concentration levels for a given leak scenario and it could be applied for deriving optimal geometry and gas detector layouts (Fig. 1.4).

Fig. 1.4 FLACS simulation example of gas dispersion

8

1 Introduction

1.3 Ignition A combustible gas cloud may be ignited if sufficient energy approaches to heat some volume of the gas cloud to the temperature required for the chemical reaction to be initiated. If the heat released in the chemical reaction is sufficient to heat the layers of gas in the immediate vicinity of ignition above the necessary initiation vault, the flame from the locally ignited gas will propagate. The minimum ignition energy is the lowest energy for a localized ignition source, such as a spark, to sufficiently ignite a premixed fuel–air cloud. As shown in Fig. 1.5, the minimum ignition energy depends on the specific fuel type and the concentration. It can be seen that sparks with energy between 0.1 and 0.3 mJ in the normal air may ignite methane. The required energy increases as the gas concentration changes towards the flammability limits. In simple summary, a very weak electrical spark may ignite methane as well as other hydrocarbon gases. Wherefore, explosion proof equipment should be in positions in the potential gas leak regions. When a flammable mixture is heated up to a critical temperature, the chemical reaction will start spontaneously. This lowest temperature to ignite a gas cloud is called the auto-ignition temperature (AIT). Figure 1.6 shows the effect of temperature for fuel oxidiser. For most pure hydrocarbon derivatives in the air, the AITs span the range from 210 to 540 °C. For instance, AIT for methane is 540 °C, propane is 450 °C, and hydrogen and ethylene are both 520 °C. For some of the heavier hydrocarbons, AIT is slightly above 200 °C such as hexane with AIT at about 200 °C. As precautious measures in practical risk management, AIT should be prevented from occurring in the areas where potential leaked gas could be incurred by malfunction of equipment. It is worth mentioning that equipment with surface temperature significantly higher than AIT does not necessarily ignite the gas since the heated gas concentration is not stable and may not satisfy the ignition requirement. The heated gas is usually dispersed away due to the density differences before the AIT is reached within the gas. Wherefore, for off stoichiometric concentrations, the AIT is higher.

10

Min. Ignition Energy, mJ

Fig. 1.5 Minimum ignition energy for methane in air (Bjerketvedt et al., 1997, permission from Elsevier)

1

0.1 5

7

9

% methane in air

11

13

1.3 Ignition 100% UFL

Fuel Concentration

Fig. 1.6 AIT and the effect of temperature on lower and upper flammable limits (Bjerketvedt et al., 1997, permission from Elsevier)

9

Flammable mixture

Autoignition

LFL 0%

25°C

Temperature

Ignition location plays a significant role in the intensity of the explosion loads. An explosion load may increase due to • ignition at walls, solid edges or corners, • increasing flame distance, • precompression effects (fast flames are burning towards confining walls or corners, pressure ahead of flame gets elevated and the pressure at flame arrival may become higher than 8 bar), whereas short distance from ignition to vent area induces low explosion loads. The influence of the ignition location may vary with the size of the geometry, the size of the gas cloud, the congestion level of the geometry and the reactivity of the gas cloud. A low- or moderate-strength explosion usually involves low congestion, small cloud size or low reactivity of fuel. Central ignition or wall ignition with long distance to the nearest significant vent opening tends to incur intensive explosion loads in view of both the pressure level and the duration. Stronger explosions in the sense of high pressure, compared to the central ignition, are often generated by the ignition occurring at the end of the gas cloud, whereby the distance between the flame to the vent opening is the maximum.

1.4 Combustion Properties of Fuel–Air Mixtures Components in natural gas, such as alkanes, methane, propane, ethane and butane, usually have a quite narrow flammable range in view of gas concentration. This range is called flammability limits and a premixed fuel–air mixture will burn only if the fuel concentration is in the range between the upper flammability limit (UFL) and the lower flammability limit (LFL). Figure 1.7 shows graphically the flammable range for some fuel–air mixtures.

10

1 Introduction

Fig. 1.7 Flammability limits for fuel–air mixtures at 1 atm and 25 °C (Bjerketvedt et al., 1997, permission from Elsevier)

Hydrogen Ethylene Propane Methane

0

20

40

60

80

100

volumtric % fuel in fuel-air

The wide flammable range of hydrogen implies that it is easy for this type of gas to form a flammable cloud in air. For propane and methane, the flammable range is much narrower. On the other hand, even if the concentration goes beyond UFL, the cloud still could possibly be ignited and exploded during the dilution process. Therefore, it is good practice to be on the safe side by maintaining the concentration below LFL. The flammable range is widened when the initial temperature is increasing as shown in Fig. 1.7. For hydrocarbons, changes in initial pressure will not affect LFL significantly, whereas the UFL increases. Flammability limits for fuel mixtures may be calculated by Le Chatelier’s law LFLMix =

100 C1 /LFL1 + C2 /LFL2 + · · · + Ci /LFLi

(1.2)

wherein C 1 , C 2 , …, C i [vol.%] are the proportion of each gas in the fuel mixture exclusive of air (Kuchta, 1985). It should be mentioned that this formula is only valid for the chemically similar components in the mixture. It does not work properly for H2 and unsaturated hydrocarbons. For a gas cloud to be practically dangerous, sufficient volumes should be within the flammable range wherein close to stoichiometric concentrations. Therefore, thorough mixing is prerequisit for the dangerous gas cloud to be formed. In a well-ventilated or low confined area, a high-momentum leak is usually required to generate the most dangerous gas clouds. Potential danger only lies in the jet interaction with small-scale equipment. On the other hand, leaks with a reduced momentum due to jet impingement on equipment, decks and walls, or liquid leak flashes typically generate smaller gas cloud with higher concentration gradients. Experiments and simulations indicate that very low explosion loads would be induced in such a situation. In a quite confined area with gross insufficient ventilation, time interval allowed for mixing will be much longer. At the end of the steady state, even leaks with a reduced momentum due to repeated impingement with walls or evaporation from pools give very homogenous and highly reactive gas clouds.

1.4 Combustion Properties of Fuel–Air Mixtures

11

It needs more comprehensive understanding and data than currently available to generate specific guidelines on how to prevent the dangerous flammable gas cloud from appearing. Herein, some critical factors are summarized below: • • • •

ventilation conditions, leak conditions, geometry, gas detections.

1.5 Deflagrations and Detonations 1.5.1 Deflagrations In an accidental gas explosion, deflagration is the most common mode of flame propagation. It is defined as an explosion wherein combustion propagates at subsonic velocities relative to the unburned gas immediately ahead of the flame. It implies that the burning velocity is smaller than the speed of sound in the unburned gas. The unburned gas ahead of the flame is expelled into motion by the expansion of the combustion products. The flame speed ranges from a few m/s up to 500–1000 m/s in the deflagration mode and the explosion pressure will range from a few mbar to several bars depending on the flame speed. As mentioned above, the flame speed and explosion pressure strongly depend on the conditions of the gas cloud, i.e. size, mixture, concentration, mixing level, etc. and the geometrical conditions confining the cloud. However, it is still not efficient to predict the flame speed and explosion pressure for a deflagration even if scenario parameters such as cloud size, fuel concentration, and ignition point are known. In a gas explosion, the flame propagation velocity spans more than three orders of magnitudes and the mechanism of flame propagation will be quite different in different velocity region. For deflagration, the flame starts as a laminar flame when a weak ignition source, such as a spark or a hot spot, ignites the flammable cloud. For a laminar flame, the basic mechanism of propagation is the molecular diffusion of heat and mass. Figure 1.8 illustrates the laminar flame structure in a premixed gas. The diffusion process of heat and mass into the unburned gas is relatively slow and the laminar flame will propagate with a velocity in the order of 3–4 m/s. The type of fuel and fuel concentration decide the propagation velocity of the laminar flame. The laminar burning velocities for methane–air, ethylene–air and hydrogen–air are shown in Fig. 1.9. The maximum possible burning velocity for methane is about 0.4 m/s, while that for hydrocarbons is 0.4–0.5 m/s. Ethylene, acetylene and hydrogen have higher burning velocities because of their fast chemical kinetics and high molecular diffusivity. It should be mentioned that different values for laminar burning velocity have been reported by different sources due to the

12

1 Introduction

Fig. 1.8 Illustration of gas distribution in laminar flame (Bjerketvedt et al., 1997, permission from Elsevier)

3

Laminar Burning Velocity (m/s)

Fig. 1.9 Laminar burning velocities for methane–air, ethylene–air and hydrogen–air (Bjerketvedt et al., 1997, permission from Elsevier)

Hydrogen

2

1

Ethylene Methane

0

0

10

20

30

40

50

60

% fuel

situational variations in apparatus and measurement techniques (Rallis & Garforth, 1980). As the flow field ahead of the flame front becomes turbulent, in most accidental explosions, the laminar flame will accelerate and transit into a turbulent deflagration. The turbulence is usually created by the interaction of the flow field with the process equipment, piping, structure, etc. One of the mechanisms increasing the burning rate in turbulent deflagration is the wrinkles of the flame front by large turbulent eddies. In this scenario, the burning rate is increased by the enlarged flame surface area. The wrinkling flame front occurs when the turbulent integral length scale lt is significantly larger than the thickness of the flame front δ. Contrarily, a thick turbulent flame brush forms when the turbulent integral length scale is of the order of the thickness of the flame front or smaller. For this type of combustion form, increased diffusion of heat and mass, thereby a higher burning rate, is induced by the turbulence. Figure 1.10 describes graphically both turbulence forms.

1.5 Deflagrations and Detonations

13

Fig. 1.10 Wrinkled flame front and turbulent flame brush of turbulent eddies (Bjerketvedt et al., 1997, permission from Elsevier)

(a) Increased flame surface

(b) Increased diffusion of heat and mass

1.5.2 Detonation Detonation is the most devastating form of gas explosion. Unlike the deflagration, confinement or congestion in order to propagate at high velocity is not necessarily required for detonation. The detonation is a combustion wave propagating at a supersonic velocity relative to the unburned gas immediately ahead of the flame. In fuel–air mixtures at atmosphere pressure, the detonation velocity is typically 100–2000 m/s and the peak pressure is up to 20 bar. The probability for the occurrence of detonation, the propagation and transmission of the detonation waves depend strongly on the reactivity of the gas cloud. A detonation is more likely to happen for higher reactive gases such as ethylene, hydrogen and acetylene. In simple terms, a detonation wave can be described as a shock wave immediately followed by a flame. The shock compression heats the gas and triggers the combustion. The energy released by the combustion maintains the detonation wave. Figure 1.11a shows the ZND model that describes the detonation wave as a shock wave immediately followed by a reaction zone. The thickness of this zone is given by the reaction rate. Figure 1.11b illustrates that an actual detonation is a threedimensional shock wave followed by a reaction zone. The leading shock consists of curved shock segments. At the detachment lines between these shock segments, the shock wave interacts in a Mach stem configuration. The size of the fish shell pattern generated by the triple point (Mach stem) of the shock wave is a measure of the reactivity of the mixture represented by the length scale characterizing the overall chemical reaction in the wave (Lee, 1984). This length λ is generally the cell size or the cell width. The smaller the cell size,

14

1 Introduction

Fig. 1.11 ZND structure and pattern of an actual structure of a detonation front (Bjerketvedt et al., 1997, permission from Elsevier)

(a) ZND theory

(b) Pattern of an actual structure

Detonation cell size (mm)

1000

Ethylene

100

Hydrogen Acetylene

10

1

0

10

20

30

40

50

60

% fuel in air (%vol)

(a) Ethylene, Acetylene, and Hydrogen Detonation cell size (mm)

Fig. 1.12 Cell size versus fuel concentration for fuel–air mixtures (Bjerketvedt et al., 1997, permission from Elsevier)

1000

Propane

Methane

100

Ethylene

10

0

5

10

15

volumtric % fuel in fir

(b) Methane, Propane, and Ethylene

20

1.5 Deflagrations and Detonations

15

the more reactive the mixture is. The detonation cell size versus fuel concentration for several fuel–air mixtures is shown in Fig. 1.12. The cell size is usually measured experimentally, and it is a parameter of practical importance because the transition from deflagration to detonation, propagation and expansion of the detonation are able to be evaluated according to the cell size of the mixture.

1.5.3 Deflagration to Detonation Transition (DDT) A sudden transition from deflagration to detonation can occur when a deflagration becomes sufficiently strong in terms of the interaction between shocks and flames. Flames generate shocks, especially in confined spaces and/or in the presence of obstacles and reinforce the shocks passing through a turbulent flame brush. In turn, shock interacts with flames to create and drive the turbulence in flames far more dynamically than standard combustion does. The mechanism of transition to detonation is not yet fully understood. It is worth noting, from a practical point of view, that transition to detonation will cause extremely high pressures in the immediate vicinity of the transition.

1.6 Facility Conditions With the rapid development of the oil and gas industry, thousands of various oil and gas facilities have been constructed. Typical oil and gas structures include refinery factories, offshore platforms, transportation pipelines, petrol stations, etc. Specific structural settings for various oil and gas facilities, such as confinements, congestions and venting conditions, have significant influences on the consequences of VCEs, whereas the triggering of VCEs occurs regardless of the specific setting. Confined gas explosions usually occur within fully confined regions, such as tanks, process equipment, pipes, culverts, sewage systems, closed rooms and underground installations, with no venting and no heat release. Figure 1.13 shows an example of a typical confined VCE. In the confined conditions, the turbulent combustion process causes a dramatic increase in overpressure. For example, the overpressures and impulses in a confined chamber in a shock-dispersed-fuel explosion can be enlarged 2–3 times by increasing the confinement volume (Kuhl & Reichenbach, 2009). Wherefore, it is of great importance to investigate the flame propagation in such confined explosions for the design of structures (Sauvan, Sochet, & Trelat, 2012; Shi, Li, & Hao, 2009; Tang et al., 2014). And it is worth mentioning that, in a confined situation, a high flame velocity is not required to generate pressure. For instance, in a closed vessel without or with very little relief of the explosion pressure, even a slow combustion process will cause pressure. Generally, explosions

16

1 Introduction

Fig. 1.13 VCE in confined area

from underground transportation, such as pipelines, can be classified as confined explosions. Partly confined explosions occur in a structure, which is partly open, such as offshore modules or the production or process area within buildings. In this situation, the explosion pressure can only be relieved through the vent areas. Thus, the vent area size and location play a significant role in modulating overpressures in the partly confined VCEs. Generally, from the fully confined VCE to the partly confined gas explosion with uncongested condition, ventilation will reduce the turbulence level, hence, lower the flame speed of the explosion. On the other hand, under the partly confined condition, the flame can still be accelerated to several hundred meters per second. The temperature will increase with the burning of the gas and the gas will be expanded by a factor of up to 8 or 9. The unburned gas is then pushed ahead of the flame and a turbulent flow field is generated. When the flame propagates into the turbulent flow field, the effective burning rate will increase, and the flow velocity and turbulence increase further. This strong mutual-strengthening mechanism yields both flame acceleration and high pressure, and in some cases, transits the deflagrations into detonations. Figure 1.14 demonstrates a typical VCE in partly confined areas. Figure 1.15 illustrates the VCE in unconfined regions. Unconfined gas explosions refer to the explosions in the open areas, such as onshore process plants. An explosion outside a storage tank, which is separated from other tanks or process areas, may be identified as an unconfined explosion. If the cloud is practically unconfined and unobstructed, the flame tends not to accelerate, wherein velocities usually do not exceed 20–25 m/s. The overpressure can be neglected. In view of that, the explosion usually turns out to be flash fires only. On the other hand, cautions also should be exerted because even unconfined areas may contain local areas being partly confined and/or obstructed. Obstacle is generally another critical factor that may have significant influence on gas explosion loads. The expansion-driven flow from combustion will generate turbulence when the fluid flows pasting the obstacles (Dorofeev, 2007; Kim et al., 2014; Na’inna, Phylaktou, & Andrews, 2013). The newly generated turbulence will

1.6 Facility Conditions

17

Fig. 1.14 VCE in partly confined area

Fig. 1.15 VCE in unconfined area

increase the burning velocity by expanding the flame area and enhancing the molecular diffusion and conduction processes, and consequently, expel the expansion flow and boost the turbulence. The generation of increasingly higher burning velocities and overpressures due to the obstacles is called the Schelkchkin mechanism (Lea, 2002). Figure 1.16 shows the Schelkchkin mechanism of flame acceleration by obstructions that constitute a strong mutual-strengthening loop (Bjerketvedt et al., 1997). A blockage ratio of obstacles is usually adopted to describe the degree of obstruction. It is an important factor that influences the flame propagation and the explosion overpressures (Oh, H. Kim, J. B. Kim, & Lee, 2001). The maximum overpressure increases constantly with the increase of the blockage ratio. The blockage rate depends on the specific geometry of the obstruction (Ibrahim & Masri, 2001). Higher

18

1 Introduction

Increased pressure

Combustion of premixed gas cloud

Expansion

Flow interacts with obstacles

Turbulence is generated

Turbulence enhances the combustion

Fig. 1.16 Mutual-strengthening mechanism of flame acceleration and turbulence (Bjerketvedt et al., 1997, permission from Elsevier)

overpressures tend to be induced by smaller-diameter objects for a specific blockage ratio. Tortuous route (flame in baffle-type obstacles) results in greater explosion overpressures compared to round obstacles. It is probably because that the turbulence boosts the burning velocity more significantly by the shear layer of the sharp obstacle (Bjorkhaug, 1986). In simple summary, other than the gas properties, the condition of confinement and obstacles affect the flame acceleration to generate VCE overpressures. On the other hand, for common oil and gas facilities, it is usually ambiguous to define structural conditions as confined or unconfined. A combination of confined, unconfined and congested situations is usually the case. Still, it is important to evaluate specific structural arrangements of each facility in order to achieve a reliable risk assessment of VCEs. Different risk analysis methods should be applied based on different situations.

1.7 Overpressure Prediction Reliable prediction of gas explosion overpressures is most critical in explosion risk analysis. It serves as the base to derive the severities of the other consequences, such as building damage, human losses and business losses. In view of that, the deterministic explosion load prediction methods, from the traditional empirical ones to the most state-of-the-art fluid dynamic simulation, are exhaustively reviewed (Lea, 2002) in the following context. DNV PHAST (DNV GL, 2016) is a leading consequence analysis tool using empirical models to simulate the hazardous gas releases, gas dispersions, fires and explosions. It is simple to use as a screening tool for rapid assessment of physical effects and consequences. The empirical models used by PHAST

1.7 Overpressure Prediction

19

include simplified TNT Equivalency model (Mannan, 2012), TNO multi-energy model (Vandenberg, 1985), and Baker–Strehlow–Tang model (Baker, Doolittle, Fitzgerald, & Tang, 1998). These models are highly simplified to circumvent the complex physics by inducing relevant correlations from experimental data. The empirical TNT Equivalency method has been extensively applied for overpressure calculation for simplicity. On the other hand, with further understanding of the mechanism of the gas explosion and flame propagation, it has been gradually phased out due to oversimplification. A multi-energy method, based on an experimental database to approximate explosion overpressures, has outstripped the TNT Equivalency method as a fast and more reliable prediction solution. Its competitive effectiveness lies in the treatment of the VCE as a set of sub-explosions instead of a single entity explosion (Bjerketvedt et al., 1997). In spite of the improvement in effectiveness in the multi-energy method, inadequacy is identified as not to efficiently represent complicated geometries. The Baker–Strehlow–Tang (BST) model was firstly published at 28th Loss Prevention Symposium in 1994. It applies the Strehlow approach to select blast curves. It is based on the flame speed and the procedure of the multi-energy model to determine explosion energy according to confinement and congestion. This model is still considered to be similar to the TNO multi-energy model. The numerical approaches generally adopt the computational fluid dynamic (CFD) codes. The fundamental partial differential equations, which govern the fluid flow and the other explosion processes, are employed in most of the numerical models for the calculation of VCEs. Compared to the empirical models, the numerical models demonstrate much higher reliability and flexibility. By discretising the solution domain in both space and time, complicated geometrical configuration and conditions in the VCEs can be accommodated with reasonable accuracy in the numerical simulations. The numerical simulations can provide comprehensive exploration into the flow behaviours, i.e. flame velocities, gas density, overpressures, etc. For the practical oil and gas safety issues, compared to conducting expensive, if possible, experiments, it is more feasible, effective, as well as economical to apply numerical models to simulate different scenarios. On the other hand, improficiency in numerical approach lies in that the effectiveness of the numerical estimation depends on the calibration of the numerical models against experiments. Errors could arise from the uncertainties in geometry, boundary conditions, mesh grids, etc. in modelling during the data validation. Sufficiently accurate overpressure calculations demand a great number of small grids in the numerical modelling to lead to intractable computational intensity for the current available computer hardware and software. One of the most popular numerical models is the FLame ACceleration Simulator (FLACS) code. The FLACS code has been developed and continually improved over the recent two decades at Christian Michelsen Research Institute in Norway (Bjerketvedt et al., 1997). The three-dimensional Cartesian mesh is employed in FLACS and the first-order discretization in both time and space is used in the finite volume approach. For sub-grid scale obstacles in FLACS, the porosity resistance approach is adopted to simulate the accelerating effects of these obstacles. The com-

20

1 Introduction

monly used k − ε turbulence model deduces the source terms through turbulent processes. And based on Arntzen’s correlations (Arntzen, 1998), a β flame model was developed in FLACS to calculate the turbulent burning velocities and other parameters (Bakke & Hjertager, 1986; Hjertager, 1986). FLACS has been widely used in the onshore/offshore explosion analysis, and extensive validations have been reported (Bakke & Hjertager, 1986; Hansen, Hinze, Engel, & Davis, 2010; Hjertager, Fuhre, & Bjørkhaug, 1988; Middha & Hansen, 2009). Ma, Li, and Abdel-Jawad (2014) applied the FLACS to predict the consequences of explosion events at large-scale oil and gas facilities with complicated and highly congested environments. Li, Ma, Abdel-jawad, and Huang (2016), Li, Ma, Hao and Huang (2016) used FLACS to evaluate safety gap effect on both gas dispersion and explosion risk for super-large and highly congested offshore floating liquefied natural gas (FLNG) platforms. Gavelli, Davis, and Hansen (2011) applied FLACS to evaluate the consequences of the ignition of a flammable vapour cloud from an LNG spill during the LNG carrier offloading process. Middha, Engel, and Hansen (2011) analysed the safety of hythane by using FLACS regarding the flame speeds and flammability limits. Bakke, van Wingerden, Hoorelbeke, and Brewerton (2010) carried out a study on the effect of trees on gas explosion and Yet-Pole, Shu, and Chong (2009) employed FLACS to evaluate the possible hazards of different worst-case scenarios within a naphtha-cracking plant. More detailed description and application of both the empirical methods and CFD modelling can be found in Chaps. 2 and 3 of this book.

References ARIA Technologies. (2007). Explosion of hydrocarbons in an urban sewerage network, April 22nd, 1992, Guadalajara, Mexico. Boulogne-Billancourt, France. Arntzen, B. J. (1998). Modelling of turbulence and combustion for simulation of gas explosions in complex geometries. Asumadu-Sarkodie, S., Owusu, P. A., & Rufangura, P. (2015). Impact analysis of flood in Accra, Ghana. Advances in Applied Science Research, 6(9), 53–78. Baker, Q. A., Doolittle, C. M., Fitzgerald, G. A., & Tang, M. J. (1998). Recent developments in the Baker-Strehlow VCE analysis methodology. Process Safety Progress, 17(4), 297–301. https:// doi.org/10.1002/prs.680170411. Bakke, J. R., & Hjertager, B. H. (1986). Quasi-laminar/turbulent combustion modelling, real cloud generation and boundary conditions in the FLACS-ICE code, CMI No. 865402–2, 1986. Bakke, J. R., van Wingerden, K., Hoorelbeke, P., & Brewerton, B. (2010). A study on the effect of trees on gas explosions. Journal of Loss Prevention in the Process Industries, 23(6), 878–884. https://doi.org/10.1016/j.jlp.2010.08.007. Bella Naija. (2015). Explosion at petrol station in Ghana leaves scores dead. https://www.bellanaija. com/2015/06/explosion-at-petrol-station-in-ghana-leaves-scores-dead/. Bjerketvedt, D., Bakke, J. R., & vanWingerden, K. (1997). Gas explosion handbook. Journal of Hazardous Materials, 52(1), 1–150. Bjorkhaug, M. (1986). Flame acceleration in obstructed radial geometries. City University. Broadcast Greenlight. (2013). Fire in the night. http://greenlight.broadcastnow.co.uk/greenlights/ 5548-fire-in-the-night.

References

21

BP. (2017). BP statistical review of world energy June 2017. https://www.bp.com/content/dam/ bp/en/corporate/pdf/energy-economics/statistical-review-2017/bp-statistical-review-of-worldenergy-2017-full-report.pdf. CSB. (2005). Investigation report—Refinery explosion and fire. Us Chemical Safety And Hazard Investigation Board, Report No. 2005-04-I-TX. CSB. (2014). Investigation report overview—Explosion and fire at the Macondo Well. US Chemical Safety and Hazard Investigation Board, Vol. 1, Report no. 2010-10-I-OS. Dadashzadeh, M., Abbassi, R., Khan, F., & Hawboldt, K. (2013). Explosion modeling and analysis of BP Deepwater Horizon accident. Safety Science, 57, 150–160. DNV GL. (2016). PHAST tutorial manual. DNV GL software, London, UK. Dorofeev, S. B. (2007). Evaluation of safety distances related to unconfined hydrogen explosions. International Journal of Hydrogen Energy, 32(13), 2118–2124. https://doi.org/10.1016/ j.ijhydene.2007.04.003. Gavelli, F., Davis, S. G., & Hansen, O. R. (2011). Evaluating the potential for overpressures from the ignition of an LNG vapor cloud during offloading. Journal of Loss Prevention in the Process Industries, 24(6), 908–915. https://doi.org/10.1016/j.jlp.2011.07.002. Hansen, O. R., Hinze, P., Engel, D., & Davis, S. (2010). Using computational fluid dynamics (CFD) for blast wave predictions. Journal of Loss Prevention in the Process Industries, 23(6), 885–906. https://doi.org/10.1016/j.jlp.2010.07.005. Hendershot, Dennis C. (2013). Process safety: Remembering Piper Alpha. Journal of Chemical Health and Safety, 20(3), 58–59. https://doi.org/10.1016/j.jchas.2013.04.004. Hjertager, B. H. (1986). Three-dimensional modeling of flow, heat transfer, and combustion. In Handbook of heat and mass transfer (Vol. 1, pp. 1303–1350). Hjertager, B. H., Fuhre, K., & Bjørkhaug, M. (1988). Concentration effects on flame acceleration by obstacles in large-scale methane-air and propane-air vented explosions. Combustion Science and Technology, 62(4–6), 239–256. Ibrahim, S. S., & Masri, A. R. (2001). The effects of obstructions on overpressure resulting from premixed flame deflagration. Journal of Loss Prevention in the Process Industries, 14(3), 213–221. https://doi.org/10.1016/S0950-4230(00)00024-3. Kuchta, J. M. (1985). Investigation of fire and explosion Accidents in the chemical, mining, and fuel-related industries—A manual. Bulletin/U.S. Department of the Interior, Bureau of Mines; Superintendent of Documents. no.: 128.3:680. Kuhl, A. L., & Reichenbach, H. (2009). Combustion effects in confined explosions. Proceedings of the Combustion Institute, 32, 2291–2298. https://doi.org/10.1016/j.proci.2008.05.001. Lea, C. J. (2002). A review of the state-of-the-art in gas explosion modelling. Buxton: Health & Safety Laboratory. Lee, J. H. (1984). Dynamic parameters of gaseous detonations. Annual Review of Fluid Mechanics, 16(1), 311–336. Li, J., Ma, G., Abdel-jawad, M., & Huang, Y. (2016). Gas dispersion risk analysis of safety gap effect on the innovating FLNG vessel with a cylindrical platform. Journal of Loss Prevention in the Process Industries, 40, 304–316. Li, J., Ma, G., Hao, H., & Huang, Y. (2016). Gas explosion analysis of safety gap effect on the innovating FLNG vessel with a cylindrical platform. Journal of Loss Prevention in the Process Industries, 44, 263–274. Liaw, H. J. (2016). Lessons in process safety management learned in the Kaohsiung gas explosion accident in Taiwan. Process Safety Progress, 35(3), 228–232. Ma, G. W., Li, J. D., & Abdel-Jawad, M. (2014). Accuracy improvement in evaluation of gas explosion overpressures in congestions with safety gaps. Journal of Loss Prevention in the Process Industries, 32, 358–366. https://doi.org/10.1016/j.jlp.2014.10.007. Mannan, S. (2012). Lees’ Loss prevention in the process industries: Hazard identification, assessment and control. Butterworth-Heinemann. Mega. (2014). Taiwan gas pipeline blasts kill 25, injures 267. https://business.mega.mu/2014/08/ 02/taiwan-gas-pipeline-blasts-kill-25-injures-267/.

22

1 Introduction

Mercx, W. P. M., & Vandenberg, A. C. (2005). Chapter 5: Vapour cloud explosion. In TNO Yellow Book: Methods for the calculation of physical effects due to releases of hazardous materials (2nd ed.). Rijswijk, The Netherlands. Middha, P., & Hansen, O. R. (2009). Using computational fluid dynamics as a tool for hydrogen safety studies. Journal of Loss Prevention in the Process Industries, 22(3), 295–302. https://doi. org/10.1016/j.jlp.2008.10.006. Middha, P., Engel, D., & Hansen, O. R. (2011). Can the addition of hydrogen to natural gas reduce the explosion risk? International Journal of Hydrogen Energy, 36(3), 2628–2636. https://doi.org/ 10.1016/j.ijhydene.2010.04.132. Na’inna, A. M., Phylaktou, H. N., & Andrews, G. E. (2013). The acceleration of flames in tube explosions with two obstacles as a function of the obstacle separation distance. Journal of Loss Prevention in the Process Industries, 26(6), 1597–1603. https://doi.org/10.1016/j.jlp.2013.08. 003. O’Byrne, C. (2011). Remembering the Piper Alpha disaster. Historical Reflections-Reflexions Historiques, 37(2), 90–104. https://doi.org/10.3167/hrrh.2011.370207. Oh, K. H., Kim, H., Kim, J. B., & Lee, S. E. (2001). A study on the obstacle-induced variation of the gas explosion characteristics. Journal of Loss Prevention in the Process Industries, 14(6), 597–602. https://doi.org/10.1016/S0950-4230(01)00054-7. Patecornell, M. E. (1993). Learning from the Piper Alpha accident—A postmortem analysis of technical and organizational-factors. Risk Analysis, 13(2), 215–232. https://doi.org/10.1111/j. 1539-6924.1993.tb01071.x. Rallis, C. J., & Garforth, A. M. (1980). The determination of laminar burning velocity. Progress in Energy and Combustion Science, 6(4), 303–329. Reader, T. W., & O’Connor, P. (2014). The Deepwater Horizon explosion: Non-technical skills, safety culture, and system complexity. Journal of Risk Research, 17(3), 405–424. https://doi.org/ 10.1080/13669877.2013.815652. Sauvan, P. E., Sochet, I., & Trelat, S. (2012). Analysis of reflected blast wave pressure profiles in a confined room. Shock Waves, 22(3), 253–264. https://doi.org/10.1007/s00193-012-0363-1. Shi, Y., Li, Z., & Hao, H. (2009). Numerical investigation of blast loads on RC slabs from internal explosion. Paper presented at the international workshop on structures response to impact and blast conference, Israel. Tang, C. L., Zhang, S., Si, Z. B., Huang, Z. H., Zhang, K. M., & Jin, Z. B. (2014). High methane natural gas/air explosion characteristics in confined vessel. Journal of Hazardous Materials, 278, 520–528. https://doi.org/10.1016/j.jhazmat.2014.06.047. The Guardian. (2010). BP plans to close its US safety watchdog. https://www.theguardian.com/ business/2010/oct/10/bp-us-safety-ombudsman-closure. The Telegraph. (2015). BP oil spill: Five years after ‘worst environmental disaster’ in US history, how bad was it really? http://www.telegraph.co.uk/news/worldnews/northamerica/usa/ 11546654/BP-oil-spill-Five-years-after-worst-environmental-disaster-in-US-history-how-badwas-it-really.html. Vertigo Politico. (2015). Explosiones de Guadalajara, a 23 años de la tragedia. http://www. vertigopolitico.com/articulo/31681/Explosiones-de-Guadalajara-a-23-anos-de-la-. Yet-Pole, I., Shu, C. M., & Chong, C. H. (2009). Applications of 3D QRA technique to the fire/explosion simulation and hazard mitigation within a naphtha-cracking plant. Journal of Loss Prevention in the Process Industries, 22(4), 506–515. https://doi.org/10.1016/j.jlp.2009.04.002.

Chapter 2

VCE Overpressure Prediction Using Empirical Methods

Abstract In this chapter, three of the most popular empirical load prediction models, i.e., TNT equivalency model, TNO Multi-Energy model and Baker–Strehlow–Tang model, are introduced. The mechanism of each method is explained in detail for reader’s reference. DNV PHAST also introduced and used for modelling. Three application examples of real gas explosion accidents are presented. The explosion events are simulated and analysed by using DNV PHAST.

2.1 TNT Equivalency Model The destructive potential of high explosives has always been interesting to the military forces as well as civilian applications. Therefore, the relationship between damage and explosion of high explosives has been continuous research attraction for many years. Various prediction methods ranging from simple empirical methods to high-proficient 3D numerical fluid dynamic simulation have been proposed and developed. Amongst, the simplest is the TNT Equivalency method. It relates an accidental explosion power of explosives other than the TNT to that of an equivalent TNT charge. The TNT Equivalency model quantifies the gas explosion source as an equivalent quantity of high-explosive TNT. In spite of the differences between the blasts from a TNT detonation and a gas explosion, the TNT Equivalency method has been widely used for gas explosion prediction for simplicity. In the TNT Equivalency method, the equivalent TNT charge of gas cloud or other explosives are derived based on the relative energy content compared to TNT and empirical pressure-distance curves for TNT explosions are applied to deduce the explosion intensity. In order to apply the TNT Equivalency model to the estimation of the effects of a VCE, the fraction of total energy from the explosion carried in the form of shock wave should be predicted. The estimated energy in shock wave is then converted into the equivalent mass of TNT W TNT in kg (Mannan, 2012): WTNT =

Wgas × η × Hc(gas) Hc(TNT)

© Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_2

(2.1)

23

24

2 VCE Overpressure Prediction Using Empirical Methods

where W TNT is the equivalent mass of TNT that would produce the similar effects as the explosion of interest; dimensionless η denotes the explosion yield factor with a value between 1 and 10% for most explosions; W gas in kg stands for the total mass of the flammable gas in the cloud; Hc(gas) in J/kg represents the lower heat of combustion of the material; and Hc(TNT) in J/kg is the specific heat of TNT combustion approximately ranging from 4190 to 4650 kJ/kg. Although the efficiency of the TNT Equivalency method is limited by its oversimplification, it is still widely used to predict overpressures at a given distance from the centre of an explosion (Kingery & Pannill, 1964). This model is based on an empirical law according to physical test data to synthesize the effects of the distance and the weight in term of the scaled distance, which is expressed as R =

R

(2.2)

1

(WTNT ) 3

wherein R  in m/kg1/3 is the scaled distance; R in m denotes the physical distance. For a scaled distance, the corresponding overpressure can be derived from the empirical diagram as shown in Fig. 2.1, which is the best fit to numerous experimental data. The TNT Equivalency model serves as a quick measure to roughly estimate the blast effects of a vapour cloud explosion. The attractiveness of this method lies in the direct empirical relation between the charge weight and the resulted structural damage. Therefore, the TNT Equivalency method is still of interest in view of its trade-off balance between simplicity and effectiveness. The Health and Safety

10000

Scaled Peak Side-on Overpressure P's

Fig. 2.1 Scaled peak side-on overpressure versus scaled distance (Kingery & Pannill, 1964, public released data, http://www. dtic.mil/docs/citations/ AD0443102)

1000

100

10

1

0.1 0.1

1

10

100

Scaled Distance

1000

2.1 TNT Equivalency Model

25

Executive (HSE) evaluated the TNT Equivalency method for both the near- and farfield explosions of explosives and energetic materials (Formby & Wharton, 1996). Rui, Lizhong, Wanghua, Jiacong and Weicheng (2002) applied the TNT Equivalency method to derive the distributed blast of fuel–air detonation. Skacel, Janovsky, Dostal and Svihovsky (2013) discussed the applicability of the TNT equivalency method for calculations of blast wave dynamics after vessel rupture in a 1D geometry detonation. On the other hand, inadequacy of the TNT Equivalency method for application in modelling the vapour cloud explosions have been observed as it does not take into account the propagation of an accelerating wavefront, rather a single-entity explosion instead. The resulted overpressure for a VCE by TNT Equivalency method is excessively high in terms of magnitude and unreasonably short for duration. It is because the shock wave from the VCE interacts with the environment in the specific form of obstructions for flame front propagation and congestion, which is not modelled in the TNT Equivalency method. As a result, this method does not fit for prediction of the consequences of a VCE if proficiency instead of simplicity is a priority. In summary, the TNT Equivalency method has the following features to • identify a proportional relationship between the equivalent TNT charge weight and the cloud energy content; • be simple and easy to use; • be incapable to represent weak gas explosions; • require selection of yield factor which is usually not readily available; • ignore the variability of explosion strengths; • neglect the geometry of vapour cloud; • overestimate the nearby pressure effects; • predict only overpressures to leave out the duration and blast wave shape; • be in conflicts of high popularity with equivalent critics for the similar reason of simplicity.

2.2 TNO Multi-Energy Model For vapour cloud explosions (VCE), the TNO Multi-Energy model is generally preferred over the TNT Equivalency method as a simple and practical measure to determine the overpressure and positive phase duration as a function of distance. The Multi-Energy concept is based on the observation that the explosion potential impact of a vapour cloud is primarily determined by the obstructed and/or confined portions of the cloud. Thus, the amount of energy released during a VCE is limited by the volume of the confined portion if the vapour cloud is larger than the confined region. The Multi-Energy model represents the hemispherical cloud as a homogeneous and stoichiometric mixture of flammable gas and air with the combustion energy of 3.5 MJ/m3 for the most hydrocarbon mixtures at stoichiometric concentration with air. This model reflects the most significant blast wave parameters, such as the side-

26

2 VCE Overpressure Prediction Using Empirical Methods

on peak overpressure, peak dynamic pressure and the positive phase duration of the blast wave, with regard to the distance to the blast centre for a hemispherical fuel–air charge. As shown in Figs. 2.2, 2.3 and 2.4, the Multi-Energy model consists of a family of blast curves for peak overpressures, dynamic pressures and duration against distance. These curves are a summary of the detailed model predictions for idealized explosion scenarios. To generate the family of curves, different constant flame speeds during the explosion are assumed. Initial blast strength is defined by a set of blast wave parameters at the location of the charge radius Ro . It is indicated by a number spanning from 1 to 10 in alignment with low to high detonative strength. The shape of the blast wave from a detonative blast is indicated by solid lines for the shock wave and dotted lines for the pressure waves of low initial strength. In between, dashed lines are identification of a state of transition. Figures 2.2, 2.3 and 2.4 reflect the characteristics of gas explosion blast. Pressure waves produced by a fuel–air charge of low strength show an acoustic overpressure decay behaviour and a constant duration in the positive phase. A much faster decay of the overpressure and a substantial increase of the positive phase duration are observed for high-initial strength shock waves. The volume and the location of the flammable vapour are prerequisite for application of the Multi-Energy method. To derive this information, source term models and dispersion analysis should be carried out beforehand. The geometrical layout or a rough description of the obstruction in the build-up area, where the cloud is located,

100

10 Scaled Peak Side-on Overpressure Ps'

Fig. 2.2 Scaled peak side-on overpressure versus scaled distance from TNO Multi-Energy model (Van den Berg, 1985, permission from Elsevier)

10

9 8 7

1

6 5 4 3

0.1

2 1

0.01

0.001 0.1

1

10

100

Combustion Energy-scaled Distance R'

2.2 TNO Multi-Energy Model

27

Fig. 2.3 Scaled dynamic pressure versus scaled distance from TNO Multi-Energy model (Van den Berg, 1985, permission from Elsevier)

Scaled Peak Dynamic Pressure Pdyn'

100

10

10

9 8 1

7 6 5

0.1

4 3 2 1

0.01

0.001 0.1

1

10

Fig. 2.4 Scaled positive phase duration versus scaled distance from TNO Multi-Energy model (Van den Berg, 1985, permission from Elsevier)

Scaled Positive Phase Duration tp'

Combustion Energy-scaled Distance R' 10

1 2 3 4 1

5 6 7 8 9 10

0.1 0.1

1

10

100

Combustion Energy-scaled Distance R'

should be provided to determine the distance, the number and the volume of the obstructed regions within the cloud. Accordingly, the values for the blast parameters can then be derived from the blast charts of the Multi-Energy method (Van den Berg, 1985). For derivation of the blast parameters, to start with, a dimensionless scaled distance R  is calculated R = 

R E exp /P0

 13

(2.3)

28

2 VCE Overpressure Prediction Using Empirical Methods

where R is the distance of the location under consideration to the centre of the explosion; E exp (J) denotes the explosion energy; P0 (101,000 Pa) is the atmospheric pressure. An explosion strength from 1 to 10 is then assumed. The dimensionless scaled  , and peak overpressure P  , the dimensionless scaled peak dynamic pressure Pdyn  the dimensionless scaled positive phase duration t p can be derived from the charts accordingly. The peak overpressure Ps in Pa, peak dynamic pressure Pdyn in Pa, and the positive phase duration t p  in s are calculated from the following equations: Ps = P  × P0

(2.4)

 Pdyn = Pdyn × P0

(2.5)

1  t p = t p × E exp /P0 3 /c0

(2.6)

wherein c0 with a value of 340 m/s is the sound velocity in air. The positive wave impulse i in Pa.s can be calculated by integrating the overpressure variable over the positive phase. The calculation is usually simplified by the following approximating 1/2 multiplication of the side-on overpressure with the positive phase duration: i=

1 × Ps × t p 2

(2.7)

To apply the Multi-Energy model to the practical projects, conditional constraints listed below should be accounted for. • The Multi-Energy model is intended for unconfined vapour cloud explosions rather than vented vapour cloud explosions or internal explosions. • It does not consider the directional blast effects due to the inhomogeneous distribution of the confinement, congestion, and the non-spherical shape of the vapour cloud. • This method assumes deflagration instead of detonation for the combustion of the VPE. It is conservative in view of that unconfined vapour cloud detonation is almost improbable with exception of only one precedent occurrence to our knowledge. The procedure of vapour cloud explosion modelling using Multi-Energy method can be divided into the below specific steps. 1. Determine the cloud size: A dispersion analysis is usually conducted to provide information on cloud size of an accidental release for the explosion calculation. If no dispersion calculation is implemented, the mass quantity is usually estimated. In a conservative manner, the whole mass inventory contained within the process unit under consideration is considered to contribute to the formation of a flammable cloud. In

2.2 TNO Multi-Energy Model

29

the case of pool evaporation, the mass quantity is calculated by the multiplication of the evaporation rate with a specified time period. Therefore, the volume V c of a cloud can be calculated by the equation below: Vc = Q ex /(ρ × cs )

(2.8)

herein ρ is the density of the gas cloud; Qex is the flammable mass quantity; cs is the volumetric percentage stoichiometric concentration. 2. Identify the potential blast sources: Potential sources of blast in the vicinity of a postulated centre of the cloud could be: • extended spatial configuration of objects, e.g., process equipment at chemical plants or refineries piles of crates; • the space between the extended parallel planes, such as the open buildings of multi-storey parking garages; • the space in tubelike structures, i.e., tunnels, corridors, sewage systems, culverts; • an intensely turbulent fuel–air mixture in a jet due to release at high pressure. 3. Define congested regions by: • identifying boundaries of congested regions and determining the free volume V r of each obstructed region; • determining the potential maximum portion of the cloud V gr inside the congested regions; • calculating the volume V o of the unobstructed portion of the vapour cloud using the following equation; Vo = Vc − Vgr

(2.9)

• calculating the energy E of each region by multiplying V gr and V o with the combustion energy per unit volume. 4. Estimate the source strength or class number for each region: A class number of 10 for each obstructed region is usually selected for a conservative prediction. Other numbers may be chosen only if additional information is available to justify. Class 1 is applied to the remaining unobstructed regions. If initial low-turbulence motion is expected in uncongested areas, for example, due to the momentum of the fuel release, a class number of 3 instead of 1 is suggested. 5. Combine congested regions: If more than one congested regions are comprehended, an additional blast source should be defined by adding all energies of the separate blast sources and assum-

30

2 VCE Overpressure Prediction Using Empirical Methods

ing a centre for the additional blast source. This centre can be a geometrical centre of the separate blast sources weighted by their respective energies. 6. Locate the centre of unobstructed part from the overall vapour cloud: A centre is identified for the uncongested vapour cloud volume by considering the geometric centres of the separate uncongested areas weighted by their respective energies. According to the number of explosion centres and their respective energies, the blast parameters as a function of distance to each centre can be calculated using the blast charts given in Fig. 2.2. 7. Calculate radius: The blast is then modelled for each source by equivalent hemispherical fuel–air charge of volume E/E v m3 , where E v of 3.5 MJ/m3 is an average value for most hydrocarbons at stoichiometric concentration. Equation (2.10) below gives the radius Ro for each blast source:  Ro =

E 3 × 2 Ev × π

 13 (2.10)

8. Calculate blast parameters: The blast parameters at a specific distance R from the centre of blast source can be calculated according to Eqs. (3.1)–(3.5). 9. Process multiple congested regions: If separate blast sources are located close to each other, they may be triggered simultaneously. Instead of superposition of the separate blast effect, the respective quantities of combustion energy of the individual source in question should be added up as a single source to derive the blast dynamics. 10. Construct the blast history at a specific location: Combustion in the uncongested region is considerably different from that in a congested area. Sharp and relatively short peaks, i.e., shock waves, will be yielded by the blasts from congested regions, while relatively slow combustion in an uncongested region results in low-pressure waves of long duration. On the other hand, whether the blast waves can have severe consequence at a specific location depend on the total energies released. The blast history at a specific location is usually dominated by the blast parameters and blast shape from the congested region, wherein the blast parameters from the uncongested region are subordinate to overlay.

2.3 Baker–Strehlow–Tang Model The Baker–Strehlow–Tang (BST) model has similarities to the TNO Multi-Energy model in the family of curves to correlate Ps with R, which are deduced from Eqs. (2.3)

2.3 Baker–Strehlow–Tang Model 100

Scaled Side-on Overpressure Ps'

Fig. 2.5 Scaled peak side-on overpressure versus scaled distance from BST model (Tang & Baker, 1999, permission from Elsevier)

31

Mf=5.2 Mf=4.0 Mf=3.0 Mf=2.0 Mf=1.4 Mf=1.0 Mf=0.7 Mf=0.35 Mf=0.2

10

1

0.1

0.01

0.001 0.1

1

10

Scaled Distance R'

and (2.4) for both models. The difference lies in the graphical relationship between  dimensionless Ps and combustion energy scaled distance R from the BST model, as shown in Fig. 2.5, is different from that by TNO. The curves in Fig. 2.5 are derived based on the numerical modelling of constant velocity flames and accelerating flames running through the spherical vapour clouds. In this method, the strength of the blast wave is proportional to the maximum flame speed within the cloud. Therefore, each curve in Fig. 2.5 is tagged with a flame velocity in the form of a Mach numbers M f . Table 2.1 shows the Mach numbers, which serve as the calibre of selection for specific situations (Baker, Doolittle, Fitzgerald, & Tang, 1998; Pierorazio, Thomas, Baker, & Ketchum, 2005).

Table 2.1 Mach numbers M f for ignition sources in Baker–Strehlow–Tang model

Flame expansion

Fuel reactivity

Congestion levels Low

Medium High

1D

High

5.2

5.2

5.2 1.77

2D

2.5D

3D

Medium

1.03

1.77

Low

0.294

1.03

2.27

High

0.59

1.03

1.77 1.6

Medium

0.47

0.66

Low

0.079

0.47

0.66

High

0.47

0.58

1.18 1.0

Medium

0.29

0.55

Low

0.053

0.35

0.50

High

0.36

0.153

0.588

Medium

0.11

0.44

0.50

Low

0.026

0.23

0.34

32

2 VCE Overpressure Prediction Using Empirical Methods

In Table 2.1, the degree of expansion can be interpreted as confinement condition. The obstructed region can be identified as 3D if the flame is free to expand in all directions, similarly, 2D or 1D if the flame can only expand in two or one dimensions. An intermediate level of confinement, 2.5D, is suggested to be applied when it is ambiguous to decide whether 2D or 3D should be chosen for the hazard assessment. For example, 2.5D will be a proper selection where the confinement is made of frangible panels or by nearly solid confining planes. Congestion in Table 2.1 is classified as low, medium and high based on area blockage ratio (ABR) and the pitch in the flame path. Projection of the practical condition onto the specification is explained below. • Low congestion level: A few obstacles in the flame’s path or ABR less than 10% with only a few layers of obstacles. • Medium congestion level: Anything falling between the low and high levels. • High congestion level: Closely spaced layers of obstacles with an ABR of 40% or higher. Fuel reactivity is defined as low, medium and high as well. Methane and carbon monoxide are considered to have low reactivity, while hydrogen, acetylene, ethylene, ethylene oxide and propylene are highly reactive. All other materials are classified as medium reactivity. It is worth mentioning that, after two updates of this table, the 1D category was excluded from the update published in 2005 (Pierorazio et al., 2005). It is because the maximum flame speed in 1D condition is usually a function of the ratio of the length to the diameter in addition to the aforementioned three required parameters. Therefore, the BST model is not recommended for the 1D vapour cloud explosions. Other than the curves for overpressure, the BST model also provides a family of curves for impulse as shown in Fig. 2.6. The impulse I can be calculated from the scale impulse I  as follows,

1

Scaled Impulse I'

Fig. 2.6 Scaled impulse I’ versus scaled distance from BST model (Tang & Baker, 1999, permission from Elsevier)

Mf=5.2 Mf=4.0 Mf=3.0 Mf=2.0 Mf=1.4 Mf=1.0 Mf=0.7 Mf=0.35 Mf=0.2

0.1

0.01

0.001 0.01

0.1

1

Scaled Distance R'

10

2.3 Baker–Strehlow–Tang Model

33



E exp /P0 I = I × P0 × c0 

 13 (2.11)

In simple summary, to apply the BST model, the real explosion scenario is converted into an equivalent vapour cloud explosion. The appropriate curve is selected to identify the values for peak overpressure and impulse in view of the specific distances of the target from the centre of the explosion. The improficiency of this method lies in that only peak overpressure and impulse can be predicted and the explosion is assumed to be symmetric.

2.4 Applications of Empirical Models Using PHAST 2.4.1 Introduction to DNV PHAST DNV PHAST (DNV GL, 2016) is a leading consequence analysis tool to apply empirical models for deduction of hazardous gas releases, fire, simple explosions and gas dispersions. It is simple and is usually used as a screening tool for rapid identification of physical effects and consequences. Similar to most other empirical models, DNV PHAST takes limited account of the influence of obstructions or congestion. However, this problem can be partly solved by manually defining the obstructive areas. PHAST adopts DNV’s unified dispersion model (UDM), which has been extensively verified to be effective to model dispersion scenarios. UDM is capable of modelling ground-level or elevated two-phase releases from pressurized vessels, wherein the release could be continuous, instantaneous, constant finite duration, or time-varying from different variety of sources. To simulate consequence of an explosion event by PHAST, the simulation procedure is shown in Fig. 2.7. In simple words, GIS map is initially input, and the geometry is defined accordingly. Weather category is then identified. Followed is defining the explosion scenarios for the specific case. The scenarios should include explosion location, equipment layout and confinement and congestion conditions. Appropriate analysis models, i.e., Multi-Energy model, TNT Equivalency model and/or Baker–Strehlow–Tang model, are then selected. Relevant data are collected or calculated from a dispersion analysis. In the case that the dispersion analysis is not implemented, assumptions based on collected data should be made. The explosion consequence calculation will be conducted by PHAST based on the data collected or calculated. Important explosion parameters, such as side-on peak overpressure, peak dynamic overpressure, explosion duration or blast impulse, can be derived from the calculation. The results are represented as consequence report and graphical charts. The structure damage level will be deduced against the overpressures according to Table 2.2 (Lobato et al., 2009).

34

2 VCE Overpressure Prediction Using Empirical Methods 1. Input the site map and collect site informations 2. Define explosion scenarios 3. Select appropriate analysis models 4. Collect relevant data for selected models 5. Input data and implement explosion analysis 6. Develop representation of Results 7. Check overpressure against structure damage table

Fig. 2.7 Flow chart for analysis with PHAST Table 2.2 Damage level against overpressure (Lobato et al., 2009) Overpressure (KPa)

Damage levels

0.204

Occasional breakage of large windows already under strain

0.275

Loud noise, breakage of windows due to sound waves

0.681

Breakage of small panes of glass already under strain

2.04

“Safety distance” (probability 0.95 no serious damage beyond this value); 10–20% windows broken; minor structural damage to houses

6.8

Partial demolition of houses, which become uninhabitable

13.6

Partial collapse of house roofs and walls

13.1–20.4

Destruction of cement walls of 20–30 cm width

16.2

1% eardrum breakage

17

Destruction of 50% brickwork of houses. Distortion of steel frame building

20.4–27.7

Rupture of storage tanks

34–47.6

Almost total destruction of houses

47.7–54.4

Breakage of brick walls of 20–30 cm width

68.9

Probable total destruction of buildings. Machines weighing 3500 kg and below displaced and highly damaged

101

1% death due to lung haemorrhage

169.2

90% death due to lung haemorrhage

2.4 Applications of Empirical Models Using PHAST

35

2.4.2 Case Study 2.4.2.1

Gulei Oil Tank Explosion Analysis Using Multi-Energy Model

The explosion at the Dragon Aromatics Plant at 6:56 p.m. on 6 April 2015 damaged a paraxylene PX facility on the Gulei Peninsula. Government investigation indicated that the blast was caused by a chemical leak and implicated four oil tanks. The South China Morning Post reported that 14 people were injured; six of them were admitted to hospital, of whom four were firefighters. Figure 2.8 gives the accident site after explosion from side and top views. Multi-Energy model is used to analyse this accident. To start the consequence analysis using PHAST, a site map is collected, and the explosion location is identified. Figure 2.9 shows the local site map from GOOGLE satellite map, and the explosion centre is indicated as the red point.

Fig. 2.8 Gulei explosion (Caixin, 2015, online open resource, link provided in reference)

Fig. 2.9 Site map of Gulei explosion accident

36

2 VCE Overpressure Prediction Using Empirical Methods

Table 2.3 Inputs for Multi-Energy model

Inputs

Value

Material

Paraxylene

Flammable mass in cloud

1.45E + 07 kg

Number of confined sources

1

Confined source 1

Strength of confined source

6

Volume of confined source

1000 m3

Fig. 2.10 GIS view of explosion radii based on overpressures

Evaluation of the site map detects that the structures around the explosion centre are constructed closely. Therefore, one congested region with strength 6 is assumed for the Multi-Energy model and the confined fuel–air cloud volume is assumed to be 1000 m3 according to the specific structural configuration. Table 2.3 lists all the inputs for PHAST analysis using Multi-Energy model. Figure 2.10 gives the calculated explosion radii according to the different overpressures on the GIS site map. Figures 2.11 and 2.12 show the explosion peak side-on overpressures and pulse durations against downwind distance. From these figures, it is observed that the safety distance is about 340 m from the explosion centre. All the storage tanks within approximately 50 m from the explosion centre will be damaged.

2.4.2.2

Texas BP Refinery Explosion Analysis Using BST Model

The Texas City Refinery explosion occurred on 23 March 2005. A hydrocarbon vapour cloud exploded at the ISOM isomerization process unit at BP’s Texas City refinery, which caused fatal blunt trauma to 15 people in and around the trailers and

2.4 Applications of Empirical Models Using PHAST

37

Confined Overpressure (gauge)[bar]

0.5 0.4 0.3 0.2 0.1 0 0

50

100

150

200

250

300

350

400

450

500

400

450

500

Distace Downwind [m] Fig. 2.11 Overpressure versus distance

Pulse Duration [s]

0.05 0.04 0.03 0.02 0.01 0 0

50

100

150

200

250

300

350

Distance Downwind [m] Fig. 2.12 Pulse duration versus distance

injured 180 others. Figure 2.13 shows the accident site during and after the explosion event. BP’s own accident investigation report stated that the accident was caused by heavier-than-air hydrocarbon vapours combustion when coming into contact with an ignition source, probably a running vehicle engine. The hydrocarbons aggregated from liquid overflow from the F-20 blowdown stack from the operation of the raffinate splitter overpressure protection system by overfilling and overheating of the tower contents. This accident is modelled using Baker–Strehlow–Tang method. Figure 2.14 gives the local site map from GOOGLE satellite map and the explosion centre is identified and indicated by the red point. Table 2.4 tabulates all the inputs for PHAST analysis using BST model.

38

2 VCE Overpressure Prediction Using Empirical Methods

Fig. 2.13 Explosion accident in Texas BP refinery (LLG, 2015 (Left); The Guardian (2010) (Right)) (online open resource, link provided in reference)

Fig. 2.14 Site map of Texas explosion accident Table 2.4 Inputs for BST model

Input parameter

Value

Material

Hydrocarbon

Flammable mass in cloud

13,644 kg

Flame expansion

3D

Obstacle density

High

Fuel reactivity

High

Mach number

0.588

2.4 Applications of Empirical Models Using PHAST

39

Fig. 2.15 GIS view of explosion radii based on overpressures

Overpressure (gauge)[bar]

0.5 0.4 0.3 0.2 0.1 0

0

200

400

600

800 1000 1200 1400 1600 1800 2000 2200

Downwind Distance [m] Fig. 2.16 Overpressure versus distance diagram

Figure 2.15 shows the explosion radii according to different overpressures on the GIS site map. Figures 2.16 and 2.17 give the explosion peak side-on overpressures and impulses against downwind distance. From the results, it can be observed that the safety distance is about 1850 m from the explosion centre. Impairment of eardrum breakage may occur within 330 m from the explosion centre.

40

2 VCE Overpressure Prediction Using Empirical Methods

Impulse [N.s/m2]

20000

15000

10000

5000

0 0

200

400

600

800 1000 1200 1400 1600 1800 2000 2200

Distance Downwind [m] Fig. 2.17 Impulse versus distance

Fig. 2.18 Streets in Kaohsiung after explosion events (The Wall Street Journal, 2014 (left); DailyMail, 2014 (right), online open resource, link provided in reference)

2.4.2.3

Kaohsiung Pipeline Explosion Analysis Using TNT Equivalency Method

On 31 July 2014, a series of gas explosions occurred in the Cianjhen and Lingya districts of Kaohsiung, Taiwan, following reported gas leaks earlier that night. Thirtytwo people were killed, and 321 others were injured. Reportedly, fireballs were sighted soaring into the sky and flames reaching 15 stories high. The blasts ripped up roads, trapped and overturned cars and fire trucks and caused a blackout to the electrical grids. About 6 km of road length was damaged. Figure 2.18 shows the streets after the explosion event. Below gives calculation of this event using TNT Equivalency method.

2.4 Applications of Empirical Models Using PHAST

41

Fig. 2.19 Street map for the Kaohsiung explosion accident Table 2.5 Inputs for TNT Equivalency method

Input parameter

Value

Material

Propane

Flammable mass in cloud

37,700 kg

TNT explosion efficiency

0.1

Explosion mass modification factor

3

Figure 2.19 gives the local site map from GOOGLE satellite map. The pipeline is indicated as blue lines and the explosion centre is shown as the red point. Table 2.5 shows all the inputs for PHAST analysis using TNT Equivalency method. Figure 2.20 shows the explosion radii according to different overpressures on the GIS site map. Figure 2.21 gives the explosion peak side-on overpressures against downwind distance. From the results, it can be observed that the safety distance is about 550 m from the explosion centre and all buildings and structures may probably be destroyed within the distance of 60 m from the explosion centre.

2.5 Summary Consequence modelling is a powerful tool to approximate damages by a vapour cloud explosion accident. Although complex software with high proficiency with intensive computation capacity demands have been developed and generally applied, simple empirical models, such as TNT Equivalency method, TNO Multi-Energy model and BST are still used for well trade-off simplicity and efficiency.

42

2 VCE Overpressure Prediction Using Empirical Methods

Fig. 2.20 GIS view of explosion radii based on overpressures

Overpressure (gauge)[bar]

1.2 1 0.8 0.6 0.4 0.2 0

0

100

200

300

400

500

600

700

800

900 1000 1100

Downwind Distance [m] Fig. 2.21 Overpressure versus distance

On the other hand, improficiencies with the application of these empirical models are observed as listed below. • Empirical models have been proved is not sufficiently accurate in modelling vapour cloud explosions as it does not take into account the propagation of an accelerating wavefront, but a single-entity explosion instead. • The empirical methods model the vapour cloud in 2D flat area without considerations of the practical elevations of the environment/terrain/structure. In view of that, they are unable to model the 3D obstructions, such as tanks, structural members, temporary buildings and blast walls.

2.5 Summary

43

• Although Multi-Energy model and BST model consider the influences of confinement and congestion, only the degree of congestion and confinement level is defined, and detailed configuration of the obstructions is left out. For example, for two highly congested offshore platforms, the level of the congestion will be defined as 10 for Multi-Energy method in spite of their totally different structural layouts. • The empirical methods are unable to assess the effects of secondary impingement induced by a blast other than the overpressure, such as the impact of projectiles from explosion or fire. In summary, empirical methods provide a simple and cost-effective tool to modelling the explosion consequence. On the other hand, due to the oversimplification of the empirical models, they may not be able to produce detailed results regarding the complicated procedure of the vapour cloud explosion. In the next few chapters, advanced computational fluid dynamic (CFD) programmes, such as FLACS, will be introduced to model a VCE and the propagation of the flame front by considering the acceleration of the flame front due to congestion. It is understood that such comprehensive modelling takes longer time, efforts and expertise compared to a simple empirical model does. But more realistic, thus, reliable results will be yielded.

References Baker, Q. A., Doolittle, C. M., Fitzgerald, G. A., & Tang, M. J. (1998). Recent developments in the Baker-Strehlow VCE analysis methodology. Process Safety Progress, 17(4), 297–301. https:// doi.org/10.1002/prs.680170411. Caixin. (2015). Fujian Gulei explosion. http://photos.caixin.com/2015-04-07/100797810_1.html. DailyMail. (2014). Taiwan port city left devastated after ‘sewage system gas leak’ explosions killed 25 people, destroying buildings and roads. http://www.dailymail.co.uk/news/article-2713259/ Five-massive-explosions-caused-gas-leak-sewage-kill-25-injure-hundreds-rip-port-city-Taiwan. html. DNV GL. (2016). PHAST tutorial manual. London, UK: DNV GL software. Formby, S. A., & Wharton, R. K. (1996). Blast characteristics and TNT equivalence values for some commercial explosives detonated at ground level. Journal of Hazardous Materials, 50(2), 183–198. Kingery, C. N., & Pannill, B. F. (1964). Peak overpressure vs scaled distance for TNT surface bursts (hemispherical charges) (No. BRL-MR-1518). ARMY BALLISTIC RESEARCH LAB ABERDEEN PROVING GROUND MD. LLG. (2015). Causes of Texas oil and gas refinery explosions. https://liggettlawgroup.com/blog/ causes-of-texas-oil-and-gas-refinery-explosions/. Lobato, J., Rodríguez, J., Jiménez, C., Llanos, J., Nieto-Márquez, A., & Inarejos, A. (2009). Consequence analysis of an explosion by simple models: Texas refinery gasoline explosion case. Afinidad, 66(543), 372–279. Mannan, S. (2012). Lees’ Loss prevention in the process industries: Hazard identification, assessment and control. Butterworth-Heinemann. Pierorazio, A. J., Thomas, J. K., Baker, Q. A., & Ketchum, D. E. (2005). An update to the Baker–Strehlow–Tang vapor cloud explosion prediction methodology flame speed table. Process Safety Progress, 24(1), 59–65.

44

2 VCE Overpressure Prediction Using Empirical Methods

Rui, H., Lizhong, Y., Wanghua, C., Jiacong, L., & Weicheng, F. (2002). Evaluation of the power of the distributed blast type explosive. Journal of Loss Prevention in the Process Industries, 15(5), 323–327. Skacel, R., Janovsky, B., Dostal, L., & Svihovsky, J. (2013). Small-scale physical explosions in shock tubes in comparison with condensed high explosive detonations. Journal of Loss Prevention in the Process Industries, 26(6), 1590–1596. Tang, M. J., & Baker, Q. A. (1999). A new set of blast curves from vapor cloud explosion. Process Safety Progress, 18(4), 235–240. The Guardian. (2010). BP plans to close its US safety watchdog. https://www.theguardian.com/ business/2010/oct/10/bp-us-safety-ombudsman-closure. The Wall Street Journal. (2014). Deadly gas-pipeline explosions rock Taiwan. https://www.wsj. com/articles/gas-explosions-in-taiwan-kill-at-least-25-1406893177. Van den berg, A. C. (1985). The Multi-Energy method—A framework for vapor cloud explosion blast prediction. Journal of Hazardous Materials, 12(1), 1–10. https://doi.org/10.1016/03043894(85)80022-4. Van den Bosch, C. J. H., & Weterings, R. A. P. M. (Eds.). (2005). Methods for the calculation of physical effects: Due to releases of hazardous materials, liquids and gases: Yellow Book. Ministerie van Volkshuisvesting en Ruimtelijke Ordening (VROM).

Chapter 3

VCE Overpressure Prediction by CFD Modelling

Abstract This chapter explores the mechanism of gas explosion and demonstrates its computational fluid dynamic (CFD) simulation procedure. The fundamental partial differential equations, which govern the fluid flow and other explosion processes employed in the numerical models for calculation of VCEs, are presented. FLACS is adopted for the evaluation of the potential VCEs in oil and gas industries. The CFD simulation theories in FLACS regarding the flame turbulence, geometry condition and fluid–obstacle interaction, etc., are discussed. The flow chart for the simulation procedure along with gas explosion examples is given.

3.1 Methodology of CFD Simulation 3.1.1 Fluid Flow Equations The mathematical models of FLACS can be found in Ferrara, Di Benedetto, Salzano, and Russo (2006), Hjertager (1984, 1993), and Arntzen (1998). A simple summary of the governing equations for a typical reactive gas dynamic calculation is presented below. For a general variable, the differential equation, which is based on Reynolds averaged mass, momentum, and energy balance equations, can be expressed as follows,    ∂ ∂ μeff ∂ ∂  ρu  − (3.1)  = S ;  = + (ρ) j  ∂t ∂xj ∂xj ∂xj σ where  denotes a general variable; ρ stands for the gas mixture density; xj represents the coordinate in j-direction; uj is the velocity component in j-direction;  denotes the effective turbulent diffusion coefficient; μeff is the effective turbulence viscosity; and S stands for a source term. The state equation of an ideal gas is pW = ρRT © Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_3

(3.2)

45

46

3 VCE Overpressure Prediction by CFD Modelling

herein p is the pressure; R denotes the universal gas coefficient; T represents temperature; and W is the molar weight of the gas mixture. The continuity equation has the expression of ∂   ∂ρ + ρuj = 0 ∂t ∂xj

(3.3)

The momentum balance equation takes the form of  ∂ ∂p ∂   ∂  ρuj ui = − σij + (ρui ) + ∂t ∂xj ∂xj ∂xj The energy balance equation is written as    ∂ ∂p ∂h ∂p ∂ ∂  h + + uj ρuj h = (ρh) + ∂t ∂xj ∂xj ∂xj ∂t ∂xj

(3.4)

(3.5)

wherein σij is the flux of momentum and h is the enthalpy.

3.1.2 Stoichiometry In the combustion, fuel is oxidated in the air with various reactants and products, which emits heat and light. The fuel mixtures, which are either too thin or too dense, will demonstrate oxidant dilution or incomplete reaction, respectively. Whereas for a complete combustion, the stoichiometric reaction can be expressed as   nh nh O2 → ncCO2 + H2 O (3.6) Cnc Hnh + nc + 4 2 For substitutive hydrocarbons, the stoichiometry is written in general as such (Kuchta, 1985):   nh − x − 2n O2 Cnc Hnh On Nm Zx + nc + 4   nh − x m H2 O + xHZ + N2 → ncCO2 + (3.7) 2 2 where Z is a halogen atom. For combustion in air, the stoichiometric concentration C sc is written as Csc =

100  1 + 4.773 nc +

nh−x−2n 4

 molpct

(3.8)

3.1 Methodology of CFD Simulation

47

Table 3.1 Dry air properties Constituent

Molecular weight

Density (0 °C) g/L

Specific heat (20 °C), cal/(g °C)

Vol pct

Nitrogen

28.01

1.251

0.249

78.09

Oxygen

32.00

1.429

0.219

20.95

Argon

39.94

1.784

0.124

20.93

Carbon dioxide

44.01

1.977

0.200

20.03

Air

28.97

1.293

0.240

100.00

Therefore, the weight ratio of fuel–air in stoichiometric concentration r f-a-sc can be determined by rf −a−sc =

Mf 100 28.97 100 − Csc

(3.9)

herein 28.97 is the molecular weight of air; Mf denotes the molecular weight of fuel. The air properties in different conditions are summarized in Table 3.1 (Kuchta, 1985).

3.1.3 Thermodynamic Relationships Proper representation of the thermodynamic properties is vital in effective computation of the energy balance of an explosion. As a function of temperature, the thermodynamic relationship is dominated by the formation enthalpy H and the heat C h (Kee, 1987). H and C h in the CHEMKIN thermodynamic database are given in a polynomial form: a3 H a2 a4 a5 a6 = a1 + T + T 2 + T 3 + T 4 + RT 2 3 4 5 T Ch = a1 + a2 T + a3 T 2 + a4 T 3 + a5 T 4 R The thermodynamic relationship between H and C h is   ∂H Ch = ∂T p

(3.10) (3.11)

(3.12)

H and C h are represented by second-order polynomials as functions of temperature in FLACS (Arntzen, 1998), so that

48

3 VCE Overpressure Prediction by CFD Modelling

Ch = a + bT

(3.13)

  H = a(T − T0 ) + 0.5b T 2 − T02 + h0

(3.14)

where T 0 and h0 are standard temperature and heat of formation, respectively; h0 equates 0 when the combustion heat is used in enthalpy calculations. The specified values for a, b and d represent the composition of H2 O, H2 , CO, CO2 and C2 H4 , as seen in Table 3.2 (Arntzen, 1998). The temperature in the thermodynamic system is then expressed as T=

1/2  −a + a2 + 2b(H + d ) b

(3.15)

3.1.4 Ignition Process Ignition triggers the combustion reaction with evolution of emission and heat. Ignition types include electrical, thermal and chemical ignitions. In FLACS, all ignition types initially adopt the H-M model (Hjertager, 1982; Bakke & Hjertager, 1986) for the early versions of FLACS-86 and FLACS-89, which assumes that half of the combustible mixture in the ignition cell is converted into combustion products at time = 0. The H-M model calculates the reaction rate for turbulent combustion from ε wr = Nc ρ min(mfu , mfo − mfu ) when τt > τid k

(3.16)

where ε denotes the dissipation rate of turbulent energy; k represents turbulent kinetic energy; wr is the reaction rate; N c takes a constant; mfu stands for the mass fraction of fuel; mfo is the initial mass fraction of fuel; τt represents the turbulent time; and τid is the ignition delay time. The ignition of combustion depends on the competition between the turbulent time τt and the ignition delay time τid . According to the shock tube tests (Hjertager, 1982; Bakke & Hjertager, 1986), the ignition delay time τn , which is in chemical time scale, is calculated by En

bn τn = Pn e RT Cfuan Cox

(3.17)

wherein C ox and C fu are the concentrations of oxygen and fuel; T represents the gas temperature; and for fuel type n, the constants E n , Pn , an and bn can be found from the database by CHEMKIN (Kee, Miller, & Jefferson, 1980). In view of that the cold front quenching criteria controls the H-M model, and the reaction rate is expressed as

C4 H10

H2

H2 S

SO2

CO

CO2

H2 O

H2 O

OH

NO

O2

N2

Hydrogen

Hyd. sulphide

Sulph. dio.

Carb.m.ox.

Carb. diox.

Water vap.

Water liq.

Hydroxyl

Nitr. oxide

Oxygen

Nitrogen

C3 H8

Propane

Butane

C2 H6

C3 H6

C2 H4

Ethylene

Propylene

C2 H2

Acetylene

Ethane

CH4

Methane

28

32

30

17

18

18

44

28

64

34

2

58.1

44.1

42.1

30.1

28.1

26

16

M (g/mole)

1

1

0

4

3

3

2

2

2

1

C

0

0

1

1

5

4

3

3

2

1

2

H

0

0.5

1.5

0.5

6.5

5

4.5

3.5

3

2.5

2

A

120

45.7

46.3

45.8

47.4

47.2

48.2

50.0

h 298 K (MJ/kg)

Table 3.2 Data of gas, enthalpy of combustion and enthalpy of formation in FLACS

824

888

1740

1002

1000

a

0.397

0.195

0.614

0.173

4.11

b

h = aT + bT 2 /2

1780 4000

−13.43 −15.88

0

0

3.010

1036

950

1040

1620

1060

−8.957

2.293

765 1050

−3.54 −3.951

925

−0.606

641

−2.147

13,600

660

−2.355 0

690

700

−2.786 0.486

740

1340

1200

a

0.118

0.112

0.087

0.200

0.550

0.515

0.157

0.115

0.14

0.40

1.719

3.48

3.55

3.1

3.70

2.85

1.40

3.40

b

0.31

0.29

−2.7

−1.80

17.1

14

9.28

4.27

4.87

0.90

4.13

2.52

2.71

−0.14

3.16

−1.52

−8.26

5.19

d × 10−6

hf = aT + bT 2 /2-d

1.867

8.722

−3.581

hf 298 K (MJ/kg)

3.1 Methodology of CFD Simulation 49

50

3 VCE Overpressure Prediction by CFD Modelling

w = Nc ρ

u min(c, 1 − c) when c > cq lt

(3.18)

where c denotes the mass fraction of products; cq is a function of the turbulent time τt and delay time of ignition τid in Eq. (3.16); u stands for the turbulent velocity fluctuation; and lt is turbulent length scale. The cold front quenching criteria requires high mass fraction of products, which implies the time between the maximum overpressure and ignition is significantly dependent on the grid dimension selected for the FLACS simulation. To evaluate the flame area in a given flame volume, the β flame model has replaced the H-M model since the version of FLACS-93. The β flame model is implemented with the flame area calculation model as below: 1

2

A = π 3 (6V ) 3

(3.19)

wherein V is the flame volume. It has the relationship with the volume fraction of products f as follows, f =c

1+τ 1 + τc

(3.20)

where τ is density/expansion ratio between the flammable product and the reactant. Ignition region in FLACS is usually recommended to be set as a point, a line or a plane in one control volume. However, the newly developed guidelines specify the ignitions at different locations and time.

3.1.5 Geometry Counting and Porosity Calculations In FLACS, the geometry of a practical structure with complex components is generally reconstructed using a series of simple objects, such as boxes and cylinders. The cofile, which is the geometry storage file in FLACS, is used to count the number of cylinders and boxes in the geometry. Boxes are identified with area porosities, size and location in x-, y- and z-directions, while cylinders are described with a diameter, a position and the length in one direction. Depending on the grid size, the geometry objects fit into either on-grid or sub-grid numerically. The area and volume porosities in the grid cells are then calculated using Porcalc, which is one of the preprocessors in FLACS. The area porosity is defined according to the mean blockage of the control volume surface area, and the mean blockage of the inside volume of the control volume defines the volume porosity. If the control volume is completely blocked, the porosity value will be 0, and the fully open control volume will result in porosity value of 1. Porcalc also computes drag factors and turbulence generation for end surface contributions in sub-grid. In the i-direction, where the object walls pointing positively or

3.1 Methodology of CFD Simulation

51

negatively, the turbulence generation factor from a sub-grid object will be calculated as (Arntzen, 1998): Ti± = γi

ai± Ai

(3.21)

where Ai and ai± are the i-direction area of the grid cell and the area of the object inside the grid cell, respectively. To account for the differences in flows around the cylinders and the boxes, γi is specified as 1.0 for boxes and 0.7 for cylinders. In the control volume, the turbulence generation factors from all sub-grid items are then summed up to derive the turbulence factor in a direction for a solo grid cell, so that cell = Ti±



Ti±

(3.22)

cv

In a grid cell, the sub-grid objects exist only if (Ti+ + Ti− ) > 0. Turbulence subgrid diameter Dicell is calculated as an averaged turbulence generation factor of the sub-grid diameter from all sub-grid objects: Di max(Ti+ , Ti− ) cell (3.23) Di = cv cv max(Ti+ , Ti− )

3.1.6 Boundary Conditions Prior to running FLACS simulations, the specific boundary conditions for the outer boundaries of the simulation domain should be specified. Potential error could be incurred in modelling the flow conditions at the numerical boundaries in FLACS, especially if boundary domain is small. The sound speed at the boundary will be overestimated when the flame transports beyond the boundary. The increase in the flow of volume over the boundary leads to inaccurate overpressure calculation. Therefore, a sufficiently large boundary domain is suggested to avoid the improficiency in modelling the flow at the outer boundaries. FLACS provides five boundary conditions for choice to prevent artefacts in calculation for different scenarios (Bjerketvedt, Bakke, & Bakke, 1997). EULER The EULER formulation is the inviscid flow equations, which are discretized for a boundary element. In the case of outflow, the governing continuity and momentum equations are applied on the outer boundary. The pressure outside the boundary is specified as ambient pressure in the simulation. While for sonic outflow and inflow, a NOZZLE boundary condition can be adopted. In unconfined scenarios, the EULER boundary condition may lead to over low overpressures. Thus, the boundary domain may be extended and the PLANE_WAVE condition should be applied.

52

3 VCE Overpressure Prediction by CFD Modelling

NOZZLE As mentioned above, the NOZZLE boundary condition is applied for both sonic outflow and sub-sonic outflow/inflow calculation. For porous objects with small sharp-edged grids or holes, such as grating and louvres, the NOZZLE formulation is the optimal choice. A discharge coefficient is computed from a drag coefficient and the area porosity. Compared to the EULER boundary condition, the NOZZLE condition may yield too high explosion overpressures. In spite of that, it is considered to be a robust formulation. Similarly, for unconfined cases, the NOZZLE boundary condition may give too low overpressures. PLANE_WAVE To eliminate the reflection of the blast waves at the open boundaries, which is ensued by the NOZZLE or EULER formulations, the PLANE_WAVE boundary condition should be specified. In the PLANE_WAVE boundary condition, the reflection of outgoing pressure waves is approximately eliminated by extrapolating the overpressure at the boundary. On the other hand, after the expansion of the gas explosion, the overpressures in the PLANE_WAVE condition may be slightly increased and stabilized at an elevated level. Thus, it is suggested to extend the grid to increase the total volume and the PLANE_WAVE formulation is applied to avoid the overpressure elevation, especially in low confinement conditions. However, for the scenarios that boundaries are in the vicinity of vents in the semi-confined condition, the PLANE_WAVE is not a feasible option. WIND The WIND boundary condition, such as the wind direction, speed and length scale, should be identified for the gas dispersion simulations. In the WIND condition, the velocity of flow is assumed perpendicular to the outer boundary. The turbulence parameters should be given manually. The designated wind velocity in FLACS is accumulated over a specified time interval to prevent artefact of strong transient response by sudden application of velocity. SYMMETRY The SYMMETRY boundary condition is applicable for the scenarios where a symmetry plane can be identified, such as the MERGE geometries. Computation intensity is significantly reduced by application of the SYMMETRY boundary condition. However, cautions much be exerted for application of the SYMMETRY boundary because the overpressures may be artificially elevated by the unrealistic symmetry plane, which practically acts as a computational boundary to reflect the overpressures over the entire geometry. If such is the case, the SYMMETRY is not recommended.

3.2 Modelling Turbulence and Combustion

53

3.2 Modelling Turbulence and Combustion 3.2.1 Turbulence Model The strength of a vapour cloud explosion depends very much on the turbulent burning velocity, which is determined by the turbulent length scale and intensity in the turbulence field. Therefore, turbulence model is indispensable to calculate efficiently the turbulent burning velocity. The k–e turbulence model given by Launder and Spalding (Launder & Spalding, 1974) is thus applied in FLACS (Hjertager, 1993; Arntzen, 1998). The equation for turbulent kinetic energy is    ∂uj ∂ μeff ∂k ∂ ∂  ρuj k = (3.24) + G − ρε; G = σij (ρk) + ∂t ∂xj ∂t σk ∂xj ∂xj The equation for dissipation of turbulent kinetic energy takes the form of    ∂ ε ∂ μeff ∂ε ε2 ∂  + C1 G − C2 ρ ρuj ε = (ρε) + ∂t ∂xj ∂xj σε ∂xj k k

(3.25)

wherein G is the generation rate of turbulence. The relationship between k and ε and the effective turbulence viscosity μeff in the Boussinesq eddy viscosity model are, respectively, given by μt = C3 ρ

k2 ε

μeff = μ + μt

(3.26) (3.27)

where μt and μ are the turbulent viscosity and laminar; the constants C 1 , C 2 , C 3 , σk and σε are 1.44, 1.92, 0.09, 1.0 and 1.3, respectively.

3.2.2 Combustion Model Two different models, namely the flame model and burning velocity model, are usually adopted to govern the combustion process. Flame model For the flame modelling, the relation between the mass fraction of products and the reaction rate of the unburned reactants to fully burned products is ρc = ∇ρ∇c + w t

(3.28)

54

3 VCE Overpressure Prediction by CFD Modelling

wherein w denotes the reaction rate; c stands for the mass fraction of products; and  represents the diffusion coefficient. In terms of mass fraction of fuel mfu in FLACS, the conservation equation in the combustion modelling gives ρmfu = ∇ρ∇mfu + mfo w t

(3.29)

The H-M model, which calculates the reaction rate for turbulent combustion as in Eq. (3.16), was adopted in the earlier versions of FLACS. The β flame model replaced it in the later versions. The reaction rate after modification of the probability density function is then expressed as

   wβ = wρmin δ c − cq , c, 9 − 9c

(3.30)

herein cq is the minimum mass fraction of products when the reaction rate is larger than 0; δ denotes the flame thickness; W stands for the dimensionless reaction rate. Corresponding to the burning eigenvalue, the diffusion coefficient and dimensionless reaction rate should satisfy the following relations to calculate the burning velocity S τ cq = 0.325

(3.31)

w = 1.37S 2

(3.32)

Burning velocity model The burning velocity, an input for the flame propagation during the explosion, varies from the laminar burning velocity to the quasi-laminar burning velocity before it reaches the congested region to become turbulent. In the beginning of the combustion, the laminar burning velocity is specified according to the fuel type, pressure and fuel–air mixture. Below a standard atmospheric pressure, the pressure in terms of the laminar burning velocity is written as (Kuo, 1986)  SL = SL0

P P0

β e (3.33)

where SL0 is the initial laminar burning velocity; P0 is the initial pressure; β e is the pressure exponent, which is approximately 0 for hydrocarbons with burning velocity in the range of 0.5 m/s–1.0 m/s. For stoichiometric methane and propane mixtures, β e is −0.18 and −0.05, respectively. For compressed gas, β e is 0.07 for methane, 0.44 for propane and ethylene.

3.2 Modelling Turbulence and Combustion

55

Whereas in the quasi-laminar regime, the burning velocity increases with the flame propagation distance from the ignition point to the end of flame radius. The correlation between the laminar and the quasi-laminar burning velocity can be described as

   R 1/2 SQL = SL 1 + χa min ,1 (3.34) 3 wherein R denotes the flame radius; χa is a fuel dependent constant, which is between 2 and 8 depending on the fuel–air mixture and the condition of ignition. In the turbulent regime, the turbulent burning velocity correlation in FLACS is expressed as (Bray, 1990) ST = 0.875u Kf−0.392

(3.35)

where u is the turbulent velocity fluctuation; K f stands for the ratio of flow strain rate to flame gradient, which is termed as the Karlovitz stretch factor (Abdelgayed, Bradley, & Lawes, 1987). The Karlovitz stretch factor is derived by the turbulent Reynolds number Rn and the integral length scale lI :   2 u Kf = 0.157 R−0.5 n SL  u lI Rn = v

(3.36) (3.37)

where v is the kinematic viscosity. Substituting the turbulent Reynolds number Rn and the strain rate or the Karlovitz stretch factor K f in Eqs. (3.36) and (3.37) into Eq. (3.35), the turbulent burning velocity in FLACS becomes ST = 15SL0.784 u 0.412 lI0.196

(3.38)

The burning velocity in FLACS is then deduced as   Su = max SQL , ST

(3.39)

FLACS solves the equations above to derive the overpressures from each previous time step. The momentum equation gives a velocity field, which will be rectified along with the updated pressure and density field with a pressure correction algorithm (Patankar, 1980). The fuel density, flame radius and initial laminar flame speed of fuel dominate the combustion, thereby govern development of overpressure. In simple summary, parameters influencing the development of explosion pressures, i.e. mechanism of turbulent reactive gas dynamics, combustion processes and

56

3 VCE Overpressure Prediction by CFD Modelling

the geometry of configurations, are all accounted for in the methodology of the CFD-based solver—FLACS.

3.3 Numerical Procedure 3.3.1 Conservation Equations in Finite Domain For a typical gas explosion, the reactive dynamic calculation is governed by all the distinct terms in Eq. (3.1). The conservation equations of convection, transient, diffusion and source terms are given as ∂ (ρ) ↔ transient ∂t  ∂  ρuj  ↔ convestion ∂xj   ∂ ∂  ↔ diffusion ∂xj ∂xj S ↔ source

(3.40) (3.41) (3.42) (3.43)

Elaboration on the equations to calculate mass, velocity, kinetic energy of turbulence, dissipation rate and enthalpy, etc., can be found in Hjertager (1984).

3.3.2 Finite Domain Calculations The governing equations in the calculation domain are integrated over a control grid volume. In the y-direction, the net convective and diffusive flux will be stored as  A N (P − N ) + AS (P − S )

(3.44)

 where N, S, P are directions and locations;  denotes a general variable. A N and AN represent coefficients expressed by   ,N |e|)(−ρ xz (3.45) A = + 0.5(e + v ) N n N yN   ,S |e|)(ρ A xz (3.46) = + 0.5(e + v ) S P S yS

herein v is velocity; 0.5(e + |e|) is the differencing of upwind for the convection terms; and  is the grid cell size.

3.3 Numerical Procedure

57

The time derivative in the numerical approximations is given as tδ =

 ρPo xyz  P − oP t

(3.47)

wherein op is the general variable value of  at the previous time step for point P. The source terms after the volume integral can then be expressed as S = So + SP P

(3.48)

where So is the source term of SP at the previous time step for point P.

3.3.3 The Continuity and Momentum Equations in Finite Domain Similarly, by integrating the momentum equations over a control volume in all directions, the momentum in the finite domain can be expressed as AU P UP =



U o o AU i Ui + b UP − zy(pP − pW ) + SU

(3.49)

i

APV VP =



AiV Vi + bV UPo − zx(pP − pS ) + SVo

(3.50)

o AiW Ui + bW UPo − xy(pP − pL ) + SW

(3.51)

i

APW UP =

 i

where bU , bV and bW can be written as b , which is the time derivative of Eq. (3.47); p denotes pressure. For the continuity equation, it is the sum of the adjacent points in all directions to point P. The continuity equation in the finite domain is written as       ρP − ρPo xyz + (ρU )E − (ρU )W yz + (ρV )N − (ρV )S xz t   − (ρW )L − (ρU )H xy = 0 (3.52) where H and L denote the directions normal to the xy-plane, while E, W, N, S directions are in the xy-plane.

58

3 VCE Overpressure Prediction by CFD Modelling

3.3.4 Calculation Procedure To derive the density and velocity fields in conformity to continuity, the following equations are adopted:  yz   pW − pP U AP    xz VP = VP∗ + pS − pP V AP    xy pL − pP WP = WP∗ + W AP

(3.54)

pP = pP∗ − pP

(3.56)

UP = UP∗ +

(3.53)

(3.55)

where p is the pressure correction. The source term in the continuity equation is then written as  

   

 ∗    ρU W − ρU ∗ E yz + ρV ∗ S − ρV ∗ N xz   ∗   

  ρ − ρPo xyz + ρW ∗ L − ρW ∗ H xy − P t

SP =

(3.57)

A tri-diagonal matrix algorithm (Patankar, 1981) is then applied to solve the pressure correction and momentum equations. The implementation of the algorithm consists of three phases:  o o I  I  I A P P = AN N + AS S + FI  , S Phase I

(3.58)

 o I o II  II  II A P P = AH H + AL L + FII  ,  , S Phase II

(3.59)

 o I II o  III  ∗  ∗ A P P = AE E + AW W + FIII  ,  ,  , S Phase III

(3.60)

By substituting all the conservation, continuity and momentum equations into the tri-diagonal matrix algorithm, the velocity, turbulence kinetic energy and pressure in the finite domain are then calculated.

3.4 Simulation Flow Chart In simple summary of the previous sections, the methodology of the CFD simulation using FLACS has been delineated. The significance of appropriate representation of the geometry and boundary conditions for the effectiveness of the three-dimensional simulation has been addressed. The mechanism of gas dispersion and explosion to

3.4 Simulation Flow Chart

59

model the turbulence and combustion has been elaborated. The numerical procedure of FLACS has been summarized. The reactive gas dynamic approximations, which use the discrete grid volumes, finite domain equations and tri-diagonal matrix algorithm, have been mathematically described. The flow chart in Fig. 3.1 demonstrates in a glance the overall procedure of FLACS to simulate gas dispersion and explosion.

Start

Geometry import

Gas dispersion simulation grids

Gas composition

Wind speed&direction

Grid & porosities calculation

Leak rate, location&direction

Gas explosion simulation grids

Ignition type&location

Gas composition

Gas dispersion simulation

Gas explosion simulation

Leak dispersion output

Gas explosion output ( overpressure, impulse and combustion product mass fraction, etc.)

Gas cloud size, shape & location

Ignition

No Ignition End

Fig. 3.1 Flow chart of gas dispersion and explosion simulation in FLACS

End

60

3 VCE Overpressure Prediction by CFD Modelling

3.5 Simulation Examples Two examples of gas dispersion and explosion following the flow chart above are presented as below.

3.5.1 Geometry An offshore liquefaction module, which can be preprocessed by MicroStation, AutoReaGas or AutoCAD, is imported into the Geo2flacs utility in FLACS. Alternatively, it can be reconstructed numerically in CASD, which is the preprocessor in FLACS to model simple 3D geometries. By creating new database, new project and new geometry in the Geo2flacs utility, the structural components are generated and represented by idealized cylinders and boxes in FLACS as seen in Fig. 3.2.

3.5.2 Definition of Grid and Calculation of Porosities The grid simulation volume is specified to be three times more than the practical module size to eliminate the boundary reflection of flow and pressure in the gas

Fig. 3.2 Geometry for the gas dispersion and explosion simulation

3.5 Simulation Examples

61

dispersion and explosion simulations. The grid simulation volume is set to be 160 × 120 × 160 m3 for the liquefaction module against the real dimension of 40 × 30 × 40 m3 . To start with, the GRID DIR X, Y and Z are set as 160, 120 and 160, respectively, to define a regular 1-m grid. For gas dispersion simulation, the grids are then stretched in three directions. The stretching starts from the leakage location in the centre of the module as seen in Fig. 3.3a. Whereas for gas explosion simulation, the grids are stretched from the edge of the module as shown in Fig. 3.3b.

3.5.3 Gas Dispersion Simulation Gas dispersion simulation is to derive the gas cloud time-history build-up, including gas monitor region control, boundary condition set-ups, release scenarios identification and the extraction of dispersion results. Monitors and output control Figure 3.4 shows a regular pattern of red points on the ground of the module, which represents the monitor deployment. 3D outputs of FMOLE, FUEL, ER and VVEC in FLACS, standing for the respective fuel mole fraction, fuel mass fraction, equivalent ratio and velocity vector for these points are monitored and extracted as the representative gas dispersion data. The default value of NPLOT is 0 to specify no data for field plots are written to file during a simulation. In the present simulation, DTPLOT is set as 5 s to extract field outputs for each time interval of 5 s. The Courant–Friedrich–Levy number based on sound velocity (CFLC) and fluid flow velocity (CFLV) are given as 10 and 1, respectively, for the calculation refinement. The total simulation time TMAX is set as 1000 s to assure that the gas dispersion reaches the steady status. Boundary conditions To approximate the wind in W-E direction (+X), the WIND boundary condition is chosen for XLO at 4 m/s wind speed and the wind build-up time is set to be 0 s. Referring to Sect. 3.1.6, the NOZZLE boundary condition, which is in conformity to sonic/sub-sonic outflow/inflow calculation, is selected for XHI, YLO, YHI and ZHI. The ground ZLO uses SYMMETRY boundary condition to reduce simulation time and computational domain. For other initial conditions, such as the relative turbulence intensity, which is the parameter to calculate the value for turbulent kinetic energy in Eq. (3.24), is set as 0.1. The turbulent length scale for calculation of the dissipation rate of turbulent kinetic energy in Eq. (3.25) is equal to 0.01. It is usually chosen as 10% of leak diameter. During the simulation, FLACS automatically updates profiles for turbulence parameters and velocities at the boundary.

62

3 VCE Overpressure Prediction by CFD Modelling

(a) Gas dispersion simulation

(b) Gas explosion simulation Fig. 3.3 Grid volume of the geometry

3.5 Simulation Examples

63

Fig. 3.4 Regular pattern of monitor points in the geometry

Selection of leak scenarios Gas dispersion from all possible leakage sources in the module is modelled by selecting the representative scenarios. The possible leak scenarios are the combinations of gas composition, gas leak location, leak direction and wind direction, which govern the gas cloud size. • Wind regimes In the determination of wind directions, the ventilation simulation, which derives the wind speeds, is implemented. Generally, for a symmetric module, 1 single wind direction is sufficient, whereas simulations of 2–3 wind directions are required for asymmetric/non-uniform modules. In the present example, 1 wind direction from east to west is chosen for demonstration purpose, and the wind speed is given as 4 m/s. On the other hand, for a gas dispersion simulation, 2–3 wind speeds will be tested. • Gas composition The natural gas usually consists of 27% methane, 33% ethane, 15% propane and 19% pentane, which are similarly specified for the gas release simulation in the present study. The thermodynamic properties and relationships of the gas composition are

64

3 VCE Overpressure Prediction by CFD Modelling 3500

Cloud size (m3)

3000 2500 2000 1500 1000 500 0

0.375

0.75

1.5

3

6

12

24

48

96

Leak rate (kg/s) Fig. 3.5 Gas cloud size from gas dispersion simulations with different constant leak rate

inherently built-in in the FLACS system. For the pure gas release modelling, the equivalent ratio (ER) is set to be 1 × 1030 . • Gas leak rate The NORSOK standard (NTS 2001) specifies that the following categories should be used (all values in kg/s) for a gas leak simulation: 0.1–0.5; 0.5–1; 1–2; 2–4; 4–8; 8–16; 16–32; 32–64; >64. Accordingly, the leak rates in this example are identified as 0.375, 0.75, 1.5, 3, 6, 12, 24, 48 and 96 kg/s, respectively. Figure 3.5 shows the resulted gas cloud sizes against the leak rate and reveals that for the leak rate 0.375 and 0.75 kg/s scenarios, the resulted gas cloud size is very small. It can be interpreted as that these leak rates have negligible contribution on explosion risk so as not to be a concern. Two types of leak rates, i.e., constant leak rate and transient leak rate, can be chosen with reference to the practical leak scenario. The traditional approach to simulate the gas release in FLACS adopts a fixed leak rate until the steady-state pattern of dispersion is reached. The output of gas dispersion simulation can then be extracted. Whereas for dispersion simulations with transient leak rates, these setting may lead to underestimated gas cloud size in view of that the gas cloud build-up due to the transient leak could reach the concentration beyond the upper flammability limit (UFL). In such a case, instead of applying the steady-state approach, it is recommended to run a series of different leak rates for the transient leaks and manually incorporate the leak profile into a time-dependent leak database (cl-files) in FLACS to derive more approximating gas cloud size. For the dispersion simulation example in this section, the conventional constant leak rate is applied for simplicity. • Gas leak location and orientation Gas dispersions with multiple leak locations as shown in Fig. 3.6 are then modelled. The leakage locations are specified to distribute equably to engage the impacts of

3.5 Simulation Examples

65

congestion and confinement. For representation of the leak orientation, the leak impinging on a solid obstacle as in Fig. 3.6a, the leak towards more congested region as of Fig. 3.6b, and the leak in less congested area as in Fig. 3.6c, d are simulated, respectively. The interaction of wind direction with jet leak direction is modelled. From Fig. 3.6a, b, the jets in alignment with the wind direction result in larger gas dispersion clouds to cover the entire modules. Whereas Fig. 3.6c, d demonstrate the smaller gas clouds ensued in the modules by the jet perpendicular to the wind direction or shooting against the wind, respectively. Running simulations With definition of the geometry, grid and scenarios, the gas dispersion simulations are then run by the FLACS Runmanager as shown in Fig. 3.7. The simulation errors can be checked in the log file windows of the Runmanager. The errors are usually rectified by eliminating large Courant–Friedrich–Levy numbers (CFLV and CFLC), locating monitor and ignition into unblocked control volumes and verifying the leak parameters. Generally, 100% CPU core will be specified for FLACS simulations. Thus, it is recommended to run parallel simulations with proper job numbers equal to the number of CPU cores to reduce the computation time. The simulations can also be stopped by using the command “kill” in Task Manager or command “TSTOP” in the runtime simulation control file of FLACS. Extracting results The outputs of equivalent stoichiometric gas cloud volume Q9, the lower and upper flammability limits LFL and UFL, blockage volume and fuel mass/mole fraction can be extracted from rt-files in FLACS as in Fig. 3.8. For a specific gas dispersion scenario, such as the case of jet leak impinging on big obstruction as in Fig. 3.6a, the equivalent stoichiometric gas cloud volume Q9 extracted from the rf-files in FLACS can be tabulated as shown in Fig. 3.9. In view of that this example is conducted for the gas release case with a constant leak rate, it is observed that the steady state is reached at time 120 s. The stoichiometric gas cloud size is derived to be 16,000 m3 from this gas dispersion simulation, which can be applied in the following simulation of gas explosion.

3.5.4 Gas Explosion Simulation In order to determine the gas explosion overpressures on specified targets under different explosion circumstances, the gas explosion simulations are conducted after the gas dispersion simulations.

66

3 VCE Overpressure Prediction by CFD Modelling

(a) Jet leak impinges on big obstruction

(b) Jet leak directs into congested region

(c) Free jet leak propagates perpendicular to the wind direction Fig. 3.6 Gas dispersion with different leak locations and orientations

3.5 Simulation Examples

(d) Free jet leak propagates against the wind direction Fig. 3.6 (continued)

Fig. 3.7 Runmanager in FLACS

67

68

3 VCE Overpressure Prediction by CFD Modelling

Fig. 3.8 rt-files containing gas dispersion output

Equivalent Stoichiometric gas cloud volume Q9 (m3)

18000 16000 14000 12000 10000 8000 6000 4000 2000 0

0

20

40

60

Time (s)

80

100

120

Fig. 3.9 Equivalent stoichiometric gas cloud profile for the jet leak impinging on big obstacle at a constant leak rate

3.5 Simulation Examples

69

Monitors and output control Similar to the gas dispersion simulation, the layout of the monitors is designed as seen in Fig. 3.4. Monitor points measuring pressures across the equipment are located one control volume away from the target object on each side. In terms of the output, fuel mole fraction, overpressure, maximum overpressure, drag pressure, pressure impulse, temperature, velocity vector and combustion product mass fraction as represented by respective quantities of FMOLE, P, PMAX, DRAG, PIMP, T, VVEC and PROD in FLACS are identified for the specified monitoring points to measure gas explosion. Boundary conditions The default boundary conditions are EULER except for open configurations where PLANE_WAVE should be used. SYMMETRY can be applied to model the ground as an alternative to solid boundary, such as sea or solid ground. If the ZLO boundary is not extended to the sea level, EULER or preferably PLANE_WAVE boundaries should be adopted. EULER can be adopted when the external explosion is not expected to contribute substantially to the explosion loads in the area of interest or all boundaries are very far away from the area of interest. On the other hand, PLANE_WAVE should be applied if the external explosion may be of interest and/or have impact on the explosion inside the area of interest. For simple demonstration purpose, in this section, EULER boundary condition is adopted for all boundaries in the gas explosion examples. Gas compositions Different types of gases are usually processed in a practical module. Thus, it will be more effective for the explosion simulation to model at least two different gas compositions. The different gas compositions are differentiated/represented by the different gas weight/density/reactivity. It is recommended that for selection of gas compositions, consideration should be based on the specific leak frequencies for the different segments/compositions in the module and the range of density/reactivity of the different gas compositions. If multiple gas compositions are modelled, the compositions should be maintained the same for both dispersion and explosion. In the current examples, the natural gas consisting of 27% methane, 33% ethane, 15% propane and 19% pentane is also applied for the gas explosion simulation. It is worth mentioning that Eqs. (3.5)–(3.9) in Sect. 3.1.2 are the governing equations for the stoichiometric reaction during the combustion. Gas cloud size, shape and location • Gas cloud location The cloud position numbering is shown in Fig. 3.10. The cloud position also represents implicitly the fill direction of the cloud. For most offshore settings, automatic script can be adopted directly without adjustment. If part of the area is

70

3 VCE Overpressure Prediction by CFD Modelling

Fig. 3.10 Ignition and gas cloud numbering

blocked by large object, adjustment may be required. In order to prevent clouds from ending up in uncongested areas, repeated adjustment/repositioning should be implemented. It is suggested to place the clouds at different locations to engage the effect of the different congestion regions. The principle for scenario variation applicable for all cloud categories in view of that the explosion pressure may decrease with increasing cloud size due to shift in ignition and cloud location. A preferable gas cloud layout is recommended to be five cloud locations with each one in the respective corner and the centre, such as P0, P1, P2, P3 and P4 in Fig. 3.10. On the other hand, since the module size in this gas explosion example is small, the gas cloud is placed only in the centre to cover the entire module. • Gas cloud size After the gas dispersion simulations, the time-dependent equivalent stoichiometric gas cloud size can be extracted from the Q9 data in FLACS output files. Consequently, the cloud category can be deducted according to the clouds frequency and the resulting pressures based on the standards ISO, NORSOK. The maximum gas cloud size is expected to be the worst-case dispersion cloud, while the minimum gas cloud size generally results in explosion pressure over 0.1 barg with return frequency close to 10−4 . The grid guidelines applying to all the cloud categories and all the clouds in a given category should have a similar net cloud size. In the present example, the gas cloud size of 40 × 40 × 10 m3 , as shown in Fig. 3.11, to fill the overall module is chosen.

3.5 Simulation Examples

71

Fig. 3.11 Gas cloud volume for gas explosion simulation

• Cloud shape Clouds do not necessarily take up the whole space within the solid confinement, such as walls, decks and floor. When filling a module or congested region in one direction, the cloud can be expanded in the other directions to reach arbitrarily specified cloud size. Ignition Equations (3.16)–(3.20) are applied to govern the ignition process in FLACS. For each gas cloud, potential worst scenario very probably comes from centre and end/corner ignitions, which should be the cases to be simulated to be on the safe side, such as the two simulation examples shown in Fig. 3.12. The centre ignition usually yields the maximized radius of the spherical flame in view of that the ignition point is located furthest away from the open cloud side. This type of ignition is representative of the ignition from discrete ignition sources. Whereas for scenarios with walls acting as symmetry planes, wall ignition or corner ignition, representative of ignitions from continuous ignition sources, will outstrip the centre ignition in view of the worst explosion scenario identification. On the other hand, for small clouds where the end/corner and centre ignition are very close to each other, the end/corner ignition can be represented by the centre ignition.

72

3 VCE Overpressure Prediction by CFD Modelling

(a) Centre ignition

(b) Corner ignition Fig. 3.12 Ignition locations

3.5 Simulation Examples

73

According to the grid guideline for ignition location, the end/corner ignition should be located 2.5 control volume, i.e. grid cells, into the cloud for open ends and at 0.5 control volume for closed ends or partially blocked area, e.g. a solid wall. Running simulations and extracting results The gas explosion simulations are then implemented with the specified explosion scenario parameters. The parameter of DTPLOT, which is the time interval in seconds for field output, is set as 0.1 s in view of that the explosion lasts only 1–3 s in the present explosion examples. During the explosion overpressure modelling in Runmanager, the inherent turbulence model as in Eqs. (3.24)–(3.27) and combustion model in Eqs. (3.28)–(3.39) are employed along with the numerical algorithm as in Eqs. (3.50)–(3.53) in FLACS. Data reporting refers to the extraction of numerical simulation results with the utility programs r1-file and r3-file in FLACS, and the 1D, 2D and 3D output can be viewed in the post-processor in FlowVis, such as the results for the two types of explosions shown in Fig. 3.13. The overpressures are the highlight for the investigation. In the present examples, four monitor points on the corners, i.e. points 27, 34, 87 and 94 as in Fig. 3.14, are identified to extract the overpressure to compare the results from corner and centre ignition explosions. The overpressures for the corner ignition and centre ignition explosions are compared in Figs. 3.15 and 3.16. Job number 310008 represents the centre ignition, while 310001 is the corner ignition. From Fig. 3.15, centre ignition explosion results in greater overpressures than corner ignition explosion does for the monitor points 34 and 87. It is probably because that the ignition starting from the centre consumes more gas cloud volume before the flame turbulence propagates to the corners, provided that the flame distances for these two types of explosion are the same or very close to each other. On the other hand, for the monitor points of 27 and 94, overpressure comparison demonstrates opposite tendency for the centre and corner ignitions. Specifically, as seen in Fig. 3.16a, the centre ignition explosion generates rather even overpressures between 0.45 and 0.6 barg at the two opposite corners indicated by points 27 and 94. For corner ignition, the overpressure at point 27 is the minimum of 0.1 barg, whereas it shows the maximum overpressure of 0.65 barg at the corner of point 94. It can be interpreted as that the different flame distances in the two types of explosion contribute to the opposite overpressure development. The above examples demonstrate the specific simulation procedure using FLACS with the specifications for the gas cloud volume, flame distance, congestion and confinement. More gas explosion output can be extracted, such as the temperature and velocity data as shown in Figs. 3.17 and 3.18. The explosion frequency and probabilistic study can be carried out by conducting a sufficiently big number of explosion simulations to achieve a statistical significance.

74

3 VCE Overpressure Prediction by CFD Modelling

(a) Centre ignition

(b) Corner ignition Fig. 3.13 2D output of overpressure distribution

3.6 Summary

75

Fig. 3.14 Monitor points for comparison

3.6 Summary In this chapter, the methodology of the CFD simulation in FLACS has been elaborated with description of the relevant mathematical formulations. The detailed procedures of using FLACS to conduct gas dispersion and explosion simulations have been demonstrated along with the implementation of examples. Specifically, the general FLACS mathematical models, i.e. Reynold’s averaged mass, continuity, momentum, energy balance equations that govern the differential equations in the fluid flow mechanism of an explosion, have been discussed. The stoichiometry condition regarding the balance of reactants and production, the thermodynamic relationships for the energy balance in explosion, and ignition process to trigger the flammable cloud into combustion, have been mathematically expressed. The equations for calculation of turbulence and combustion, the boundary conditions and geometry modelling for gas dispersion and explosion simulation in FLACS have been illustrated. Each step in the numerical procedure of FLACS using the finite domain equations, discrete grid volumes and tri-diagonal matrix algorithm has been explained with the examples of gas dispersion and explosion simulations.

76

3 VCE Overpressure Prediction by CFD Modelling 0.95

P87 P34

Pressure (barg)

0.75 0.55 0.35 0.15 -0.05 -0.25

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Time (s)

(a) Centre ignition 0.32

P87 P34

Pressure (barg)

0.22

0.12

0.02

-0.08

-0.18

0

0.5

1

1.5

2

Time (s)

(b) Corner ignition Fig. 3.15 Comparison of 1D output of overpressure for monitors 34 and 87

The advanced CFD methods as of in FLACS have the flexibility to accommodate complex conditions to yield effective load prediction. On the other hand, some improficiencies of CFD analysis are observed as follows. • The CFD analysis may ensue very high computation intensity for very complicated practical scenario, which could be intractable for the current computer software and hardware.

3.6 Summary

77 0.75

P94 P27

Pressure (barg)

0.55

0.35

0.15

-0.05

-0.25

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Time (s)

(a) Centre ignition 0.8

P94 P27

Pressure (barg)

0.6

0.4

0.2

0

-0.2 0

0.5

1

1.5

2

Time (s)

(b) Corner ignition Fig. 3.16 Comparison of 1D output of overpressure for monitors 27 and 94

• It demands profound knowledge and understanding of the physics of the problem for an investigator to implement CFD analysis. • The CFD modelling is still under development. Therefore, further improvement is required to ensure the reliability and effectiveness of the CFD codes. More application of CFD-based consequence modelling will be presented in the following three chapters.

78

3 VCE Overpressure Prediction by CFD Modelling 2400

Temperature (K)

2100 1800 1500 1200

P27 P87 P34 P94

900 600 300 0.4

0.6

0.8

1

1.2

1.4

1.6

Time (s) Fig. 3.17 1D output of temperature for monitors 27, 34, 87 and 94 180

Velocity (m/s)

150 120

P27 P87 P34 P94

90 60 30 0 0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

Time (s) Fig. 3.18 1D output of velocity for monitors 27, 34, 87 and 94

References Abdelgayed, R. G., Bradley, D., & Lawes, M. (1987). Turbulent burning velocities—A general correlation in terms of straining rates. Proceedings of the Royal Society of London Series aMathematical Physical and Engineering Sciences, 414(1847), 389–413. https://doi.org/10.1098/ rspa.1987.0150. Arntzen, B. J. (1998). Modelling of turbulence and combustion for simulation of gas explosions in complex geometries. (Ph.D.), The Norwegian University Norway.

References

79

Bakke, J. R., & Hjertager, B. H. (1986). The effect of explosion venting in obstructed channels. In Modeling and simulation in engineering (pp. 237–241). Amsterdam: Elsevier Science Publication. Bjerketvedt, D., Bakke, J. R., & vanWingerden, K. (1997). Gas explosion handbook. Journal of Hazardous Materials, 52(1), 1–150. Bray, K. N. C. (1990). Studies of the turbulent burning velocities. Proceedings of the Royal Society of London, Series A: Mathematical and Physical Sciences, 431, 315–335. Ferrara, G., Di Benedetto, A., Salzano, E., & Russo, G. (2006). CFD analysis of gas explosions vented through relief pipes. Journal of Hazardous Materials, 137(2), 654–665. https://doi.org/ 10.1016/j.jhazmat.2006.03.037. Hjertager, B. H. (1982). Simulation of transient compressible turbulent reactive flows. Composites science and Technology, 27, 159–170. Hjertager, B. H. (1984). Computer-simulation of turbulent reactive gas-dynamics. Modeling Identification and Control, 5(4), 211–236. Hjertager, B. H. (1993). Computer modeling of turbulent gas-explosions in complex 2d and 3d geometries. Journal of Hazardous Materials, 34(2), 173–197. https://doi.org/10.1016/03043894(93)85004-X. Kee, R. J., Miller, J. A., & Jefferson, T. H. (1980). CHEMKIN: A general purpose, problemindependent, chemical kinetics code package. Sandia report, SAND80-8003, Livermore, California 94551. Kee, R. J. (1987). The Chemkin thermodynamic data base. Sandia report, SAND87-8215B. UC-4, Livermore, California 94551. Kuchta, J. M. (1985), Investigation of fire and explosion accidents in the chemical, mining and fuel-related industries—A manual. United States Department of the interior, Bureau of Mines, Bulletin 680. Kuo, K. K. (1986). Principles of combustion. New York: Wiley. Launder, B. E., & Spalding, D. B. (1974). The numerical computation of the turbulent flows. Computer Methods in Applied Mechanics and Engineering, 3, 269–289. NTS. (2001). Risk and emergency preparedness analysis, NORSOK STANDARD in Z-013, Rev. 2, 2001-09-01, Norway. Patankar, S. V. (1980). Numerical heat transfer and fluid flow. London: Hemisphere Publishing corporation. Patankar, S. V. (1981). A calculation procedure for two-dimensional elliptic situations. Numerical Heat Transfer, 4(4), 409–425.

Chapter 4

CFD-Based Overpressure Prediction for Single Modules-Extended GAME Correlation

Abstract In this chapter, a newly developed correlation for estimation of boundary overpressures in and around congested regions subject to vapour gas explosions is presented. The developed model (confinement specific correlation), which consists of parameters of volume blockage ration, the density of the gas, the flame path distance, the confinement ratio and the laminar flame speed of the flammable, shows a closer correlation with detailed CFD simulation in general, especially for realistic geometries.

4.1 Introduction This chapter conducts gas explosion analysis on practical modules. Specifically, the explosion overpressures are calculated for a series of single modules that consist of continuous congestions. A new correlation is developed to estimate boundary overpressures in and around congested regions subject to vapour gas explosions. The congested homogenous single modules in conformity to the experimental modules from the Guidance for the Application of the Multi-Energy method (GAME) (Eggen, 1998) and the realistic single modules from practical oil and gas trains are modelled. Similar to the correlation between structural geometries and gas explosion mechanism in GAME, the current gas explosion analysis on the single modules aims to develop a more proficient correlation between the geometric and the physical parameters in terms of versatility and accuracy. Although the empirical methods in GAME correlation have long been in use, significant improficiency and uncertainty have been identified due to inadequacy in accounting for important parameters, specifically the congestion and confinement, which boost flame acceleration and, thus, elevate overpressures. In particular, methods such as the Multi-Energy method (MEM) from GAME could be erroneous by more than an order of magnitude because it does not take into account the details of the practical geometrical layouts in view of that the input data for the explosion strength and congestion are estimated by the engineer’s observation. Conservation is usually exerted in applying these methods so as to yield overestimated overpressure, which could be much higher than the practical counterpart, leading to significant financial © Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_4

81

82

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

overspends. On the other hand, underestimation is not absolutely eliminated, which could lead to catastrophic consequence. In contrast to GAME, the CFD simulation, which is the most proficient methodology for quantifying the explosion risk, is applied in the current development of a new correlation to derive the explosion overpressures. The parametric study regarding the flame propagation, turbulence acceleration, congestion, confinement environment, etc., within a continuous congested region is implemented. The new correlation, integrating details of complex geometries into prediction, is more proficient than the existing analytical and experimental methods while saving the expense of high computation intensity in implementing CFD analysis.

4.2 The GAME Correlations and CFD Case Studies 4.2.1 GAME Correlations Guidance for the Application of the Multi-Energy method (GAME) (Eggen, 1998) provides additional guidance and extends the applicability of the MultiEnergy method (MEM). Phenomenological approach (Alfred, 1976; Edmund, 1989; Gurwitsch & Garcia-Gomez, 2009) is applied to derive the GAME correlation based on the experiments performed during the MERGE and EMERGE projects (EMEG, 1997; Harris & Wickens, 1989; Mercx, Johnson, & Puttock, 1995; Schumann, Haas, & Schmittberger, 1993; Vanwingerden, 1988, 1989) at the Dutch research institute TNO. From the report by Eggen (1998), corroboration of the GAME correlation against limited experiments data is positive. It is a conservative approach in the determination of the overpressure in most situations as characterized by assumption of homogenous congestion and confinement. Experimental tests for vapour cloud explosion are very expensive, whereas the relevance is not very promising in view of the difficulty in reproducing the realistic fields of congestion and confinement at the physical scale. The reliability and repeatability of the physical tests are very difficult to achieve because some factors, such as initial turbulence, the stability of the wind direction and speed, and the flexibility of structural components, are usually not repeatable. Therefore, the predictions from the proposed correlation and the results from the GAME correlation are verified against the results from simulations by CFD software FLACS so as to examine hundreds of cases including those with practical geometries at physical scales to spare the expense of physical tests. Derived from experiments, two variants of the GAME correlation were presented to determine the vapour cloud explosion overpressure. For low ignition energy and no confinement in 3D flame expansion conditions, there is

4.2 The GAME Correlations and CFD Case Studies

 Po = 0.84 ·

VBR · Lf D

83

2.75 Sl2.7 · D0.7

(4.1)

Low ignition energy and confinement between parallel plates (2D expansion) gives  Po = 3.38 ·

V BR · Lf D

2.25 Sl2.7 · D0.7

(4.2)

where Po the overpressure [barg]; VBR the volume blockage ratio, which is defined as the ratio of the total volume of the obstacles inside an obstructed region over the total volume of the region; the maximum distance of flame propagation, the radius of an equivalent hemiLf sphere with a volume equal to that of the configuration [m]; D the average obstacle diameter by assuming a homogeneous distribution of obstacle types and diameters [m]; the laminar flame speed of the flammable gas by assuming a homogenous Sl stoichiometric flammable cloud in all assessments [m/s].

4.2.2 Modules Tested in CFD Simulations CFD simulations are applied for single modules to validate the results from both the GAME correlation and the proposed correlation. The overpressures from the CFD code FLACS are compared with the results from both correlations. CFD simulations are performed for three artificial cases with homogenous congestion along with five practical inhomogeneous configurations as shown in Figs. 4.1 and 4.2. For the artificial modules 1–3 in Fig. 4.1, all module sizes are specified as 80 × 80 × 80 (m) and the obstacles in the configurations are deployed orthogonally by filling the pipes with diameter of 0.5 m. The five realistic modules come from a liquefied natural gas (LNG) train. All the structural components are modelled with boxes and cylinders. The preprocessor Porcalc is adopted to calculate the on-grid and sub-grid geometry objects by geometry counting and porosity calculations as described in Sect. 3.2.5. The volume blockage ratio (VBR) in Table 4.1 is read into the cofiles in FLACS. Propane and methane are applied in the explosion simulations. The properties of propane and methane are given in Tables 3.1 and 3.2, respectively. Equivalent stoichiometric gas clouds are assumed to perform the explosion modelling. The stoichiometric reaction during the combustion is expressed as in Eqs. (3.6)–(3.9). EULER formulation in Sect. 3.1.6, which is the inviscid flow equations, is specified for the boundary condition. The ignition process is modelled according to Eqs. (3.16)–(3.20) for the fractionation area, the pipe rack area, the combination

84

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

Artificial Module 1

Artificial Module 2

Artificial Module 3 Fig. 4.1 Artificial modules 1–3 (permission from Elsevier)

areas of the pipe racks and the mercury removal, and dehydration areas as shown in Fig. 4.2. For the explosion overpressure calculation in the FLACS simulator Runmanager, the turbulence model in Eqs. (3.24)–(3.27) and combustion model in Eqs. (3.28)–(3.39) are applied to derive the flame and velocity fields, which govern the development of explosion overpressure by engaging the numerical algorithm in Eqs. (3.53)–(3.60). In simple summary, totally three assumed modules subject to propane vapour explosions, five methane and five propane vapour explosions in realistic modules as given in Table 4.1 are studied. The values of volume blockage ratio, the laminar flame velocity, the characteristic average obstacle diameter and the gas composition as given in Table 4.1 are identified to calculate the overpressure.

4.2 The GAME Correlations and CFD Case Studies

85

Realistic Module 4

Realistic Module 5

Realistic Module 6

Realistic Module 7

Realistic Module 8 Fig. 4.2 Realistic modules 4–8 (permission from Elsevier)

4.2.3 Verification of GAME Correlation by Case Study With the CFD parameters listed above, the applicability and accuracy of GAME correlation are evaluated. Three CFD simulation models for the artificial modules are created with a uniform distribution of cylinders similar to those in the experiments upon which the GAME correlation is defined. Due to the homogeneity of the obstacle

86

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

Table 4.1 Parameters for different modules Case No.

Module No.

Gas composition

D(m)

VBR

Sl (m/s)

Gas density (kg/m3 )

Cm

1

1

Pure propane

0.50

0.070

0.46

1.8

1.000

2

2

Pure propane

0.50

0.070

0.46

1.8

0.925

3

3

Pure propane

0.50

0.070

0.46

1.8

0.888

4

4

Pure methane

0.37

0.040

0.4

0.65

0.716

5

5

Pure methane

0.45

0.058

0.4

0.65

0.707

6

4

Pure propane

0.37

0.040

0.46

1.8

0.716

7

5

Pure propane

0.45

0.058

0.46

1.8

0.707

8

6

Pure methane

0.12

0.080

0.4

0.65

0.917

9

7

Pure methane

0.34

0.103

0.4

0.65

0.980

10

8

Pure methane

0.31

0.096

0.4

0.65

0.903

11

6

Pure propane

0.12

0.080

0.46

1.8

0.917

12

7

Pure propane

0.34

0.103

0.46

1.8

0.980

13

8

Pure propane

0.31

0.096

0.46

1.8

0.903

arrangement and meshing grid, the CPU time for each calculation for the artificial configuration is reasonably short, i.e., within one hour. The confinement of the single modules is approximated with the insertion of parallel plates to test Eq. (4.2) in the GAME correlations. Figure 4.3 shows the values of the correlation R-squared factor between the GAME predictions and the numerical data for the first two homogenous cases, which are 0.78 and 0.51, respectively. These values imply that the GAME correlation gives prediction of overpressure as accurate as the CFD simulation does for congestion configurations that are filled with the regular-patterned pipes within a specific range of the confinement. On the other hand, as the area of the top plate decreases, negative R-squared values of −0.32 are derived for the homogenous case 3 with partially confined roof as shown in Fig. 4.3. It is because that the confinement effect is not accounted for in the GAME correlation. By varying the confinement as shown in Fig. 4.1 with the other parameters kept constant, the overpressures for the three homogenous scenarios from GAME correlation remain the same, while the overpressures are reduced with the decreasing confinement. It indicates that the overpressures are overestimated by the GAME correlation in the low confinement case. From the above verification, it can be stated that inappropriate definition of confinement within the GAME equation could result in an ineffective prediction. GAME correlation can only give conditional satisfactory results with the confinement is within a specific range. For simulation of realistic cases, the CPU time is increased to the range of one hour to three hours in view of the complexity of the geometries and longer time for the flame turbulence development in the irregularly congested regions. The corresponding GAME results and FLACS data do not match well as seen in Fig. 4.4,

6

6

y=x 5 R2=0.78

5

FLACS (Barg)

FLACS (Barg)

4.2 The GAME Correlations and CFD Case Studies

4 3 2

y=x 2

R =-0.51

4 3 2 1

1 0

87

0

2

4

6

0

0

2

4

GAME (Barg)

GAME (Barg)

(a) Case 1

(b) Case 2

6

6 y=x

FLACS (Barg)

5

2

R =-0.32

4 3 2 1 0

0

2

4

6

GAME (Barg)

(c) Case 3 Fig. 4.3 Comparison of GAME correlation overpressure results with FLACS results for homogenous cases subject to propane vapour explosions (permission from Elsevier)

which can be attributed to the improficiency in GAME not to accommodate the geometric inhomogeneity of the realistic geometric configurations and the variable confinement. It is because that the GAME correlations are derived from MERGE experiments with a regular pattern of obstacles, all of which can be idealized as homogeneously distributed cylinders in the obstructed region (Fig. 4.4). From the above comparison, it is found that the GAME correlation deduces results moderately close to those predicted by experiments only when the obstacles are regularly deployed with equal obstacle spacing and specific obstacle diameter, whereas for realistic modules with inhomogeneous congested volumes, the GAME correlation yields poor prediction of overpressures with more often overprediction than grossly underpredicts. In simple summary, the GAME equations deduce ineffective results for all realistic modules. It only gives a moderately efficient prediction for the idealized case with a homogenous distribution of congestion. The improficiency lies in lack of consideration of congestion inhomogeneity and the definition criteria of confinement, which prevent the general applicability of GAME correlation for practical problems. To

88

4 CFD-Based Overpressure Prediction for Single Modules-Extended … 100

case 1 case 2 case 3 case 4 case 5 case 6 case 7 case 8 case 9 case 10 case 11 case 12 case 13

y=x

FLACS (Barg)

10

1

0.1

0.01 0.01

0.1

1

10

100

GAME (Barg)

Fig. 4.4 Results from GAME correlation versus FLACS simulation for all cases (permission from Elsevier)

circumvent this improficiency, the following section will introduce a new correlation for calculating the explosion overpressure in congested single modules.

4.3 Parametric Studies and Development of a New Correlation Confinement is introduced into the proposed confinement specific correlation (CSC). Other critical parameters are chosen similarly as those for the GAME correlation. The derivation of the CSC is based on the linear least square method with a set of CFD simulations. In order to address appropriately the effect of individual parameter, all the approximately 400 CFD cases in this set are specified as homogeneous models by distributing pipes in regular patterns manually.

4.3.1 Conceptual Definition of Confinement and Congestion Both confinement and congestion affect turbulence-induced flame acceleration, which has a significant impact on overpressure (Bradley, Lawes, & Liu, 2008; Harrison & Eyre, 1987; Moen, Donato, Knystautas, & Lee, 1980; VandenBerg & Mos, 2005). For the GAME correlation, the congestion is defined using the volume blockage ratio (VBR) over the average pipe diameter. The inadequacy in application of this parameter lies in that the confinement variation by the change in congestion is

4.3 Parametric Studies and Development of a New Correlation

89

Place a pipe inside

1m*1m*1m Open air cube

Pipe (1m length, 0.4m diameter)

Fig. 4.5 Conceptual definition of confinement and congestion (permission from Elsevier)

unaccounted. And it is not appropriate to integrate VBR with the characteristic pipe diameter since the effect of the VBR is more prominent. Thus, in this chapter, a unique confinement parameter and different weighting factor for VBR in relation to the characteristic pipe diameter are evaluated. The conceptual confinement ratio is defined as the total blocked edge area of a space divided by the total volume of the space, i.e., ABlocked /ATotal . For a cubic volume with six open sides with dimension of 1 m x 1 m x 1 m, the conceptual confinement ratio is derived as ABlocked /ATotal = 0/6 (m2 /m2 ), while the fully confined cube has the confinement ratio of ABlocked /ATotal = 6/6 = 1. It implies the more surface area being blocked, the greater the confinement of the cube. For example, a partially confined volume with two sides fully blocked yields a ratio of 1/3. For the same cube with six open sides of ABlocked /ATotal = 0, placing a pipe with dimension of 1 m length and 0.4 m diameter in the centre of the cube, as shown in Fig. 4.5, the congestion volume in the cube becomes the volume of the pipe (0.126 m3 ). Commensurately, the volume blockage ratio, VBR = V blockage /V total , increases from 0/1 (m3 /m3 ) to 0.126/1 (m3 /m3 ) and the conceptual confinement ratio of the cube increases from 0/6 (m2 /m2 ) to 0.25/6 (m2 /m2 ). 0.25 m2 is the total area of the top and bottom cross sections of the pipe overlaying the surfaces of the cube on two sides. It is demonstrated that a change in congestion yields concomitant change in confinement of the configuration. The confinement and congestions should be correlated as two interactional factors for calculation of the explosion pressure.

4.3.2 Definition of Parameters Six parameters, i.e., the confinement ratio, the volume blockage ratio, the characteristic obstacle diameter, the flame propagation path, the laminar flame speed, and the gas density, are specified for calculation of the overpressure in a vapour cloud explosion event. In order to investigate the respective effect of each individual parameter, parametric studies are conducted and compared to the output from CFD simulations using FLACS.

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

Pressure from correlation Po (Barg)

90 4

Lf=64.3m

3.5

Lf=67.3m

3

Lf=70.9m

2.5

Lf=76.8

2

Lf=64.3m trendline Lf=67.3m trendline Lf=70.9m trendline Lf=76.8m trendline

1.5 1 0.5 0 0

0.1

0.2

0.3

Confinement: Exp (8.5*Cm) Fig. 4.6 Simulation results and the trendlines for confinement effect (permission from Elsevier)

Confinement effect In alignment with the 2D expansion of the GAME correlation, simulations are conducted using a geometric configuration that has parallel plates. The confinement is then defined as Cm =

AB AT

(4.3)

where the blocked area AB is the sum of all the obstructed areas on the top and bottom of the simulated domain; AT is the total area of the top and bottom surfaces. As shown in Fig. 4.1, the confinement parameter is modulated by reducing the blocked surface on the top of the geometries in FLACS, while all the other parameters are fixed. Twenty-four CFD simulations with six different confinement levels are performed to investigate the effect of the confinement parameter on overpressure. Figure 4.6 shows the pressure variation with varying confinement. Best fitting reclusive curve gives Po ∼ exp(8.5 · Cm )

(4.4)

where Po is the overpressure calculated at different monitor points along the explosion flame path. The L f in the legend of Fig. 4.6 denotes the direct distance from the ignition location to the target point of overpressure. Effects of VBR and average obstacle diameter D The GAME correlation integrates the volume blockage ratio (VBR) with the characteristic obstacle diameter (D) into one parameter as the indicator of congestion. In

4.3 Parametric Studies and Development of a New Correlation

91

order to address the effect of irregular congestion, the VBR and the averaged obstacle diameter are separately specified to explore the effect of irregularity. The volume blockage ratio herein is defined as the ratio of obstruction volume within the domain from the ignition point to the target point to the total configuration volume. Thus, for each specific target of interest, there is a unique VBR to calculate the overpressure. Applying the results from the above 8 cases with 4 different VBR results in totally 32 CFD cases. The parameter of VBR is varied, while all the other parameters are kept constant, e.g., constant C m = 1, to determine the effect of VBR on overpressure. Similarly, 25 CFD simulations with 5 different averaged obstacle diameters (D) are conducted to investigate the effect of the averaged obstacle diameter on the overpressure, while all the other parameters are fixed. The predicted results and the recursive curves are given in Fig. 4.7a and b, respectively, which can be expressed as Po ∼ 1.6 ln(VBR) + 6  Po ∼

D H

(4.5)

−1.5 (4.6)

where H is the height of the configuration. Maximum distance of flame propagation The maximum distance of flame propagation L f in the CSC is defined as the direct distance from the ignition location to the target point of overpressure in contrast to the assumption in the GAME project that L f is equal to the radius of an equivalent hemisphere with a volume equal to the volume of the configuration. Totally 300 CFD simulation cases are implemented to identify the effect of L f . The recursive trendlines for all cases show similar slopes in Fig. 4.8 with the maximum distance of flame propagation powered by 2.2, namely  Po ∼

Lf H

2.2 (4.7)

Mass density and laminar flame speed of gas For the two parameters of mass density and laminar flame speed of gas, 13 explosion scenarios extrapolating to approximately 1100 simulation cases are conducted for methane and propane with respective two different mass densities and two different laminar flame speeds. The approach taken here is the phenomenological method to analyse all the available data. The correlation for mass density and laminar flame powered by 0.5 and 2 is subsequently derived by minimizing the total variance of the complete correlation against CFD results.

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

Pressure from correlation Po (Barg)

92 3.5

Lf=39.9m

3

Lf=42.3m Lf=45.6

2.5

Lf=46.7m

2

Lf=39.9m trendline Lf=42.3m trendline Lf=45.6m trendline Lf=46.7m trendline

1.5 1 0.5 0 1

1.5

2

2.5

3

Volume blockage ratio: 1.6ln(VBR)+6

Pressure from correlation Po (Barg)

(a) Effect of VBR 14

Lf=45.6m

12

Lf=48.7m

10

Lf=49.7m Lf=51.7m

8

Lf=64.3m

6 4 2 0 0

5

10

15

Lf=45.6m trendline Lf=48.7m trendline Lf=49.7m trendline Lf=51.7m trendline

Averaged obstacle diameter: (D/H)-1.5

(b) Effect of D Fig. 4.7 Simulation results and trendlines for the effects of VBR and D (permission from Elsevier)

4.3.3 Proposition of New Correlation With derived parameters of confinement, volume blockage ratio, the average obstacle, laminar flame velocity and gas density, the new dimensionless correlation (CSC) is given as Po = 0.037 · e8.5Cm · [1.6 ln(VBRt ) + 6] · Pair



Lf H

where Po

the escalation overpressure [barg];

 2  2.2  −1.5  ρgas 0.5 D Sl · · · (4.8) H ρair Ss

4.3 Parametric Studies and Development of a New Correlation

93

Pressure from correlation Po (Barg)

5

Corner ignition Cm=1

4.5

Corner ignition Cm=0.925

4

Edge ignition Cm=0.85

3.5 3

Corner ignition Cm=1 trendline

2.5 2 1.5 1 0

20

40

60

80

Corner ignition Cm=0.925 trendline Edge ignition Cm=0.85 trendline

Maximum distance of flame propagation Lf 2.2 (m) Fig. 4.8 Simulation results of flame propagation maximum distance effect and recursive trendlines (permission from Elsevier)

Pair D Lf Sl Ss Cm VBRt ρgas ρair H

standard atmospheric pressure 101.325 kPa [1 barg]; the average obstacle diameter [m]; the direct distance from the ignition location to the target point [m]; the laminar flame speed of the flammable gas [m/s]; the speed of sound [m/s]; the confinement ratio; the volume blockage ratio of configuration region from the ignition point to the target; mass density of gas (kg/m3 ), which is assumed ideal under one standard atmosphere pressure at temperature 26 degrees; mass density of air (kg/m3 ); the height of the configuration (m).

4.3.4 Verification of the New Correlation For the verification of the proposed CSC, about 1100 realistic and idealized vapour cloud explosion simulations for the single modules as in Figs. 4.1 and 4.2 are conducted by CSC and FLACS. The obstacle configurations are filled with equivalent stoichiometric flammable gas cloud in the simulations. Methane and propane are adopted as fuels. The corresponding parameters are given in Table 4.1. The comparison of the results between the overpressures by the CSC and the FLACS simulation is given in Fig. 4.9. The R-squared value yielded by this comparison is 0.8394 to indicate generally very close approximation. For individual case as shown in Fig. 4.10, the correlation factor R-squared is within the range from 0.44 to

94

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

Pressure in FLACS (Barg)

16 14

y=x R2=0.836

12 10 8 6 4 2 0 0

2

4

6

8

10

12

14

16

CSC correlations calculated pressure (Barg) Fig. 4.9 Overall R-squared values of CSC predictions versus FLACS results for 13 simulation cases subject to two types of gas vapour explosions (permission from Elsevier)

0.90, which implies the CSC predicts overpressures for all the realistic and idealized modules as accurately as FLACS simulation does. In terms of the homogeneous cases 1–3 with different confinement, the factors of R-squared in CSC are 0.886, 0.608, and 0.464, respectively, which are much closer to the CFD results than the results from GAME correlation. It can be stated that the CSC equation with the refined confinement definition can be effectively applied to the practical problems with different congestion and confinement conditions.

4.3.5 Discussions In the gas explosion overpressure analysis on single modules, the GAME correlation includes the combined volume blockage ratio and the characteristic pipe diameter to account for congestion, which leaves out the definition of the confinement. It is shown in Sect. 4.3.1 that a change in congestion ensues change in confinement. Confinement and congestion play equally important roles in the evolution of combustion and the overpressure. Thus, in realistic explosion situations, the two parameters should be both explicitly accounted for in the calculation of explosion overpressure. Another important parameter to determine the vapour explosion pressure is the flame path length, which is assumed in the GAME correlation to be equal to the radius of a hemisphere with a volume equal to the volume of the obstructed region (Mercx, van den Berg, & van Leeuwen, 1998). Thus, for cases that ignition locations are at the edge/corner of the configurations and the aspect ratio of configurations is larger than 1, derivation of the flame path length based on the GAME assumption

4.3 Parametric Studies and Development of a New Correlation 6

7

y=x R² = 0.886

5 4 3 2 1 2

4

1 0

2

4

(1) Case 1

(2) Case 2

6

0.025

0

2

0.015 0.01 0.005 0

4

(4) Case 4 0.05

0.01 0.005 0.01

0.02

0.03 0.02 0.01 0

0.03

0.03

y=x R² 0.585

0.04

0

0.02

0.04

CSC (Barg)

CSC (Barg)

(5) Case 5

(6) Case 6 7

y=x R² = 0.767

y=x R² = 0.765

6

FLACS (Barg)

0.06

0.02

(3) Case 3

0.015

0.07

0.01

CSC (Barg)

0.02

0.08

0

CSC (Barg)

y=x R² = 0.486

0

y=x R² = 0.441

0.02

FLACS (Barg)

FLACS (Barg)

2

CSC (Barg)

y=x R² = 0.464

0.025

FLACS (Barg)

3

CSC (Barg)

0.03

0

4

0

6

FLACS (Barg)

FLACS (Barg)

5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0

0

y=x R² = 0.608

5

FLACS (Barg)

FLACS (Barg)

6

0

95

0.05 0.04 0.03 0.02

5 4 3 2 1

0.01

0

0 0

0.05

0

2

4

CSC (Barg)

CSC (Barg)

(7) Case 7

(8) Case 8

6

Fig. 4.10 CSC overpressure results versus FLACS results for 13 cases subject to methane and propane vapour explosions

96

4 CFD-Based Overpressure Prediction for Single Modules-Extended … 7

FLACS (Barg)

FLACS (Barg)

y=x R² = 0.485

6 5 4 3 2 1

20 18 16 14 12 10 8 6 4 2 0

0

2

4

6

y=x R² = 0.561

0

2

4

6

CSC (Barg)

CSC (Barg)

(9) Case 9

(10) Case 10

8

14

y=x R² = 0.644

y=x R² = 0.787

12

FLACS (Barg)

FLACS (Barg)

0

9 8 7 6 5 4 3 2 1 0

10 8 6 4 2

0

5

10

15

20

0

0

5

CSC (Barg)

CSC (Barg)

(11) Case 11

(12) Case 12

10

14

y=x R² = 0.90

FLACS (Barg)

12 10 8 6 4 2 0

0

5

10

CSC (Barg)

(13) Case 13 (Permission from Elsevier)

Fig. 4.10 (continued)

will yield inappropriate conversion, which ensues error in the overpressure. This improficiency further hinders the general application of the GAME correlation for realistic scenarios. It should be mentioned that the GAME correlation is dimensionally unbalanced. The dimensions of the right-hand side of the equation do not match those for the left-hand side, which is probably resulted from the insufficient parameters to address the realistic problems. Compared to the GMAE correlation, the applicability of the newly developed correlation for overpressure calculation in continuously congested single modules

4.3 Parametric Studies and Development of a New Correlation

97

is greatly improved because confinement and congestion are more realistically addressed. It is understood that the striking difference between the two correlations is that the confinement C m , which is not accounted for in the GAME correlation, is introduced into CSC. In order to trigger the two different GAME equations to calculate the overpressure, the gas explosions are classified into three categories, namely 1D, 2D and 3D expansions, depending on the degree of confinement. However, it is ambiguous to distinguish 2D from 3D flame expansion in view of insufficient details for configurations. For example, explosion charge confined by parallel planes is usually identified to be 2D expansion to engage Eq. (4.2) for calculation, which results in very large errors because of inappropriate representation of the partially confined top. To amend this improficiency, a criterion is proposed in the CSC that the confinement is defined by considering the ratio of blocked area to the total surface area of the configuration. The correlation of the CSC equation’s results and the FLACS data are compared in Fig. 4.10. In the CSC, the volume blockage ratio and the average diameter are addressed separately with unequal weightages to quantify the congestion. The volume blockage ratio of configuration is defined as the total obstructed volume over the total volume from the ignition point to the target. The overpressure calculation in the configuration with inhomogeneous congestion can then be more specifically represented for the individual target of interest. Advantage of the CSC over GAME lies in also the relatively easy derivation of the maximum distance of flame propagation (L f ) as the direct distance from the ignition location to the target point of overpressure, whereas in the GAME project, L f is assumed to be equal to the radius of an equivalent hemisphere with a volume equal to the volume of the configuration. Thus, for edge/corner ignition cases and the configurations with aspect ratio of larger than 1, the GAME correlation may not be applicable. Improvement in the new correlation is also reflected in the balanced input dimension to represent different fuels and the gas mass density as well as the laminar flame speed of gas (S l ) is introduced into the CSC. On the other hand, all the derivations conducted so far by the two correlations have been limited to methane and propane as fuels for explosions. More tests using different flammable gases and mixed gases are expected to further validate and extend the applicability of CSC model. It is noteworthy that for CFD calculations, application for mixed gases and multiple species is a new development in recent 10 years, while pure propane or pure methane are the general industrial standard practices. The effect of carbon dioxide in explosion reactions is mainly as a thermal sink which slows down the combustion rate and, thereby, reduces overpressures. On the other hand, a notable effect requires large amounts of the gas to present. The same is true for humidity, i.e., water content in the vapour cloud. Hence, the impact of air humidity is not thoroughly explored in the present simulations. The cases modelled here and the current common practice with CFD in industry ignore humidity in the most majority of cases as its effect on overpressure is relatively negligible. The exception is for mitigation measures involving deluge, or events involving rain where

98

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

the evaporation of the water has a significant effect. The inclusion of this effect would make an interesting expansion to the current work, which requires in-depth exploration. It is not affordable and a deviation for the current study. Similarly, the effect of carbon dioxide mixed with the reactants has not been considered. CO2 is oftener than not ignored in CFD modelling of explosions unless it presents in significant quantities mixed with the reactants, which is possible but not frequent.

4.4 Evaluation of Explosion Overpressures in Irregularly Structured Configurations Using New Correlation The newly developed correlation CSC is then applied to predict the explosion overpressures for irregularly structured single modules with deliberately varied geometrical parameters. The main purpose of modelling irregularly structured modules is to approximate the large-scale realistic oil and gas structures and conduct the explosion simulation on such complex geometries. These realistic onshore/offshore facilities typically display a high degree of inhomogeneity in confinement and congestion.

4.4.1 Definition of Regularity or Irregularity of Confinement and Congestion This subsection describes two types of geometries—regular and irregular arrangements of congestion and confinement. In terms of the congestion, the artificial module in Fig. 4.1 features uniform obstacle diameter and a regular pattern of obstacles. In contrast, the Module 1 and Module 4 in Fig. 4.11 are modelled with irregularities. Unlike the previous study (Li, Abdeljawad, et al., 2014) where the simulations modules are extracted from an existing liquefied natural gas (LNG) train with realistic layouts of structural components with random irregularities, the geometries in Fig. 4.11 are artificial modules with adjustable irregularities. For example, modules 1–4 in Fig. 4.11 are organized with increasing obstacle diameters, equidistant separation distances, mixed intersecting obstacle arrangements, etc. These artificially irregular modules are large-scale modules, while those artificial ones in the previous study (Li, Abdel-jawad, & Ma, 2014) are in small scale. Using the previously proposed definition of confinement, all simulations are conducted for the configurations with the parallel plates in semi-3D overpressure expansion. The confinement ratio is defined as the ratio of the blocked area on the top and bottom plates to the total area of the top and bottom surfaces. Thus, a configuration covered with two solid top and bottom plates, such as the module in Fig. 4.12a, is considered to be fully confined in the z-direction. The one without top plate as in

4.4 Evaluation of Explosion Overpressures in Irregularly Structured …

Module 1 - irregular-arranged

Module 3 - irregular-arranged

99

Module 2 -irregular-arranged

Module 4 -irregular-arranged

Fig. 4.11 Modules 1–4 with irregularities (permission from ACSE)

Fig. 4.12c is defined as open in the +z-direction. The partial confinement between the open air and the full confinement is specified to test the correlations under the conditions of irregular confinement.

4.4.2 Application of CSC to Irregularly Arranged Modules Using the CSC, overpressures are estimated for configurations with congestion of an irregular arrangement subject to vapour cloud explosions. The four modules in Fig. 4.11 with inhomogeneous obstacles are evaluated to estimate totally 400 new explosions. High confinement is specified for all the four modules. In the explosion models, a stoichiometric flammable gas cloud is assumed to fill the obstacle configurations; methane and propane are both used as fuels in this study. The parameters are given in Table 4.2.

100

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

(a) Fully confined module

(b) partially confined module

(c) Open in +z-direction Fig. 4.12 Artificial modules with varying confinement (permission from ACSE)

Figure 4.13 shows the proposed correlation pressure predictions as the x-ordinates against the pressures calculated with FLACS as the y-ordinates. The R-squared (R2 ) value is extracted for each of the cases. From Fig. 4.13, the R-squared value for each simulation case goes between 0.66 and 0.90, which implies the CSC correlation is effectively applicable to practical geometries with varying confinement ratios, irregular pattern of VBR, and varying obstacle diameters in the configurations. For comparison, GAME correlation is also adopted for predictions of the single modules in Fig. 4.11 with the irregular obstacles as well as high degrees of confinement. The results from the CSC correlation are also compared to results from GAME correlations as shown in Fig. 4.14. It is seen that GAME correlations give a poor agreement with the FLACS results. Specifically, the data obtained by GAME correlations tend to overestimate the overpressure significantly, whereas the CSC correlation result agrees well with FLACS simulations.

4.4 Evaluation of Explosion Overpressures in Irregularly Structured …

101

y=x R2=0.803

1.5

0.8

FLACS (Barg)

FLACS (Barg)

2

1 0.5

y=x R2=0.657

0.6 0.4 0.2 0

0 0

0.5

1

1.5

0

2

0.2

0.4

0.6

0.8

New correlation (Barg)

New correlation (Barg)

(1) Case 1

(2) Case 2 2

y=x R2=0.82 FLACS(Barg)

FLACS (Barg)

4 3 2 1 0

1

2

3

0.5

4

0

1.5

2

(3) Case 3

(4) Case 4 0.7

y=x

1.2

R2=0.878

y=x R2=0.801

0.6

1 0.8 0.6 0.4

0.5 0.4 0.3 0.2 0.1

0.2 0

0.5

1

0

1.5

0

(5) Case 5 1.4

0.4

0.6

(6) Case 6 0.7

y=x R2=0.666

y=x R2=0.895

0.6

FLACS(Barg)

1.2

0.2

New correlation (Barg)

New correlation (Barg)

FLACS (Barg)

1

New correlation (Barg)

1.4

1 0.8 0.6 0.4

0.5 0.4 0.3 0.2 0.1

0.2 0

0.5

New correlation (Barg)

FLACS (Barg)

FLACS (Barg)

1

0 0

0

y=x R2=0.861

1.5

0

0.5

1

1.5

0

0

0.2

0.4

0.6

New correlation (Barg)

New correlation (Barg)

(7) Case 7

(8) Case 8

Fig. 4.13 Comparison of CSC correlation overpressure data to FLACS results for the irregular configurations subject to methane and propane vapour explosions (permission from ACSE)

102

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

Table 4.2 Parameters in modules 1–5 with irregularities Case No.

Gas composition

D (m)

VBR*

S l (m/s)

Gas density (kg/m3 )

Cm

1. Module 1

Pure methane

0.37

0.11

0.40

0.65

1.00

2. Module 1

Pure propane

0.37

0.11

0.46

1.80

1.00

3. Module 2

Pure methane

0.31

0.14

0.40

0.65

0.96

3. Module 2

Pure propane

0.31

0.14

0.46

1.80

0.96

4. Module 3

Pure methane

0.33

0.13

0.40

0.65

0.90

6. Module 3

Pure propane

0.33

0.13

0.46

1.80

0.90

7. Module 4

Pure methane

0.21

0.04

0.40

0.65

0.90

8. Module 4

Pure propane

0.21

0.04

0.46

1.80

0.90

*VBR is the volume blockage ratio of the entire obstructed region for modules 1–4

Pressure in FLACS (Barg)

3.1 2.6

GAME correlation calculated pressure New correlation calculated pressure

2.1 1.6 1.1 0.6 0.1 0.1

0.6

1.1

1.6

2.1

2.6

3.1

The new correlation and GAME correlation calculated pressure (Barg) Fig. 4.14 Comparison of the new correlation and the GAME overpressure data to FLACS results for the irregular configurations subject to methane and propane vapour explosions (permission from ACSE)

4.4 Evaluation of Explosion Overpressures in Irregularly Structured …

103

(a) Monitors within the congestion

(b) Monitors in the open space Fig. 4.15 Specified monitor points for different gas explosion scenarios (permission from ACSE)

4.4.3 Rapid Prediction of Structural Damage The CSC correlation has been validated (Li, Abdel-jawad, et al., 2014; Li, Ma, Abdeljawad, & Hao, 2014) with very favourable agreement with pressures predicted by CFD modelling. In this subsection, a rapid structural damage level prediction process is presented. Two different configurations with 8 monitor points, as shown in Fig. 4.15, are numerically modelled using FLACS (GexCon, 2011). The pressure versus time history is derived for the specific structure members at the specified monitor points. Figure 4.16 gives the overpressure for the continuously congested single module as in Fig. 4.15a and the double modules with a safe gap as in Fig. 4.15b. For both configurations, the explosion starts at the centre of the left module. The flame propagates through the fuel away from the ignition point till the fuel is exhausted.

104

4 CFD-Based Overpressure Prediction for Single Modules-Extended … 250 Point 1 Point 2 Point 3 Point 4

Overpressure (kPa)

200 150 100 50 0 -50 0.02

0.025

0.03

0.035

0.04

Time (s)

(a) Monitors within the congestion 80

Point 1 Point 2 Point 3 Point 4

Overpressure (kPa)

60 40 20 0 -20 -40 0.02

0.03

0.04

0.05

Time (s)

(b) Monitors in the open space Fig. 4.16 Overpressure time histories for the specified monitor points (permission from ACSE)

The monitor points 1 to 4 are placed in the central line along the flame propagation direction from left to right. It is noted in Fig. 4.16a that the magnitude of the maximum overpressure increases from 125 kPa to 230 kPa as the flame path from the ignition through congestion increases. The maximum overpressure is detected at monitor point 3.

Impulse

0.03

0.04

0.05

180 160 140 120 100 80 60 40 20 0 -20 -40 0.01

0.2 0.15 0.1 0.05

0.02

0.15 0.1

50

0.05

0 0.05

0 -0.05 0.06

Overpressure (kPa)

0.2

100

Impulse (kPaxs)

Overpressure (kPa)

0.25

150

0.04

0.04

0.05

250

0.3

0.03

0.03

0 -0.05 0.06

(b) Monitor point 2 0.35

Perssure Impulse

0.02

0.25

Time (s)

250

-50 0.01

0.3

Impulse

Time (s)

(a) Monitor point 1 200

0.35

Pressure

200

Pressure Impulse

150 100 50 0 -50 0.01

0.02

0.03

0.04

0.05

Time (s)

Time (s)

(c) Monitor point 3

(d) Monitor point 4

0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 -0.05 0.06

Impulse (kPaxs)

0.02

0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 -0.05 0.06

105

Impulse (kPaxs)

Pressure

Overpressure (kPa)

140 120 100 80 60 40 20 0 -20 -40 0.01

Impulse (kPaxs)

Overpressure (kPa)

4.4 Evaluation of Explosion Overpressures in Irregularly Structured …

Fig. 4.17 Overpressure and impulse time histories at different monitor points in the congested area (permission from ACSE)

The overpressure increase observed above is attributed to flame acceleration, which has been described by Eggen (1998), Li, Abdel-jawad, et al. (2014), and Bjerketvedt, Bakke, and vanWingerden (1997). The geometry of the gas explosion scenario and flame propagation distance both contribute to the development of the flame acceleration and overpressure. In a gas explosion scenario, turbulence is generated when the flame interacts with the obstacles. It ensues flame acceleration and more turbulence is generated as the flame propagates further in the congested area. It is a self-feeding or mutual-strengthening mechanism for the increasing flame speed and overpressure. This is in contrast to an explosion pressure field where the maximum blast load is seen at the minimum stand-off distance decreasing with distance from ignition point. It is because a flame propagating in a premixed air–fuel cloud in an uncongested open space, as shown in Fig. 4.16b, simply does not speed up in the open uncongested space. The separation space in Fig. 4.15b reduces the congestions, and thus the intensity of turbulence, which results in the decrease in the overpressure. An explosion in open space is not affected by a separation space, and hence, the TNT explosion overpressure is only a function of the stand-off distance. For gas explosions, the pressure time history is typically a triangular-shaped wave with an extremely short time duration as shown in Fig. 4.16. For each monitor point, the impulse versus time data is derived by integrating the pressure time history as shown in Fig. 4.17. The maximum impulse is observed after the peak of the overpressure when the pressure attenuates to 0 kPa.

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

Fig. 4.18 Non-dimensional p–i diagram for an equivalent SDOF structural model (permission from ACSE)

9 8

Non-dimensional p

106

ymax = yc

7 6 5 4

Damage

3 No Damage 2

ymax > yc

ymax < yc

1 0

0

1

2

3

4

Non-dimensional i

By applying the data above to the structural members, the final states of damage, which is of major concern, can be assessed. Specifically, a structural member in an offshore module subject to gas explosion is simplified as an equivalent single degree of freedom (SDOF) system to assess its structural response in view of that the maximum deflection rather than the detailed deflection-time history of the structure determines the damage severity. In order to evaluate the structural damage level, a pressure–impulse (p–i) diagram of the equivalent SDOF structural model (Mays & Smith, 1995; Smith & Hetherington, 1994) is developed as shown in Fig. 4.18. Once the critical deflection (maximum allowable deflection) yc of the structure is specified, a critical curve is obtained as the dashed line shown in Fig. 4.18. The dashed line indicates various combinations of the non-dimensional initial peak overpressure p and the impulse i of the external load that will cause the allowable deflection of the structure. It should be mentioned that the non-dimensional pressure and impulse are defined as p = Po A/(kyc /2) and √ i = Io /yc kmse , respectively. The impulsive asymptote of the curve is i = 1.0, and the quasi-static asymptote is p = 1.0. Po is the initial peak pressure of the blast load, and I o is the impulse of the blast load as shown in Fig. 4.17. A is the equivalent cross-sectional area of the SDOF structure; mse is the equivalent mass of the SDOF structure; and k is its stiffness. In this study, taking the gas explosion scenarios at the four monitor points in the congested configuration as examples, the steel material is used to simulate the offshore structural members which are modelled as simply supported beams. The cross-sectional area, the equivalent mass and the stiffness are set as 1 m2 , 1 kg and 3 × 106 N/m, respectively. Therefore, the p–i combinations of the gas explosion blast load are determined. The four points indicated in Fig. 4.18 represent the blast load results obtained from Fig. 4.17. For the four monitor points, any data below the dashed curve (overpressure and impulse at point 1 and point 2) will not result in any damage of the structure, while those above the curve (overpressure and impulse at point 3 and point 4) will induce failure of the structure.

4.5 Summary

107

4.5 Summary In this chapter, a series of homogeneously or irregularly configured single modules with continuous congestion are calculated by CFD simulations. A new correlation to quantify the gas explosion overpressure is developed based on the linear least square method by fitting results from 400 CFD simulations of homogeneous geometries. The developed correlation is applicable to propane and methane to register the first step to develop versatile robust correlations for overpressure prediction of an arbitrary vapour cloud explosion. The proposed correlation, named as CSC, gives satisfactory results for idealized as well as practical inhomogenous single modules for two different explosion blast materials. CSC specifies a relation between parameters for the obstructed region, such as the average obstacle diameter, volume blockage ratio, confinement and the fuel properties. CSC is the recursive curve form data of numerical simulations using the CFD software FLACS. Approximately 700 CFD simulations results for practical geometries are compared with counterparts from both GAME and CSC for comprehensive verification. Highlight in CSC is its difference from GAME not to integrate geometrical confinement in prediction. Confinement is modelled in CSC by specifying quantitative congestion and the volume blockage ratio of different confined scenarios. The gas mass density is also taken into account in the CSC calculation. The new correlation has been tested against over 1100 simulation monitor points to demonstrate that it has more versatile applicability than the GAME correlation does. On the other hand, it is recommended that both correlations are applied as a supplementary benchmarking tools and GAME correlation should not be extended beyond the specifications of the experiments from which it was derived. Other 400 scenarios in irregularly structured single modules are further investigated by using CSC. These irregularly structured single modules approximate the scenarios in MERGE experiments with the only distinctive exception: the confinement and congestion are varied in a controlled manner so to engage the geometries with inhomogeneity for both parameters. The results further demonstrate that the CSC correlation is suitable for prediction of realistic geometries in large scale. It is worth mentioning that the numerically calculated pressure and impulse time histories by CFD codes are correlated to structural damage level by simplifying the offshore structural component as an equivalent SDOF model. The structural damage level is determined against the p–i diagram. It is observed from the results that more proficient calculation is demanded for the cases examined in Sect. 4.4.3 as they are at critical severity to cause permanent damage to structural members. In this chapter, a new VCE approximation correlation-CSC has been developed to be applied to evaluate effectively VCE overpressure in the irregularly structured single modules. The congestions with double modules or modules with gaps in between will be explored in the following chapter. Specifically, a data-dump tech-

108

4 CFD-Based Overpressure Prediction for Single Modules-Extended …

nique to improve the overpressure calculation accuracy in the safety gap modelling of double modules will be proposed.

References Alfred, S. (1976). The phenomenology of the social world. London: Heineman Educational Books. Bradley, D., Lawes, M., & Liu, K. X. (2008). Turbulent flame speeds in ducts and the deflagration/detonation transition. Combustion and Flame, 154(1–2), 96–108. https://doi.org/10.1016/j. combustflame.2008.03.011. Bjerketvedt, D., Bakke, J. R., & van Wingerden, K. (1997). Gas explosion handbook. Journal of Hazardous Materials, 52(1), 1–150. Edmund, H. (1989). Ideas pertaining to a pure phenomenology and to a phenomenological philosophy. Second book: Studies in the phenomenology of constitution. Dordrecht: Kluwer. Eggen, J. B. M. M. (1998). GAME: Development of guidance for the application of the multi-energy method. TNO Report PML 1995-C44, HSE Books. EMEG. (1997). Explosion Model Evaluation Group, specifications of test cases for gas explosions—Test case C1. EME project, DGXII, Brussels, Belgium. GexCon. (2011). FLACS v9.1 user’s manual. Norway: Doxygen. Gurwitsch, A., & Garcia-Gomez, J. (2009). The foundation of phenomenology: Edmund Husserl and the quest for a rigorous science of philosophy. Collected works of Aron Gurwitsch (1901–1973) (Vol. 1, Series Volume 192, pp. 463–470). https://doi.org/10.1007/978-90-481-2831-0_17. Harris, R. J., & Wickens, M. J. (1989). Understanding vapour cloud explosions—An experimental study. In 55th Autumn Meeting of The Institution of Gas Engineers. Harrison, A. J., & Eyre, J. A. (1987). The effect of obstacle arrays on the combustion of large premixed gas air clouds. Combustion Science and Technology, 52(1–3), 121–137. https://doi.org/ 10.1080/00102208708952572. Li, J. D., Abdel-jawad, M., & Ma, G. W. (2014). New correlation for vapor cloud explosion overpressure calculation at congested configurations. Journal of Loss Prevention in the Process Industries, 31, 16–25. https://doi.org/10.1016/j.jlp.2014.05.013. Li, J. D., Ma, G. W., Abdel-jawad, M., & Hao, H. (2014). Evaluation of gas explosion overpressures at configurations with irregularly arranged obstacles. Journal of Performance of Constructed Facilities, B4014003. Mays, G. C., & Smith, P. D. (1995). Blast effects on buildings—Design of buildings to optimize resistance to blast loading. London, UK: Thomas Telford Publications. Mercx, W. P. M., Johnson, D. M., & Puttock, J. (1995). Validation of scaling techniques for experimental vapor cloud explosion investigations. Process Safety Progress, 14(2), 120–130. https:// doi.org/10.1002/prs.680140206. Mercx, W. P. M., van den Berg, A. C., & van Leeuwen, D. (1998). Application of correlations to quantify the source strength of vapour cloud explosions in realistic situations. Final report for the project: ‘GAMES’. TNO Report PML 1998-C53. Moen, I. O., Donato, M., Knystautas, R., & Lee, J. H. (1980). Flame acceleration due to turbulence produced by obstacles. Combustion and Flame, 39(1), 21–32. https://doi.org/10.1016/ 0010-2180(80)90003-6. Schumann, S., Haas, W., & Schmittberger, H. (1993). Dust explosion venting—Investigation of the secondary explosion for vessel volumes from 0.3 M(3) to 250 M(3). Staub Reinhaltung Der Luft, 53(12), 445–451. Smith, P. D., & Hetherington, J. G. (1994). Blast and ballistic loading of structures. Oxford, UK: Butterworth-Heinemann. van den Berg, A. C., & Mos, A. L. (2005). Research to improve guidance on separation distance for the multi-energy method. TNO Prins Maurits Laboratory Research Report 369.

References

109

Vanwingerden, C. J. M. (1988). Investigation into the blast produced by vapour cloud explosions in partially confined areas. TNO Prins Maurits Laboratory Report No. PML 1988-C195. Vanwingerden, C. J. M. (1989). Experimental investigation into the strength of blast waves generated by vapour cloud explosions in congested areas. In 6th International Symposium on Loss Prevention and Safety Promotion in the Process Industries.

Chapter 5

CFD-Based Overpressure Prediction for Double Modules—Data-Dump Technique

Abstract When FLACS is applied to conduct the numerical simulations of large gaps with sizes approaching one or two times of the module size, a large error in the overpressures in the acceptor module is observed. In this chapter, a data-dump technique is developed to address this issue. The overall results indicate that the datadump technique is very effective for the evaluation of gas explosion overpressures in areas with large separation gaps.

5.1 Introduction In Chap. 4, the gas explosion evaluation is performed for the configurations with continuous congestion in single units. However, more complexity arises in the calculation of overpressure using FLACS for a congestion interruption between two modules. This chapter proposes a data-dump technique to segregate the complexity and improve the accuracy of simulation for the double modules with a safety gap using FLACS (Ma et al. 2014). The safety gap is one of the most efficient and state-of-the-art overpressure mitigation measures. Reports on the studies of safety gap effect on gas explosion using CFD simulation are limited. In this chapter, in order to study the effect of the safety gap, a data-dump technique is developed to improve the CFD simulation accuracy in overpressure estimation for the double modules. The data-dump technique resets the turbulence length scale for the double modules with different separation distances. The gas explosion assessment on the double modules with variable safety gaps, which are placed between the donor and acceptor, is carried out with application of the proposed data-dump technique. Five sets of the double modules containing obstructed regions with different separation spaces as in the TNO Prins Maurits experimental program are numerically simulated. Comparison of numerical simulation results and the published data from experiments indicate that the proposed technique improves the accuracy of the FLACS simulation significantly.

© Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_5

111

112

5 CFD-Based Overpressure Prediction for Double Modules—Data-Dump …

5.2 Numerical Models for Separated Congestions in Double Modules 5.2.1 Experimental Set-up The scenarios of double modules with safety gap extracted from the Research to Improve Guidance on Separation Distance for the Multi-Energy Method (RIGOS) research program (VandenBerg & Versloot, 2003) are modelled using FLACS. The configuration set-up parameters are given in Table 5.1. As seen in Fig. 5.1, the tested double modules consist of a number of tubes in two separated modules, the donor and acceptor modules. A plastic sheet is used to cover the two obstacles, which are placed on a concrete pad and filled with a flammable fuel–air mixture. The gas clouds are ignited at the ground centre of the congestion in

Table 5.1 Definition of obstacle configurations Case no.

Fuel type

VBR* (%)

Separation Cylinder distance diameter (m) (m)

Pitch (m)

Dimension No. of of the tubes in a donor row (m)

1

Ethylene

10.1

2.11

0.0191

0.089

1.408

2

Methane

10.1

0.22

0.0191

0.089

1.76

20

3

Ethylene

10.1

0.70

0.0191

0.089

1.408

16

4

Ethylene

10.1

0.35

0.0191

0.089

1.408

16

5

Ethylene

10.1

1.60

0.0191

0.089

1.06

12

6

Ethylene

10.1

0.35

0.0191

0.089

1.408

16

7

Ethylene

10.1

0.27

0.0191

0.089

1.06

12

8

Ethylene

10.1

0.27

0.0191

0.089

1.06

12

9

Methane

10.1

0.35

0.0191

0.089

1.408

16

10

Methane

10.1

0.35

0.0191

0.089

1.408

16

11

Methane

10.1

2.11

0.0191

0.089

1.408

16

12

Methane

10.1

0.70

0.0191

0.089

1.408

16

13

Methane

14

0.20

0.0191

0.134

1.596

12

14

Methane

14

0.20

0.0191

0.134

1.596

12

15

Ethylene

4.6

0.33

0.0191

0.134

1.33

10

16

Ethylene

4.6

0.40

0.0191

0.134

1.596

12

17

Ethylene

4.6

1.33

0.0191

0.134

1.33

10

18

Ethylene

4.6

1.60

0.0191

0.134

1.596

12

19

Methane

4.6

0.40

0.0191

0.134

1.596

12

16

*VBR is the volume blockage ratio, which is the ratio of the summed volume of the obstacles in an obstructed region to the overall volume of that region

5.2 Numerical Models for Separated Congestions in Double Modules

113

Fig. 5.1 Obstacle configurations in experiments (permission from Elsevier)

Fig. 5.2 Test layout of double modules (permission from Elsevier)

one module, which is identified as the donor module. The flame propagates through the donor module to reach and pass through the safety gap before advancing through the other module, which is termed as acceptor. In the experiments as in Fig. 5.2, nine overpressure sensors are deployed at equal distances along the axis of the donor–acceptor double configurations. The entire setup, including the locations of the sensors, which are represented in the simulations by monitor points, is numerically simulated. The pressures from sensors at the edge of each module obtained from experiments are compared with the counterparts from the numerical simulations.

5.2.2 CFD Modelling Using FLACS Three different volume blockage ratios (VBRs) are specified as shown in Fig. 5.3. The donor modules are numerically modelled in FLACS with varying obstacle diameters and arrangements. Specifically, all the cylinders are of the same diameter, i.e. D

114

5 CFD-Based Overpressure Prediction for Double Modules—Data-Dump …

(a) VBR = 10.1%

(b) VBR = 4.6%

(c) VBR = 14% Fig. 5.3 Obstacle configurations in FLACS (permission from Elsevier)

= 19.1 mm, and orientates orthogonally and regularly. By specifying two different pitches of P = 4.65D and P = 7D, two different volume blockage ratios of VBR = 10.1% and VBR = 4.6% modules are derived as shown in Figs. 5.3a, b. A third type of configuration with VBR = 14% is modelled by adding 24 regularly patterned vertical tubes with diameter of 114 mm as shown in Fig. 5.3c. The three obstacle modules are identified as type 1, type 2 and type 3, respectively. The configurations of the acceptors are identical in all the simulations with the volume blockage ratios of VBR = 10.1% and pitch P = 4.65D. All the simulations with FLACS are conducted using the grid cell size of 0.03 m, which equates to 33% of the smallest pitch length P = 4.65D (0.089 m). It is worth mentioning that the grid cell size of 0.03 m is determined based on a mesh size sensitivity calibration using a series of different grid sizes. Each simulation model in FLACS consists of two separate configurations of obstacles as seen in Fig. 5.4. The separation distances in the double modules are in the range of 0.20 m–2.11 m. The width and the length of the donor are set equally

5.2 Numerical Models for Separated Congestions in Double Modules

115

Fig. 5.4 Simulation model for two obstacle configurations with a safe gap (permission from Elsevier)

Fig. 5.5 Gas cloud transformation during explosion (permission from Elsevier)

116

5 CFD-Based Overpressure Prediction for Double Modules—Data-Dump …

from 1.06 m to 1.76 m by changing the number of the cylinders in a row as seen in Table 5.1, while the height of the configuration is half of the donor width or length. In the numerical simulations, the configurations are fully enveloped by a stoichiometric gas cloud as seen in Fig. 5.5 and the fuel types in the simulations are pure ethylene or pure methane in different tests. All explosions are initiated and simulated by igniting the cloud in the centre at ground level of the donor. In FLACS, the overpressures are monitored along the central axis within the configurations corresponding to locations of the pressure sensors in the RIGOS test layout as seen in Fig. 5.5.

5.3 Results and Discussions 5.3.1 Gas Explosion Overpressures with a Small Separation Distance in the Double Modules A small separation distance is defined as the open space with the separation distanceto-donor dimension ratio spanning 0.125–1. Large gaps refer to a greater than 1 separation distance-to-donor dimension ratio. Simulations of explosions in these congested configurations with the small separation distances are conducted using FLACS. The evolution of the overpressure after ignition of the gas cloud is presented in Fig. 5.6. High pressures are observed at the boundaries of the obstacles of the donor. The overpressure decreases substantially and continually along the separation gap due to flame deceleration in un-obtruded passage of the gap. Comparison of simulation results and experimental results from the measurement sensors at the edge of the donor module is shown in Fig. 5.7. The numerical maximum pressures and evolution of the pressure agree very well with the respective counterparts from the experiments. On the other hand, the simulated overpressure arrival time is sooner than that from the experimental observations. The reason could be that, in FLACS simulation, the gas cloud is ignited instantly at the specified time spot for ignition in the FLACS simulation, whereas in the experiments, the combustion occurs only when the fuel reaches the flammable limits. Therefore, the formation of stoichiometric gas delays the ignition in experiments. From Fig. 5.7b, the overpressure in the acceptor at point 9 is significantly greater than the pressure at the donor module point 1 (see Fig. 5.7a). It is because a very small-time interval is observed for the propagation of the blast in the donor and the acceptor due to the very short separation distance between the modules so as not to dissipate significantly the turbulence. The evolution of overpressures agrees well between the numerical simulation and the experiment observations. Taking the maximum overpressure at the measurement sensor at the edge of the obstacles, the overall comparison of the experimental and FLACS overpressure data is presented in Fig. 5.8, wherein a reasonably good agreement is displayed.

5.3 Results and Discussions

Fig. 5.6 Overpressure evolution in simulation (permission from Elsevier)

117

5 CFD-Based Overpressure Prediction for Double Modules—Data-Dump …

Fig. 5.7 Experimental records of overpressure-time history and FLACS simulation results at sensors (permission from Elsevier)

30 Experiment P1 Simulation P1

25

Overpressure [kPa]

118

20 15 10 5 0 -5 -10

0

50

100

150

200

Time [ms] (a) Case no.2

Overpressure [kPa]

400

Experiment P1 Experiment P9 Simulation P1 Simulation P9

350 300 250 200 150 100 50 0 -50

0

20

40

60

Time [ms] (b) Case no. 4 1000

Experimental Overpressure (kPa)

Fig. 5.8 Comparison of experimental and numerical overpressures (permission from Elsevier)

100

10

1

1

10

100

FLACS Overpressure (kPa)

1000

5.3 Results and Discussions

119

5.3.2 Gas Explosion Overpressures with a Large Separation Distance in the Double Modules Despite the good agreement for the test cases of short safe gap is demonstrated between the experiment and numerical simulation, opposite is observed when the separation distance (SD) to the donor dimension (DD) ratio is equal to 1.5. The simulations overpredict overpressures in the acceptor module. The test AE08 is taken from the RIGOS program (VandenBerg & Mos, 2002) to be simulated by FLACS. The donor has the similar dimensions as of the acceptor of (1.4 × 1.4 × 0.7 m3 ) in the test. The overall configuration, as shown in Fig. 5.9, is set up with the SD/DD ratio of 1.5, VBR of 10.1% and obstacle diameter of 0.019 m. The fuel type is specified as stoichiometric ethylene–air in FLACS. From Fig. 5.10, the overpressures observed at the monitors within and near the donor from FLACS simulation are approximately 80 kPa, coinciding with the experimental data. On the other hand, remarkably different evolution of the overpressures is observed after the flame propagates to the acceptor module between experiment and numerical simulation. The pressure values in the safe gap in both experiment and numerical simulations drop almost to zero because of that the obtrusion does not present, which is the booster for turbulence generation. When the flame reaches the acceptor module, the flame reaccelerates to ensue increased overpressures. The experimental pressures monitored in acceptor as in Fig. 5.10b of the RIGOS test are lower than the overpressures in the donor due to dissipation along the separation distance. In contrast, a significant jump of overpressures is obtained by using FLACS as seen in Fig. 5.10a. The reason for this discrepancy is postulated as related to the turbulence length scale. In the combustion process, the increment of pressure is mainly due to the increase of turbulent burning velocity. An important parameter used in the calculation of this velocity in the turbulence model is the turbulence length scale lLT , which is a typical length scale of the boundary. The turbulence length scale is also used to calculate an initial value for dissipation of turbulent kinetic energy.

Fig. 5.9 Simulation model with a separate distance of SD/DD = 1.5 (permission from Elsevier)

120

5 CFD-Based Overpressure Prediction for Double Modules—Data-Dump … 800 700

Pressure (kPa)

600 500

P1

P2

P3

P4

P5

P6

P7

P8

P9

400 300 200 100 0 -100 20

30

40

50

60

70

80

Time (ms)

(a) Overpressures-time histories from FLACS simulation

(b) Overpressures-time histories from experimental test AE08 of RIGOS (VandenBerg & Mos 2005) Fig. 5.10 Comparison of numerical and experimental overpressures for test with large separation distance (permission from Elsevier)

5.3 Results and Discussions

121

The equation for dissipation of turbulent kinetic energy (Hjertager, 1984) is given as  ∂  ∂ ∂ (ρε) + ρu j ε = ∂t ∂x j ∂x j



μeff ∂ε σz ∂ x j

ε=



Cμ k 3/2 lLT

ε ε2 + 1.44 G − 1.79ρ k k

(5.1) (5.2)

where ε is the rate of dissipation; k denotes the kinetic energy of turbulence; Cμ is the constant in the k–ε equation (typically Cμ = 0309); ρ stands for the density; x is the length coordinate in j-direction; u denotes the velocity (jth component); μeff is the effective viscosity; σz represents the Prandtl–Schmidt number, which is given the value of 1.3 (Launder & Spalding, 1974); and G is the generation rate of turbulence. In the numerical simulation process, the critical parameters, such as e, k, u, meff and G, are variables and will be updated according the inherent turbulent and combustion evolution in FLACS. On the other hand, the turbulence length scale lLT is specified as the length from the ignition point to the edge boundary of the congestion. It is kept constant throughout the entire explosion simulation. Specifically, for a configuration with a large separation distance, such as the case in Fig. 5.9, the turbulence length lLT is equal to 3DD, which is the distance from the ignition point to the donor boundary (0.5DD) plus the separation distance (1.5DD) and the dimension of acceptor (DD). While this specification is appropriate for a continuously congested region, in the large separation distance simulation case, the turbulence length scale may not be appropriately measured from the initial ignition point in the donor module, as the flow would have gone through a relaminarization process over the length of the large safety gap. Probably, a more appropriate datum for the calculation of the turbulence intensity in the acceptor module is the distance from the upstream end of the acceptor boundary where the flame begins to accelerate again. The exclusive turbulence length scale value of 3DD in the default overpressure calculation results in smaller dissipation of turbulent kinetic energy for cases with large separation gaps, and thus leads to the significant overprediction of the overpressure at the boundary of the acceptor. To amend this improficiency, a technique is developed, whereby two different turbulence length scales are used: one for the donor modules is specified to start at the ignition point and the other for the acceptor module is to start at the most upstream location of that module. This can be implemented by data-dumping the results immediately before the flame reaches the acceptor module. It is termed as the data-dump technique by extracting and reloading the data before the flame enters the acceptor module.

5.3.3 Data-Dump Technique The data-dump technique will be introduced herein through a simulation example using the same configuration AE08 to demonstrate its efficiency.

122

5 CFD-Based Overpressure Prediction for Double Modules—Data-Dump …

As best practice for FLACS simulations is recommended to apply minimum of 13 grid cells across the partially confined gas cloud FLACS (GexCon, 2011). The simulation grids are supposed to be meshed in the range of 0.03–0.06 m uniformly in all directions. A mesh sensitivity study using a series of different grid sizes has been conducted in the present study. Each grid size is tested by the comparison of the numerical overpressures with the experimental results. The satisfactory agreement as seen in Fig. 5.8 indicates the sufficient accuracy by 0.03 m grid. Therefore, 0.03-m grid is applied throughout the simulations in this section. By creating a cc-file in FLACS, the explosion results of scenario AE08 are dumped at time 0.026 s when the flame exits the edge of the first congested region (the donor) as seen in Fig. 5.11. A new explosion file (AE08’) with the same set-up as AE08 is created. The explosion AE08’ starts from 0.026 s by loading the previous data dumped from AE08. And in order to restart the flame acceleration process in the acceptor module with a new turbulence length scale datum at the upstream edge of the acceptor module, the ignition is relocated to the edge of the second congested region (the acceptor) as seen in Fig. 5.11b. By the data-dump command, it is noted in the comparison in Fig. 5.12 that the overpressures obtained in the second congested region are significantly reduced from 700 kPa to 85 kPa, which is very close to the experimental result of 60 kPa in Fig. 5.10b. The proposed data-dump technique proves to work for this case from the RIGOS experiments with an updated turbulent length scale. In order to assess the overall performance of the proposed data-dump technique, a total of five sets of experimental tests in RIGOS research program (VandenBerg & Versloot, 2003) are numerically simulated by using FLACS with the data-dump technique. As shown in Table 5.2, the separation distances vary from 0.5DD to 1.5DD and both ethylene and methane as fuel are tested. For all cases, the grid cell size in FLACS is uniformly kept as 0.03 × 0.03 × 0.03 m, and the data-dump procedure as demonstrated above is applied to all the simulations. For the explosion cases with wide separation gap from Table 5.2, the peak gas explosion overpressures at both the donor and acceptor modules are illustrated in Fig. 5.13 and Fig. 5.14, respectively, with the overpressures calculated using FLACS with and/or without the data-dump technique. From Figs. 5.13 and 5.14, the overpressures in the donor module from the FLACS simulations are the same regardless of application of the data-dump technique. All the donor explosion overpressures calculated by FLACS are in good agreement with results from the experimental tests. For the acceptor module, the two sets of data with or without the application of the data-dump technique are presented. The results in the acceptor module for FLACS simulations without the application of the data-dump technique are seen to be significantly higher than the experimental results. It can be said the software predicts a detonation in the acceptor module as the flame propagates into the acceptor module, which is in conflict with the observation from the experiments. On the other hand, when the data-dump technique is applied, the magnitudes of the overpressures from the simulations for the wide gap configurations are found to be in good agreement with data from the RIGOS experiments.

5.3 Results and Discussions

123

(a) Case AE08 before the data-dump command with ignition at the centre of donor

(b) Case AE08’ after the Data-dump command with relocation of ignition at the edge of acceptor Fig. 5.11 Flame propagation during explosion with data-dump technique (permission from Elsevier)

Therefore, the overall results in the assessment of FLACS with or without the application of the data-dump technique manifest that the calculation accuracy of FLACS is verified for the congested regions without separation distance, such as the single module described in the last chapter. On the other hand, for the large separation gaps in double modules, overprediction of overpressures by FLACS can be amended by using the data-dump technique. The present data-dump technique is proposed to circumvent the improficiency in the FLACS simulation for the large safe gap between modules without attempt for the modification of the software.

124

5 CFD-Based Overpressure Prediction for Double Modules—Data-Dump … 800

P1 P2 P3 P4 P5 P6 P7 P8 P9

700 600

Pressure (kPa)

500 400 300 200 100 0 -100

0

20

40

60

80

Time (ms)

(a) Overpressures before data-dump 110 90

Pressure (kPa)

70

P1

P2

P3

P4

P5

P6

P7

P8

P9

50 30 10 -10 -30 25

35

45

55

65

75

85

Time (ms) (b) Overpressures after data-dump Fig. 5.12 Comparison of overpressures before and after data-dump command (permission from Elsevier)

5.4 Summary

125

Table 5.2 Configurations simulated using FLACS Case no.

Exp. no. * in RIGOS

Fuel type

VBR (%)

Separation distance (m)

Cylinder diameter (m)

Dimension of the donor (m)

1

AE07

Ethylene

10.1

0.70(0.5DD*) 0.019

1.4 × 1.4 × 0.7

2

AM04

Methane

10.1

0.70(0.5DD)

0.019

1.4 × 1.4 × 0.7

3

BE05

Ethylene

4.6

1.33(1.0DD)

0.019

1.33 × 1.33 × 0.7

4

AM01

Methane

10.1

2.11(1.5DD)

0.019

1.4 × 1.4 × 0.7

5

AE08

Ethylene

10.1

2.11(1.5DD)

0.019

1.4 × 1.4 × 0.7

*AE: type A donor with the volume blockage ratio of 10.1% and ethylene as fuel *AM: type A donor with the volume blockage ratio of 10.1% and methane as fuel *BE: type B donor with the volume blockage ratio of 4.6% and ethylene as fuel *DD: maximum length of the donor dimensions

Explosion overpressures (kPa)

90 80 70 60

Overpressures at the donor in the experiments Overpressures at the donor in FLACS

50 40 30 20 10 0

1

2

3

4

5

Case Number Fig. 5.13 Peak donor explosion overpressure (permission from Elsevier)

5.4 Summary This chapter reports a comparison of simulation results with the corresponding published experimental data from TNO Prins Maurits Laboratory for cases that involved safety gaps of various sizes in the double modules. The CFD software FLACS has been comprehensively verified to give generally effective results in agreement with

126

5 CFD-Based Overpressure Prediction for Double Modules—Data-Dump …

Numerical explosion overpressures (kPa)

1000

Case 4 Case 3

Case 5

100

Case 2 Case 1

10

1

1

Overpressures at the acceptor in the experiments Overpressures at the acceptor before DUMP in FLACS Overpressures at the acceptor after DUMP in FLACS

10

100

1000

Experimental explosion overpssures (kPa) Fig. 5.14 Peak acceptor explosion overpressure (permission from Elsevier)

experimental data for single module or double modules with short safe gap shorter than the width of module. On the other hand, when the separation distance is very large between the two modules, numerically overpressures in the acceptor module are overpredicted. Large discrepancy in the overpressure between the numerical and experimental results is observed. As the flame propagates into the acceptor, a detonation in the acceptor is predicted by FLACS while only the deflagrations are detected in the experiments. It is probably because the inappropriate application of turbulence length scale across the separation gap leads to exaggerated flame speeds and, thus, the unrealistic high overpressures. A technique termed data dump is tested for these double modules with different separation gaps. During the data dump, the simulation is interrupted purposely before the flame propagates into the acceptor module. It is restarted with the ignition point reset to the upstream end of the acceptor module to modulate the turbulence length scale. Using the data-dump technique on five sets of explosion scenarios with two types of fuels and three different separation distances, the derived data are then compared with experimentally observed explosion overpressures. The comparison indicates that FLACS gives remarkably good agreement with experimental data even for largescale safety gaps when this technique is used. It can be stated this technique effectively improves the accuracy of FLACS to evaluate the gas explosion overpressures in the region with large separation gaps. In the next chapter, the gas explosion simulation and overpressure calculation will be carried out for the multi-modules with different sets of safety gap from both realistic offshore structures and artificial configurations. A cylindrical FLNG will also be numerically modelled based on a traditional ship-shaped FLNG with application of the proposed data dump technique.

References

127

References GexCon. (2011). FLACS v9.1 user’s manual. Norway: Doxygen. Hjertager, B. H. (1984). Computer-simulation of turbulent reactive gas-dynamics. Modeling Identification and Control, 5(4), 211–236. Launder, B. E., & Spalding, D. B. (1974). The numerical computation of the turbulent flows. Computer Methods in Applied Mechanics and Engineering, 3, 269–289. Ma, G., Li, J., & Abdel-jawad, M. (2014). Accuracy improvement in evaluation of gas explosion overpressures in congestions with safety gaps. Journal of Loss Prevention in the Process Industries, 32, 358–366. van den Berg, A. C., & Mos, A. L. (2002). RIGOS—The critical separation distance. TNO Prins Maurits Laboratory Report No. PML 2002-C50. https://doi.org/10.1016/s0950-4230(02)001122. van den Berg, A. C., & Versloot, N. H. A. (2003). The multi-energy critical separation distance. Journal of Loss Prevention in the Process Industries, 16(2), 111–120. https://doi.org/10.1016/ S0950-4230(02)00112-2. van den Berg, A. C., & Mos, A. L. (2005). Research to improve guidance on separation distance for the multi-energy method (RIGOS). TNO Prins Maurits Laboratory, Research Report 369.

Chapter 6

CFD-Based Overpressure Prediction for Congested Multi-Modules—Safety Gap Effect

Abstract This chapter presents a thorough investigation of the effect of the gas explosion on a cylindrical floating liquefied natural gas (FLNG) platform by using CFD simulation. The effect of safety gap on overpressure mitigation of gas explosion is evaluated. The data-dump technique is applied to ensure simulation accuracy. Two sets of different safety gaps are designed for the cylindrical FLNG platform. The overall results indicate that the safety gap is effective in reducing overpressure in two adjacent congestions.

6.1 Introduction This chapter evaluates the gas explosion overpressures for a cylindrical floating liquefied natural gas (FLNG) vessel. The conventional ship-shaped and innovative circular FLNG platforms are studied by detailed CFD-based analysis. The gas explosion safety on the cylindrical FLNG is investigated, particularly the safety gap effects on different configurations (Li et al. 2016). In the process industry, the safety gap, which is an open space with no congestion placed between the congested process areas, is one of the most effective and widely used safety-in-design measures. The principle behind the operation of the safety gap is that it interrupts the mutual-strengthening mechanism in congested areas. The mutual-strengthening mechanism consists of the generation and intensification of turbulence, enhancement of thermal and chemical mixing between combustion products and reactants, higher flame speeds and higher pressures. The absence of obstacles in a safety gap eliminates the fluid–obstacle interaction, thereby preventing the generation of turbulence. It is very effective in reducing pressures prior to the onset of detonation. Investigations of flame acceleration and overpressure in gas explosions, in most available studies, have mainly focussed on setups of multi-obstacle groupings with successively and regularly spaced obstacles (Alekseev, Kuznetsov, Yankin, & Dorofeev, 2001; Chan, Moen, & Lee, 1983; EMEG 1997; Kindracki, Kobiera, Rarata, & Wolanski, 2007; Molkov, Makarov, & Puttock, 2006; Wen et al. 2013). Limited experimental studies have been reported to examine the effect of the safety gap on gas explosions (Gubba, Ibrahim, Malalasekera, & Masri, 2008; © Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_6

129

130

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

Moen, Donato, Knystautas, & Lee, 1980; Rudy, Porowski, R., & Teodorczyk, 2011; Na’inna, Phylaktou, & Andrews, 2013; VandenBerg & Versloot 2003). In most experimental explosion programs, investigation on the safety gap is conducted in highly confined chambers. Tubes are arranged such that cylindrical flames propagate in one direction only. Exceptions are the two separate configurations of obstacles, which are tested to develop guidelines for critical safety gap by VandenBerg and Mos (2002), VandenBerg and Versloot (2003). The experimental configuration consists of orthogonally arranged obstacles enclosed in plastic sheeting. Therefore, the flames propagate three-dimensionally in the tests. The configurations from VandenBerg and Versloot (2003) with hemispherical flame propagation and multiple separation distances are modelled in this study. The separation distance is defined as the space between the boundaries of the congested regions. Specifically, the separation distance starts from the downstream end of the first module, where ignition is initiated, to the upstream end of the second module towards which the flame propagates after passing through the separation space. This chapter considers a series of different safety gap configurations to derive the overpressure mitigation effect. The data-dump technique proposed in previous chapter is applied.

6.2 Numerical Models The ship-shaped and cylindrical FLNG cases are modelled using the preprocessor CASD of FLACS (GexCon, 2011). All main structural components are converted into boxes and cylinders as congestion elements in FLACS. The reason to choose boxes and cylinders as the simulation elements is the complexity of the geometries, which are realistic large-scale modules. The porosity calculation in CFD simulation for boxes and cylinders is time efficient, whereas the CPU time for other element types, such as ellipsoids, general truncated cones and complex polyhedron, is much longer for such complicated geometries. In specifying congestion, walls and decks are assumed to be rigid during the CFD simulations. In terms of the grid models in the dispersion simulations, the quadrate grids are adopted. Refined cubical grid cells in the vicinity of the leak are applied, while the grid beyond the main area of interest is stretched towards the boundaries. The aspect ratio of the grid is controlled as not below 20%. There are total 108,900 and 72,576 cell volumes for the ship-shaped and cylindrical FLNG platforms. The grid size sensitivity test is also conducted based on the previous work from Ma, Li, and Abdel-Jawad (2014), where the CFD simulation results were calibrated with the experiments. However, the geometries in current study are more than 100 times larger than their laboratory modules. Since the grid size of 0.03 m was adopted in the simulation by Ma et al. (2014), the current grid size is therefore increased in the range of 0.3–3 m. A series of different grid sizes is applied to assess the convergence. The grid size of 1 m near the leakage is eventually chosen for all the following gas dispersion simulations.

6.2 Numerical Models

131

Fig. 6.1 Ship-shaped FLNG 3D geometry in FLACS (permission from Elsevier)

The cylindrical FLNG platform is modelled based on the prototype of the shipshaped FLNG vessel as shown in Fig. 6.1. Volume block ratio (VBR) within a given zone in the model is defined as the ratio of the total volume of the geometrical objects in that zone to the total volume of the zone. Homogeneous cubical grid cells are applied as the grid models in the explosion simulations.

6.2.1 Ship-Shaped FLNG In recent years, more oil and gas projects are constructed in remote fields without convenient access and under severe weather conditions, such as those in northern area of Darwin, Australia, which often experiences cyclonic conditions. Floating LNG vessels are relatively new concepts that are increasingly built as cost-effective alternatives to fixed platforms. The main advantage of the ship-shaped vessel is the mobility to be towed out and anchored at arbitrary locations in the ocean (Shimamura, 2002; Suardin, McPhate, Sipkema, Childs, & Mannan, 2009). The PRICO (Pwaga, 2011; Xu, Liu, Jiang, & Cao, 2013; Xu, Liu, & Cao, 2014) FLNG units are explored numerically in this study. A realistic ship-shaped FLNG model including liquefaction trains, dehydration and mercury removal modules and compressors, etc. is built as shown in Fig. 6.1.

6.2.2 Cylindrical FLNG Most of the available ship-shaped vessels can drift in one direction only in the water and cannot provide a stable platform due to hull deflection (sagging/hogging)

132

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

caused by the large aspect ratio. Thus, under the extreme weather conditions, such as typhoons, these vessels have to suspend operation. Suspension of the operations for more than 5% of the year can convert a profitable operation into an unprofitable one. To improve the stability, cylindrical FLNGs are considered as promising substitute. The cylindrical FLNGs, with symmetrically designed platforms, have eliminated the wave-induced fatigue loads and the large bending moments experienced in rough seas for ship-shaped vessels. The improvement in hydrodynamic stability allows for less suspension of the operations. Thus, the cylindrical platform is usually preferable especially in areas where cyclonic conditions are a concern. Very little studies have been reported on investigation of the cylindrical FLNGs and other cylindrical vessels. Instead, researches so far have been focused on construction, operation studies, hydraulic and the hydrodynamic analysis (Hirdaris et al., 2014; Kvamsdal et al., 2010; Wang et al., 2013; Zhao et al., 2011). No topsides safety evaluation for cylindrical platforms subject to gas dispersion and explosion has been published in the open literature. On-duty cylindrical-hulled FPSOs are mainly designed for cruel oil drilling and production, whereas the design and construction of FLNG with circular hull are still conceptual or under development. In this chapter, a cylindrical FLNG model will be investigated and compared with traditional vessel FLNG. The cylindrical FLNG platform, which is composed of six liquefaction unit modules and six other natural gas processing modules with detailed equipment and piping layout, is analysed in this study. A series of different safety gaps are simulated among the obstructed configurations to derive the optimal safety gap in view of vapour cloud explosion overpressure controlling. As shown in Fig. 6.2, the three-dimensional cylindrical FLNG platform is modelled based on the model for the ship-shaped FLNG. The order of the 12 modules on the cylindrical FLNG platform is kept the same as that in the ship-shaped FLNG. The modules are organized in a U shape to fit into the cylindrical hull. Consequently, the cylindrical FLNG platform has a more compact area with 12 modules in the same process order while the turret area is eliminated. One of the liquefaction trains Module 7 is zoomed-in for a close view in Fig. 6.3. Figure 6.4 shows the topside modules on the platform including 1. power generation (Module 1), 2. three Trent gas turbines and two essential diesel generators (Module 2), 3. nitrogen package, hot oil, mono-ethylene-glycol (MEG) processing and inlet facilities (Module 3), 4. boil off gas compressor and fuel gas system (Module 4), 5. acid gas removal unit and end flash gas compressor (Module 5), 6. dehydration and mercury removal (Module 6), 7. liquefaction modules (Module 7 to Module 12).

6.2 Numerical Models

Fig. 6.2 FLACS model for cylindrical FLNG platform (permission from Elsevier)

Fig. 6.3 Zoomed-in view of liquefaction train (permission from Elsevier)

133

134

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

Fig. 6.4 Topside layout of modules on cylindrical FLNG platform (permission from Elsevier)

6.3 Safety Gap Effect on Gas Explosion A series of gas explosion scenarios are then simulated on the cylindrical FLNG platform. The effects of different safety gaps on the platform are investigated in both near-field and far-field explosion. The data-dump technique is applied to improve the calculation accuracy of FLACS.

6.3.1 Near-Field Gas Explosion Simulation of Cylindrical FLNG Platform On the cylindrical FLNG platform, a near-field gas explosion is defined as the scenario where flames propagate through two adjacent congestions with one safety gap. Two different safety gaps of 12.5 and 20 m are modelled as shown in Fig. 6.5. Figure 6.6 shows the major equipment in the liquefaction module including the turbine air intake, turbine bundle removal equipment, scrubber and cold box on the lower level and two heat exchangers on the top level. The detectors are placed on the lower level to monitor overpressures. The gas composition is 27 vol% methane, 33 vol% ethane, 15 vol% propane, 19 vol% pentane and 6 vol% nitrogen.

6.3 Safety Gap Effect on Gas Explosion

135

Fig. 6.5 Two configurations with different safety gaps in near field (permission from Elsevier)

(a) 3D view of liquefaction lower level

(b) Monitor regions

Fig. 6.6 Lower level of liquefaction train (permission from Elsevier)

6.3.1.1

Application of Data-Dump in Gas Explosion Simulations of Cylindrical FLNG Platform

For the comparison of the effect of different sets of the safety gaps, the data-dump technique (Ma et al., 2014) is applied in this study. The data-dump technique is an effective tool to reset the turbulence length scale for gas explosion scenario with a safety gap, thereby increases the calculation accuracy of FLACS. In this study, the 25%, 50% and 100% filled gas cloud cases are investigated. The overpressures before and after application of the data-dump are extracted under or on the surfaces of the turbine air intake, turbine bundle removal equipment, scrubber and cold box as shown in Table 6.1. For each gas explosion simulation, the explosion results at the time when the flame exits the edge of the donor are dumped. Herein, donor refers to the module where the gas cloud is ignited. By creating a cc-file and executing a duplication command,

136

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

Table 6.1 Overpressures before and after application of data-dump technique Gas cloud coverage

Monitor region

20 m safety gap Pressure before data dump (bar)

Pressure after data dump (bar)

Pressure before data dump (bar)

Pressure after data dump (bar)

100% filled

Turb air intake

7

6.4

10.5

10

Turb bun removal

3.6

2.6

4.1

4

Cold box

2.8

2.5

3.2

3.1

Scrubber

6.2

5

11.5

10.5

Turb air intake

6.5

6.4

9.2

9.1

Turb bun removal

3.1

2.9

3.8

3.7

Cold box

2.5

2

2.8

2.6

50% filled

25% filled

12.5 m safety gap

Scrubber

5.8

4.5

10.5

9.2

Turb air intake

3.4

2.6

4.3

3.8

Turb bun removal

2.1

2

2.3

2.2

Cold box

0.29

0.28

0.9

0.8

Scrubber

1.65

1.1

5

3.9

a new explosion file for the receiving module is created along with the data of the overpressure resulted from the donor. In order to reset the turbulence length scale and restart the flame acceleration in the acceptor module, the ignition is relocated at the upstream edge of the acceptor module that is opposite to the donor. The overpressure percentage change due to data-dump technique is depicted in Figs. 6.7 and 6.8. Generally, the application of the data-dump technique decreases the overpressures in the acceptor module. Specifically, for the 12.5 m safety gap scenario, the overpressure shift percentages at different locations with application of the data-dump technique tend to be below 12% except for the overpressures observed in the scrubber area. The overpressures in the scrubber area are modulated to larger extents. For the 20 m safety gap case, the percentage change for each case, as in Fig. 6.8, is greater than the respective counterpart in Fig. 6.7. In other words, the greater the safety gap, the more overprediction of overpressure by FLACS, which could be amended by application of the data-dump technique. Therefore, in order to assure the overpressure calculation accuracy in safety gap modelling, the data-dump technique is applied for all the following simulations in this study.

6.3 Safety Gap Effect on Gas Explosion

137

Overpressure reduction percentage

100% 25% filled gas cloud 50% filled gas cloud 100% filled gas cloud 22.00% 12.38% 10%

11.63%

11.11% 4.35%

4.76%

2.44%

7.14%

8.70%

3.13%

2.63%

1%

1.09% Turbine air intake

Turbine bundle removal

Cold box

Scrubber

Fig. 6.7 Overpressure reduction percentages with application of data-dump technique for 12.5 m safety gap scenario with different gas cloud coverage (permission from Elsevier)

Overpressure reduction percentage

100% 25% filled gas cloud 50% filled gas cloud 100% filled gas cloud 23.53% 10%

33.33% 27.78% 20.00%

22.41% 19.35%

10.71%

8.57% 6.45% 4.76%

3.45% 1.54% 1%

Turbine air intake

Turbine bundle removal

Cold box

Scrubber

Fig. 6.8 Overpressure reduction percentages with application of data-dump technique for 20 m safety gap scenario with different gas cloud coverage (permission from Elsevier)

6.3.1.2

Different Safety Gaps Subject to Gas Explosion Under Different Gas Cloud Coverage

With the application of data-dump technique in the gas explosion overpressure calculation in FLACS, the safety gap effect on overpressure mitigation is then investigated in the adjacent (near field) modules-liquefaction trains. Two sets of safety gap configurations, i.e. 12.5 and 20 m, are modelled. Figure 6.9 gives an overview of the 100% gas-filled simulation case with the maximum overpressure up to 9 bar. The

138

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

results indicate that nearly identical pressures are observed in the donor modules, where ignition occurs on the left-hand side of the safety gap, for both the 20 m gap and 12.5 m gap configurations. On the other hand, pressures are lower in the acceptor module in the 20 m gap case than that in the 12.5 m gap case. The coverage percentage in modelling the 25%, 50% and 100% gas cloud filled cases is controlled by adjusting the volume height of the gas cloud. Figures 6.10, 6.11 and 6.12 demonstrate three similar scenarios with different safety gaps, where a reduction in the overpressure is observed in the acceptor modules when the safety gap increases from 12.5 to 20 m. Maximum overpressures for 50% and 25% filled cloud configurations with 12.5 m and 20 m safety gaps are given in Figs. 6.11 and 6.12, respectively. It is interesting to note that overpressures near the turbine air intake and scrubber are higher than those in the cold box and turbine bundle removal areas regardless of the safety gap size for all the simulation cases. The reason could be that the flame propagation distances from the ignition in the donor to the targets of the turbine air intake and scrubber are longer than the flame propagation distances to the cold box and turbine bundle removal areas as shown in Fig. 6.6b. The turbine air intake and scrubber are placed on the right-hand side, which is further away from the ignition coming from left than the cold box and turbine bundle removal areas are. The longer the flame path length is, the longer time the flame turbulence develops within the congestion, which induces greater overpressures (Li, Abdel-jawad, & Ma, 2014; Li, Ma, Abdel-jawad, & Hao, 2014). The fact that the turbine air intake and scrubber region are more congested with small dimension objects could be another add-on factor. Therefore, the smaller average diameter of the obstacles and greater congestion ratio contribute more to the turbulence induced flame acceleration, which again boosts overpressures increase (Bradley et al., 2008; Van den Berg & Mos, 2005). For the safety gap effect, it is seen that the overpressure difference between the 12.5 m and 20 m safety gap cases is most prominent in the turbine air intake and scrubber areas, which implies that the safety gap reduces more overpressures where the flame path is longer; smaller average obstacle diameter ensues greater congestion ratio. In addition to the above analysis of the safety gap effect on overpressure mitigation in near field, the investigation into the safety gap effect in far field is conducted in the following context.

6.3.2 Far-Field Gas Explosion Simulation of Cylindrical FLNG Platform To investigate the explosion consequence for modules far away from the ignition module on the cylindrical platform, explosion simulations are performed using varying parameters, such as different gas cloud location, size and ignition locations. The far-field gas explosion scenario is defined as the region where flame propagates

6.3 Safety Gap Effect on Gas Explosion

139

(a) Simulation case of 12.5 m safety gap

(b) Simulation case of 20 m safety gap Fig. 6.9 Maximum overpressures for 100% filled gas cloud spanning the modules with different safety gaps (permission from Elsevier)

140

6 CFD-Based Overpressure Prediction for Congested Multi-Modules … 12 12.5m gap configuration with 100% filled gas cloud

Overpressure (bag)

10

20m gap configuration with 100% filled gas cloud

8 6 4 2 0

Turbine air intake Turbine bundle removal

Cold box

Scrubber

Fig. 6.10 Maximum overpressures for 100% filled cloud configurations with 12.5 m and 20 m safety gaps (permission from Elsevier)

Overpressure (bag)

10 9

12.5m gap configuration with 50% filled gas cloud

8

20m gap configuration with 50% filled gas cloud

7 6 5 4 3 2 1 0

Turbine air intake

Turbine bundle removal

Cold box

Scrubber

Fig. 6.11 Maximum overpressures for 50% filled cloud configurations with 12.5 m and 20 m safety gaps (permission from Elsevier)

through more than one safety gap. Four different gas clouds with the same size of 140 × 140 × 10 m3 are placed at different locations to spread all over the modules. For each gas cloud, six ignition scenarios at the respective ground centre of each module are simulated as shown in Fig. 6.13. Thus, for each cylindrical FLNG platform, totally 24 simulations are carried out in the far-field gas explosion investigation. Two different cylindrical FLNG platforms, as shown in Fig. 6.14, are modelled in order to compare the effect of different safety gaps on gas explosion mitigation. One configuration is the platform with all modules moved 10 m inwards within the pipe rack circle to form the safety gap of 10 m in north–south direction as shown in

6.3 Safety Gap Effect on Gas Explosion 4.5 4

141

12.5m gap configuration with 25% filled gas cloud 20m gap configuration with 25% filled gas cloud

Overpressure (bag)

3.5 3 2.5 2 1.5 1 0.5 0

Turbine air intake Turbine bundle removal

Cold box

Scrubber

Fig. 6.12 Maximum overpressures for 25% filled cloud configurations with 12.5 m and 20 m safety gaps (permission from Elsevier) Fig. 6.13 Overview of explosion scenarios ignited in each module (permission from Elsevier)

Fig. 6.14b, while the other configuration has no gap between the modules and pipe rack as shown in Fig. 6.14a. Figure 6.14 shows an explosion scenario with the gas cloud is ignited in Module 10, and the flame propagates through all the surrounding modules to reach far-field ones, such as Module 3 and Module 4. The gas explosion path running through two gaps in north–south direction is defined as Path 1, i.e. the space from Module 10 to Module 3 or from Module 9 to Module 4, whereas Path 2 is identified as the flame path after three gaps. For each module, about ten monitor points are uniformly deployed on the ground level to detect the overpressures. The recorded overpressures, as tabulated in Table 6.2, are averaged from the modules in the far end of the flame

142

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

(a) Modules in cylindrical FLNG without safety gap

(b) Modules in cylindrical FLNG with 10 m safety gap Fig. 6.14 Layout of cylindrical FLNG for gas cloud ignited in Module 10 (permission from Elsevier)

path. Table 6.2 and Fig. 6.15 show the comparison of results for the cylindrical FLNG platform with two different safety gap setups. From Fig. 6.15, for simulation group cases 1–8, the gas explosion in the configuration without safety gap produces greater overpressures in the far field than the respective counterparts in the 10 m safety gap configuration. Opposite is observed for the simulation group cases 9–16, where the overpressures in the 10 m safety gap configuration are comparatively higher. It may be explained that the gaps between three adjacent modules, e.g. Module 3, Module 9 and Module 10, in north–south direction have different distance, which results in different flame turbulence interruption effects. For instance, as shown in Fig. 6.14a, the flames propagating from ignition point in Module 10 towards to Module 3 passes through a small safety gap first and then a longer safety gap between Module 9 and Module 3. If the flames start oppositely from ignition in Module 3 towards to Module 10, opposite turbulence path order ensues different turbulence acceleration, and thus different overpressures.

6.3 Safety Gap Effect on Gas Explosion

143

Table 6.2 Results of far-field gas explosion simulations Case no.

Ignition location

Gas cloud no.

Overpressure without safety gap

Overpressure with 10 m safety gap

Flame propagation path 1 1

Module 1

4

4.11

3.84

2

Module 3

4

3.96

3.72

3

Module 3

3

5.00

4.71

4

Module 5

3

5.77

5.48

5

Module 7

2

6.55

6.47

6

Module 9

2

8.04

7.68

7

Module 9

1

5.17

4.79

8

Module 11

1

3.90

3.68

9

Module 12

4

1.47

1.89

10

Module 10

4

1.47

2.04

11

Module 10

3

3.52

4.48

12

Module 8

3

3.80

5.07

13

Module 6

2

3.51

4.43

14

Module 4

2

3.14

4.01

15

Module 4

1

2.15

2.97

16

Module 2

1

1.98

2.66

Flame propagation path 2 17

Module 2

1

2.30

2.22

18

Module 4

1

2.47

2.03

19

Module 4

2

3.06

2.31

20

Module 6

2

1.72

1.67

21

Module 8

3

2.04

1.78

22

Module 10

3

2.37

1.97

23

Module 10

4

1.80

1.62

24

Module 12

4

2.18

1.76

Unlike the flame path 1 simulation, the flame path 2 simulation experiences the similar safety gap order regardless of the flame propagation direction in north–south due to the geometrical symmetry of the cylindrical layout. Therefore, from case 17 to case 24, similar tendency is observed, namely the far-field overpressures for the no safety gap configuration outstrip the respective counterpart in the 10 m safety gap setting. In simple summary, the arrangement order and the distance of the safety gaps between the modules play critical roles. In order to investigate its effect systematically and quantitatively, the following two sets of artificial configurations with regularly adjusted safety gaps are modelled.

144

6 CFD-Based Overpressure Prediction for Congested Multi-Modules … 9.00 8.00

Overpressures without safety gap Overpressures after 10m safety gap

Overpressure (Bar)

7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Case number Fig. 6.15 Overpressures at far end of flame (permission from Elsevier)

Fig. 6.16 Congested configurations with two safety gaps (permission from Elsevier)

6.3.2.1

Safety Gaps Between Three Congested Regions

The artificial configurations with three congestions are firstly modelled to investigate the corresponding explosion scenarios in Sect. 6.3.2 where the flame path 1 runs through two safety gaps as shown in Fig. 6.16. The FLACS models are reproducing tests from the Research to Improve Guidance on Separation Distance for the MultiEnergy Method (RIGOS) research programme (Van den Berg & Versloot, 2003). The overpressure calculations for these artificial models had been validated in the previous work (Ma et al., 2014). In FLACS, the three modules in Fig. 6.16 are modelled with the same obstacle diameter of 19.1 mm and volume blockage ratios of 10.1%. All obstacles are orientated orthogonally and regularly. The ignition location in this study is the centre of the donor on the left-hand side of the configuration.

6.3 Safety Gap Effect on Gas Explosion

145

Fig. 6.17 Overpressures in different configurations with two safety gaps (bar) (permission from Elsevier)

Eight simulation scenarios with eight different safety gaps are specified. For all the simulations, the overall distance from the donor to acceptor 2 is fixed, while acceptor 1 is located at varying locations in between the donor and acceptor 2. Herein, a safety gap distance ratio is defined as the ratio of the safety gap 1 distance to that of safety gap 2. The safety gap distance ratios are specified to increase from 0.1 in case 1 to 4 in case 8 as shown in Fig. 6.17. Overpressure monitors are positioned at regular distances along the axis of the donor–acceptor configurations. Figure 6.17 gives the overpressures in different configurations with two safety gaps. From Fig. 6.17, the overpressures in the farthest module—acceptor 2 increases with shorter distance from acceptor 1 to acceptor 2. The maximum overpressures are observed in simulation case 8, which has the greatest safety gap distance ratio. The reason could be that the congestion volume in case 8 is increased once acceptor 1 connects to acceptor 2. Larger congestion volume leads to longer flame turbulence path. From the analysis of the simulation results, if the target of protection is acceptor 2 in the farthest end, a potential solution in such case is to maximize the distance of safety gap 2, which can substantially decelerate the flame turbulence in the open space, such as the example case 1 as shown in Fig. 6.17. For each acceptor module, the overpressures at the monitor points are averaged and presented in Fig. 6.18. Unlike the overpressure increase in acceptor 2, it is interesting to note that the averaged overpressure in acceptor 1 decreases constantly

146

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

Fig. 6.18 Overpressures in acceptor modules subject to two safety gaps (permission from Elsevier)

0.12

Overpressure (bar)

0.1 0.08 0.06 0.04 0.02

Overpressue in Acceptor 1 Overpressue in Acceptor 2

0

0

1

2

3

4

Satefy gap distance ratio

from case 1 to 7. It is probably because the increasing safety gap 1 between the donor and acceptor 1 has stronger interruption effect on the flame turbulence. According to the overpressure safeguarding targets, different strategies should be adopted in explosion mitigation. In order to minimize the overpressure in acceptor 2, the safety gap 2 should have the greatest distance to discharge the flame turbulence and overpressure. If the safeguarding target is acceptor 1, sufficient distance of safety gap 1 should be assigned. To balance the overpressures in acceptor 1 and acceptor 2, the optimal solution is probably to set the safety gap distance ratio to be 1, namely safety gap 1 and safety gap 2 have the same distances and the intersection points as shown in Fig. 6.18.

6.3.2.2

Safety Gaps Between Four Congested Regions

The second set of artificial configurations with four congestions is modelled to investigate the corresponding explosion scenarios in Sect. 3.2, where the flame path 2 runs through three safety gaps as shown in Fig. 6.19. The obstacle diameter, arrangement and volume blockage ratio are the same as they are in Sect. 6.3.2.1. The distance from the donor to acceptor 2 is fixed, while the two congested regions in the middle move oppositely so that safety gap 1 is equal to safety gap 3. As shown in Fig. 6.20, eight simulation scenarios with eight different safety gap distance ratios are simulated. Similarly, the safety gap distance ratio is defined as the distance of safety gap 1 to the distance of safety gap 2. Figure 6.20 gives the simulation results for cases 1–8. The overpressures in the three safety gap configurations have obvious decreasing tendency in acceptor 2 compared to the respective overpressures in Fig. 6.17. Figure 6.21 gives the overpressures in acceptor modules subject to three safety gaps. From Fig. 6.21, the averaged overpressures in acceptor 1 significantly increase

6.3 Safety Gap Effect on Gas Explosion

147

Fig. 6.19 Four congested configurations with three safety gaps (permission from Elsevier)

Fig. 6.20 Overpressures in different configurations with three safety gaps (bar) (permission from Elsevier)

with the increasing safety gap distance ratio from case 1 to 7. It implies that although the safety gaps 1 and 3 effectively reduce the overpressures in the farthest field in acceptor 2 by adjusting the middle congestions acceptor 1, acceptor 1 is susceptible to greater explosion overpressures. In view of the protected target under such circumstances, the arrangement of safety gaps should be target orientated depending on the specific congestions and the gaps in between. For example, larger safety gap 2 provides overpressure mitigation for acceptor 1, such as in case 1, whereas greater distance of safety gap 1 and 3 significantly reduce the explosion overpressures in acceptor 2. The balanced overpressures in both acceptor modules occur at the intersection point in Fig. 6.21, where safety gap 2 is about 1.5 times of safety gap 1 or 3.

148

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

Overpressure (bar)

0.25

0.2

0.15

0.1

0.05

Overpressure in Acceptor 1 Overpressure in Acceptor 2

0

0

1

2

3

4

5

Safety gap distance ratio Fig. 6.21 Overpressures in acceptor modules subject to three safety gaps (permission from Elsevier)

In such scenario, the overpressures are alleviated in both acceptor 1 and 2 as shown in the example of case 5.

6.4 Summary The current simulations of the two sets of artificial configurations demonstrate the effect of different safety gap layout on the overpressure distribution in the cylindrical FLNG platform in Sect. 6.3.2. For the artificial configurations with two safety gaps, i.e. cases 1–8 in Table 6.2, the cylindrical FLNG platform without safety gap is the same as case 8 in Fig. 6.17, while the 10 m safety gap cylindrical FLNG platform is similar to case 7. In those scenarios, the gas cloud is ignited in the donor, i.e. Module 3 on the cylindrical FLNG platform. Acceptors 1 and 2 are Modules 9 and 10, respectively. It is shown in Fig. 6.17 that if acceptor 1 connects to acceptor 2 as in case 8, the overpressures in acceptor 2 become greater than that in case 7. In other words, longer safety gap distance with regard to acceptor 2 as in case 7 reduces overpressures in the acceptor modules, which conforms to the general observation that the far-field overpressures are smaller on the cylindrical FLNG platform with 10 m safety gap. On the other hand, for simulations of cases 9–16 in Table 6.2, the overpressure distribution is analogue to those from the corresponding scenarios 1 and 2 in Fig. 6.17. Comparing scenario 2 to scenario 1 in Fig. 6.17, the shorter distance between the donor and acceptor 1 in scenario 1 induces lower overpressures in the far end acceptor 2 due to the longer distance of safety gap 2. It is the equivalent scenario on the cylindrical FLNG platform where the ignition is relocated to Module 10 except

6.4 Summary

149

for that the far-field overpressures on the cylindrical FLNG platform with 10 m safety gap are greater. For the simulation cases from 17 to 24 in Table 6.2, the flame propagates from the edge module through three safety gaps to the farthest module, which is the equivalent artificial model in Sect. 6.3.2.2. The cylindrical FLNG with safety gap of 10 m mimics the simulation case 8 in Fig. 6.20 where the safety gaps between the donor and acceptor 2 effectively interrupt the flame turbulence, and thus, mitigate overpressures in the far field. In summary, 16 different artificial configurations are numerically simulated with different ignition locations and gas cloud on the cylindrical FLNG platform. It is concluded that the 10 m safety gaps on the cylindrical FLNG platform effectively mitigate overpressures in the far-field modules in most cases. On the other hand, exception is identified in some scenarios that the overpressures are increased in the far field due to the flame turbulence interaction with the adjacent modules. Therefore, balanced solution to reduce overpressures for a specific target or targets should be derived with target-orientated CFD simulations by parametric study regarding safety gap distance ratios. From Chaps. 4–6, the CFD-based overpressure predication method is applied to single, double and multiple modules, wherein Chap. 3 introduces both theory and application of empirical overpressure estimation methods. Based on these numerical and empirical VCE overpressure evaluation approaches, gas explosion accidents for artificial and realistic modules have been comprehensively explored. In the next chapter, the risk analysis theories and applications will be presented.

References Alekseev, V. I., Kuznetsov, M. S., Yankin, Y. G., & Dorofeev, S. B. (2001). Experimental study of flame acceleration and the deflagration-to-detonation transition under conditions of transverse venting. Journal of Loss Prevention in the Process Industries, 14(6), 591–596. https://doi.org/10. 1016/S0950-4230(01)00051-1. Bradley, D., Lawes, M., & Liu, K. X. (2008). Turbulent flame speeds in ducts and the deflagration/detonation transition. Combustion and Flame, 154(1–2), 96–108. https://doi.org/10.1016/j. combustflame.2008.03.011. Chan, C., Moen, I. O., & Lee, J. H. S. (1983). Influence of confinement on flame acceleration due to repeated obstacles. Combustion and Flame, 49(1–3), 27–39. https://doi.org/10.1016/00102180(83)90148-7. EMEG. (1997). Explosion Model Evaluation Group, specifications of test cases for gas explosions—Test case C1. Brussels, Belgium: EME project, DGXII. GexCon. (2011). FLACS v9.1 user’s manual. Norway: Doxygen. Gubba, S. R., Ibrahim, S. S., Malalasekera, W., & Masri, A. R. (2008). LES modeling of premixed deflagrating flames in a small-scale vented explosion chamber with a series of solid obstructions. Combustion Science and Technology, 180(10–11), 1936–1955. https://doi.org/10.1080/ 00102200802261852.

150

6 CFD-Based Overpressure Prediction for Congested Multi-Modules …

Hirdaris, S. E., Bai, W., Dessi, D., Ergin, A., Gu, X., Hermundstad, O. A., et al. (2014). Loads for use in the design of ships and offshore structures. Ocean Engineering, 78, 131–174. https://doi. org/10.1016/j.oceaneng.2013.09.012. Kindracki, J., Kobiera, A., Rarata, G., & Wolanski, P. (2007). Influence of ignition position and obstacles on explosion development in methane-air mixture in closed vessels. Journal of Loss Prevention in the Process Industries, 20(4–6), 551–561. https://doi.org/10.1016/j.jlp.2007.05. 010. Kvamsdal, H. M., Hetland, J., Haugen, G., Svendsen, H. F., Major, F., Karstad, V., et al. (2010). Maintaining a neutral water balance in a 450 MWe NGCC-CCS power system with post-combustion carbon dioxide capture aimed at offshore operation. International Journal of Greenhouse Gas Control, 4(4), 613–622. https://doi.org/10.1016/j.ijggc.2010.01.002. Li, J. D., Abdel-jawad, M., & Ma, G. W. (2014). New correlation for vapor cloud explosion overpressure calculation at congested configurations. Journal of Loss Prevention in the Process Industries, 31, 16–25. https://doi.org/10.1016/j.jlp.2014.05.013. Li, J. D., Ma, G. W., Abdel-jawad, M., & Hao, H. (2014). Evaluation of gas explosion overpressures at configurations with irregularly arranged obstacles. Journal of Performance of Constructed Facilities, B4014003. Li, J., Ma, G., Hao, H., & Huang, Y. (2016). Gas explosion analysis of safety gap effect on the innovating FLNG vessel with a cylindrical platform. Journal of Loss Prevention in the Process Industries, 44, 263–274. Ma, G. W., Li, J. D., & Abdel-Jawad, M. (2014). Accuracy improvement in evaluation of gas explosion overpressures in congestions with safety gaps. Journal of Loss Prevention in the Process Industries, 32, 358–366. https://doi.org/10.1016/j.jlp.2014.10.007. Moen, I. O., Donato, M., Knystautas, R., & Lee, J. H. (1980). Flame acceleration due to turbulence produced by obstacles. Combustion and Flame, 39(1), 21–32. https://doi.org/10.1016/ 0010-2180(80)90003-6. Molkov, V., Makarov, D., & Puttock, J. (2006). The nature and large eddy simulation of coherent deflagrations in a vented enclosure-atmosphere system. Journal of Loss Prevention in the Process Industries, 19(2–3), 121–129. https://doi.org/10.1016/j.jlp.2005.05.006. Na’inna, A. M., Phylaktou, H. N., & Andrews, G. E. (2013). The acceleration of flames in tube explosions with two obstacles as a function of the obstacle separation distance. Journal of Loss Prevention in the Process Industries, 26(6), 1597–1603. https://doi.org/10.1016/j.jlp.2013.08. 003. Pwaga, S. S. (2011). Sensitivity analysis of proposed LNG liquefaction processes for LNG FPSO (Master’s thesis, Institutt for energi-og prosessteknikk). Rudy, W., Porowski, R., & Teodorczyk, A. (2011). Propagation of hydrogen-air detonation in tube with obstacles. Journal of Power Technologies, 91, 122–129. https://doi.org/10.1016/00102180(80)90003-6. Shimamura, Y. (2002). FPSO/FSO: State of the art. Journal of Marine Science and Technology, 7(2), 59–70. https://doi.org/10.1007/s007730200013. Suardin, J. A., McPhate, A. J., Sipkema, A., Childs, M., & Mannan, M. S. (2009). Fire and explosion assessment on oil and gas floating production storage offloading (FPSO): An effective screening and comparison tool. Process Safety and Environmental Protection, 87(3), 147–160. https://doi. org/10.1016/j.psep.2008.12.002. van den Berg, A. C., & Mos, A. L. (2002). RIGOS—The critical separation distance. TNO Prins Maurits Laboratory Report No. PML 2002-C50. https://doi.org/10.1016/s0950-4230(02)001122. van den Berg, A. C., & Mos, A.L. (2005). Research to improve guidance on separation distance for the multi-energy method (RIGOS). TNO Prins Maurits Laboratory, Research Report 369. van den Berg, A. C., & Versloot, N. H. A. (2003). The multi-energy critical separation distance. Journal of Loss Prevention in the Process Industries, 16(2), 111–120. https://doi.org/10.1016/ S0950-4230(02)00112-2.

References

151

Wang, F., Zhang, Y. K., Wang, Y., Xu, Z. Y., Zhang, Z. Y., Ni, T., et al. (2013). Floating non-traditional manufacture of floating drilling storage and offloading units-study on modeling and optimization method for the underwater rotating technology. Marine Structures, 31, 15–23. https://doi.org/10. 1016/j.marstruc.2012.12.002. Wen, X. P., Yu, M. G., Liu, Z. C., Li, G., Ji, W. T., & Xie, M. Z. (2013). Effects of cross-wise obstacle position on methane-air deflagration characteristics. Journal of Loss Prevention in the Process Industries, 26(6), 1335–1340. https://doi.org/10.1016/j.jlp.2013.08.006. Xu, X., Liu, J., & Cao, L. (2014). Optimization and analysis of mixed refrigerant composition for the PRICO natural gas liquefaction process. Cryogenics, 59, 60–69. Xu, X., Liu, J., Jiang, C., & Cao, L. (2013). The correlation between mixed refrigerant composition and ambient conditions in the PRICO LNG process. Applied Energy, 102, 1127–1136. Zhao, W. H., Yang, J. M., Hu, Z. Q., & Wei, Y. F. (2011). Recent developments on the hydrodynamics of floating liquid natural gas (FLNG). Ocean Engineering, 38(14–15), 1555–1567. https://doi. org/10.1016/j.oceaneng.2011.07.018.

Chapter 7

Risk Analysis Methods for Gas Explosion

Abstract This chapter gives a broad literature review on the state-of-the-art explosion risk analysis methods including both qualitative and quantitative approaches, such as risk checklist, HAZOP, HAZIP, event tree, fault tree and Bayesian network. A 3 × 3 risk matrix is used to classify the risk level by considering both likelihood and consequence of an explosion event. For the quantitative methods, detailed calculation procedure of each approach is presented, and the strengths and weaknesses of each method are discussed.

7.1 Introduction Risk analysis usually adopts analytical techniques to assess the probability/frequency and severity of hazardous incidents. The techniques could be either qualitative or quantitative depending on the required proficiency level of examination. Risk analysis normally takes the following four steps: • • • •

identifying incident occurrences or scenarios; estimating the frequency of the occurrences, evaluating the consequences of each occurrence, developing risk estimates in view of the identified frequency and consequence.

The basic methodology for evaluation of explosion risks in the petroleum and related industries, for both existing facilities and new projects, typically includes the following steps. • Define the facility: the facility is generally specified with the information regarding inputs and outputs of the facility, production, manning levels, basic process control system, emergency shutdown arrangements, explosion protection philosophy, assumptions and hazardous material compositions. • Identify hazards: potential sources for incidents can be exhaustively explored by listing the processes and storage of combustible materials and the process of chemistry. • Develop incident events: analyse chains of events leading to explosions and further consequences. © Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_7

153

154

7 Risk Analysis Methods for Gas Explosion

• Estimate frequency: the probabilities or likelihoods of the explosion to occur is assessed based on the historical records of similar incidents. • Evaluate consequence: the severity of the explosion is estimated based on the nature of the accident and the historical records regarding the similar accident. • Assess impact: derive the potential severity of the explosion events in terms of genetic evaluation quantities, such as injuries and deaths, structural damage, environmental impact and social influences. • Sum risk: combine the severity and probability estimates for all the identified potential explosion events. • Assess the safety measures: evaluate the effects of protective systems in different integrities on mitigation of the explosion risks. • Review against risk acceptance criteria: compare the explosion risk against the specified criterion to identify the appropriate safety measures to achieve the required risk management levels. Both qualitative and quantitative analysis methods may be applied to evaluate the risk of a facility is potentially subject to. To be cost-effective, the proficiency level of reviews should be commensurate with the comparative risk consequence that the facility may ensue, which implies high-value critical facilities require more comprehensive reviews, whereas unmanned low-hazard facilities may only need a checklist review. Generally, major process plants and offshore facilities require more comprehensive evaluation than a checklist type of review regarding the explosion events because they represent considerable capital investment and are susceptible to a big number of severe hazards. Quantitative analysis methods should be applied to demonstrate that the risk of those facilities is managed within the regulated requirements as well as public, national, industry and corporate expectations.

7.2 Qualitative Risk Analysis Qualitative risk analysis is normally implemented by team studies based on the generic experiences of knowledgeable personnel and does not involve any mathematical estimation. Overall qualitative evaluations are essentially checklist reviews, in which questions or process parameters are used in prompt discussions of the process design and operations that would develop into an incident scenario. The most common qualitative methods are introduced and three of them are explained in details in the following sections. • Checklist or worksheet: a standard listing, which identifies common protection features required for typical facilities, is checked against facility design and operations. Risks are expressed, at this stage, purposely exclusive of safety systems or system features. An example of checklist will be introduced in Sect. 7.2.1.

7.2 Qualitative Risk Analysis

155

• Hazard and Operability Study (HAZOP): a qualitative investigative safety review method performs a systematic critical examination of the process and engineering operations of new or existing facilities, which will be explained in Sect. 7.2.2. • Hazard Identification Study (HAZID): it is a particular form of hazard identification generally applied to offshore installations (Spouge, 1999). It is a systematic review of the possible causes and consequences of hazardous events. Elaboration on this approach can be found in Sect. 7.2.3. • Preliminary Hazard Analysis (Vincoli, 2006): a qualitative investigative safety review technique contains a disciplined analysis of the event sequences that possibly ensue a potential hazard into an incident. The possible undesirable events are identified first and then analysed separately by using this method. For each event or hazard, improvements or preventive measures are formulated. The result from this method provides a basis for determining which categories of hazard are relatively critical and which analysis methods are most suitable. With the aid of a frequency and consequence diagram, the identified hazards can be ranked according to the risks. Measures can then be prioritized to prevent accidents. • Safety flow chart (Spouge, 1999): a general flow chart identifies events which may occur at a facility during an incident. The flow chart is used to derive possible consequences that the event may lead to and the protection measures to protect the facilities. The flow chart helps the personnel who are not familiar with industry risk and safety measures to understand the event because it portrays a step-by-step scenario which is easy to follow and interpret. • What-if analysis (CCPS, 2011): a safety review method uses “what-if” investigative questions, which are asked by an experienced team of the system or components under review, where there are concerns about potential undesired events. Recommendations for the mitigation of identified hazards are provided.

7.2.1 Hazard Checklists for Offshore Installations A hazard checklist is a systematic inquiry into the full range of safety issues (Spouge, 1999). The checklist is usually brought up by an individual or a group of practitioners who are experts at dealing with hazards in view of the design and operating practices. In order to be applied to general installations, the checklist is composed of generic terms. Customized forms for individual installation or operation could be specified by reducing irrelevant terms while highlighting particular process-oriented items. Table 7.1 shows an example of a simplified checklist for explosion events. The strengths and weaknesses of hazard checklist are listed in Table 7.2.

156

7 Risk Analysis Methods for Gas Explosion

Table 7.1 Hazard checklist for low-risk facilities • Is the detective system installed according to standards? • Is the ignition source isolated? • Does the emergency shutdown system work properly? • Are the routine checks conducted regularly? • Have the vent systems been adequately sized and properly located? • Are there sufficient training against explosion safety provided to all staffs? • Are there any changes of structures, materials or equipment that may require extra review regarding explosion safety? Table 7.2 Strengths and weaknesses of hazard checklist Strengths

• • • • •

The checklist is simple to conduct Existing experience and knowledge of previous problems can be used It is easy for operating personnel to understand Standard practice and design intentions are checked It fully explores the known hazard areas

Weaknesses

• New or unusual hazards may not be considered • The contents in the list may not be comprehensive • It does not generate data for a QRA

7.2.2 HAZOP The hazard and operability study is a systematic critical review of a process facility design (Spouge, 1999). The HAZOP should involve a team of people who have the experiences of the facility or knowledge of the design that is under review. A HAZOP leader will normally guide the sessions and the conclusions should be recorded for future actions to be taken. Guidance on HAZOP can be found in CIA (1992), CCPS (2011) and Kletz (1999). To conduct HAZOP, each sub-system of the facility is considered in turn by subjectively evaluating the consequences of deviations from the specified performance that the design is intended to put up. A specific set of guide words are applied to examine the deviations, which ensure complete coverage of all the potential problems while allowing sufficient flexibility for an imaginative approach. The potential hazards and operating problems can then be identified, and recommendations can be made to amend the problems or clarify the issues where the terms are uncertain. The implementation procedure of HAZOP involves a series of repeated steps as shown in Fig. 7.1. The team should repeat steps 3–6 until all guide words have been exhaustively explored and all meaningful deviations have been addressed. The whole process should then be repeated from step 1 for the next section of the facility. Table 7.3 gives the typical guide words for a vent system used to reduce gas explosion risks. The record of the HAZOP can be displayed in various forms, such as “complete” recording covering each deviation as it is examined, “by exception” including only

7.2 Qualitative Risk Analysis

157

Indentify a section

Establish the design intent

Decide a deviation

Record the discussion

Decide the action

Define the potential causes and consequences

Fig. 7.1 HAZOP procedure Table 7.3 HAZOP guide words for a venting system Guide words

Meaning

Example deviation

No

No part of the intended result is achieved

Not working

Less

Quantitative decrease

Flow speed lower than normal

More

Quantitative increase

Flow speed higher than normal

Part of

Qualitative decrease

Artificial vent not working

As well as

Qualitative increase

Blast relief

Reverse

Opposite of intent

Confining an area

Other than

Something completely different than intended

Leakage

Table 7.4 Example of HAZOP worksheet SYSTEM: Vent system Guide words

Deviation

Possible causes

Consequences

Recommendations

No

Not working

Ventilation route blocked

Gas leakage accumulated

Check if the vent route is obstructed

Part of

Artificial vent not working

Gas detection signal not received

No emergency gas exhaust

Check detection system

the significant deviations or those requiring actions. Table 7.4 shows an example of HAZOP worksheet.

158

7 Risk Analysis Methods for Gas Explosion

Table 7.5 Strengths and weaknesses of HAZOP Strengths

• It is widely used and well understood • The experience of the operating personnel is used • It is systematic, comprehensive, and exhaustive of all hazardous process deviations

Weaknesses

• Its proficiency depends largely on the experience of the leader and the knowledge of the team • It is prioritized for process hazards and modifications are required to accommodate other types of hazards • The linkage between HAZOP and QRA is loose • Documentation is either lengthy and difficult to audit

The strengths and weaknesses of HAZOP are listed in Table 7.5.

7.2.3 HAZIDs HAZID is systematic hazard identification of the potential causes and consequences of hazardous events and normally is applied to offshore installations (Spouge, 1999). Similar to HAZOP, HAZID requires a team, who are familiar with the installation and led by a specialist of the HAZID technique. The main differences of the HAZID from HAZOP are: • The discussion process is conducted systematically through the installation’s modules or operations rather than its individual sub-systems. • Guide words are defined in advance according to the safety objectives for the installation. • Hazard scenarios are derived by combining guide words for the generic hazard, the cause and the consequence. Table 7.6 gives examples of HAZID guide words related to gas explosion events. HAZID focuses on how events happen rather than the possibility of their occurrence and what their consequences would be. Table 7.7 shows an example of HAZID worksheet with an emergency evacuation system in face of fire or an explosion event.

Table 7.6 Typical HAZID guide words Generic hazard

Cause

Consequence

• • • • •

• • • • • • •

• • • • • •

Leakage Fire Explosion BLEVE Structure failure

Operation/maintenance Human error Hardware failure Control system failure Structural failure Blowout Earthquake, etc.

Gas and smoke ingress Projectiles Pollution Structural collapse Safety systems impaired Human loss

7.2 Qualitative Risk Analysis

159

Table 7.7 Example of HAZID worksheet Area: emergency evacuation systems Generic hazard

Event

Consequences

Mitigating factors

Recommendations

Fire

Fire blocks escape routes

Personnel trapped

Fire extinguisher

Make escape routes to be fire resistant

Explosion

Explosion damages the escape route

Personnel trapped

Additional exit

Structural strengthening

Table 7.8 Strengths and weaknesses of HAZID Strengths

• • • •

It can be applied to any type of installation, operation or process Experience of operating personnel is used It avoids repetitive consideration of deviations It relates better to QRA than most other hazard assessment techniques do

Weaknesses

• Since specific guide words are required for each installation, some hazards may be omitted • Its proficiency largely depends on the experience of the leader and the knowledge of the team

The strengths and weaknesses of HAZID are tabulated in Table 7.8.

7.2.4 Qualitative Methods for Explosion Risks Qualitative risk analysis can be simple to identify roughly the risk levels of an installation or a compartment. Detailed structural assessment is normally not required for those facilities in consideration that they are at low risk from explosion events. It should be mentioned that risk is the product of consequence and frequency of occurrence. When quantitative values are not available, a qualitative risk screening method can be applied in case of that the required accuracy degree is sufficiently specified to make a decision on the assessment approach to be adopted. This section introduces an example of qualitatively evaluating the likelihood and consequence of explosion risks for offshore platforms according to UKOOA (2003). It is the most frequently adopted simple approach for qualitative risk analysis using a 3 × 3 matrix of potential consequence and likelihood for an explosion event. The matrix of the risk category is described in Table 7.9. The definitions of consequence and likelihood levels are discussed below. Consequence Evaluation The consequence evaluation assesses the influences of credible explosion scenarios including potential escalation. For human beings, the direct consequences of an

160

7 Risk Analysis Methods for Gas Explosion

Table 7.9 3 × 3 risk matrix Consequence Likelihood/frequency

Low

Medium

High

High

Medium risk

High risk

High risk

Medium

Low risk

Medium risk

High risk

Low

Low risk

Low risk

Medium risk

explosion are possibly considerably less than those impinges by indirect hazards of the explosion, such as the explosion incurred debris/projectiles impact. It is because the human physiology can withstand relatively high overpressures. The major risk to human beings from an explosion event is blown over by the blast wind, struck by missiles picked up by the blast, oxygen depletion and burns from the flames and hot gases. It is more likely that the escalation will cause major consequences, such as: • fires resulted from loss of inventory due to damaged equipment, pipework and vessels, • loss of TR integrity by failure of the boundary partitions, • structural failure, • insufficient outlets for escape and/or evacuation. To identify the direct and indirect effects of explosion overpressure, some critical parameters listed below should be assessed: • the vulnerability of safety critical elements to the dynamic pressure, overpressure, missiles and strong shock response, • occupancy of area immediately affected, • vulnerability of people in adjacent areas, • the relative location of the TR, • levels of congestion and confinement, • the suitability of the layout, • the dimensions of potential explosive gas clouds, • hazardous inventories, both isolatable and non-isolatable, • the operating and controlling philosophy, influencing manning levels and occupation frequency, • the operating and controlling philosophy influencing the extent of operator intervention and the potential of human error and inventory loss. Some guidance on the assessment of consequences is given below. Low consequence would be identified when the overpressure level is predicted to be relatively low, and the immediate and secondary consequences are also low. The relevant equipment count should be low that is limited to well heads and manifold with no vessels to result in low congestion and inventory. Confinement should also be low with no more than two solid boundaries including solid decks. Manning would be commensurated with a normally unattended installation with a low attendance frequency, specifically, less frequent than 6-weekly.

7.2 Qualitative Risk Analysis

161

A medium-consequence installation would be typically a platform or compartment where the congestion and confinement exceed what is defined for the lowconsequence cases but still with a low manning level consistent with a normally unattended installation. Manning would involve a moderate attendance frequency, more frequent than 6-weekly. Alternatively, a medium-consequence installation may be a processing platform requiring permanent manning but with low escalation potential to quarters, utilities and control areas, which are located on a separate structure. A high-consequence installation would encompass all remaining installations and compartments where there is significant processing on board leading to significant congestion and potential confinement with populated areas within the consequence range of escalation scenarios. This may typically be characterized by a PDUQ/PUQ installation with quarters on the same structure as the process. Likelihood Evaluation The likelihood for the occurrence of a gas cloud and delayed ignition decides the probability of an explosion. Specifically, the following factors will influence the probability of occurrence of an explosion: • hazardous inventory complexity, i.e. the number of flanges, valves, compressors and other potential gas leak sources, • the type of flanges, valves or pipework; some generic types of flange tend to have lower leak frequencies, e.g. hub type flanges, • the number of ignition sources within the potential gas cloud, • the ventilation conditions, • the equipment reliability and the maintenance philosophy. The likelihood considerations tend to align closely with the consequence factors. Low-consequence installations are usually small, not complex, and therefore, not likely to be susceptible to the explosion risk. Large installations have more potential leak and ignition sources, and thus, greater requirement for intervention and maintenance. Low likelihood installations have a low equipment count. The intervention frequency of 6-weekly or more is also recommended as a criterion because this will be a surrogate for equipment count and reliability as well as a measure of maintenance risk with respect to explosion. Medium likelihood is suggested by a NUI with equipment count greater than that for low likelihood case. Or the planned frequency of maintenance is greater than a 6-weekly basis, which suggests a higher or less reliable class of equipment with medium level of potential for explosion. A high likelihood installation will normally be a permanently manned installation with a high equipment level, and thus, a large number of potential leak sources and high ignition potential.

162

7 Risk Analysis Methods for Gas Explosion

7.3 Quantitative Risk Analysis In general, quantitative risk analyses are numerical estimations based on historical data or estimates of failures to predict the possibility of an event occurrence. Such quantitative estimations can be expressed by two basic forms, frequency or probability. Frequency is an expected number of likelihood of an event per unit time, usually a year. The frequency has dimensions as 1/time. The probability refers to the estimated possibility of occurrence of an event in a given time period or the conditional probability of the event given that such event has previously occurred. The probability is dimensionless and in the range from 0 to 1. In this book, both forms of likelihood expressions are applied. The main approaches of quantitative analysis of explosion events are given as below. • Historical accident frequency data: this method uses archived record of accidents. It is simple and relatively easy to understand. However, it may only be applicable to existing technology with significant experience of accidents. • Event tree analysis: this method shows the potential path an accident may develop from an initiating event through several branches to one of several possible outcomes. This technique is usually used to extend the initiating event frequency into a failure case frequency, which is suitable to combine with the consequence modelling (Sect. 7.3.1). • Fault tree analysis: this method breaks down an accident into its component causes, including human error, and estimates the frequency of each component from a combination of generic historical data and informed judgement (Sect. 7.3.1). • Bayesian network: it is a probabilistic graphical model and provides a systematic approach to combine historical data with judgments (Sect. 7.3.2).

7.3.1 Event Tree and Fault Tree Analysis An event tree analysis (ETA) and a fault tree analysis (FTA) are quantitative risk analysis methods that are most widely used to estimate explosion risks for process safety. Aven, Sklet and Vinnem (2006) developed a barrier and operational risk analysis (BORA) method which adopted event tree and fault tree methods to assess leak frequency of offshore platforms. Huang, Chen and Wang (2001) proposed a standard procedure for the application of fuzzy theories to evaluate human errors to be integrated into event tree analysis. Dong and Yu (2005) used fuzzy fault tree analysis to assess the failure of oil and gas transmission pipelines and a weighting factor was introduced to represent experts’ elicitations based on their different backgrounds of experience and knowledge. Ferdous, Khan, Sadiq, Amyotte and Veitch (2011) integrated fuzzy set theory and evidence theory into the traditional event tree and fault tree analysis in order to provide robust method to handle the uncertainty in

7.3 Quantitative Risk Analysis

Initiating Event

Branch 1 Immediate Ignition

Branch 2 Delayed Ignition

163 Branch 3

Branch 4

Explosion

Escalation

End Event

Probability

YES Fire (

YES Gas Releases

)

Damage from both explosion overpressures and escalation such as fire, BLEVE, projectile, etc. ( )

YES

NO

Damage from explosion overpressures only ( )

YES

NO Fire (

)

NO

NO

Unignited gas cloud ( )

Fig. 7.2 Event tree analysis of explosion event

QRA for the process systems. Wang, Zhang and Chen (2013) proposed a hybrid method of fuzzy set theory and fault tree analysis to quantify the crude oil tank fire and explosion occurrence probability. Event Tree Analysis An event tree is a visual model describing probable event sequences which may be developed from a hazardous situation (Vinnem, 2014). It uses branches to show the various possibilities that may arise at each step and often relates a failure event to various consequence models (Spouge, 1999). A detailed procedure for constructing and conducting the ETA for a process system can be found in Mannan (2012). For explosion events, the event tree construction starts from a hydrocarbon release event and works through each branch in turn as shown in Fig. 7.2. It is relatively straightforward to quantify an event tree by hand in spite of that spreadsheets or computer models are increasingly used to automate the multiplication task. A probability is associated with each branch, and the probability of each outcome is the products of the probabilities at each branch leading to them, so that

164

7 Risk Analysis Methods for Gas Explosion

Table 7.10 Strengths and weaknesses of event tree Strengths

• It has been broadly accepted • It is widely applied for many hazards in QRA which arise from sequences of successive failures • It provides a clear and logical form of presentation • It is simple to understand

Weaknesses

• Efficiency is a problem if many events occur in combination as many redundant branches will be created • All events are assumed to be independent • The clarity can be reduced when applied to systems without clear specification for failure or working states

N    P Aj = P(Bi )

(7.1)

i=1

where P(A j ) = outcome probability; P(Bi ) = branch probabilities on route to outcome; and N = number of branches on route to outcome. If the frequency of the initiating event is specified, the frequencies of the outcomes are derived by multiplying the outcome probabilities with the initiating event frequency:     F Aj = F × P Aj

(7.2)

where F(A j ) = outcome frequency; F = initiating event frequency. The strengths and weaknesses of event tree are tabulated in Table 7.10. Fault Tree Analysis FTA identifies basic causes of occurrence of an unwanted event and estimates the likelihood as well as the contribution of different causes leading to the unwanted event. In FTA, the basic causes are termed as basic events, and the unwanted event is called the top event. Construction of a fault tree generally starts from the top event and works down towards the basic events (Spouge, 1999). For each event, the conditions required to produce the event are explored. If any individual event may cause higher event, an “OR” gate is applied, while an “AND” gate is used when two or more events occur in combination. Kumamoto and Henley (2000) provide a detailed description of fault tree development and analysis for a process system. Figure 7.3 shows a fault tree analysis model of gas explosion by Wang et al. (2013). A gate-by-gate method can be used for quantification of the top event probability if all events are independent and there are no common cause failures (Spouge, 1999). If the input probabilities are smaller than 0.1, the “OR” and “AND” gates are calculated as follows: OR gate: P( A) =

N  i=1

P(Bi )

(7.3)

7.3 Quantitative Risk Analysis

165

Fig. 7.3 Fault tree analysis of explosion event (Wang et al., 2013, Permission from Elsevier)

AND gate: P( A) =

N 

P(Bi )

(7.4)

i=1

where P(A) is output event probability; P(Bi ) denotes the input event probabilities; and N stands for the number of input events. If the probability is larger than 0.1, gates with two independent inputs should be calculated as: OR gate: P(A) = P(B1 ) + P(B2 ) − P(B1 )P(B2 )

(7.5)

AND gate: P(A) = P(B1 ) × P(B2 )

(7.6)

The strengths and weaknesses of fault tree are listed in Table 7.11. It shares some common strengths and weaknesses from event tree.

166

7 Risk Analysis Methods for Gas Explosion

Table 7.11 Strengths and weaknesses of fault tree Strengths

• It has been widely applied and is broadly accepted • It is applicable to many hazards in QRA which arise from sequences of successive failures • It provides a clear and logical form of presentation

Weaknesses

• It can be complicated and inefficient when large systems are considered • The assumptions and conditional probabilities for each gate may not be clear due to the diagrammatic format • It may cause overlook of failure modes and common cause failures. • All events are assumed to be independent • The clarity can be reduced when applied to systems without simple failure or working states

It should be mentioned that ETA and FTA have only simple Boolean functions and sequentially dependent failures (Khakzad, Khan, & Amyotte, 2011). It is unable for them to clarify the complicated mechanisms of interrelationships between risk factors. Therefore, an advanced quantitative analysis method, the Bayesian network (BN), is usually implemented for risk assessments of explosion accidents when multiconsequences and complex interrelationships are required to be considered.

7.3.2 Bayesian Network Modelling Introduction The Bayesian network (BN) is a probabilistic graphical model, which represents a group of random variables and conditional dependencies between them. It can deal with multi-state variables with different causal relationships, which is superior to that of the traditional event tree and fault tree approaches, which have only simple Boolean functions and sequentially dependent failures. Figure 7.4 shows an example of a BN model for gas explosion events at petrol stations. Details of this method are described by Nielsen and Jensen (2009) and Pearl (2014). The key concepts and terms of a Bayesian network are introduced according to Spouge (1999) in the following context. • Distinction between observable and unobservable quantities. In general, data or observation is used to measure observable quantities while judgments or estimates from theoretical models can be applied to represent unobservable ones. The BN analysis aims at predicting unobservable quantities, e.g. probabilities, or unknown observable ones, e.g. future numbers of failures, based on available data, such as past numbers of failures. • Observed data can be used to modify initial/prior estimates of parameter and provide improved posterior estimates. The BN analysis can systematically blend new information with old information.

7.3 Quantitative Risk Analysis

167

Ignition Source

Leak Severity Leak Rate Direct Ignition

Fire

Delayed Ignition

Explosion

Building Type

No. of People

Building Damage

Human Loss

Fig. 7.4 Bayesian network for explosion event

• Probability is used to represent uncertainty about the occurrence of an event or the value of observable quantities. Therefore, the BN analysis uses the probability to represent a “degree of belief” that the event will occur instead of an annual frequency from other QRA methods. • Subjective judgments can be used in BN analysis. Quantification The calculation of BN depends on conditional probability and Bayes’ theorem. Johnson, Freund and Miller (2011) stated that it was only meaningful to obtain a probability of an event if it had been referred to a sample space S. The notation P(A|S) is used to refer to the probability of a specific sample space S. It should be interpreted as that P(A|S) is the conditional probability of A relative to S. If A and B are both events in S and P(B) > 0, the equation for conditional probability can be written as P( A|B) =

P(A ∩ B) P(B)

(7.7)

If A and B do not have any mutual influence on each other, A and B are defined as independent events. The multiplication rule can be written as:

168

7 Risk Analysis Methods for Gas Explosion

P( A|B) = P(A) ∗ P(B)

(7.8)

If there are mutually exclusive events B1 , B2 , . . . Bn , then one of the events must occur. According to Bayes’ theorem (Johnson et al., 2011), the rule of total probability equation can be written as P(A) =

n 

P(Bi ) ∗ P(A|Bi )

(7.9)

i=1

The probability of a particular state can be obtained using the following equation: P(Bi |A) =

P(A ∩ Bi ) P(A)

Substituting P(Bi ) ∗ P(A|Bi ) for P(A ∩ Bi ) and P(A) yields P(Bi |A) = n i=1

(7.10) n i=1

P(Bi ) ∗ P(A|Bi ) into

P(Bi ) ∗ P(A|Bi ) P(Bi ) ∗ P(A|Bi ) for P(A)

(7.11)

where r = 1, 2, …, n. Figure 7.5 shows a simple example of BN calculation regarding the sub-network of explosions. The two basic nodes are leak severity (L) and ignition source (I). The leak severity and ignition source affect the probability of delayed ignition (D). The probabilities of explosion severities (E) can then be calculated by the leak severity and delayed ignition probabilities. This sub-network contains four nodes and four links. The prior probability of major explosion severity can be calculated by the following formulation,

Fig. 7.5 Sub-network for estimating explosion loads

Ignition Source

Leak Severity

Delayed Ignition

Explosion

7.3 Quantitative Risk Analysis

P(E = Major) =

8  2 3  

169

P(E = Major, L = L i , I = I j , D = Dk )

(7.12)

i=1 j=1 k=1

where P is the probability; E denotes the explosion loads; L represents the leak severity; L i stands for the state of node A; I is the ignition source; I j denotes the states of node I; D represents the delayed ignition; and Dk stands for the state of node D. Based on the theorem of BN (Nielsen & Jensen, 2009), the joint probability can be determined as P(x1 , . . . , xn ) =

n 

P(xi |Pa(xi ))

(7.13)

i=1

where Pa(xi ) is the parent set of xi . The function remains an unconditional probability of P(xi ) if there are no parents of xi . In this sub-network, the node of delayed ignition has parents of leak severity and ignition source, and the node of explosion loads has parents of delayed ignition and leak severity. Therefore, the following equation can be derived   P E = Ma jor, L = L i , I = I j , D = Dk = P(E = Ma jor |L = L i , D = Dk )     × P D = Dk |L = L i , I = I j × P I = I j × P(L = L i )

(7.14)

Comparison with Conventional Analysis The applicability and reliability of BN analysis have not yet prevailed generally over the conventional/frequentist approaches. It is understandable for a new concept and methodology to spread and be accepted, especially when very severe consequences and economic costs are involved. The striking difference between them, which are discussed as follows, will give hints of the specific reasons. • Traditional analysis deals with risks in a statistical way which assumes that an event would occur if the situation were historically repeated many times. In contrast, the BN analysis uses probabilities to represent uncertainties of the occurrence of an event. • Conventional QRA assumes that the frequencies are “true” as they are calculated based on historical data. The BN interprets probabilities as unobservable quantities that are estimated according to data collected from other observable quantities combined with a model of how they affect the required probabilities. • Classical QRA method is often claimed to provide objective results such as annual frequencies, while BN uses subjective probabilities and the result represents a “degree of belief”. • Traditional QRAs normally represent uncertainties by using external methods such as confidence range of the estimated probabilities. The BN considers uncertainty

170

7 Risk Analysis Methods for Gas Explosion

as an intrinsic part of the analysis. The result of the BN analysis is a combination of probability and its uncertainty. • Only uncertainties due to limited data are considered by confidence analysis of classical QRAs, while BN analysis involves all types of uncertainties caused by data and models. The BN analysis is applied to VCE risks in this study because of the following advantages. • Evaluation of gas explosion risks is complicated when multi-factors, e.g. leak severity, vent condition, structural complexity, multi-consequences, such as building damage, human loss, environmental effect and complex inter-relationships are considered. Compared to traditional QRAs, each node of the BN model can be related by conditional probabilities. Thus, the BN analysis is evaluated to have higher proficiency to deal with a complicated mechanism with inter-dependent factors. • It is easier and more intuitive for engineers to understand a BN than conventional event tree or fault tree when complex mechanism are involved in the analysis. • The BN allows a quick and simple evaluation of each risk factor which offers a clearer review of the criticality of each risk factor. Based on this kind of review, decisions of further detailed assessments and risk mitigation measures can be made straighter forward. • Updating the BN analysis is more readily and systematically when more data become available. However, although the BN analysis is more advanced than conventional QRAs in some aspects, it also faces many problems when applied to oil and gas industries. • The traditional QRAs are simpler and more readily comprehensible to most engineering. It is difficult for most analysts to understand and use BN. • The conditional probabilities for each arrow may not be clear due to the diagrammatic format. • Due to the lack of practical applications, it is difficult for BN results of probabilities to meet the acceptability criteria as most industrial standards set the criteria based on frequency analysis.

7.4 Summary The risk analysis methods reviewed in this chapter include both qualitative and quantitative approaches. For qualitative methods, checklist, HAZOP, HAZID, etc. are briefly introduced. In terms of quantitative risk analysis methods, event tree, fault tree and a BN model are reviewed in details. On the other hand, improficiencies of current risk analysis to evaluate gas explosions have been identified to fail to predict and efficiently. For various process facility types, such as small petrol stations, large onshore process factories, highly congested

7.4 Summary

171

offshore platforms, specific site characteristics need to be considered in explosion risk analyses. However, most QRA methods are oversimplified not to accommodate the complex mechanism of gas explosions, while some advanced computational methods are too computationally intensive to be cost-effective. Therefore, this book proposes to apply advanced gas explosion risk analysis approaches on different oil and gas facilities in the following chapters. A confidence-level-based event tree analysis is applied to an offshore platform in Chap. 8. Both qualitative and quantitative explosion risk analyses explore various uncertainties induced by subjective judgments that may significantly affect the validity of risk evaluations. The proposed method is applied to integrate uncertainties into conventional risk analysis using the concept of fuzzy theory. It yields a more reliable risk evaluation by reducing the impact of subjective judgment-related uncertainties caused by data shortage. To improve the effectiveness and efficiency of the explosion risk analysis when multiple factors and multiple consequences are involved, an advanced Bayesian network-based quantitative explosion risk analysis method is adopted to model VCE risks of petrol stations. The method systematically specifies events from an initial release to vapour cloud explosions, till consequences. For a risk analysis of gas explosion in process facilities close to residential areas, a grid-based risk mapping method is developed to provide a detailed explosion risk analysis for large areas under complicated circumstances. Compared with traditional explosion risk analysis, the proposed grid-based method simplifies complicated conditions through the gridding process. More reliable risk analysis has been ensued. When computational fluid dynamic (CFD) software is used to calculate overpressures, high computational intensity is usually involved for structures to have enormous sizes and highly complicated designs. In Chap. 11, a multi-level explosion risk analysis method is established to provide a more cost-efficient explosion risk assessment of such super-large oil and gas facilities with highly congested environments, such as floating liquefied natural gas (FLNG) platforms. Chapter 12 introduces CFD-based detailed frequency analysis for FLNG. The exceedance frequency of overpressure at the living quarter is calculated using the monitored overpressures along with the leak frequencies and ignition probabilities. The effects of blast wall on explosion overpressures are also discussed.

References Aven, T., Sklet, S., & Vinnem, J. E. (2006). Barrier and operational risk analysis of hydrocarbon releases (BORA-Release): Part I. Method description. Journal of Hazardous Materials, 137(2), 681–691. CCPS. (2011). Guidelines for hazard evaluation procedures (3rd Ed.). Center for Chemical Process Safety. CIA. (1992). A guide to hazard and operability studies. Chemical Industries Association. Dong, Y., & Yu, D. (2005). Estimation of failure probability of oil and gas transmission pipelines by fuzzy fault tree analysis. Journal of Loss Prevention in the Process Industries, 18(2), 83–88.

172

7 Risk Analysis Methods for Gas Explosion

Ferdous, R., Khan, F., Sadiq, R., Amyotte, P., & Veitch, B. (2011). Fault and event tree analyses for process systems risk analysis: Uncertainty handling formulations. Risk Analysis, 31(1), 86–107. Huang, D., Chen, T., & Wang, M. J. J. (2001). A fuzzy set approach for event tree analysis. Fuzzy Sets and Systems, 118(1), 153–165. Johnson, R., Freund, J., & Miller, I. (2011). Probability and statistics for engineers. Boston, MA: Pearson. Khakzad, N., Khan, F., & Amyotte, P. (2011). Safety analysis in process facilities: Comparison of fault tree and Bayesian network approaches. Reliability Engineering & System Safety, 96(8), 925–932. Kletz, T. A. (1999). HAZOP and HAZAN: Identifying and assessing process industry hazards. IChemE. Kumamoto, H., & Henley, E. J. (2000). Probabilistic risk assessment and management for engineers and scientists. Wiley-IEEE. Mannan, S. (2012). Lees’ Loss prevention in the process industries: Hazard identification, assessment and control: Butterworth-Heinemann. Nielsen, T. D., & Jensen, F. V. (2009). Bayesian networks and decision graphs. Springer Science & Business Media. Pearl, J. (2014). Probabilistic reasoning in intelligent systems: Networks of plausible inference. Morgan Kaufmann. Spouge J. (1999). A guide to quantitative risk assessment for offshore installations. The Centre of Marine and Petroleum Technology (CMPT), United Kingdom. UKOOA. (2003). Fire and explosion guidance part 1: Avoidance and mitigation of explosions (Issue 1). United Kingdom: UK Offshore Operation Association. Vincoli, J. W. (2006). Chapter 6 preliminary hazard analysis. In Basic guide to system safety (2nd Ed., pp. 65–84). Hoboken, N.J.: Wiley-Interscience. Vinnem, J. E. (2014). Offshore risk assessment principles, modelling and applications of QRA studies. London: Springer. Wang, D., Zhang, P., & Chen, L. (2013). Fuzzy fault tree analysis for fire and explosion of crude oil tanks. Journal of Loss Prevention in the Process Industries, 26(6), 1390–1398.

Chapter 8

Event Tree Analysis of Offshore Hydrocarbon Release Events

Abstract This chapter presents event-tree-based risk analysis of hydrocarbon release accidents, which is the potential source for the VCE formation. A fuzzytheory-based confidence level method for reducing uncertainties is explained and a barrier and operational risk analysis (BORA-Release) method is introduced as the basic model to illustrate the proposed methodology. One case study is provided as well to demonstrate implementation of this method.

8.1 Introduction Hydrocarbon release is the very potential initiation of a chain of events leading to explosion accident in process facilities. This chapter starts with the introduction of the mechanism for the hydrocarbon releases. Event tree analysis is applied to calculate the frequency of hydrocarbon releases. To improve the reliability of risk evaluation, a confidence-based quantitative risk analysis method is developed by integrating fuzzy set theory into event tree analysis (Huang et al. 2015). Risk analysis of offshore hydrocarbon release events is presented in this chapter. Since barrier and operational risk analysis (BORA) method (Aven, Sklet, & Vinnem, 2006) has proved to be one of the most applicable and practicable event tree-based QRA in the offshore oil and gas industry, the BORA method is selected for the current analysis. The improficiency in BORA is identified and the confidence-level-based method is adopted to modify BORA. Generally, uncertainties refrain QRA from general application. The uncertainties arise mainly from two aspects for offshore QRA (Spouge, 1999). QRA is a relatively new technique. There is no general consensus on a QRA approach to be the most preferable among large variations due to lack of supporting experimental data. Although QRA is identified to be objective, subjective judgements are often indispensable in offshore risk assessments due to the complexity in the configuration of the oil and gas platforms and comprehensive nature of the problem. These subjective judgements based on experts’ elicitation may lead to inaccurate risk estimates. In

© Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_8

173

174

8 Event Tree Analysis of Offshore Hydrocarbon Release Events

addition, the extent of simplification made in the modelling of risks overlays on the uncertainties to violate the objectivity (Vinnem, 2007). The most prevailing approaches for representing and reasoning with uncertainties are Monte-Carlo simulation (Vose, 1996), Bayesian probability theory (Bernardo & Smith, 2009) and fuzzy set theory (Zadeh, 1965). In this study, the uncertainties from subjective judgements will be the main focus of exploration. The fuzzy set theory is selected due to its comparative objectivity for decision-making with estimated values or experience-based judgements according to imprecise information (Liu, Yang, Wang, & Sii, 2003). Thus, fuzzy-set-theory-based confidence level is adopted to address the uncertainties from the experts’ subjective judgements. Improficiency in the probability estimation for an accidental risk also arises from safety experts’ uncertainty in judgement. The proposed confidence level method mitigates the influence of uncertainties and improves the reliability of QRA. Compared to previous methods, this proposed method focuses on subjective judgements and divides the expert’s confidence into five levels by introducing a new form of fuzzy number function. This new L-R bell-shaped fuzzy number can be specified as a group of modified fuzzy membership curves that represent different confidence levels of the experience-based judgements. Several existing methods have integrated fuzzy set theory into conventional decision-making and reasoning methods. Huang et al. (2001) demonstrated the procedure for the application of fuzzy theories to evaluate human errors in event tree analysis. Cho et al. (2002) introduced new forms of fuzzy membership curves to represent the degree of uncertainties involved in both probabilistic parameter estimates and subjective judgements. Dong and Yu (2005) used fuzzy fault tree analysis to assess the failure of oil and gas transmission pipelines and a weighting factor was introduced to represent experts’ elicitations based on their different backgrounds of experience and knowledge. For the application of fuzzy concepts to the risk analysis of the oil and gas industry, Markowski, Mannan, and Bigoszewska (2009) developed a fuzzy-set-theory-based “bow-tie” model for process safety analysis (PSA) to address the uncertainties of information shortages. Wang, Xie, Habibullah, and Ng (2011) proposed a hybrid causal logic model to assess the fire risks on an offshore oil production facility by mapping a fuzzy fault tree into a Bayesian network. Recently, Sa’idi, Anvaripour, Jaderi and Nabhani (2014) developed a fuzzy-riskbased maintenance (RBM) method for risk modelling of process operations in oil and gas refineries to yield more sensible results compared to that from the traditional RBM model. Rajakarunakaran, Kumar, and Prabhu (2015) presented a fuzzy-logicbased method for the reliability analysis of a liquid petroleum gas (LPG) refuelling station in order to model inaccuracy and uncertainty when quantitative historical failure data are scarce or unavailable.

8.2 BORA-Release Method

175

8.2 BORA-Release Method The BORA project was conducted from 2003 to 2006 to apply barrier functions including human, technical and organizational barriers to traditional risk analysis of offshore facilities (Seljelid, Haugen, Sklet, & Vinnem, 2007). A BORA-Release method has been developed from the project to analyse the hydrocarbon release risks of the offshore structures from a set of hydrocarbon release scenarios based on the combined fault and event trees, barrier block diagrams, fault trees and risk influence diagrams (Aven et al., 2006). In contrast to the general QRA method, in the BORA-Release method, the risk analysis experts specify conditions of offshore platforms with technical, human, operational, as well as organizational risk influence factors (RIFs). The performance of the initial events and barriers will be modulated by the RIFs. The industry average frequencies/probabilities will be adjusted with regard to the final score of the RIFs. Based on the identified RIFs, a relatively more realistic frequency/probability can be derived in view of the platform-specific conditions. To implement the BORA method, Aven et al. (2006) specified eight steps: (1) developing a basic risk model; (2) modelling the performance of barrier functions; (3) assigning the industry average frequencies/probabilities to the initiating events and basic events; (4) developing risk influence diagrams; (5) scoring RIFs; (6) weighting RIFs; (7) adjusting industry average frequencies/probabilities; (8) determining the platform-specific risk by recalculating the risk. The first two steps are the development of a basic risk model, which is illustrated by an event tree including initiating events, barrier functions and end events as shown in Fig. 8.1. The initiating events are specified by a set of 20 hydrocarbon release scenarios developed by the BORA project group to cover most of the potential hydrocarbon release sources based on review of the release statistics (Sklet, 2006a). A safety barrier refers to a physical and/or non-physical method or a device to mitigate undesired events (Sklet, 2006b). The frequency of the initiating event is then defined based on recorded historical data or industrial statistics. The failure probability of the barrier functions is evaluated by fault tree analysis. In the fault tree analysis, basic events are identified and industrial average frequencies are assigned to compute the total probability of the failure of the barriers. For the quantitative risk analysis of the event tree, specific frequencies/probabilities are assigned to all of the main components, e.g. initiating event and safety barriers, of the event tree. The probabilities of each path of the event tree are then calculated accordingly as illustrated in Fig. 8.1, where F A represents the industrial average frequency of an initiating event A and P f Bi stands for the probability of failure of safety barrier i. In step 4, risk influence diagrams are developed to apply the platform-specific human, operational, organizational and technical RIFs to the initiating events and basic events. RIFs are categorized into five groups, i.e. personal characteristics, task characteristics, characteristics of the technical system, administrative control and

176 Initiating Event

8 Event Tree Analysis of Offshore Hydrocarbon Release Events Barrier Functions 1 to n

End Event

Probability

Safe: Initiating event did not occur

Safe: Barrier 1 succeeded

Event A

…… Barrier 1

Safe: Barrier n succeeded

…… Barrier n

Failure: All Barriers failed

Fig. 8.1 Basic event tree model in BORA-Release method (permission from Elsevier)

organizational factors, with each group containing several detailed RIFs (Aven et al., 2006). The next two steps score and weight the risk influence factors. All RIFs are evaluated and sorted into six levels (from A to F), which is adapted from the technical condition safety (TTS) by Thomassen & Sorum, 2002. Usually, the scoring analysis can be implemented with three methods, i.e. direct assessment using behavioural checklists and behaviourally anchored rating scales (BARS) (Jacobs & Haber, 1994), and assessment based on results from TTS project (Thomassen & Sorum, 2002), and risk level based on the Norwegian Continental Shelf (RNNS) project (PSA, 2013). The RIF’s value differentiates the significance of the different basic events and the process based on general discussions with platform personnel and analysts. The most highlighted RIF is assigned to a value of 10, and the other RIFs span a scale range of 10–8–6–4–2 in view of the relative importance. The industry’s average probabilities/frequencies of initiating event and basic events are revised in step 7 by modification factor (MF), which is an index of the evaluation of the platform-specific operational standard based on the scoring and weighting of RIFs. The revised platform-specific probabilities/frequencies are applied to all the main components in the event tree model to derive the platformspecific frequency of hydrocarbon release. The platform-specific probability or the revised probability Prev (X ) can then be calculated as Prev (X ) = MF × Pave (X )

(8.1)

8.2 BORA-Release Method

177

MF =

n 

wi Q i

(8.2)

i=1

where Pave (X ) is the industry average probability/frequency of initiating events or basic events of the safety barriers; wi denotes the weight of RIF no. i for event A; Q i represents a measure of the status of RIF no. i; and n stands for the number of RIFs. To calculate Q i , Plow (X ) and Phigh (X ) should be identified by experts as the lower and higher limits for Prev (X ). It is understood uncertainties are indispensable in the analysis using the BORA method. Uncertainties arise in scoring and weighting RIFs because it is mainly based on the subjective judgements of risk analysis experts according to their previous experiences, so as Sklet, Vinnem, and Aven (2006) pointed out that the RIF scoring was not valid due to the subjective scoring methods. The imprecision and/or lack of data further deteriorates the problem of uncertainties. Thus, it is expected that the confidence level methodology, introduced as below, will improve the accuracy of the BORA method by categorizing uncertainties.

8.3 Application of the Confidence Level Method to BORA Method This section will demonstrate that a confidence-level-based methodology can effectively incorporate the uncertainties into the QRA model. A schematic framework for the implementation of the proposed method is depicted in Fig. 8.2. As mentioned in Sect. 8.2, since the RIF scoring and weighting in the BORA method highly depend on the expert’s judgements, the result may not be valid if the data are insufficient or the scoring method is inappropriate. Thus, the proposed confidence-based method provides experts with a measurement of their confidence levels to assist them in defining the probability of hydrocarbon release accidents more objectively. The following context gives the main steps in the application of the confidence-level-based method to the BORA model.

8.3.1 Analysis Using L-R Bell-Shaped Fuzzy Number The adjusted results from the BORA method is then assigned an L-R bell-shaped fuzzy number, which can be represented by a group of modified fuzzy membership curves to represent different confidence levels of the experience-based judgements.  = (a1 , a2 , a3 ) and the membership The fuzzy number is defined by a triplet A function is given as

178 Probability of the lower bound

8 Event Tree Analysis of Offshore Hydrocarbon Release Events Revised average probability

Define confidence level

Probability of the higher bound

Value of triplet A a11,a22,a33

Confidence factor, n Bell-shaped fuzzy membership function Define optimism level

α-cut calculation

Total integral defuzzification

Final probability

Fig. 8.2 Schematic framework for proposed confidence level method (permission from Elsevier)

⎧ 0 ⎪ ⎪ ⎪ a −x n ⎪ b a 2−a ⎪ ⎪ 2 1 ⎨e μ A(x) = 1  n ⎪ x−a ⎪ b a −a2 ⎪ ⎪ 3 2 e ⎪ ⎪ ⎩ 0

⎫ ⎪ ⎪ ⎪ ⎪ ⎪ for a1 ≤ x < a2 ⎪ ⎬ for x = a2 ⎪ ⎪ ⎪ for a2 < x ≤ a3 ⎪ ⎪ ⎪ ⎭ for x > a3 for x < a1

(8.3)

where a2 is the centre of a fuzzy membership curve, which represents the expert judgement value; a1 and a3 represent the values of the upper and lower bounds, respectively; n denotes the confidence factor; and b stands for a boundary index to control the boundary of the membership function to ensure the membership is smaller than or equal to α when x = a1 or a3 . To achieve this, the boundary factor b should be equal or smaller than ln α. An example of the bell-shaped curve is depicted in Fig. 8.3. To derive the triplets for the BORA models, the RIF-revised probability can take the value of a2 to represent the centre of the bell-shaped curve and the lower limit and higher limit of the probabilities are specified by expert.

8.3 Application of the Confidence Level Method to the BORA Method

179

1.2

Fig. 8.3 Example of bell-shaped fuzzy number curve (permission from Elsevier)

a2

Membership

1 0.8 0.6 0.4 0.2 0

a3

a1 0

0.05

0.1

0.15

0.2

Probability

8.3.2 Fuzzy Calculations Initiating event and safety barriers in the BORA event tree are analysed separately using the L-R bell-shaped fuzzy number. A total fuzzy number is derived using α-cut arithmetic operations. When multiple events require modification according to the confidence-based bell-shaped fuzzy number, the α-cut operation is applied to execute the arithmetic operations of fuzzy numbers based on the extension principle by Zadeh (1965). The basic rules of the α-cut arithmetic operations are described as follows,  Aα = { x|x ∈ R, μ A (x) ≥ α} ≡ the α-cut of A, B, Bα = { x|x ∈ R, μ B (x) ≥ α} ≡ the α-cut of  then, 

 

Aα (+)Bα = a1α , a2α (+) b1α , b2α = a1α + b1α , a2α + b2α

(8.4)



 

Aα (−)Bα = a1α , a2α (−) b1α , b2α = a1α − b2α , a2α − b1α

(8.5)



 

Aα (×)Bα = a1α , a2α (×) b1α , b2α = a1α × b1α , a2α × b2α

(8.6)



 

Aα (÷)Bα = a1α , a2α (÷) b1α , b2α = a1α ÷ b2α , a2α ÷ b1α

(8.7)

 α  a1 ≥ 0, b1α > 0 where (+), (−), (×), and (÷) represent fuzzy addition, subtraction, multiplication and division, respectively.

180

8 Event Tree Analysis of Offshore Hydrocarbon Release Events 1.2

Membership

1 0.8 0.6 0.4 0.2 0

0

0.5

1

1.5

2

2.5

3

Probability for Each Component

(a) Fuzzy number curves for different components 1.2

Membership

1 0.8 0.6 0.4 0.2 0

0

2

4

6

8

Total Probability

(b) Final curve of the total fuzzy number after the -cut calculation Fig. 8.4 Illustration of α-cut operations for multiple components (permission from Elsevier)

The fuzzy membership curve of each component is then depicted and a final diagram is derived using an α-cut operation as shown in Fig. 8.4.

8.3.3 Defining Confidence Level of RIF-Scoring Process The confidence levels are classified into five categories. The values of the fuzzification factors are given in Table 8.1. A diagram of each confidence level can be obtained by assigning different confidence factors to the fuzzy membership equation. The bell-shaped fuzzy number realizes the implication of different confidence levels via

8.3 Application of the Confidence Level Method to the BORA Method Table 8.1 Category of confidence levels

181

Confidence level

Description

Confidence factor

1

Very confident

0.1

2

Confident

0.5

3

Neutral

1

4

Unconfident

2

5

Very unconfident

3

1.20E+00

Fig. 8.5 L-R bell-shaped fuzzy number curves with different confidence factors (permission from Elsevier)

n=0.5

Membership

1.00E+00

n=1 n=2

8.00E-01

n=3

6.00E-01 4.00E-01 2.00E-01 0.00E+00

0

0.05

0.1

0.15

0.2

Probobility

transforming its shapes with regard to confidence factors as shown in Fig. 8.5. The larger value of the confidence factor n, the lower the confidence level, which implies more prominent uncertainties. In general, safety experts define the confidence level of judgements based on the degree of uncertainties from four aspects (Cho, Choi, & Kim, 2002): (1) the complexity of the judgemental condition; (2) the level of education, assurance and experience; (3) the availability of data (sufficient/insufficient/none); (4) the standard of the analysis method: the higher the degree of uncertainties, the lower the confidence level.

8.3.4 Deciding Degree of Optimism and Defuzzifying Final Fuzzy Number In order to identify a matching α-cut operation and to derive complete information, the defuzzification method with a total integral value (Liou & Wang, 1992) is chosen and a factor δ for the optimism levels is assigned to reflect the confidence level in the decision-maker. Thus, for the L-R bell-shaped fuzzy number, the total defuzzified integral value will be

182

8 Event Tree Analysis of Offshore Hydrocarbon Release Events

Table 8.2 Category of optimism factors

Optimism level

Description

Optimism factor δ

A

Very optimistic

0.1

B

Optimistic

0.3

C

Neutral

0.5

D

Pessimistic

0.7

E

Very pessimistic

0.9

       = (1 − δ)I L A  + δ IR A  δ ∈ [0, 1] ITδ A

(8.8)

1       =  α αi A IL A

(8.9)

   = IR A

α=0 1 

   α αi A

(8.10)

α=0

     and I R A  are the left and right integral values of A,  respectively; δ where I L A   δ  denotes the optimism factor; and IT A represents the total integral value with the influence of δ. When δ = 0, the nought value indicates the optimistic viewpoint of the decisionmaker. Alternatively, for a pessimistic or moderate decision-maker, δ is set to be 1 or 0.5, respectively. The degree of optimism is also classified into five categories and the corresponding values of the optimism factor δ are given in Table 8.2. With the evaluation of the confidence in the decision-maker, the safety engineers/managers are able to find an appropriate probability for hydrocarbon release risks for their offshore facilities. The total integral method is used to defuzzify the final result of the fuzzy analysis and to apply the final probability to the BORA risk model.

8.3.5 Assumptions for Practical Implementation of Proposed Method For the proposed method, the following assumptions should be inferred in the practical implementation. • The application of the proposed method is limited to uncertainties ensued from subjective judgements.  = • The proposed fuzzy membership function is developed via a triplet A (a1 , a2 , a3 ). For the implementation of the proposed method, the experts are required to define the probabilities for the lower and higher bounds.

8.3 Application of the Confidence Level Method to the BORA Method

183 Safe: Initiating event did not occur

Event B0 Incorrect fitting of flanges during maintenance

Safe: Failure revealed and corrected

Barrier B1 Self-control of work

Safe: Failure revealed and corrected Barrier B2 3rd party control of work

Barrier B3 Leak test

Safe: Failure revealed and corrected

Release

Fig. 8.6 Scenario B: release due to incorrect fitting of flanges during maintenance (permission from Elsevier)

• The confidence and optimism levels in this study are divided into five levels subject to the specific conditions of real projects. • When the memberships of the two bounds a1 and a3 are assumed to be 0, it is suggested that α is set to 0.01 or less in the defuzzification process to ensure the membership of both bounds are approximating 0.

8.4 Case Study—Application of Proposed Method to BORA Model To demonstrate the application of the suggested procedure, case study is presented with risk scenario B from the work of Sklet et al. (2006). The barrier block diagram of risk scenario B is shown in Fig. 8.6 and all the relevant results for scenario B (Sklet et al., 2006) are listed in Table 8.3. Figure 8.6 illustrates the basic event tree model for the incorrect fitting of flanges during maintenance. There are three safety barriers in this scenario and the relationship between the initiating event and all the barriers are in series. Thus, the only arithmetic calculation required in this event tree analysis is multiplication. From Table 8.3, it can be observed that the frequency of the higher bound

0.11

0.04 0.0012

PFailure (B2)c

PFailure (B3)d

e f Btotal

Probability of failure to reveal failure by third-party control

Probability of failure to detect release by leak test

Total release frequency from scenario B per year

0.34

PFailure (B1)b

Probability of failure to reveal failure by self-control

0.84

f (B0)

Average probability: Pave

Frequency of incorrect fitting of flanges or bolts after inspection per year

a

Event

Description

Table 8.3 Scenario B: industry probabilities/frequencies from BORA method

2.04E−6

0.008

0.022

0.069

0.168

Lower bound: Plow

0.32

0.2

0.55

0.69

4.2

Higher bound: Phigh

0.0038

0.066

0.15

0.37

1.064

Revised probability: Prev

184 8 Event Tree Analysis of Offshore Hydrocarbon Release Events

8.4 Case Study—Application of Proposed Method to BORA Model Table 8.4 Values for triplets of initiating event and basic events

Events

185

Triplets’ value b1

b2

b3

 B(B0)  B(B1)

0.168

1.064

4.2

0.069

0.37

0.69

 B(B2)  B(B3)

0.022

0.15

0.55

0.008

0.066

0.2

Table 8.5 Modified industry frequencies based on confidence levels only Confidence level

1

2

3

4

5

Modified industry probabilities/frequencies

0.0039

0.0048

0.0085

0.021

0.034

0.32 is approximately 100 times larger than the revised average frequency of 0.0039 after the multiplication calculations, which implies the range of uncertainties. Table 8.4 shows that the values of Plow , Prev and Phigh for the initiating event B0 and safety barriers B1, B2 and B3 are applied to triplets  B = (b1 , b2 , b3 ). Equation (8.1) and the α-cut operations are adopted to implement the analysis of the fuzzy numbers with α = 0.01 and b = −7. After defuzzification, the final modified probabilities based on confidence levels and optimism levels will be deduced from Eq. (8.6). Figure 8.7 plots the final results using different confidence factors for scenario B. From Fig. 8.7, the right-hand side has a more significant dispersion than the left-hand side does with the decreases of the confidence level, which implies that the righthand side governs the influence of the uncertainties. This is due to the probability deviation between Prev and Phigh , i.e. approximately 0.3, is much larger than that between Prev and Plow of about 0.003. Table 8.5 gives the final modified industry probabilities after defuzzification according to the five confidence levels with a moderate attitude only. From Table 8.5, it is observed that the modified probability with the highest confidence level is equal to the revised frequency from the BORA method; and the probability of hydrocarbon release increases with the decreasing confidence level. On the other hand, since there are four components that require adjustments in the event tree analysis of scenario B, the deviations among the modified frequencies by different confidence levels become conspicuous as shown in Table 8.5. There are about ten-time difference in respective frequencies from confidence level 1 (0.0039) to level 5 (0.034). From Table 8.3, it can be stated that the advance of the BORA method lies in the quantification of the platform-specific operational factors to modify the industrial average value. On the other hand, according to the degree of data adequacy and the RIF-scoring method standard, the revised industrial frequencies still do not eliminate the uncertainties completely. These uncertainties could affect the result significantly in view of that the higher bound of the industrial probabilities is found to be much

8 Event Tree Analysis of Offshore Hydrocarbon Release Events 1.2

1.2

1

1

0.8

0.8

Membership

Membership

186

0.6 0.4

0

0.1

0.2

0.3

0

0.4

0

0.2

Probability

Probability

(b) n = 0.5

1.2

1.2

1

1

0.8

0.8

0.6 0.4

0.3

0.4

0.3

0.4

0.6 0.4 0.2

0.2 0

0.1

(a) n = 0.1

Membership

Membership

0.4 0.2

0.2 0

0.6

0 0

0.1

0.2

0.3

0.4

0

0.1

0.2

Probability

Probability

(d) n = 2

(c) n = 1 1.2

Membership

1 0.8 0.6 0.4 0.2 0

0

0.1

0.2

0.3

0.4

Probability

(e) n = 3 Fig. 8.7 Final fuzzy number curves with different confidence levels for scenario B (Sklet et al., 2006) (permission from Elsevier)

8.4 Case Study—Application of Proposed Method to BORA Model

187

higher than the revised frequencies, such as the 100-time difference between the higher bound and the revised value in scenario B from Sklet et al. (2006). Therefore, it is not effective to determine the likelihood of an initial event or the failure of barrier functions by one definite value. From Table 8.5, it can be observed that the confidence level divides the revised industrial average probabilities into five groups to form the confident level 1–5. With the highest confidence level, the experts can derive directly the similar probability as that from the BORA method. On the other hand, for the lower level of confidence, more uncertainties are engaged to ensue a higher frequency of hydrocarbon release risks. It is observed that the prominent risk difference by the BORA method between the revised industrial average probability 0.0039 and the higher bound probability 0.32 is decreased drastically by defining the probability with the lowest confidence level (0.034). This difference reduction mitigates the influence of uncertainties by ten times for the situation with insufficient experiences and data. Based on the five levels of confidence classification, the proposed method quantifies the uncertainties into five ranges to assist the experts to estimate the risks more objectively based on their specific confidence levels.

8.5 Summary In summary, the accuracy and validity of conventional event tree analysis are significantly susceptible to uncertainties. This chapter proposed a new methodology to incorporate uncertainties into event tree analysis in terms of confidence levels. A new form of the bell-shaped fuzzy number is devised to consider the degree of uncertainties indicated by confidence levels in view of that it is unrealistic to define the probability of an event by one single explicit value. As a result, the influence of the uncertainties has been reduced by approximately ten times. Therefore, it is concluded that the proposed confidence level improves the reliability of risk evaluations by reducing the impact of subjective-judgement-related uncertainties. It helps safety experts to make more realistic and sensible risk estimates based on the confidence level evaluations for the risk assessments. For large-scale complex process systems, the final risk may accumulate significantly and unrealistically if the experts are very unconfident about their risk assessments at every step or most steps of the large-scale process. On the other hand, if they are only unconfident about a few steps of the whole process, the final risk estimates may not show large difference from the initial results. By accommodating the confidence level of the safety expert, the proposed method yields more realistic results. This chapter focuses on the hydrocarbon release events which is the very initiation of a chain of events leading to a vapour cloud explosion. From next chapter, explosion risks at different types of oil and gas facilities will be assessed using various methods.

188

8 Event Tree Analysis of Offshore Hydrocarbon Release Events

Systematic approach for analysis of various risk factors from an initial release to the final explosion overpressure including ignition, release severity, wind effects, site conditions and congestions will be presented. To start with, the Bayesian network method is applied to deal with the complicated mechanism of VCEs in the next chapter.

References Aven, T., Sklet, S., & Vinnem, J. E. (2006). Barrier and operational risk analysis of hydrocarbon releases (BORA-Release): Part I. Method description. Journal of Hazardous Materials, 137(2), 681–691. Bernardo, J. M., & Smith, A. F. (2009). Bayesian theory (Vol. 405). Wiley. Cho, H. N., Choi, H. H., & Kim, Y. B. (2002). A risk assessment methodology for incorporating uncertainties using fuzzy concepts. Reliability Engineering & System Safety, 78(2), 173–183. Dong, Y., & Yu, D. (2005). Estimation of failure probability of oil and gas transmission pipelines by fuzzy fault tree analysis. Journal of Loss Prevention in the Process Industries, 18(2), 83–88. Huang, D., Chen, T., & Wang, M. J. J. (2001). A fuzzy set approach for event tree analysis. Fuzzy Sets and Systems, 118(1), 153–165. Huang, Y., Ma, G., Li, J., & Hao, H. (2015). Confidence-based quantitative risk analysis for offshore accidental hydrocarbon release events. Journal of Loss Prevention in the Process Industries, 35, 117–124. Jacobs, R., & Haber, S. (1994). Organisational processes and nuclear power plant safety. Reliability Engineering and System Safety, 45, 75–83. Liou, T. S., & Wang, M. J. J. (1992). Ranking fuzzy numbers with integral value. Fuzzy Sets and Systems, 50(3), 247–255. Liu, J., Yang, J. B., Wang, J., & Sii, H. S. (2003). Review of uncertainty reasoning approaches as guidance for maritime and offshore safety-based assessment. Journal of UK Safety and Reliability Society, 23(1), 63–80. Markowski, A. S., Mannan, M. S., & Bigoszewska, A. (2009). Fuzzy logic for process safety analysis. Journal of Loss Prevention in the Process Industries, 22(6), 695–702. PSA. (2013). Trends in risk level in the petroleum activity, summary report 2012—Norwegian continental shelf. Stavanger: The Petroleum Safety Authority. Rajakarunakaran, S., Kumar, A. M., & Prabhu, V. A. (2015). Applications of fuzzy faulty tree analysis and expert elicitation for evaluation of risks in LPG refuelling station. Journal of Loss Prevention in the Process Industries, 33, 109–123. Sa’idi, E., Anvaripour, B., Jaderi, F., & Nabhani, N. (2014). Fuzzy risk modelling of process operations in the oil and gas refineries. Journal of Loss Prevention in the Process Industries, 30, 63–73. Seljelid, J., Haugen, S., Sklet, S., & Vinnem J. E. (2007). Operational risk analysis—Total analysis of physical and non-physical barriers. In BORA handbook, Rev 00, Norway. Sklet, S. (2006a). Hydrocarbon releases on oil and gas production platforms: Release scenarios and safety barriers. Journal of Loss Prevention in the Process Industries, 19(5), 481–493. Sklet, S. (2006b). Safety barriers: Definition, classification, and performance. Journal of Loss Prevention in the Process Industries, 19(5), 494–506. Sklet, S., Vinnem, J. E., & Aven, T. (2006). Barrier and operational risk analysis of hydrocarbon releases (BORA-Release): Part II: Results from a case study. Journal of Hazardous Materials, 137(2), 692–708. Spouge J. (1999). A guide to quantitative risk assessment for offshore installations. United Kingdom: The Centre of Marine and Petroleum Technology (CMPT).

References

189

Thomassen, O., & Sorum, M. (2002, March). Mapping and monitoring the technical safety level. In Society of petroleum engineers international conference on health, safety and environment in oil and gas exploration and production, Kuala Lumpur, Malaysia: March (pp. 20–22). Vinnem, J. E. (2014). Offshore risk assessment principles, modelling and applications of QRA studies. London: Springer. Vose, D. (1996). Quantitative risk analysis: A guide to monte carlo simulation modelling author: David Vose. USA: Wiley. Wang, Y. F., Xie, M., Habibullah, M. S., & Ng, K. M. (2011). Quantitative risk assessment through hybrid causal logic approach. Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability, 225(3), 323–332. Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8(3), 338–353.

Chapter 9

Bayesian Network Analysis of Explosion Events at Petrol Stations

Abstract This chapter illustrates a Bayesian-network-based quantitative risk analysis method for VCE accidents at small oil and gas facilities, such as petrol stations. Meanwhile, to reduce uncertainties by data shortage, three types of data, i.e. practical information, computational simulations and subjective judgements are introduced to quantify the proposed BN. A case study using the proposed method to model the complete explosion process is presented.

9.1 Introduction In this chapter, the Bayesian network is applied to assess explosion risks in petrol stations. In oil and gas industry, very little research on explosion risk assessment has been conducted for petrol stations. It is probably because that consequences of accidents at service stations are not so disastrous as those at large-scale oil and gas facilities are. On the other hand, petrol stations are usually located close to residential or commercial areas. During an explosion event, not only the process facilities would be impaired, but also human casualties may be ensued in the neighbouring populated areas. Evaluation of gas explosion risks is complicated with the consideration of multifactors, e.g. leak severity, vent condition, structural complexity; multi-consequences, i.e. building damage, human casualties, environmental contamination; and the complex interrelationships. Therefore, to model the complicated mechanisms of gas explosion events and consequent losses, a Bayesian-network-based (BN) QRA is developed in this study to apply the exceptional strength of the BN in addressing the complex interrelationships between risk factors. In the process industry, BNs have been increasingly applied for risk and safety assessments. Pasman and Rogers (2013) integrated BN into a layer of protection analysis (LOPA) for gas risk analysis at a hydrogen tank station. It was found that the BN had great potential in describing scenarios, processing uncertainties and supporting decision making effectively. Haugom and Friis-Hansen (2011) built a BN for gas risks at a hydrogen refuelling station to consider gas leak, jet fire and loss of life. They concluded that a BN has greater efficiency and flexibility in analysing the dependence between the different variables than a standard ET does. Khakzad, © Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_9

191

192

9 Bayesian Network Analysis of Explosion Events at Petrol Stations

Khan, and Amyotte (2011) adopted BN to conduct safety analysis of a feeding control system that transfers propane from a propane evaporator to a scrubbing column. It proved that a BN is more proficient than a traditional FT model for complicated systems. Zarei, Azadeh, Khakzad, Aliabadi, and Mohammadfam (2017) developed a BN-based dynamic and comprehensive QRA (DCQRA) method for risk modelling and safety assessment of natural gas stations. Xin, Khan, and Ahmed (2017) used the BN method to implement real-time dynamic hazard identification and scenario mapping for further risk analysis in process industry. Norazahar, Khan, Veitch, and MacKinnon (2017) proposed a method to identify critical human and organisational factors in the escape, evacuation and rescue systems and applied BN to assess the criticality of those factors. Wu, Zhou, Xu, and Wu (2017) employed the BN and the Dempster–Shafer evidence theory to probabilistically analyse natural gas pipeline network accidents. On the other hand, the proficiency of BN modelling is limited by data shortage for quantification. To improve the reliability and efficiency of the proposed method, three types of data are included in the current study: practical information, subjective logical judgements and computational simulation derivations. Practical information includes historical data of basic risk factors, such as probabilities of leak scenarios, ignition sources and specific information of the target petrol stations. Subjective logical judgements are adopted when no effective data can be derived. Such judgements help to decide conditional dependencies when logic between nodes is simple and straightforward. Numerical simulation results are also applied to implement detailed explosion assessment to provide data for BN quantification. PHAST (DNV GL, 2016) is selected to simulate explosions in this study as it offers an efficient and reliable explosion assessment. Potential-explosion-induced human losses are also assessed in this chapter. Human safety of a petrol station relates to two groups of people, those inside the station, e.g. staffs and customers, and people outside the station, such as nearby residents or passers-by. Human safety analysis is restrained to people inside the station to eliminate the comparatively more intractable uncertainty from the people outside the station. The risk of explosion accidents during the refuelling process from a fuel tanker to a petrol station is explored. This scenario is selected because a tanker usually stores a large quantity of flammable materials, which may cause much more significant consequences if the tanker explodes. A domino effect may be implicated by the initial fire or explosion accidents. Both release-induced direct explosion and fire-induced tanker explosion are considered.

9.2 Methodology Implementation of the proposed method consists of the following steps: • Modelling to deduce BN based on risk factors and their interrelationships; • Quantification to collect data to quantify the established BN; • Calculation to derive the probabilities of target nodes of BN.

9.2 Methodology

193

Fig. 9.1 Proposed BN for explosion risks (permission from Elsevier)

9.2.1 Bayesian Network Modelling A Bayesian network is an illustrative diagram which features nodes and links of conditional probabilities. As shown in Fig. 9.1, a Bayesian network is proposed in this chapter to evaluate risks of explosions and losses of life when a leak occurs at petrol stations. The network consists of 14 nodes and 18 links which indicate risks of leaks, explosions and potential consequences. Nodes and states are listed in Table 9.1.

9.2.2 Quantification of BN In the quantification of the BN, the probabilities of the basic nodes and the conditional probabilities of the interrelationship between nodes should be specified. Since quantification based on historical statistical data is most convenient and reliable, in this study, a total of 27 cases of explosion accidents at service stations are selected and recorded as quantification data, which are listed in Table 9.2.

194

9 Bayesian Network Analysis of Explosion Events at Petrol Stations

Table 9.1 Nodes and states of proposed BN Node

States

No.

Name

No.

States

A

Release scenario

5

Overfilling, misconnection, hose rupture, coupling failure, vapour recovery

B

Leak severity

3

Major, moderate, minor

C

Ignition source

8

Smoking, arcing, hot ember or ash, spark or flame, unclassified heat, static discharge, friction, lightning

D

Ignition

2

Yes, no

E

Initial explosion

4

Major, medium, minor, no

F

Leak rate

3

Major, moderate, minor

G

Fire

4

Major, medium, minor, no

H

Tanker explosion

2

Yes, no

I

Building damage

4

Major, medium, minor, no

J

Evacuation time

3

Sufficient, short, little

K

Evacuation

3

Evacuated, shelter in store, failed

L

Time of day

5

8:00–9:00 a.m., 9:00 a.m.–4:00 p.m., 4:00–5:00 p.m., 5:00–10:00 p.m., 10:00 p.m.–8:00 a.m.

M

No. of people in station

3

High, medium, low

N

Human loss

4

Major, medium, minor, no

From Table 9.2, for most available cases, only fatalities, injuries and/or estimated economic losses are reported. Quantifying the interrelationships between nodes by using historical data only is not feasible. Thus, numerical simulations and logical judgements are also resorted to in this study to address the limitation of statistic data. DNV PHAST is adopted to quantify the interrelationships between leaks and consequent explosions. Running PHAST analysis has four steps: inputting data, building model, calculating and outputting result, which will be briefly introduced below. More details about how to use PHAST can be found in Chap. 3. Besides the numerical simulations, logical judgements are also indispensable to quantify the interrelationships of the proposed BN. If the logical relationship between nodes is straight forward, subjective judgements are sufficiently reliable. On the other hand, such quantification requires regular examination if the site condition changes. Adjustment is required to ensure that the logical relationship is updated to the date. If logical relationships are complicated and uncertain, a confidence-based method could be applied to reduce the uncertainties in the subjective judgements (Huang, Ma, Li, & Hao, 2015).

Year

2015

2015

2016

2015

2009

2015

2016

2016

2016

2015

No.

1

2

3

4

5

6

7

8

9

10

2100

1330

1700

1825

1100

1345

1240

1751

2200

1800

Time

Cobar, NSW

Kuala Lumpur, Malaysia

Port-au-Prince, Haiti

Kizlyar, Russia

Birmingham, UK

Maddington, Australia

Al-Ghubra, Oman

Kaduna, Nigeria

Accra, Ghana

Riverton, Australia

Location

Table 9.2 Recorded explosion accidents

Truck explosion by unidentified cause

Explosion caused by cell phone while filling the car

Fuel tanker caught fire and exploded

Explosion occurred when fuel truck is discharged LPG into tank

Electric fan heater in retail area heated up a gas cylinder

Tanker caught fire and exploded during discharging

Car caught fire while fuelling

Fire ignited and escalated over 3 h while tanker was discharged

Flood-swept fuel to nearby fire caused explosion

Explosion of gas cylinders caused major fire

Description

7

3

152

Death

1

30

40

4

2

Injury

Daily Liberal (2015)

FMT News (2016)

Yahoo News (2016)

RT News (2016)

BBC News (2015)

(continued)

Department of Mines and Petroleum (2009a, 2009b)

Times of Oman (2015)

Daily Post (2016)

Asumadu-Sarkodie et al. (2015)

ABC News (2015)

Reference

9.2 Methodology 195

Year

2016

2015

1989

1993

1997

1985

2003

2004

2004

2004

1958

No.

11

12

13

14

15

16

17

18

19

20

21

1100

1500

1600

1200

1200

1205

1540

Time

Table 9.2 (continued)

Paris, France

Aubigny-sur-Nere, France

Valleiry, France

Montlucon, France

Les Cheres, France

Compiegne, France

Annecy, France

La Gueriniere, France

Laval, France

Vienna, Austria

Southern Khatlon, Tajikistan

Location

UST leaked and exploded a few hours later, caused by spark from electrical switch

Explosion caused by vapour leak from UST

Flash occurred while filling car

New UST exploded while filling

Car crashed into dispenser

Explosion caused by worker-lit cigarette

Explosion caused by welding in tank manhole

Explosion while discharging caused by an electrical switch ignition

UST exploded when degassing and cleaning

Explosion caused by leak of acetylene

Explosion

Description

17

1

1

1

1

2

1

Death

30

1

2

1

1

2

1

17

Injury

(continued)

ARIA Technologies (2009)

ARIA Technologies (2009)

ARIA Technologies (2009)

ARIA Technologies (2009)

ARIA Technologies (2009)

ARIA Technologies (2009)

ARIA Technologies (2009)

ARIA Technologies (2009)

ARIA Technologies (2009)

WJLA (2015)

Asia-Plus (2016)

Reference

196 9 Bayesian Network Analysis of Explosion Events at Petrol Stations

Year

2007

1999

1991

2014

2016

2007

No.

22

23

24

25

26

27

0800

1515

0230

0100

0200

Time

Table 9.2 (continued)

Shanghai, China

Kuala Lumpur, Malaysia

New Orleans, USA

South Carolina, USA

Mississippi, USA

Sotteville-les-Rouen, France

Location

Exploded during maintenance work

8500 gallons of tanker collided with car and crashed into petrol station concrete pillar

Car crashed into pipes of two 4000 gallons above ground fuel tanks; fire lasted an hour before it was put out

Tanker truck overfilled and spilled 2839 L of fuel. The spill spread outside the catchment and reached an adjacent road and was then ignited by unknown ignition sources

Car fire triggered an explosion of 16 LPG cylinders, which lasted 4 h at the service station

Description

4

1

1

1

5

Death

40

Injury

ABC News (2007)

Channel News Asia (2016)

FOX8 (2014)

Evarts (2011)

Evarts (2011)

ARIA Technologies (2009)

Reference

9.2 Methodology 197

198

9 Bayesian Network Analysis of Explosion Events at Petrol Stations

9.2.3 Calculation of BN The subnetwork of human loss is taken as an illustrative example to demonstrate the BN calculation. As shown in Fig. 9.2, this subnetwork contains four nodes and three links. The severity of human loss (node N) is governed by the severity of the initial explosion (node E) and the number of people inside the petrol station (node M), which varies at different times of the day (node L). The prior probability of explosion loads can be calculated to be P(N = major) =

5  3 4  

P(N = major, E = E i , L = L j , M = Mk ) (9.1)

i=1 j=1 k=1

where P = probability; N = human loss; E = initial explosion; E i = states of node E; L = time of the day; L j = states of node L; M = number of people in station; M k = states of node M (see Table 9.1). Based on the theorem of the Bayesian network (Nielsen & Jensen, 2009), the joint probability can be derived as P(x1 , . . . , xn ) =

n 

P(xi |Pa(xi ))

(9.2)

i=1

where Pa(xi ) is the parent set of xi . Equation (9.2) remains an unconditional probability of P(xi ) if there are no parents of xi . In this subnetwork, the node of human loss has parents of evacuation and jet fire; the node of evacuation has a parent of evacuation time; and the node evacuation time has a parent of jet fire. Therefore, the following equation can be deduced

Fig. 9.2 Example of Bayesian subnetwork

9.2 Methodology

199

  P N = major, E = E i , L = L j , M = Mk     = P(M = major|L = L k , E = E i ) × P M = Mk |L = L j × P L = L j × P(E = E i )

(9.3)

where the conditional probabilities are determined by the interrelationship quantification.

9.3 Case Study A case study is conducted to illustrate the proposed method and explain the quantification process in detail. Figure 9.3 shows an example GIS map of a petrol station in Australia. An explosion accident occurred in this petrol station when a tanker was refuelling the station. This site is selected as an example to demonstrate the quantification procedure of the proposed method. The refuelling area is shaded in Fig. 9.3, and the bund size is about 4 m × 6 m. In this case study, each node of the proposed BN will be described and quantified.

Fig. 9.3 GIS map of analysis area

200

9 Bayesian Network Analysis of Explosion Events at Petrol Stations

9.3.1 Quantification of BN 9.3.1.1

Quantification of Release Scenarios

Table 9.3 shows 18 cases of previous spill incidents during the refuelling processes from fuel tankers to petrol stations. These cases were collected from dangerous goods incident reports of Western Australia between 1996 and 2008 (Department of Consumer and Employment Protection, 2000, 2002, 2004, 2006; Department of Minerals and Energy, 1996, 1998–2000; Department of Mines and Petroleum, 2008). Table 9.4 shows the probability of each scenario based on the 18 cases from Table 9.3. Since this case study is an illustrative example of the proposed method, only 18 cases are selected to quantify leak scenarios. It should be mentioned that if more data can be collected and adopted, the more accurate and reliable evaluation of the BN is expected.

Table 9.3 Cases of spills No.

Year

Location

Volume of spill

Cause

1

1996

Dardanup

40

Misconnection

2

1998

Coolgardie

200

Overfilling

3

1998

Bassendean

22

Misconnection

4

1998

Kalgoorlie

315

Coupling failure

5

1998

Jarrahdale

70

Coupling failure

6

1998

Moora

80

Misconnection

7

1999

Swanbourne

50

Vapour recovery

8

1999

Upper Swan

50

Misconnection

9

2000

Australind

50

Overfilling

10

2000

Geraldton

300

Hose rupture

11

2002

Mt. Pleasant

No record

Overfilling

12

2002

Dampier Port

No record

Overfilling

13

2002

North Dandalup

750

Misconnection

14

2004

Canning Vale

20

Misconnection

15

2004

Kwinana

5000

Hose rupture

16

2006

Rivervale

No record

Overfilling

17

2006

Christmas Island

400

Overfilling

18

2008

Collie

8400

Hose rupture

9.3 Case Study

201

Table 9.4 Probabilities of release scenarios Scenario

Description

Number

Probability (%)

Overfilling

Gauging error, driver over fill underground storage tank

6

33.33

Misconnection Driver error

6

33.33

Hose rupture

Mechanical failure of the unloading hose

3

16.67

Coupling failure

Result in disconnection of unloading hose

2

11.11

Vapour recovery

Stage 1 vapour recovery connection propped open

1

5.56

Fig. 9.4 Example of catchment at petrol station

9.3.1.2

Quantification of Leak Severity

The probability-related leak severities are classified into three categories: outside catchment, inside catchment and inside cesspit. Figure 9.4 shows an example of a catchment of fill points. Major spill is identified when the spill reaches outside the catchment, while minor spill is specified when the spill can be held within the cesspit. The size of the catchment at the target site is approximately 6 m × 4 m based on the measurement from the GIS map, and the height of the bund is about 50 mm.

202

9 Bayesian Network Analysis of Explosion Events at Petrol Stations

Table 9.5 Classification of severity of spills Volume of spill

Scenario

Height (mm)

40

Misconnection

1.67

200

Overfilling

8.33

22

Misconnection

0.92

315

Coupling failure

13.13

70

Coupling failure

2.92

✓

80

Misconnection

3.33

✓

50

Vapour recovery

2.08

✓

50

Misconnection

2.08

✓

50

Overfilling

2.08

✓

300

Hose rupture

12.50

✓

750

Misconnection

31.25

✓

20

Misconnection

0.83

5000

Hose rupture

400

Overfilling

8400

Hose rupture

208.33

Major

Minor ✓

✓ ✓ ✓

✓ ✓ ✓

16.67 350.00

Medium

✓

The size of a typical cesspit is about 200 L, and the specific size of the cesspit at the target site cannot be determined. Therefore, for a conservative estimate, the cesspit is assumed to be able to contain 120 L of spill, which is about 10% of the total volume of the catchment. Consequently, the severity of the 15 cases of spills can be determined as shown in Table 9.5.

9.3.1.3

Quantification of Ignition Sources

For the basic nodes of ignition sources, the most common heat sources ignited at the service stations in the USA from 2004 to 2008 were recorded in Table 9.6 (Evarts, 2011). The results indicate that smoking and heat generated from power equipment are the most probable sources of ignition.

9.3.1.4

Quantification of Ignition

The number of ignition sources is adopted to determine the probability of ignitions, which depends on the size of spills. If the spill spreads outside the catchment, the chance of reaching ignition sources is higher. Therefore, the incident of the spill spreading outside the catchment is assigned with all the possible ignition sources, while the ignition sources of the spill incident inside the catchment may be triggered

9.3 Case Study

203

Table 9.6 Heat sources at service station ignited from 2004 to 2008 Ignition source

Abbreviation

Cases

Probability

Smoking

S

160

0.213

Arcing

A

90

0.120

Hot ember or ash

HE

140

0.187

Spark or flame from operating equipment

SF

70

0.093

Unclassified heat from powered equipment

UH

180

0.240

Static discharge

SD

40

0.053

Heat or spark from friction

F

60

0.080

Lightning

L

10

0.013

only by S, A, HE, SD or L. Thus, S, SD and L are considered for minor spills because only these sources are accessible to spills inside the cesspit. According to the Purple Book (Uijt & Ale, 2005), the probability of ignition is 6.5% for K1-liquid. K1-liquid refers to the flammable liquid, which has a flash point less than 21 °C and a vapour pressure at 50 °C is less than 1.35 bar. The gasoline has a flash point at −43 °C and a vapour pressure of approximately 0.79 bar at 51 °C. Therefore, when ignition sources are located inside the spill range, the probability of ignition is determined as 6.5%.

9.3.1.5

Quantification of Initial Explosion

Release-related initial explosions are simulated by DNV PHAST subject to 15 leak severities. In this study, wind effects, such as wind directions and wind speeds, are not considered. Therefore, the explosive cloud is assumed to spread by diffusion only from the leak point. The Baker–Strehlow–Tang model is selected in conducting the explosion analysis. A medium level of obstacle density and fuel reactivity is assigned based on the specific condition of the target station. Table 9.7 lists the results of PHAST analyses. Figure 9.5 shows an example of a PHAST output of an explosion range of 0.689 bar with 400 L of petrol spills. To determine the severity of an explosion, an overpressure of 0.689 bar is used as an indicator, which is the vault value for the overpressure starts to inflict direct human death or severe injury (Lobato et al., 2009). If an area with overpressure higher than 0.689 bar reaches the nearest dispenser (about 10 m away from the spill) and the store area (20 m away from the spill), the severity is identified as medium and major, respectively. The simulated severities of explosions are classified as in Table 9.7.

204

9 Bayesian Network Analysis of Explosion Events at Petrol Stations

Table 9.7 Classification of severities of initial explosion Spill severity (L)

Distance of 0.689 bar (m)

Explosion severity Major

Medium

Minor ✓

40

7.8

200

13.33

22

6.39

315

15.51

70

9.4

✓

80

9.82

✓

50

8.4

✓

50

8.4

✓

50

8.4

✓

300

15.14

750

20.55

20

6.19

5000

38.82

400

16.67

8400

46.35

9.3.1.6

✓ ✓ ✓

✓ ✓ ✓ ✓ ✓ ✓

Quantification of Leak Rate

Leak rate should be specified for a jet fire analysis. It can be calculated based on the leak volume and time. Since no specific data of leak time for each case can be identified, leak times of 60 s and 300 s are assumed to calculate the leak rate according to the rough estimation that the driver has a react time in the range of 60 s–300 s to stop the spill in view of that the driver may not observe the leak promptly. For the 15 cases with recorded spill volumes, 300 s instead of 60 s is specified because, normally, only a severe spill with a longer leak time is recorded. On the other hand, 60 s is also applied to serve as another set of data of leak rates to conduct a more conservative evaluation. Based on the assumptions of 60 and 300 s of the leak time, 30 leak rates can be calculated as listed in Table 9.8. According to HSE (2015), the severity of the leak rate can be considered major when it is more than 10 kg/s for a liquid and minor when it is less than 0.2 kg/s. The density of normal gasoline is approximately 0.71 kg/l. Accordingly, Table 9.9 gives the severities of leak rates.

9.3.1.7

Quantification of Jet Fire

In this study, fire severity is assumed to affect only the probabilities of tanker explosion. Evacuation conditions and fire-induced fatality or injury are not considered. To

9.3 Case Study

205

Fig. 9.5 GIS output of explosion range of 0.689 bar with 400 L of petrol spills Table 9.8 Leak rates

Volume of spill (L)

Leak rate (L/s) 60 s

300 s

40

0.67

0.13

200

3.33

0.67

22

0.37

0.07

315

5.25

1.05

70

1.17

0.23

80

1.33

0.27

50

0.83

0.17

50

0.83

0.17

50

0.83

0.17

300

5.00

1.00

750

12.50

2.50

20

0.33

0.07

83.33

16.67

5000 400 8400

6.67

1.33

140.00

28.00

206

9 Bayesian Network Analysis of Explosion Events at Petrol Stations

Table 9.9 Severities of leak rates

Table 9.10 Severities of jet fires

Leak rate severity

Leak rate (l/s)

Major

14.08–

Moderate

0.28–14.08

Minor

0–0.28

Volume of spill

Jet fire distance 60 s

Number of cases 4 18 8

300 s

40

23.49

11.27

200

48.88

23.49

22

17.89

8.58

315

60.1

28.89

70

30.31

14.56

80

32.22

15.47

50

26.01

12.48

50

26.01

12.48

50

26.01

12.48

300

58.78

28.26

750

89.11

42.89

20

17.13

8.21

210.28

101.52

5000 400

66.99

32.22

8400

265.71

128.43

derive the interrelationship between leak rates and jet fires, a numerical simulation using PHAST is implemented. Jet fires are calculated and quantified based on 30 leak rates as listed in Table 9.8. Figure 9.6 shows the PHAST GIS output of the effect zone of a thermal radiation of 2 kW/m2 when the total volume of release is 20 L and the leak time is 60 s. Minor injury, major injury or fatality may be incurred based on the distance from the fire centre to the edge of the circled region, as shown in Fig. 9.6. According to the Federal Emergency Management Agency (1990), people may suffer severe pain or second-degree burn if they are exposed to a thermal radiation of 2 kW/m2 for over 45 s or 187 s, respectively. People who stay outside the circled region may have around 1 min of evacuation time, which is assumed to be sufficient for people to seek shelter or escape. Therefore, the region outside the circled area is considered as safety zone. Details of the derived spreading distances of jet fires with a 2 kW/m2 thermal radiation based on 30 release rates is listed in Table 9.10. The severities of fire are deduced based on the sizes of specific sites. In this case study, the size of the site is 40 m × 32 m and the refuel points are located at the edge of the site. Accordingly, the classification of the severity of the jet fire is shown

9.3 Case Study

207

Fig. 9.6 Example of GIS output with thermal radiation of 2 kW/m2 Table 9.11 Classification of severity in view of tanker fire distance

Jet fire severity

Fire distance (m)

No. of cases

Major

>34

10

Moderate

>16

12

Minor

35 kPa)

Medium (17–35 kPa)

Minor (1.01 bar, probable death due to lung haemorrhage

F

Building type

4

Residential; tank; process facilities; no building

G

Population

4

Large; medium; small; little

H

Building damage

4

Major; medium; minor; no damage

I

Human loss

4

Major; medium; minor; little

222

10 Grid-Based Risk Screening for Explosion Accidents at Large …

Wind Direction

Release Severity

Wind Speed

Congestion

Building Type

Explosion Loads

Building Damage

Population

Human Loss

Fig. 10.2 Proposed BN for explosion risks (permission from Elsevier)

10.3.1 Quantification of Bayesian Network As mentioned in the previous chapter, the quantification of a BN can be divided into two parts: deriving the probabilities of the basic nodes and defining the conditional probabilities of the interrelationship between the nodes. Quantification of Basic Nodes The proposed BN has five basic nodes: wind direction, wind speed, release severity, building type and population. Information about wind direction and wind speed can be found from local weather data resources online. As for the release severity, hydrocarbon release data from the Health and Safety Executive (HSE) annual report (2015) is adopted. Table 10.2 shows the HSE recorded number of accidents from 2006 to 2015 and the summary of the probability of each state. The basic nodes of site information for each grid, such as building damage and population, depend on the specific condition within the grid area and are specified by subjective judgments. Quantification of Interrelationships For quantification of interrelationships, the proposed BN is divided into two subnetworks: explosion loads including nodes A, B, C, D, E, building damage and human loss including nodes E, F, G, H and I. Numerical simulation and logical judgments are used to quantifying the interrelationships between nodes.

10.3 Bayesian Network Modelling

223

Table 10.2 HSE data of hydrocarbon releases Year

06

07

08

09

10

11

12

13

14

15

Probability (%)

Minor

113

110

93

95

109

82

58

70

47

49

58.33

73

71

52

81

73

57

39

42

30

32

38.84

4

4

2

3

4

3

8

6

3

3

2.83

Significant Major

10.3.2 Calculation of Bayesian Network Calculation of sub-network of explosion loads Figure 10.3 shows the sub-network of the explosion loads. The three basic nodes are wind direction, wind speed and release severity. The release severity and wind speed define cloud sizes, while the location is decided by the wind direction. The condition of congestion can then be deduced based on the cloud size and location. The frequency of each explosion load level is calculated by the release severity and congestion conditions. This sub-network contains five nodes and five links. The prior probability of explosion loads can be calculated as follows, P(E = a) =

3  3  3 4  

P(E = a, A = Ai , B = B j , C = Ck , D = Dh ),

i=1 j=1 k=1 h=1

(10.1)

Wind Direction

Wind Speed

Release Severity

Congestion

Explosion Loads

Fig. 10.3 Sub-network for estimating explosion loads (permission from Elsevier)

224

10 Grid-Based Risk Screening for Explosion Accidents at Large …

Building Type

Explosion Loads

Building Damage

Population

Human Loss

Fig. 10.4 Sub-network for estimating building damage and human loss (permission from Elsevier)

where P is the probability; E stands for the explosion loads; a is the state “a” of node E; A represents the wind direction; Ai denotes the states of node A; B is the wind speed; Bj represents the states of node B; C is the release severity; Ck stands for the states of node C; D denotes the congestion; and Dh is the states of node D (see Table 10.1). Based on the theorem of BN (Nielsen & Jensen, 2009), the joint probability can be derived as below: P(x1 , . . . , xn ) =

n 

P(xi |Pa(xi ))

(10.2)

i=1

whereas Pa(xi ) is the parent set of xi . The function remains an unconditional probability of P(xi ) if there are no parents of xi . In this sub-network, the node of congestion has parents of wind direction, wind speed and release severity, while the node of explosion loads has parents of congestion and release severity. Accordingly, the following equations should be applied:   P E = A, A = Ai , B = B j , C = Ck , D = Dh = P(E = A|D = Dh , C = Ck )   × P D = Dh |C = Ck , B = B j , A = Ai   × P(C = Ck ) × P B = B j × P(A = Ai ). (10.3) Calculation of sub-network of building damage and human loss The sub-network for the two consequences—building damage and human loss—is illustrated in Fig. 10.4. For building damage, only building type is applied as a basic factor. Different types of buildings provide different resistant levels to the explosion overpressures. The total human loss is then deduced by explosion loads, building damage and population within each grid. Similar to the sub-network of explosion loads, this network also has five nodes and five links. Building damage has parents of building type and explosion loads. The parents for human loss are explosion loads, building damage and population. Therefore, human loss can be calculated by

10.3 Bayesian Network Modelling Fig. 10.5 Typical BN for estimating building damage (permission from Elsevier)

225

Building Type

Explosion Loads

Building Damage

P(K = Major) =

4  4  4 5  

P(K = Major, E = Ei , F = F j , G = Gk , H = Hh ),

i=1 j=1 k=1 h=1





(10.4)

P K = Major, E = Ei , F = F j , G = Gk , H = Hh   = P(K = Major|H = Hh , G = Gk , E = Ei ) × P H = Hh |F = F j , E = Ei   × P(G = Gk ) × P F = F j × P(E = Ei ), (10.5) where K is the human loss; E denotes the explosion loads; Ei stands for the states of node E; F represents the building type; Fj is the states of node F; G stands for the population; Gk represents the states of node G; H is the building damage; and Hh denotes the states of node H (see Table 10.1).

10.3.3 Matrix Calculation and Result Display For convenient calculation, the aforementioned equations are synchronized into a matrices form. All the data from each node and interrelationships fit into the matrices as the corresponding items. A MATLAB script is written to conduct the calculations to give the value to each grid. For example, Fig. 10.5 shows a simple illustrative BN of building damage. The probability of major building damage can be calculated by the following equation: P(H = Major) =

4 5     P H = H1 , E = Ei , F = F j i=1 j=1

=

5  4      P H1 |Ei , F j × P(Ei ) × P F j i=1 j=1

= P(H1 |E1 , F1 ) × P(E1 ) × P(F1 ) + P(H1 |E1 , F2 ) × P(E1 ) × P(F2 ) + · · · + P(H1 |E5 , F4 ) × P(E5 ) × P(F4 ). (10.6)

226

10 Grid-Based Risk Screening for Explosion Accidents at Large …

Fig. 10.6 Example of 3D result histogram presentation (permission from Elsevier)

The MATLAB script to transfer this equation to the matrix calculation is written as:   a = P(H1 |E1 , F1 ), P(H1 |E1 , F2 ), . . . , P(H1 |E5 , F4 )] %P H1 |Ei , F j %

(10.7)

b = [P(E1 ), P(E2 ), . . . , P(E5 )] %P(Ei )%

(10.8)

  c = [P(F1 ), P(F2 ), P(F3 ), P(F4 )] %P F j %

(10.9)

   P(H = Major) = sum a  · reshape(repmat(b, 4, 1), 20, 1) · repmat c , 5, 1 (10.10) For each grid, a “for” loop is used to conduct the calculation automatically and the result is depicted as a 3D bar plot in Fig. 10.6. In Fig. 10.6, the height of each bar represents the probability of related states at each grid and a total risk profiling of target area is formed by the combination of risks from all of the grids. Such a result display provides a clear risk indicator for each local area and protection measures can be specified based on the risk map.

10.4 Case Study A case study is conducted to demonstrate the procedure to implement the proposed method. Figure 10.7 shows a GIS map of a gas refinery factory with the adopted grids. This factory is surrounded by a residential area. From Fig. 10.7, the closest

10.4 Case Study

227

Fig. 10.7 GIS map of target area (permission from Elsevier) Table 10.3 Probabilities of wind direction and wind speed Wind direction

East

South

West

North

Wind speed

3 m/s

1.5 m/s 0.1/s

Probability

0.203

0.284

0.284

0.229

Probability

0.698

0.2

0.102

residential building is located only about 100–200 m from a gas storage tank. Within this distance, consequences may be detrimental if an explosion occurs. A 50 m * 50 m grid size over a domain range of 2 km * 2 km is selected based on the result of mesh convergence (see Sect. 10.3.3). The BN model introduced in Sect. 10.2 is applied.

10.4.1 Quantification of Bayesian Network Quantification of basic nodes As mentioned above, there are five basic nodes in the proposed BN. Data on wind direction and wind speed are collected from a website that records local weather data daily. All the information since 2015 are collected and analysed. As a result, four wind directions and three wind speeds are considered in this study with their probabilities in 2015 are listed in Table 10.3. As to the release severity, the probability of each state from the HSE database can be found in Table 10.2.

228

10 Grid-Based Risk Screening for Explosion Accidents at Large …

(a) Population

(b) Building types

Fig. 10.8 Site information (permission from Elsevier) Table 10.4 Input data for PHAST analysis

Material

Hydrocarbon

Flammable mass in cloud

300 kg; 30 kg; 3 kg

Wind direction

East; North; West; South

Wind speed

0.1 m/s; 1.5 m/s; 3 m/s

Congestion

High; Medium; Low

Explosion load

A: 70 kPa+; B:20–70 kPa; C: 2–20 kPa; D: 0–2 kPa

Site information is depicted in Fig. 10.8. Figure 10.8a shows population information with red, yellow, green and blue representing large, medium, small and little populations, respectively. Similarly, Fig. 10.8b describes building type with red, yellow, green and blue indicating residential buildings, tanks, process facilities and no buildings, respectively. Microsoft Excel is then applied to read all the colours and output numerical data for further analysis. Quantification of interrelationships In quantifying interrelationships, for the sub-network of the explosion loads, DNV PHAST is applied to calculate explosion loads under different conditions and provide data for BN calculation. The leak point is set at the tank that is nearest to the residential area to be on the safe side. Huang et al. (2016) developed a multi-level explosion risk analysis method to screen the whole site to qualitatively determine the most dangerous leak source. Quantitative analysis is then conducted based on the results of risk screening. Table 10.4 gives the input data for PHAST analysis.

10.4 Case Study

229

Fig. 10.9 GIS output of cloud formation (permission from Elsevier)

PHAST calculations give GIS output of gas cloud dispersion. Consequently, the congestion level can be deduced based on the cloud size and location. Figure 10.9 shows an example of GIS output of cloud formation. Based on the cloud size and location from Fig. 10.9, the congestion level for this scenario is identified as high. With the specification of the congestion, the final GIS output of explosion loads can be depicted, and different levels of load can be specified for each grid as shown in Fig. 10.10. To quantify the sub-network of building damage and human loss, logical judgment is made based on the specific site information. For instance, the interrelationship among building type, explosion loads and building damage are identified as shown in Table 10.5. For different facilities and buildings, the standards for resistance to dynamic blasting loads are different. Generally, under a blast load of 17 bar (Lobato et al., 2009), 50% of the brickwork of houses will be destroyed and steel frame building will be distorted. Therefore, for residential buildings and process facilities, major damage is identified when explosion loads are larger than level C. Storage tanks normally have higher resistance levels than those of the residential buildings and process facilities. Thus, medium damage is defined for storage tanks under level

230

10 Grid-Based Risk Screening for Explosion Accidents at Large …

Fig. 10.10 GIS output of explosion loads (permission from Elsevier)

C blasting loads. Similarly, the interrelationship between basic nodes and human loss is quantified. The table of logical judgments between basic nodes and human loss is not presented because it is too large and complicated to be described in detail.

10.4.2 Results and Discussion Based on the equations in Sect. 2.4 and network quantification, the probability of each state of explosion loads, building damage and human loss can be derived and output as a 3D risk map of the risk level for each grid as in Figs. 10.11, 10.12, and 10.13, respectively. Figure 10.11 shows the risk map of explosion loads based on the five states. From Fig. 10.11e, overpressures may cause direct human death and complete building damage within a few grids around the explosion centre. The level “B” blast load has a high probability of occurring within a radius of 500 m from the explosion centre, while beyond 500 m, the probability of level “B” or higher overpressures is less than half. A “safety” zone with a level “A” explosion loads is defined as approximately 1800 m away from the explosion centre and beyond.

10.4 Case Study

231

(a) State “A”

(b) State “B”

(c) State “C”

(d) State “D”

(e) State “E” Fig. 10.11 Risk map of explosion loads (permission from Elsevier)

232

10 Grid-Based Risk Screening for Explosion Accidents at Large …

Table 10.5 Interrelationship between nodes E, F and H

Residential

E

Major

Residential

D

Major

Residential

C

Major

Residential

B

Medium

Residential

A

Minor

Tank

E

Major

Tank

D

Major

Tank

C

Medium

Tank

B

Minor

Tank

A

Minor

Process facilities

E

Major

Process facilities

D

Major

Process facilities

C

Major

Process facilities

B

Medium

Process facilities

A

Minor

No structures

E

No

No structures

D

No

No structures

C

No

No structures

B

No

No structures

A

No

Figure 10.12 shows building damage. From Fig. 10.12, there is high probability of major building damage at the process site and the residential areas that are close to the explosion centre. Therefore, the buildings within this area should be reinforced to resist high blast overpressures. Figure 10.12b indicates the area with probable medium building damage. For the region with probabilities over 50%, the buildings should be examined and strengthened. Figure 10.13 gives the risks of human loss. From Fig. 10.13a, the most dangerous region for human safety is located in the residential area close to the explosion centre because of the large population are expected within that area. There is a gap between the residential area and the factory with a very low chance for major human loss because no structures are present in that area. Therefore, if projectiles and fires do not present in the explosion, evacuation to this area without buildings is probably a better choice than sheltering inside the buildings within the dangerous region. Figure 10.13b shows that even far from the explosion centre, there is still a chance for injury and medium human loss. The main reason could be that the storage tank is located close to the residential area and partial building damage may ensue within 1800 m under level “B” explosion loads. Therefore, appropriate design of protection barriers and structure strengthening of buildings are required for human safety considerations.

10.4 Case Study

233

(a) State “Major”

(b) State “Medium”

(c) State “Minor”

(d) State “No”

Fig. 10.12 Risk map of building damage (permission from Elsevier)

10.4.3 Mesh Convergence A mesh convergence study is conducted to determine an optimal balance between accuracy and computational intensity. In the study of mesh convergence, all the information from each grid are synchronized as the total input into the BN. Four sizes of grid, i.e. 200, 100, 50, 25 m, are tested and the results are listed in Table 10.6. Figure 10.14 shows the results for explosion loads, building damage and human loss from different grid sizes. From Fig. 10.14a, there is not much difference among the four grid sizes for the probabilities of explosion load levels. However, for building damage and human loss, the probabilities of each state show a large difference until the grid size is reduced to 50 m. When the grid size is reduced from 50 m to 25 m, the difference in probability is approximately less than 5%. Therefore, a grid size of 50 m * 50 m is considered to be the convergence size and applied in this study.

234

10 Grid-Based Risk Screening for Explosion Accidents at Large …

(a) State “Major”

(b) State “Medium”

(c) State “Minor”

(d) State “Little”

Fig. 10.13 Risk map of human loss (permission from Elsevier) Table 10.6 Results with different mesh sizes

Grid Size Load

Building Damage

Human Loss

200 m

100 m

50 m

25 m

A

0.408

0.434

0.45

0.453

B

0.464

0.464

0.457

0.458

C

0.0514

0.046

0.0426

0.0403

D

0.0315

0.0269

0.0252

0.0247

E

0.045

0.0291

0.0255

0.0245

Major

0.118

0.0954

0.0608

0.0539

Medium

0.122

0.084

0.0672

0.0635

Minor

0.33

0.248

0.178

0.165

No

0.43

0.572

0.694

0.718

Major

0.134

0.094

0.0611

0.0574

Medium

0.167

0.119

0.086

0.0803

Minor

0.251

0.193

0.147

0.1356

Little

0.448

0.594

0.705

0.727

Probability

10.4 Case Study

235

0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

10*10(200m) 20*20(100m) 40*40(50m) 80*80(25m)

A

B

C

D

E

States of Explosion Loads

(a) Probabilities of explosion loads 0.8

Probability

0.7

10*10(200m) 20*20(100m)

0.6

40*40(50m)

0.5

80*80(25m)

0.4 0.3 0.2 0.1 0 Major

Medium

Minor

No

States of Building Damage

(b) Probabilities of building damage 0.8 0.7

Probability

0.6 0.5 0.4

10*10(200m) 20*20(100m) 40*40(50m) 80*80(25m)

0.3 0.2 0.1 0 Major

Medium

Minor

States of Human Loss

(c) Probabilities of human loss Fig. 10.14 Results with different mesh sizes (permission from Elsevier)

Little

236

10 Grid-Based Risk Screening for Explosion Accidents at Large …

10.5 Summary A detailed grid-based risk mapping method for explosion events is proposed in this chapter. This method uses a Bayesian network (BN) as a risk analysis tool to estimate the consequences and related probabilities for each grid. Based on the results of all the grids, the 3D histogram is depicted to represent the risks of explosion loads, building damage and human loss. A case study is conducted to demonstrate the applicability of the proposed method. From the case study, it can be stated that the method provides a detailed and proficient risk analysis of a large site with complex conditions. The results of 3D risk mapping charts provide a clear view of the potential risks, which is useful for risk and safety management during planning, construction and operation stages. A mesh convergence study is implemented, and a grid size of 50 m * 50 m is identified to be the most appropriate over a domain range of 2 km * 2 km. A simple BN with basic risk influence factors is constructed to evaluate the risks of explosion loads, building damage and human loss. The case study proves that BN is capable of dealing with complicated interrelationships between basic factors and consequences. Since BN is flexible, extra consequences or risk factors, such as environmental concerns, human factors and safety barriers can be added to the proposed BN. Chapters 9 and 10 use DNV PHAST as an overpressure prediction tool for gas explosions for relatively simple structural arrangement of petrol station and onshore gas refinery factory. In the next two chapters, risk analysis will be conducted on offshore platforms, which are highly congested with complex structure arrangements. For such offshore facilities, it is improficient to adopt simple empirical methods to estimate overpressures. Therefore, the advanced CFD method is applied to the risk assessment of offshore facilities.

References Dnv, G. L. (2016). PHAST tutorial manual. London, UK: DNV GL software. Huang, Y., Ma, G., & Li, J. (2016). Multi-level explosion risk analysis (MLERA) for accidental gas explosion events in super-large FLNG facilities. Journal of Loss Prevention in the Process Industries. (In Press). HSE. (2015). Offshore statistics & regulatory activity report 2015. United Kingdom: Health and Safety Executive. Lobato, J., Rodríguez, J., Jiménez, C., Llanos, J., Nieto-Márquez, A., & Inarejos, A. (2009). Consequence analysis of an explosion by simple models: Texas refinery gasoline explosion case. Afinidad, 66(543), 372–279. Nielsen, T. D., & Jensen, F. V. (2009). Bayesian networks and decision graphs. Springer Science & Business Media. Pula, R., Khan, F. I., Veitch, B., & Amyotte, P. R. (2006). A grid based approach for fire and explosion consequence analysis. Process Safety and Environmental Protection, 84(B2), 79–91. Seo, J. K., & Bae, S. Y. (2016). An approach to grid-based fire frequency analysis for design accidental loads in offshore installations. Journal of Applied Mechanical Engineering, 5(2), 1–7.

References

237

Zohdirad, H., Ebadi, T., Givehchi, S., & Meysami, H. (2016). Grid-based individual risk calculation in the classification of hazardous area with a risk-based approach. Journal of Loss Prevention in the Process Industries, 43, 98–105.

Chapter 11

Multi-Level Explosion Risk Analysis for VCEs in Super-Large FLNG Facilities

Abstract This chapter illustrates a multi-level explosion risk analysis method for super-large oil and gas facilities, so as of the FLNG platform. Three levels of risk analyses, i.e., a qualitative risk screening, a semi-quantitative risk classification and a quantitative risk assessment, are implemented. The CFD method is applied for detailed risk quantification, and an as low as reasonably practical (ALARP) method is adopted as a calibration tool. Safety barriers are introduced as extra risk indicators and a case study is conducted based on a cylindrical FLNG model.

11.1 Introduction This chapter develops a multi-level explosion risk analysis (MLERA) procedure for the super-large FLNG platforms as described in Chap. 6. Figure 11.1 shows the world’s first FLNG facility designed by Shell Global, the Prelude FLNG. It is 488 m long and 74 m wide, weighing more than 600,000 tons when fully ballasted. It is roughly six times the weight of the largest aircraft carrier (Shell Global, 2016). Explosion risks are governed by three critical conditions, i.e., confinement, congestion and ventilation. Since a FLNG facility processes and stores large amount of flammable gas in a relatively small and congested area compared to onshore LNG plants, much higher explosion risk is expected on FLNG platforms. Compared to other congested offshore structures, explosion events may yield much more severe consequences due to the super-large space on board, which allows for a large volume of gas cloud to be accumulated. Therefore, for this kind of large and highly congested structure, explosion risks should be addressed during the design process and restrained to an acceptable level. Among all the explosion safety assessment methods, explosion risk analysis (ERA) is one of the most prevailing approaches to derive the accidental loads for design purposes. Description of ERA could be found in Vinnem (2011) and detailed guidelines on how to perform ERA are given in NORSOK Z013 (2001) and ISO 19901-3 (2014). Due to the complex geometry and obstacles of the offshore structures, computational fluid dynamics (CFD) tools, such as FLACS (GEXCON, 2011), are usually adopted in ERA. On the other hand, Hocquet from © Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_11

239

240

11 Multi-Level Explosion Risk Analysis for VCEs …

Fig. 11.1 Shell prelude FLNG (Shell Global, 2016) (permission from Elsevier)

Technip (2013) pointed out that one critical issue in applying ERA to FLNGs is the numerous CFD dispersion and explosion calculations. It will render intractable computational intensity due to the large size, complex structures of FLNGs and various potential uncertainties. This chapter introduces a multi-level explosion risk analysis method (MLERA) for FLNGs, which classifies the FLNG into different subsections of different risk levels before the detailed CFD simulations are conducted. The advantage of this method lies in the highly reduced CFD computational intensity since detailed computation is limited to the identified areas with the highest risks. The MLERA includes three levels: qualitative risk screening, semi-quantitative risk classification and quantitative risk assessment. Through the three levels of analyses, an exceedance curve of frequency versus overpressure will be developed. An as low as reasonably practical (ALARP) method is adopted to ascertain if the explosion risk is acceptable (NOPSEMA, 2015). Risk mitigations are required until the explosion risk of the target area is reduced as low as reasonably practical. Another challenge in assessing explosion risks for an FLNG facility is that there are neither design rules nor industry standards available as references and benchmarks because FLNG is a new technology (Paris & Cahay, 2014). Current standards, such as UKOOA (2003), HSE (2003) and API (2006), provide detailed guidelines on how to perform offshore explosion analyses with a specific procedure. Most of these guidelines are proposed based on fixed platforms; it may not be appropriate to comply in conducting an explosion risk analysis for FLNG platforms. For example, if the risk screening process used for fixed platforms is extended straightforwardly to FLNG facilities, all FLNG platforms are identified at the highest risk level, which implies the inefficacy of the risk screening process.

11.1 Introduction

241

To amend the aforementioned improficiency, other than the usual contributors from current standards, such as confinement, congestion and ventilation, safety barriers are engaged in the risk screening and classification procedures in the currently proposed method. Safety barriers are usually applied for both likelihood reduction and consequence mitigation. Some of the safety barriers used in the MLERA are listed and briefly introduced in the following section.

11.2 Multi-Level Explosion Risk Analysis A multi-level explosion risk analysis (MLERA) method (Huang, Ma, & Li, 2017) is proposed in this chapter by integrating a multi-level risk assessment method into the traditional ERA method for offshore platforms. The multi-level risk assessment method is an extension from the corresponding framework of the Department of Planning & Infrastructure of New South Wales Government (2011) to formulate and implement risk assessment and land-use safety planning processes. It is a balanced trade-off between the derivation cost and the quality of the results. To achieve that, both qualitative and quantitative approaches are engaged. Some key factors of the three levels of analysis from NOPSEMA (2012) are shown in Table 11.1. The traditional ERA for offshore platforms is one of the most prevailing approaches to derive the accidental loads for design purposes. As mentioned above, one of the critical issues in applying ERA to FLNG platforms is the intractable computational intensity. Due to the huge size of the FLNG facilities, numerous CFD dispersion and explosion simulations are required to acquire sufficient data to derive realistic design explosion loads.

Table 11.1 Key factors of multi-level risk analysis Level 1 of preliminary qualitative risk screening • • • •

Likelihood and consequence are expressed on a scale and described in words There is no numerical value for risk output Often used as a preliminary risk assessment or screening tool Rapid assessment process and relatively easy to use

Level 2 of semi-quantitative risk classification and prioritization • Generate a numerical value, but not an absolute value of risk • Provides greater capacity to classify hazards on the basis of risk • Better for evaluating cumulative risk Level 3 of detailed quantitative risk assessment • Provides a calculated value of risk based on estimates of consequence (usually software modelling) and likelihood (estimates based on failure rate data—site or industry) • Appropriate for complex decision making or where risks are relatively high • More intensive and expensive than other prevailing methods

242

11 Multi-Level Explosion Risk Analysis for VCEs …

Therefore, the multi-level method is developed in this chapter to improve the ERA to modulate the computational cost to a manageable level. The proposed MLERA method is a systematic risk analysis approach that includes three assessment stages, i.e., qualitative explosion risk screening as the first level, semi-quantitative explosion risk classification as the second level and quantitative explosion risk analysis as the third level. It aims to provide an efficient risk analysis method for explosion accidents on offshore super-large structures, such as FLNG facilities. The key aspects in multi-level risk analysis, as given in Table 11.1, are then briefly described for the proposed MLERA for FLNG platforms. Related analysis features of each level are listed, and detailed explanations of each step are discussed in the following context. Level 1 of qualitative risk screening: • qualitative description of critical risk contributors, • taking the overall FLNG facility as the analysis target, • using a risk matrix diagram to rank the risk level of a FLNG platform. Level 2 of semi-quantitative risk classification: • using a score and weight system to quantify each risk contributor, • estimating the risk of each FLNG subsection, • classifying the subsections using a cumulative density function diagram. Level 3 of quantitative risk assessment: • combining ERA and FLACS to derive the quantitative results of explosion frequency and consequences, • assessing the subsections with the highest risk levels, which are identified from the previous two analyses, • the final result is indicated by an overpressure versus frequency exceedance curve, • the ALARP concept is applied to check if the explosion risk of the corresponding subsection is as low as reasonably practical. Since the proposed MLERA considers also the safety barriers, some of the safety barriers that are engaged in the proposed method are briefly introduced in Table 11.2.

11.2.1 First Level of Qualitative Risk Screening The first-level risk screening defines the total qualitative risk level of a FLNG platform and provides guideline for the next two levels of explosion risk analyses. At this stage of risk screening, general risk indicators, as well as safety barriers, design, operation and maintenance measures, are engaged to define a relative risk level for FLNG in view of that unreasonably high explosion risk will be yielded if traditional risk screening methods are applied for such super-large and highly congested structure. Based on API (2006) and UKOOA (2003), most of the qualitative risk indicators for the traditional risk screening process are listed in Table 11.3.

11.2 Multi-Level Explosion Risk Analysis

243

Table 11.2 Explosion safety barriers Blast relief panels

The overpressure can be diverted away from potential escalation sources by blast relief panels. Blast relief panels will open quickly during an explosion in order to reduce peak overpressures

Emergency shutdown systems (ESD)

An effective ESD system will limit the inventory released in an incident, and thus, the size and duration of any resulted fire. The location of the ESD valves is usually determined based on the judgement where each particular inventory could be released

Isolation and blowdown

A leak may be reduced by isolating it manually or using the ESD system and depressurizing the leaking section using the blowdown system. Damage or fatality risk in escalation can be reduced by isolation and blowdown, so that evacuation may be avoided

Blast wall

Blast walls have long been used to protect adjacent areas from the impingement of overpressure. These walls are designed to absorb blast energy through displacement

Water deluge

Deluge has been found suitable for reducing overpressure in congestion-generated explosions. If explosion mitigation is considered critical, a deluge flow-rate of at least 13–15 L/min/m2 is recommended for general area coverage

Artificial vent

Artificial ventilation is defined as the ventilation not supplied from the action of the environmental wind alone. Upon detection of flammable gas, the standby fan(s) should be started to give maximum possible ventilation in order to aid dilution of the leak to prevent or limit the generation of an explosive cloud

Inert gas

Inert gas can be used to dilute the flammable mixture by flooding the volume within which the gas has been detected. For example, CO2 or N2 is typical inert gas. The explosive gas can then be diluted below its lower explosive limit

Detection device

Detection measures can be used to identify hazardous conditions on the plant, such as excess process pressure, an unignited release of flammable gas or a fire. Detection devices enable controlling or mitigating measures and emergency response to be activated

Alarm

The alarm system may allow operators to mitigate leaks before they ignite or to, at least, evacuate the area

Soft barriers

Progress has been made in devising soft barriers such as the micro-mist device, which consists of a cylinder of superheated water that is released quickly as a fine mist in response sirening pressure or flame sensors during an explosion. This device suppresses the explosion and significantly reduces overpressures

Safety gap

In the process industry, the safety gap is an open space with no congestion, deliberately placed in between congested process areas. The absence of obstacles in a safety gap eliminates the fluid–obstacle interaction, thereby preventing the generation of turbulence. It can be very effective in reducing pressures prior to the onset of detonation

244

11 Multi-Level Explosion Risk Analysis for VCEs …

Table 11.3 Traditional risk screening indicators from explosion risk standards Consequence Low consequence

• Low congestion level due to the low equipment count, being limited to wellheads and manifold with no vessels (i.e., no associated process pipework) • No more than two solid boundaries, including solid decks • Unattended facilities with low maintenance frequency, less frequent than 6-weekly

Medium consequence

• Medium congestion level due to the greater amount of equipment installed compared to those of the low consequence cases • Higher confinement level than that for the low consequence cases • Unattended facilities with a moderate maintenance frequency, more frequent than 6-weekly • A processing platform necessitating permanent manning but with low escalation potential to reach quarters, utilities and control areas located on a separate structure

High consequence

• High congestion level due to the significant processing on board, which leads to a high equipment count • High confinement level of the potential gas release point • Permanent manning with populated areas within the consequence range of escalation scenarios

Likelihood Low likelihood

• Low equipment and inventory count, which align closely with the consequence scenarios • Low frequency of intervention, less frequent than 6-weekly • No ignition sources within the potential gas cloud

Medium likelihood

• Greater amount of equipment installed than those for the low likelihood • Medium frequency of intervention, more frequent than 6-weekly • Weak ignition sources, such as a hot surface, exist within the potential gas cloud

High likelihood

• A high equipment and inventory count • Permanently manned installations with frequent processing on board • Strong ignition sources exist within the potential gas cloud

Table 11.4 describes exhaustively risk screening process that uses safety barriers, design, operation and maintenance measures as screening contributors. A corresponding modified risk matrix diagram is illustrated in Table 11.5. From the modified diagram, only a relative risk category is defined. The results from this category will be applied as guideline for further assessment in the next stage of the proposed MLERA.

11.2 Multi-Level Explosion Risk Analysis

245

Table 11.4 Risk indicators based on safety barriers Consequence No.

Risk level

Description

A

Moderate

• Safety barriers covering most or all parts of the FLNGs • High design capacity of the structure to counteract dynamic pressure, overpressure, missiles and strong shock response. No or minor structural damages would occur

B

Major

• Safety barriers covering the structural critical elements only • Medium design capacity of the structure to counteract dynamic pressure, overpressure, missiles and strong shock response. A medium level of structural damages would occur without affecting the overall structural integrity

C

Catastrophic

• No or only safety barriers for human living quarters • Low design capacity of the structure to sustain dynamic pressure, overpressure, missiles and strong shock response. Significant structural damages would occur and undermine the structural integrity

No.

Risk level

Description

1

Almost certain

• No or only safety barriers for human living quarters • Low level of operation and the maintenance measure corresponding to a level considerably lower than industry average

2

Likely

• Safety barriers covering only critical potential release points • Medium level of operation and maintenance philosophy corresponding to the industry average

3

Possible

• Safety barriers covering all or most of potential release points of the FLNG structures • High level of operation and the maintenance philosophy corresponding to the best standard in industry

Likelihood

Table 11.5 Risk matrix diagram for further risk screening of FLNGs Consequence of Failure Likelihood of Moderate Major Failure A B Relatively medium Almost certain 1 Relatively high risk risk Relatively medium Likely 2 Relatively low risk risk Possible

3

Relatively low risk

Relatively low risk

Catastrophic C Relatively high risk Relatively high risk Relatively medium risk

246

11 Multi-Level Explosion Risk Analysis for VCEs …

11.2.2 Second Level of Semi-Quantitative Risk Classification The analysis at this level estimates the risk level of each subsection of a FLNG facility to identify assessment prioritization for the third-level ERA. A score and weight system is applied to each selected risk contributor, so that the subsections can be classified according to the respective accumulated value. Only some of the main risk contributors for offshore explosion events are selected and briefly described in Table 11.6. Each contributor is evaluated by two elements, i.e., weight and score. The weight of each risk factor is subjectively specified by the authors based on the relevant standards as shown in below Eq. (11.1) (API, 2006; Bjerketvedt, Bakke, & Van Wingerden, 1997; UKOOA, 2003). This may be adjusted by the safety engineers according to their own experience and the practical conditions of their projects. Flammability limits for fuel mixtures can be calculated by Le Chatelier’s law: LFLMix =

100 C1 /LFL1 + C2 /LFL2 + · · · + Ci /LFLi

(11.1)

where C 1 , C 2 , …, C i are the volumetric proportions of each gas in the fuel mixture without air (Kuchta, 1985). Safety barriers are employed as extra risk contributors in the semi-quantitative risk classification procedure. All safety barriers are divided into three categories: likelihood reduction, consequence mitigation and those for both functions. Based on the classifications, safety barriers are given different weights as shown in Table 11.8. Score is specified by the quantity of each barrier deployed in each module. From Table 11.8, for safety barriers to reduce likelihood, two different weights, 6 and 4, are defined. It is because although water deluge and inert gas are able to reduce the flammable limit of the cloud and consequently prevent the explosion, they may simultaneously aggravate the consequence if the explosion eventually occurs. Inert gas can pose a significant asphyxiation risk to personnel and water deluge without proper design may increase turbulence of the affected area, and thus, the blast loads. Therefore, these two barriers are given lower weight than normal prevention barriers unless more persuasive design is presented. The total weighted score of each subsection can then be calculated as ST = SC − S B SC = SB =

n  i=1 n  j=1

(11.2)

wei sei

(11.3)

wbj sbj

(11.4)

11.2 Multi-Level Explosion Risk Analysis

247

Table 11.6 Weight and score of explosion risk contributors Risk Contributor

Description

Equipment count

Leak frequency is proportional to the amount of process equipment on the platform

Weight 3

Score = number of equipment count

Ignition

In general, the main ignition sources are welding/hot work, compressors, electrical equipment and engines/exhausts. A weak, continuous ignition source can stay and wait for the gas cloud to reach its flammable range

7

= 3 if continuous ignition source exists = 2 if only discrete ignition source exists = 1 if no or few ignition source exists

Flammable limit of process material

The higher the upper flammable limit of a certain fuel, the easier it is usually to get a flammable cloud in the air Eq. (11.1).

4

= 3 if upper flammable limit >40% = 2 if upper flammable limit is between 10 and 40% = 1 if upper flammable limit 75 cm/s = 2 if laminar burning velocity is between 45 cm/s and 75 cm/s = 1 if laminar burning velocity 40%

3 or more

Medium High

High

2

Low

Medium

High

1

Low

Low

Medium

Table 11.8 Weight of barriers based on function classification

Barrier

Classification

Weight

Emergency shutdown (ESD) system

Likelihood reduction

6

Detection device

Likelihood reduction

6

Water deluge

Likelihood reduction

4

Inert gas

Likelihood reduction

4

Safety gap

Consequence mitigation

3

Blast wall

Consequence mitigation

3

Blast relief panels

Consequence mitigation

3

Soft barriers

Consequence mitigation

3

Artificial vent

Both

9

Isolation and blowdown

Both

9

Alarm

Both

9

11.2 Multi-Level Explosion Risk Analysis

249

wherein St refers to the total weighted score for each subsection; SC and S B are the weighted scores of risk contributors and barrier functions of each subsection, respectively. After the total score of each subsection is calculated, the total weighted scores are derived with a cumulative density function and the identified the level in the risk category as shown in Fig. 11.2. The cumulative percentage is calculated from the total weighted scores of all the subsections from the target FLNG platform. Figure 11.3 illustrates the analysis procedure of the proposed MLERA and identifies the subsections requiring third-level risk quantification.

100%

S1

Cumulative density

90% 80% 70%

S2

60%

S3

50% 40% 30% 20% 10% 0% 50

100

150

200

250

Total risk score

Fig. 11.2 Cumulative density function (CDF) for total risk score (permission from Elsevier) First Level Qualitative risk screening

Second Level Semi-quantitative risk classification

Relatively low risk

Category S1: 10% of sub-sections

Relatively medium risk

Category S2: 50% of sub-sections

Relatively high risk

Category S3: 90% of sub-sections

Third Level Quantitative explosion risk assessment

Fig. 11.3 Implementation procedure of MLERA (permission from Elsevier)

Detailed CFD assessment

250

11 Multi-Level Explosion Risk Analysis for VCEs …

The first-level risk screening process divides the qualitative results into three risk levels, i.e., relatively low, medium and high risks. If the FLNG facility is categorized with a relatively low explosion risk level, only the subsections with the highest risks, which belong to category S1 (top 10%), are required for additional detailed quantitative explosion risk assessment. From Fig. 11.2, for an FLNG facility with relatively low explosion risks, the number of category S1 subsections is two. The numbers of subsections with relatively medium risk in categories S2 (50%) or high risk in S3 (90%) are 10 and 18, respectively, which require risk quantification assessment. If all the subsections in one category fail the ERA, then the next level subsections require further ERA as well.

11.2.3 Third Level of Quantitative Risk Assessment Third-level QRA is a CFD software-based quantitative analysis procedure. The process includes four main steps: leak frequency analysis, flammable gas dispersion simulation, ignition probability modelling and flammable gas explosion simulation. Figure 11.4 shows the detailed quantitative analysis procedure applied to offshore structures using CFD tools, such as FLACS. From the quantitative ERA analysis, an overpressure versus frequency exceedance curve is deduced. The risk calibration method ALARP is adopted to define the risk acceptance criteria. The ALARP framework of risk criteria is divided into three regions as shown in Fig. 11.4. • An unacceptable region: in this region, risks are not acceptable except for extraordinary circumstances. Risk reduction measures should be deployed. • A tolerable region: it is usually identified as an ALARP region, which implies that the risks are considered acceptable providing that they have been made as low as reasonably practicable. In this region, risk reduction measures are desirable but may not be necessary if a cost–benefit analysis shows that their cost is disproportionate to the benefit to be achieved. • A broadly acceptable region: risks in this region are acceptable and no more risk reduction measures are required. Figure 11.5 shows an example of application of ALARP to the overpressure versus frequency exceedance curve. As shown in the diagram, if the design strength of the primary components of the FLNG comes to cross the predicted explosion load in the unaccepted zone, more risk reduction measures are required to be deployed until the design strength is proved to be sufficient to resist the explosion loads. No further reductions are required if the design strength falls in the accepted zone. For the ALARP zone, reduction measures should be deployed unless the cost is proved to be disproportionate to the potential expectant benefit.

11.2 Multi-Level Explosion Risk Analysis

251 START

Data Entry

Release Frequency Analysis

Leak size/ Location

Geometry

Dispersion Simulation

Flammable Volume Exceedance Curve

Cloud Size/ Location

Ignition type/ location

Explosion Simulation

Overpressure Exceedance curves

Overpressure on targets

Fig. 11.4 Quantitative assessment procedure (permission from Elsevier)

Fig. 11.5 Application of the ALARP to final results of MLERA (permission from Elsevier)

252

11 Multi-Level Explosion Risk Analysis for VCEs …

11.3 Case Study For FLNG structures, cylindrical FLNG vessels introduced in Chap. 6 are applied to improve hydrodynamic stability. Figure 11.6 shows the geometry of a cylindrical FLNG platform in FLACS. The cylindrical platform has a smaller area than a usual rectangular one. All the highly congested subsections are concentrated on the board, which may increase the explosion risks. Little research has been reported on the gas explosion risk analysis for cylindrical platforms. In this section, a cylindrical FLNG structure proposed by Li, Ma and Abdel-jawad (2016) is adopted as the basic model to illustrate the proposed MLERA. Figure 11.7 shows the arrangement of topside modules including totally 12 modules. A brief introduction of each module is listed as below. • Module 1: power generation, • Module 2: Trent gas turbines and two essential diesel generators, • Module 3: nitrogen package, hot oil, mono-ethylene-glycol (MEG) processing and inlet facilities, • Module 4: boil-off gas compressor and fuel gas system, • Module 5: acid gas removal unit and end flash gas compressor, • Module 6: dehydration and mercury removal, • Module 7–Module 12: liquefaction modules.

Fig. 11.6 Geometry of cylindrical FLNG (permission from Elsevier)

11.3 Case Study

253

Module8

Module10

Module12

Module7

Module9

Module11

Module5

Module3

Module1

Module6

Module4

Module2

Fig. 11.7 Topside arrangement of modules (permission from Elsevier)

11.3.1 Qualitative Risk Screening of Cylindrical FLNG For the first level of the risk screening process, based on the concepts from API (2006) and UKOOA (2003), the selected FLNG module is specified as a high-risk platform because it is a permanently manned and highly congested offshore structure with a large amount of equipment and inventories. The second level of risk screening analysis is then conducted directly. The conditions of safety barriers, design, operation and maintenance philosophies are defined as below. • Safety barriers: as shown in Fig. 11.7, safety gaps are applied to every module. In view of grossly insufficient information about other safety barriers, such as alarms, detection devices, ESDs and water deluges on this FLNG model, a medium level for the condition of the safety barriers on this FLNG platform is assumed. • Design philosophy: this FLNG is a recently designed offshore structure to have a high level of design philosophy under the most recent design standards. • Operation and maintenance philosophy: the standard of operation and maintenance philosophy is assumed to be medium implying that the average industry standard is employed because that no such FLNG facility has yet been operated throughout the world.

254

11 Multi-Level Explosion Risk Analysis for VCEs …

Based on the aforementioned conditions, this cylindrical FLNG is specified as of a medium risk, which indicates that all subsections belongs to category S2 from the second level of semi-quantitative risk classification. Thus, it is subject to detailed assessment in the third step.

11.3.2 Semi-Quantitative Risk Classification The subsections of the selected model are defined by 12 modules. Each module is assessed in this risk classification process by specifying the explosion and safety barrier contributors, which are defined in Tables 11.6 and 11.8. The target area of consequence analysis is the human living quarters. As shown in Fig. 11.8, a cumulative density function diagram can be calculated based on the final scores from Table 11.9. During the second level of risk classification process, some of the contributors have the same score for different subsections. For instance, as in Table 11.9, the final 100% 90%

Cumulative density

80% 70% 60% 50% 40% 30% 20% 10% 0% 30

40

50

60

70

80

90

Total weighted score

Fig. 11.8 Cumulative density function diagram of subsections (permission from Elsevier) Table 11.9 Scores and weights of risk contributors Subsections

1

2

3

4

5

6

7

Total score of explosion contributors

93

64

66

66

81

69

101 101 108 108 115 115

8

9

10

11

12

Total score of safety barrier contributors

30

30

30

30

27

30

27

30

30

30

30

30

Final score

63

34

36

36

54

39

74

71

78

78

85

85

11.3 Case Study

255

scores of safety barriers are the same for most of the modules. It is probably because this second level of risk classification process is still an abbreviated assessment of each module. Thus, it may not be sufficiently proficient to differentiate the modules by one particular contributor. It is understood that lack of detailed information also contributes to this improficiency. For example, in this case study, module 5 and 7 have less safety gaps than the other modules based on design drawings of the proposed model. This is the only difference of safety barriers that can be defined and the other scores of barriers for each module are assumed to be the same due to the insufficiency of data. Therefore, the total scores of barriers for most of the subsections remain the same. It is expected that more detailed information of the target structure will lead to a higher level of accuracy for this classification. From the cumulative density function diagram, the S2 category includes six subsections, i.e., modules 7–12. Therefore, six subsections require further detailed assessment.

11.3.3 Detailed Quantitative Risk Assessment As a medium risk level is identified for the target FLNG during the first level of the qualitative risk screening process, modules 7–12, which belong to category S2, require further detailed assessment. Therefore, detailed quantitative risk assessment for FLACS is conducted in this section. However, due to the limitation of computer capacity, a simplified analysis model, which is assumed to be sufficient to demonstrate the proposed method, is built and applied. In this model, three leak locations on subsections 7, 9 and 11 are specified for assessment. Final results of this assessment are deduced by combining the analyses on these three locations. The three selected locations are shown in Fig. 11.9. Other specific assumptions for this model are described as below. • Four leak rates (12, 24, 48, 96 kg/s) are simulated to study the potential gas volume build-up in order to compare the blast wall configurations. • In the simulations of dispersion leaks and explosion gas clouds, the inventory of the gas composition inside the cylindrical FLNG platform is summarized as in Table 11.10.

Table 11.10 Gas composition for dispersion and explosion study

Component

Export gas (%)

Methane

27

Ethane

33

Propane

15

Hexane

19

CO2

6

256

11 Multi-Level Explosion Risk Analysis for VCEs …

Fig. 11.9 Selected leak locations on cylindrical FLNG platform (permission from Elsevier) Table 11.11 Leak cases identified for dispersion study Case

Wind direction

Wind speed (m/s)

Leak rate (kg/s)

Leak position

Leak orientation

1

Due east

4

12, 24, 48, 96

West end

Along and opposite wind

2

Due east

4

12, 24, 48, 96

Middle

Along and opposite wind

3

Due east

4

12, 24, 48, 96

East end

Along and opposite wind

• In this study, the assessment focuses on the living quarters with a protective blast wall on the west side. The living quarters are located at the very east side of the FLNG (Fig. 11.7). • Wind speed and wind direction are specified to be constant of +4 m/s in the due east direction to examine the worst gas dispersion scenarios. • Leak directions are modelled in both eastern and western directions. Dispersion analysis Based on the aforementioned assumptions, the overall leak cases adopted in this chapter are listed in Table 11.11. The gas monitoring region for dispersion analysis covers all the modules on the cylindrical FLNG platform.

11.3 Case Study

257

Figure 11.10 demonstrates examples of dispersion simulation outputs for gas releases with a leak rate of 48 kg/s. The releases are simulated from both release directions and leak locations are set on the ground centre of modules 7, 9 and 11.

(a) Leaks from module 11 with both wind directions Fig. 11.10 Gas dispersion simulations for leaks with leak rate of 48 kg/s (permission from Elsevier)

258

11 Multi-Level Explosion Risk Analysis for VCEs …

(b) Leaks from module 9 with both wind directions Fig. 11.10 (continued)

11.3 Case Study

259

(c) Leaks from module 7 with both wind directions Fig. 11.10 (continued)

260

11 Multi-Level Explosion Risk Analysis for VCEs … 100%

Cumulative density

90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 0

2

4

6

8

10

12

Stoichiometric gas cloud size (104 m3)

Fig. 11.11 Cumulative curve of gas cloud sizes for all leak rate scenarios (permission from Elsevier)

To investigate all the potential leak rate cases, the overall cumulative curve of gas cloud sizes within the gas monitor region is depicted in Fig. 11.11. The cumulative curve is derived by sorting the gas cloud size from small to large with assumption of equal leak frequencies. Explosion simulations Explosion simulations are performed using gas cloud data from dispersion simulations with leak rates of 12 to 96 kg/s. The gas clouds are specified in four different locations covering the entire platform. So that the overall gas explosion consequences for all modules can be analysed. For all the gas clouds, the plan view sizes are all fixed at 100 × 80 m2 , while the heights of the clouds are variable in consistency to the corresponding gas dispersion results. For each gas explosion simulation, the gas cloud is ignited at the ground centre of each module. Figure 11.12 gives an overview of the gas cloud coverage and ignition locations. From Fig. 11.12, each gas cloud covers four modules. About 200 monitor points are homogeneously deployed on the ground to record the overpressures in a gas explosion simulation. Totalling all the different gas leak rate scenarios, gas cloud sizes and locations, more than 3000 VCE overpressures are monitored in this probabilistic study on gas explosion. Since the main purpose of this study is to assess the condition of the living quarters, ten monitor points are assigned near the living quarter to record the overpressures for each gas explosion scenario. Figure 11.13 shows three explosion examples simulated based on different leak rates: 96, 48 and 24 kg/s. The explosive gas clouds are set at the north and east ends of the model. The ignition is assigned at the centre of the gas cloud located at the east and north ends of the platform. The gas explosion blast is observed to spread from the ignition point to all the surrounding objects. The maximum overpressures are found near the edge of the gas cloud in the congested region. To consider the influence of blast walls, a blast wall is modelled in front of the west end of the living quarter. Two monitors are set at both sides of the blast wall.

11.3 Case Study

261

Fig. 11.12 Overview of gas cloud coverage and ignition locations (permission from Elsevier)

A large overpressure of approximately 1.8 bar is detected at the left side of the wall under the leak rate of 96 kg/s as shown in Fig. 11.14. However, after reduction by the blast wall, the overpressure at the right side of the wall is about 0.2 bar to prove the effectiveness of the blast wall. To consider all the gas dispersion outputs as input for the gas explosion simulations, 120 explosion cases are numerically simulated, which are commensurating with the former dispersion simulations for four leakage rates, two leakage directions, three gas release locations and five different series of blast wall layout designs (Li, 2017). The overpressure of each case is calculated by FLACs and the overall cumulative curve of gas explosion simulations is summarized in Fig. 11.15. Equal frequencies are allocated to all monitored overpressures for the living quarters, which are sorted from small to large. Frequency analysis A simple illustrative explosion frequency calculation is then conducted in this subsection. The exceedance curve of frequency against overpressure at the living quarters is specified using the monitored overpressures for over 1000 scenarios. To simplify the analysis process, the leak frequencies of different leak rates are assumed to be the same. Based on the data from the Purple Book (Uijt & Ale, 2005), the leak frequency is taken as 3.33 × 10−1 per year. Based on the ignition intensities and the previously performed dispersion simulations, the ignition probability is determined to be 0.36%. The explosion frequency is consequently calculated by multiplying the leak frequency with the ignition probability. Therefore, the total explosion frequency is derived as approximately 1.2 × 10−3 per year.

262

11 Multi-Level Explosion Risk Analysis for VCEs …

(a) Leak rate of 96 kg/s

(b) Leak rate of 48 kg/s

(c) Leak rate of 24 kg/s Fig. 11.13 Gas explosion simulation examples with different leak rates (permission from Elsevier)

11.3 Case Study

263

Fig. 11.14 Demonstration of function of blast wall (permission from Elsevier) 100%

Cumulative density

90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

0

0.1

0.2

0.3

0.4

Overpressure (bar)

Fig. 11.15 Cumulative curve of overpressure for living quarters (permission from Elsevier)

Consequently, the explosion risk regarding the living quarters subject to overpressures of VCE from the liquefaction modules is evaluated and the probability of exceedance curves with a frequency of 10−4 /year is shown in Fig. 11.16. From Fig. 11.16, the maximum overpressure is about 0.4 bar in the living quarters. In view of the ALARP criterion, the acceptable zone starts from 0.2 bar, which implies that no further risk reduction method is required if the maximum strength of the primary components of the FLNG is designed to be larger than 0.2 bar with a corresponding frequency of 10−11 . Otherwise, risk reduction measures should be

264

11 Multi-Level Explosion Risk Analysis for VCEs …

Exceedance frequency (/year)

1.00E-03

1.00E-04

1.00E-05

1.00E-06 0

0.1

0.2

0.3

0.4

Overpressure (barg)

Fig. 11.16 Exceedance curve of overpressures around living quarters for all leak rate scenarios (permission from Elsevier)

deployed until the design strength demonstrates greater than 0.05 bar with corresponding frequency of 10−4 and also proves to be as low as reasonably practical.

11.4 Summary In summary, a more efficient multi-level explosion risk analysis method (MLERA) is proposed in this chapter. This method includes three levels of assessment, i.e., qualitative risk screening for an FLNG facility at the first level, semi-quantitative risk classification for subsections at the second level and quantitative risk calculation for the target area with the highest potential risks at the third level. Since the current design standards for normal offshore platforms are not sufficiently robust for effectively assessing explosion risks of super-large offshore structures, during the risk screening and risk classification processes, safety barriers are adopted as extra risk indicators besides the traditional ones such as congestion, confinement and ventilation. From the aforementioned analyses, with only traditional standards, FLNG platforms will always be identified as high risk. However, with the extra contributors of safety barriers, the target FLNG facility is distinguished into relatively low, medium or high risks for different subsections, which identifies the subsections for further assessments. For detailed quantitative risk assessment, a CFD software, FLACS, is applied to model and simulate the target FLNG platform. The results are shown as an exceedance curve, which describes the possibilities of overpressure at the target area. An ALARP method is then adopted as a calibration tool to determine if the explosion loads from the exceedance curve can be accepted. If the overpressure exceeds the acceptable limitation, more safety barriers should be installed. Further assessments are required

11.4 Summary

265

until the final results show that the risk is reduced to an acceptable level or/and as low as is reasonably practical. Through the three levels of risk assessments, the areas with the highest level of potential risks are identified to be further assessed. From the case study, only half of the subsections on the selected model require detailed assessment using FLACs with regard to the living quarters. It implies the potential computational intensity is reduced to half by the proposed method. In the next chapter, a more detailed CFDbased ERA is introduced with consideration of blast wall effect to explore potential risk reduction methods.

References API. (2006). Recommended practice for the design of offshore facilities against fire and blast loading (1st ed.). United States: American Petroleum Institude. Baker, Q. A., Tang, M. J., Scheier, E. A., & Silva, G. J. (1996). Vapor cloud explosion analysis. Process Safety Progress, 15(2), 106–109. Bjerketvedt, D., Bakke, J. R., & Van Wingerden, K. (1997). Gas explosion handbook. Journal of Hazardous Materials, 52(1), 1–150. Department of Planning & Infrastructure of New South Wales Government. (2011). Assessment Guideline Multi-level Risk Assessment. New South Wales, Australia: Department of Planning & Infrastructure. GexCon. (2011). FLACS v9.1 user’s manual. Norway: Doxygen. HSE. (2003). Fire, explosion and risk assessment topic guidance (Issue 1). United Kingdom: Health and Safety Executive. Hocquet, J. (2013). Explosion risk analysis ‘ERA’ for FLNG facilities: The main challenges. Chemical Engineering Transactions, 31, 595–600. Huang, Y., Ma, G., & Li, J. (2017). Multi-level explosion risk analysis (MLERA) for accidental gas explosion events in super-large FLNG facilities. Journal of Loss Prevention in the Process Industries, 45, 242–254. ISO 19901-3. (2014). Petroleum and natural gas industries—specific requirements for offshore structures—part 3: Topsides structure (2nd ed.). Switzerland. Kuchta, J. M. (1985). Investigation of fire and explosion accidents in the chemical, mining and fuel-related industries—A manual. United States Department of the interior, Bureau of Mines, Bulletin 680. Li, J., Ma, G., & Abdel-jawad, M. (2016). Gas dispersion risk analysis of safety gap effect on the innovating FLNG vessel with a cylindrical platform. Journal of Loss Prevention in the Process Industries, 40, 304–316. Li, J., Ma, G., Hao, H., & Huang, Y. (2017). Optimal blast wall layout design to mitigate gas dispersion and explosion on a cylindrical FLNG platform. Journal of Loss Prevention in the Process Industries. NOPSEMA. (2012). Risk assessment guidance note. Australia: National Offshore Petroleum Safety and Environmental Management Authority. NORSOK Z-013. (2001). Risk and emergency preparedness analysis. Norway: Norwegian Technology Centre. Paris, L., & Cahay, M. (2014). Challenges in a multi-disciplinary approach for explosion design of FLNG. Paper presented at the Offshore Technology Conference, Houston, Texas, USA. Shell Global. (2016). Prelude FLNG—Overview. Retrieved from http://www.shell.com/about-us/ major-projects/prelude-flng/prelude-flng-an-overview.html.

266

11 Multi-Level Explosion Risk Analysis for VCEs …

Uijt, P., & Ale, B. (2005). Guidelines for quantitative risk assessment. VROM: Ministerie van Verkeer en Waterstaat. UKOOA. (2003). Fire and explosion guidance part 1: Avoidance and mitigation of explosions (Issue 1). United Kingdom: UK Offshore Operation Association. Vinnem, J. E. (2011). Offshore risk assessment principles, modeling and applications of QRA studies (2nd ed.). Springer Series in Reliabilities Engineering, London, UK.

Chapter 12

CFD-Based Explosion Risk Analysis of Blast Wall Effects on FLNG Platforms

Abstract This chapter presents a comprehensive safety design of blast wall layout on a cylindrical floating liquefied natural gas (FLNG) platform. The computational fluid dynamics (CFD) simulation results of more than 120 gas cloud sizes and 16,000 gas explosion overpressures indicate that blast walls are exclusively beneficial for mitigating flammable gas cloud and explosion overpressure only if the initial gas leak rates are highly momentous. A series of different blast wall layouts are designed for the cylindrical FLNG.

12.1 Introduction This chapter investigates overpressure mitigation by blast walls on the cylindrical FLNG platform using CFD-based explosion risk analysis. In contrast to Chap. 6, where the safety gaps were designed to separate the congested modules exclusively in the north–south direction, blast wall is explored in this chapter to be a more feasible alternative safety measure in the space-limited east–west direction (Li, Ma, Hao, & Huang, 2017). Blast wall provides protection for specific targets by restricting the spread of flammable fuel–air cloud and isolating the generation of turbulence and explosion waves from the neighbouring modules. It is generally adopted as one of the explosion mitigating approaches in the oil and gas industry (Kang, Choi, Choi, Choi, & Choi, 2016; Syed, Mohamed, & Rahman, 2016). Blast walls could be installed very close to or as part of the identified explosion protection targets/temporary refuge (HSE, 2006). In other words, because of the space limitation, blast walls instead of safety gaps are chosen in this study to investigate the explosion mitigation effect on the cylindrical FLNG platform in the east–west direction. Extensive studies on the structural analysis and design of blast walls have been reported (Langdon & Schleyer, 2005, 2006; Louca, Boh, & Choo, 2004; Schleyer, Lowak, Polcyn, & Langdon, 2007). However, most the studies are the blast wall material and structural response analyses. The overpressure protection effects of blast wall have not been well addressed. The existing offshore applications blast walls tend to be traditional structures (Boh, Louca, & Choo, 2007; Kang et al., 2016; © Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_12

267

268

12 CFD-Based Explosion Risk Analysis of Blast Wall …

Louca, Punjani, & Harding, 1996; Sohn et al., 2013). No research has been carried out on the state-of-the-art offshore structure—FLNG, letting alone the cylindricalshaped FLNG. In this chapter, the same cylindrical FLNG platform investigated in previous chapters is applied. Contrasting to the deterministic study of safety gap, a CFD-based probabilistic study on a series of blast wall configurations is currently carried out to accommodate more variabilities. The widely accepted and highly validated CFD software FLACS (Hansen & Johnson, 2015; Hansen, Gavelli, Ichard, & Davis, 2010; Li, Abdel-jawad, & Ma, 2014; Silvestrini, Genova, & Trujillo, 2008) is employed to perform the gas dispersion and explosion simulations. The current risk analysis are composed of totally 120 CFD simulations with different variables of four varying leakage rates, two opposite leakage directions, three different gas release locations and four sets of blast wall configurations plus the original platform configuration without blast walls. For each module on the cylindrical FLNG platform, more than 50 monitor points are assigned to record the overpressures in the gas explosion analysis. Therefore, over 3000 gas explosion simulation results are taken into account in the overpressure mitigating evaluation of blast walls.

12.2 Numerical Models The PRICO® technology-designed FLNG units (Black & Veatch, 2016) are installed on the cylindrical FLNG topside. The layout of all modules, including dehydration and mercury removal facilities, compressors and liquefaction trains, etc., is shown in Fig. 12.1. The numbering order of these 12 modules is kept the same as that in the ship-shaped FLNG except for that a U-shape pipe rack is used to connect all the modules. The turret area, which usually presents in the ship-shaped FLNG, is eliminated in this cylindrical hull (Li, Ma, Abdel-Jawad, & Huang, 2016). The living quarter and workshop are located in the far end of platform away from the other modules. Each topside module is 30 m wide and 40 m long. All the main structural components of these modules are converted into boxes and cylinders in the geometry modelling using the preprocessor CASD of FLACS (Gexcon, 2015). The total number of the simplified components is 72,576. The experimentally calibrated sub-grid models of FLACS are applied to account for the influence of small objects, while the large objects and walls are represented on-grid in the porosity calculation of FLACS.

12.3 Inputs for Gas Dispersion and Explosion Simulations The variables in the CFD simulations and risk analyses usually include leak rate, leak direction, leak location, gas composition and wind condition. In this study, the inves-

12.3 Inputs for Gas Dispersion and Explosion Simulations

269

Fig. 12.1 Model wireframe of cylindrical FLNG platform geometry in FLACS (permission from Elsevier)

Fig. 12.2 Leak locations on cylindrical FLNG platform (Permission from Elsevier)

270

12 CFD-Based Explosion Risk Analysis of Blast Wall …

tigation into blast wall’s explosion mitigating effect is limited to east–west direction. It is because the protection target—living quarter, which was investigated and identified as the area with the highest level of potential risks by Huang, Ma, Li and Hao (2016), is at the furthest east side of the platform, as shown in Fig. 12.2. Therefore, west-to-east wind direction (+x coordinate in CASD of FLACS) is selected as the worst wind scenario to derive the gas dispersion of a leakage. The wind speed is fixed as +4 m/s, and both of the east and west leak directions are considered in the gas dispersion simulation. Three critical leak locations in the liquefaction train modules are specified as in Fig. 12.2. It is based on the conclusion of previous work of Li, Ma, Abdel-Jawad, and Huang (2016) that gas leaks in the more congested regions are more likely to generate bigger volume of stoichiometric gas clouds and transmit more gas cloud to the far field. Accordingly, the representative leak rates from 0.25 to 96 kg/s (Norsok, 2001) for these three leak locations are specified in the gas dispersion simulations, which are shown in Fig. 12.3. The inventory of gas composition inside the cylindrical FLNG is summarized in Table 12.1.

Leak rate (kg/s)

120 100 96

80 60 40

48

20 0

0.25

0.75

1.5

3

6

12

24

Leak rate range (kg/s) Fig. 12.3 Leak rate data on cylindrical FLNG platform (permission from Elsevier) Table 12.1 Gas composition for dispersion and explosion simulations

Component

Export gas (%)

Methane

27

Ethane

33

Propane

15

Hexane

19

CO2

6

12.4 Design of Blast Walls

271

12.4 Design of Blast Walls Figure 12.4 gives the general overview of the geometry with blast walls installed in front of the congested modules. In FLACS, all structural components are assumed to be unyielding during the entire explosion, which implies the blast walls remain in place even for gas explosion with extreme overpressures. Therefore, the structural design and analysis of the blast walls themselves are not implemented in this study. Instead, its functions in interrupting gas cloud propagation from nearby modules and protecting target unit against explosion loads from neighbouring modules are investigated. In view of that, this risk analysis explores the consequences of different gas dispersions and explosions in different congestion and confinement conditions with the instalment of the blast walls. In view of that, blast walls are designed as barriers immediately next to the congested modules to account for different confinement scenarios, as shown in Fig. 12.5. Since only west-to-east wind direction is specified to investigate the highrisk region—living quarter, blast walls are mainly installed on the east side of the module to block the flammable gas cloud propagating from more congested areas in west. Four essential blast wall configurations are designed. Figure 12.5 shows blast walls as the red lines in the plane view of the cylindrical FLNG platform.

Fig. 12.4 Blast wall installed in front of modules of cylindrical FLNG (permission from Elsevier)

272

12 CFD-Based Explosion Risk Analysis of Blast Wall …

Fig. 12.5 Blast wall arrangement (permission from Elsevier)

12.5 Dispersion and Explosion Analyses 12.5.1 Dispersion Analyses Totally, nine categories leak rates in Fig. 12.3 are modelled. For the specific leak rate, the effective cross-sectional area of a leak outlet in FLACS is derived as: m  = ρu A

(12.1)

wherein m  is the leak/flow rate; u stands for the velocity at the outlet; ρ represents the gas density; and A denotes outlet areas. The gas density of gas composition, as shown in Table 12.1, is 1.6 kg/m3 . Thus, for a representative low–medium momentum leak rate of 6 kg/s and outlet leakage velocity of 5 m/s, the effective cross-sectional area of leak outlet is calculated to be 0.75 m2 . Consequently, the theoretical grid size of the leak outlet is determined as 0.87 m, i.e. the square root of 0.75 m2 , which is smaller than the initial choice of 1-m grid size near leakage. In order to model small leakage area size (8 m in height, while the dimension of the smallest topside module is

12.5 Dispersion and Explosion Analyses

273

30 × 40 × 20 m3 . Therefore, in this study, the grid resolution is kept as 1 m2 starting from the leak area to avoid the unnecessarily increased simulation time. For gas leakages with small leak rates on the large-sized cylindrical FLNG platform, the generated steady state gas cloud size is too small that it could be confined in just one module. After the ignition of such small size gas cloud, the subsequently induced explosion overpressures are not high enough to threaten neighbouring modules, e.g. the living quarter with a safety gap away other congested modules. Therefore, the low momentum leak rate case (≤6 kg/s) is not considered in this study. Four categories of leak rates, which are 12, 24, 48 and 96 kg/s, are taken into account to study the possible gas cloud volume and explosion overpressure build-ups in all blast wall layout designs. Table 12.2 presents the overall leak cases including the variations of three leak locations and two leak orientations modelled in this chapter. The gas monitor region covers all the modules on the cylindrical FLNG platform as shown in Fig. 12.6.

Table 12.2 Leak cases for dispersion simulations Case

Wind direction

Wind speed (m/s)

Leak rate (kg/s)

Leak position

Leak orientation

1

West to East

4

12, 24, 48, 96

West end

Along and opposite wind

2

West to East

4

12, 24, 48, 96

Middle

Along and opposite wind

3

West to East

4

12, 24, 48, 96

East end

Along and opposite wind

Fig. 12.6 Gas cloud monitor region on cylindrical FLNG platform (permission from Elsevier)

274

12 CFD-Based Explosion Risk Analysis of Blast Wall …

12.5.2 Dispersion Simulation Results and Discussions Figure 12.7 demonstrates four gas dispersion simulation outputs for gas releases with different blast wall configurations. For these simulations, the leak rates are set as 48 kg/s; leak direction is assigned the same as that of the wind from west to east; and leak sources are assumed to be on the ground centre of the module. The comparison for the protection effect of different blast walls layout against the flammable gas cloud is conducted in view of the equivalence ratio (ER) ranging from lower flammability limit (LFL) 0.5 to upper flammability limit (UFL) 2.5. The simulation is terminated at 210 s when the stoichiometric gas cloud reaches the steady state. It is shown in Fig. 12.7a, b that the blast walls at the east end effectively restrain the propagation of the gas cloud. As for the other two blast wall configurations in Fig. 12.7c, d, the additional blast walls on the left-hand side of the east end further reduce the cloud size near the living quarter. On the other hand, the gas cloud is more condensed in the middle module of the FLNG platform. The high

(a) Design 1: blast walls on 1 side

(c) Design 3: blast walls on 2 sides

(b) Design 2: blast walls on 2 sides

(d) Design 4: blast walls on 3 sides

Fig. 12.7 Gas dispersion simulation results for leaks on ground centre with different blast wall configurations (permission from Elsevier)

12.5 Dispersion and Explosion Analyses

275

gas concentrations in these two configurations result in the ER going beyond the upper flammability limit. Thus, it reduces the overall stoichiometric gas cloud size. From the comparison, the extra blast walls mitigate the total gas cloud size for the target area in gas dispersion simulations. Since the comparison is conducted for four typical cases of overall 24 scenarios with leak rate of 48 kg/s, all the other leak rate cases are thoroughly investigated to derive more persuasive conclusions. The overall exceedance curve of gas cloud sizes within the gas monitor region is summarized in Fig. 12.8. For comparison, the gas dispersion simulations for the original cylindrical FLNG without blast wall are also performed. The exceedance curve is derived by sorting the gas cloud size from small to large with equal leak frequency assumption for all the leaks. From Fig. 12.8, for stoichiometric gas cloud size smaller than 3 × 104 m3 , all blast wall designs would yield greater frequencies compared to the original cylindrical FLNG without blast wall. In other words, the blast walls magnify the gas cloud size, which is adversarial. On the other hand, all these enlarged gas clouds are from the leakages at leak rates of 12 and 24 kg/s, which develop into much smaller clouds than those from the leak rate of 48 and 96 kg/s gas release cases. Thus, the enlargement may not necessarily effectuate high overpressures in gas explosion to threaten structures and people. In contrast, for the gas dispersions at leak rates 48 and 96 kg/s, a distinct tendency of cloud size reduction by blast walls is observed as in Fig. 12.9. It is seen that most of the blast wall configurations ensue lower frequencies at the similar stoichiometric gas cloud size with high leak rates. For instance of gas cloud size of 4.7 × 104 m3 , the exceedance probability for blast wall design 4 is 12% against 24% from the original configuration without blast wall. From another perspective of view, at the same exceedance probability of 24%, the gas cloud size of 4.7 × 104 m3 can be reduced to 4.1 × 104 m3 , which is 12.8% reduction, by the blast wall design 4. 100.00% Original- No blast wall Design 1 Design 2 Design 3 Design 4

90.00% 80.00% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00% 0

2

4

6

8

10

12

Stoichiometric gas cloud size (104 m3) Fig. 12.8 Exceedance curve of gas cloud sizes for all leak rate scenarios (permission from Elsevier)

276

12 CFD-Based Explosion Risk Analysis of Blast Wall … 70.00%

Original- No blast wall Design 1 Design 2 Design 3 Design 4

60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00% 3

4

5

6

7

8

9

Stoichiometric gas cloud size (104 m3) Fig. 12.9 Exceedance curve of gas cloud sizes over 3 × 104 m3 for different leak rate scenarios (permission from Elsevier)

To further demonstrate the gas cloud mitigating effect of blast walls at different leak rates, the stoichiometric gas clouds at all locations are averaged and displayed in Fig. 12.10. From Fig. 12.10, the blast walls reduce gas cloud when leakage rate is 48 kg/s or higher. In such condition, blast wall design 2 and 4 ensue the smallest stoichiometric gas cloud sizes. On the other hand, blast wall design 2 is the most undesirable configuration for the leak rate of 24 kg/s. In view of that, blast wall design 4 is identified as the optimal design among all the configurations in terms of reducing the overall gas cloud size.

Averaged Stoichiometric gas cloud size (104 m3)

7 6 5

Original - No blast wall Design 1 Design 2 Design 3 Design 4

4 3 2 1 0 12

24

48

96

Leakage rate (kg/s) Fig. 12.10 Averaged gas cloud size for different leak rate scenarios (permission from Elsevier)

12.5 Dispersion and Explosion Analyses

277

The above gas dispersion simulations evaluate the effect of the different blast wall designs on the stoichiometric gas cloud sizes. From the derived exceedance curve and the graph of the averaged gas could sizes for these four leak rate scenarios, it can be stated that the blast walls are not effective and even adversarial with low leak rate (6 kg/s)

1.00

1.00

0.60

0.60

0.60

/

Technology

1.00

1.00

0.60

0.60

0.60

Technology

0.57

1.00

0.49

0.49

0.46

/

Overall

0.57

1.00

0.49

0.49

0.46

Overall

0

4

16

16486

6882

16486

16486

/

Module

Items or sq. metre

16486

6882

16486

16486

0

4

16

Module

Items or sq. metre

0.57

1.00

0.49

0.49

0.46

0.46

0.46

/

Adjust

0.57

1.00

0.49

0.49

0.46

0.46

0.46

Adjust

2.82E-2

8.95E-3

2.08E-2

2.08E-2

0.00E + 0

4.22E-3

7.05E-4

1.14E-2

Total

Discrete

3.76E-4

1.17E-4

1.68E-5

2.16E-4

0.00E + 0

9.36E-6

1.54E-6

Total

Discrete

12.5 Dispersion and Explosion Analyses 285

12 CFD-Based Explosion Risk Analysis of Blast Wall …

Exceedance frequency (/year)

286 1.00E-03

Original- No blast wall Design 1 Design 2 Design 3 Design 4

1.00E-04

1.00E-05

1.00E-06 0

0.1

0.2

0.3

0.4

0.5

Overpressure (barg) Fig. 12.21 Exceedance curve of overpressures around living quarter for all leak rate scenarios (permission from Elsevier)

12.6 Summary Unlike the worst scenario analysis of VCE overpressure mitigation design in Chap. 6, a CFD-based probabilistic study on the gas dispersion and explosion risk regarding the blast wall configurations is conducted in this chapter. FLACS is adopted to model the blast walls on the cylindrical FLNG platform. Four sets of different blast wall configurations are studied for gas dispersion and explosion comparisons. The explosion risk assessment is carried out as a follow-up overpressure analysis on the cylindrical FLNG platform after Chap. 7. Since the N-S direction with safety gaps has been investigated previously, blast walls are installed in the east–west direction in view of the space limitation. Some critical assumptions are made in this chapter that wind direction and speed are fixed, leak directions are also assigned in the east–west direction correspondingly as the potential worst scenario. Over 120 gas dispersion simulations are carried out. The results indicate that all blast wall designs are not beneficial for stoichiometric gas cloud size restriction if the leakage rate is 24 kg/s and below. Only slight reduction in cloud size can be derived for blast wall configuration 2 and 4 in large rate scenarios. The investigation of gas explosions with over 3000 scenarios is performed subsequently. The gas dispersion data are used as the input in the explosion simulations. Generally, the simulations of gas explosion overpressures show that blast walls can effectively mitigate overpressures on the entire FLNG platform. In spite of that overpressure and exceedance frequency increase can be observed in low leak rate scenarios (12 and 24 kg/s), it is of negligible consequence for structural integrity. In contrast, in the large gas releases with leak rate of 48 and 96 kg/s, blast walls mitigate significantly VCE overpressure. In some cases, No. 4 blast wall design can decrease the overpressure exceedance frequency by 20%.

12.6 Summary

287

Local optimization design of blast wall for the living quarter is conducted using the detailed explosion frequency calculation to address the greatest concern in the offshore industry. The leak frequency and ignition probability are taken into account to define the exceedance curve of overpressures for the living quarter. The effectiveness in overpressure mitigation by using blast walls is further confirmed, while blast wall configuration 4 is again identified to be the optimum. In summary, performance of different blast wall designs on the cylindrical FLNG platform is derived globally and locally for different objects. The results from the gas dispersion analyses indicate that blast walls may have adversarial effect on the structures in the gas release scenarios with low leak rate, whereas in the gas explosion analysis of large leak rate scenarios, blast walls mitigate the explosion overpressure substantially. Comparing to the worst case/deterministic study, the probabilistic study in this chapter consider uncertainties and provide more convincing derivation in the general process of explosion design. Overall, this chapter provides the practical information of safety evaluation procedure for engineers to optimize the safety design in gas explosion mitigation. The relevant studies on the material testing, structural behaviour of the blast wall and the structural response of the integrated structure are not attempted here because of the space limitation of this book.

References Black & Veatch. (2016). Floating liquefied natural gas. https://en.wikipedia.org/wiki/Floating_ liquefied_natural_gas. Boh, J., Louca, L. A., & Choo, Y. (2007). Finite element analysis of blast resistant structures in the oil and gas industry. Gexcon. (2012). FLNG concept explosion study—Explosion risk analysis. Ref. No.: GexCon-2011F40836-RA-01. Gexcon. (2015). FLACS v10.4 user’s manual (Doxygen: Norway). Hansen, O. R., Gavelli, F., Ichard, M., & Davis, S. G. (2010). Validation of FLACS against experimental data sets from the model evaluation database for LNG vapor dispersion. Journal of Loss Prevention in the Process Industries, 23(6), 857–877. Hansen, O. R., & Johnson, D. M. (2015). Improved far-field blast predictions from fast deflagrations, DDTs and detonations of vapour clouds using FLACS CFD. Journal of Loss Prevention in the Process Industries, 35, 293–306. HSE. (2006). Structural strengthening of offshore topsides structures as part of explosion risk reduction methods. In Research Report 489. UK: Health & Safety Executive. Huang, Y. M., Ma, G. W., Li, J. D., & Hao, H. (2016). Multi-level explosion risk analysis (MLERA) for accidental gas explosion events in super-large FLNG facilities. Journal of Loss Prevention in the Process Industries. In Press. Kang, K.-Y., Choi, K.-H., Choi, J., Choi, Y., & Choi, J.-M. (2016). Explosion induced dynamic responses of blast wall on FPSO topside: Blast loading application methods. International Journal of Naval Architecture and Ocean Engineering. In Press. Langdon, G. S., & Schleyer, G. K. (2005). Inelastic deformation and failure of profiled stainless steel blast wall panels. Part I: Experimental investigations. International Journal of Impact Engineering, 31(4), 341–369.

288

12 CFD-Based Explosion Risk Analysis of Blast Wall …

Langdon, G. S., & Schleyer, G. K. (2006). Deformation and failure of profiled stainless steel blast wall panels. Part III: Finite element simulations and overall summary. International Journal of Impact Engineering, 32(6), 988–1012. Li, J. D., Abdel-jawad, M., & Ma, G. W. (2014). New correlation for vapor cloud explosion overpressure calculation at congested configurations. Journal of Loss Prevention in the Process Industries, 31, 16–25. Li, J. D., Ma, G. W., Abdel-Jawad, M., & Huang, Y. M. (2016). Gas dispersion risk analysis of safety gap effect on the innovating FLNG vessel with a cylindrical platform. Journal of Loss Prevention in the Process Industries, 40, 304–316. Li, J., Ma, G., Hao, H., & Huang, Y. (2017). Optimal blast wall layout design to mitigate gas dispersion and explosion on a cylindrical FLNG platform. Journal of Loss Prevention in the Process Industries. Louca, L. A., Boh, J. W., & Choo, Y. S. (2004). Design and analysis of stainless steel profiled blast barriers. Journal of Constructional Steel Research, 60(12), 1699–1723. Louca, L. A., Punjani, M., & Harding, J. E. (1996). Non-linear analysis of blast walls and stiffened panels subjected to hydrocarbon explosions. Journal of Constructional Steel Research, 37(2), 93–113. Norsok. (2001). Norsok standard—Risk and emergency preparedness analysis Z-013. Norway: Norwegian Technology Centre. Schleyer, G. K., Lowak, M. J., Polcyn, M. A., & Langdon, G. S. (2007). Experimental investigation of blast wall panels under shock pressure loading. International Journal of Impact Engineering, 34(6), 1095–1118. Silvestrini, M., Genova, B., & Trujillo, F. J. L. (2008). Correlations for flame speed and explosion overpressure of dust clouds inside industrial enclosures. Journal of Loss Prevention in the Process Industries, 21(4), 374–392. Syed, Z. I., Mohamed, O. A., & Rahman, S. A. (2016). Non-linear Finite Element Analysis of Offshore Stainless Steel Blast Wall under High Impulsive Pressure Loads. Procedia Engineering. 145, 1275–1282. Uijt, P. A. M., & Ale, B. J. M. (2005). Guidelines for quantitative risk assessment. Ministerie van Verkeer en Waterstaat, VROM.

Chapter 13

Standard-Based Lifecycle Risk Management of Explosion Events

Abstract This chapter outlines the prevailing explosion risk management methods based on worldwide industrial standards. Wherein, a lifecycle risk management procedure for explosion accidents based on structural integrity management (SIM) is introduced. A quantitative risk assessment process including frequency analysis and consequence modelling is explained, and risk reduction methods including explosion risk reduction and structural strengthening are demonstrated.

13.1 Introduction to Explosion Risk Management In this chapter, explosion risk management according to industrial standards are introduced. A lifecycle risk management process is proposed against explosion accidents on oil and gas facilities based on the concepts of a structural integrity management (SIM) for offshore structures.

13.1.1 Risk Management In general, risk management is the identification, assessment and prioritization of risks followed by coordinated and economical application of resources to monitor, control and minimize the probability and the consequence of accidental events. The objective of explosion risk management is to minimize the effects of the potential explosion events to ensure the personnel safety and structural integrity of an oil and gas facility. According to ISO 31000 (2009), a risk management process contains five steps: • • • • •

communication and consultation, establishing the context, risk assessment, risk treatment, monitoring and review.

© Springer Nature Singapore Pte Ltd. 2019 G. Ma et al., Risk Analysis of Vapour Cloud Explosions for Oil and Gas Facilities, https://doi.org/10.1007/978-981-13-7948-2_13

289

290

13 Standard-Based Lifecycle Risk Management …

Communication and consultation with external and internal stakeholders should be performed at an early stage, which should be implemented throughout the stages of the risk management process. It ensures that those personnel implementing the risk management process and the relevant stakeholders understand the rationales behind the decisions and actions. To establish the context, the safety organization specifies objectives and defines the external and internal factors to be considered when managing risk, setting the scope and risk criteria for the remaining process. The factors, which are similar to those considered in the design of the risk management framework, and particularly, how they are related to the scope of the particular risk management, should be addressed in greater details. Details for establishing each context are seen in Sect. 5.3, ISO 31000 (2009). Risk assessment usually involves risk identification, risk analysis and risk evaluation as described below. • Risk identification is to specify sources of risk, areas of impacts, events and their causes and potential consequences. This step aims to generate a comprehensive list of risks based on the events that may ensue, enhance, prevent, degrade, accelerate or delay the pursuit of objectives. Exhaustive identification is critical because any risk that is not identified at this stage will not be included in the followed further analysis. • Risk analysis develops an understanding of the risks. It provides inputs to the risk evaluation and decisions on whether specific risks should be addressed and what are the most appropriate risk treatment strategies and methods. Risk analysis involves consideration of the causes/sources of risk, their positive and negative consequences, and the corresponding likelihood. Consequences and their likelihood can be derived by modelling the outcomes of an event or sets of events, or by extrapolation of experimental studies or recorded data. Analysis can be qualitative, semi-quantitative or quantitative according to the specific circumstances. • Risk evaluation is based on the outcomes of risk analysis to identify the risks to be addressed and prioritize the relevant treatment implementation. Risk evaluation compares the derived level of risk against the specified risk criteria during the establishment of the context. Reduction measures should be deployed for the risk if it outstrips the criteria. Risk treatment addresses options for modulating risks and implementing options. Selecting the most appropriate risk treatment option can be fulfilled by balancing the costs and efforts of implementation against the benefits derived with regard to legal, regulatory and other requirements, such as social responsibility and the protection of the natural environment. A number of treatment options can be considered and applied either individually or in combination and a treatment plan should clearly specify the priority order, where the individual risk treatment should be implemented sequentially. Both monitoring and review should be planned integral parts of the risk management process to engage regular checking or surveillance. The results of monitoring

13.1 Introduction to Explosion Risk Management

291

and review should be recorded, externally and internally reviewed and reported regularly, and used as inputs to the review of the risk management framework.

13.1.2 Structural Integrity Management Structural integrity management (SIM) is a continuous risk management process for maintaining the fitness-for-purpose of the facility during their lifespan, especially for offshore platforms (API RP2SIM, 2014). The SIM process was initiated in 1980s and has been continuously developed over 30 years to provide industry a systematic approach to guarantee the continuous safe and reliable operation of the ageing fleet of offshore platforms. The SIM process consists of four main steps, i.e. data collection, evaluation, strategy development and program implementation, as illustrated in Fig. 13.1. The SIM database is constituted of all the information from the original designs, fabrication and installation process, inspections, evaluations, modifications, strengthening and repairing works of the target facility. The SIM process requires the data to be kept up to date and as detailed as possible. Generally, the SIM data can be divided into two main categories of facility characteristic data and structural condition data. The data collection in the SIM process is followed by the evaluation of the facilities to provide the basis for the future strategy and program stages. It should be implemented throughout the lifespan of the facility. Whenever new data is collected, the evaluation process should be executed to identify the possible problems and ensure

DATA

EVALUATION

PROGRAM

STRATEGY

Fig. 13.1 Flow chart of SIM process

292

13 Standard-Based Lifecycle Risk Management …

the fitness-for-purpose of the structure. Structural assessment is a detailed analysis process to evaluate a structure’s fitness-for-purpose. When the owner requires this assessment or any initiator has been triggered from the evaluation process, a more complex structural analysis probably should be implemented. The SIM strategy provides the overall inspection and risk reduction philosophy for the facilities in view of the analysis results of the structural risks. The SIM program refers to the implementation of the detailed work scope and should be executed to accomplish the activities identified in the SIM strategy. To fulfill the SIM process, the data collected in the SIM program should be incorporated into the SIM data management system. Subsequently, the data are applied for next cycle of the SIM process.

Table 13.1 Lifecycle explosion risk-related factors Design stage

Design factors affecting explosion risk

Concept selection and definition

• Processing content (amount of equipment and potential leak and ignition sources) • Structure type • Location of living areas • Equipment layout • Confinement and congestion • Operation philosophy

FEED

• • • • • • •

Philosophy for engineering, piping, etc. Facility sizing Nominal explosion loading Fire area sizing, firewall and blast wall location Determination of constructability Element specific or low-level performance standards set Safety critical element (SCE) categorization—identification of high critical items

Detailed design

• • • • • • •

Design against overpressures and dynamic pressures Finer points of layout Firewater and vent piping, location and schedule Supports for safety critical elements determined Control systems designed Verification of constructability Assembly/writing of maintenance and inspection procedures

Construction

• • • •

Small bore piping runs located Changes to ensure constructability Competence of construction Assembly of decommissioning procedure and assurance of integrity during decommissioning

Operation

• • • •

Maintenance Inspection Change control Hot work procedures

Decommissioning

• Implementation of decommissioning procedures • Implementation of disposal procedures

13.1 Introduction to Explosion Risk Management

293

13.1.3 Lifecycle Explosion Risk Management Explosion risk management is to identify and evaluate the risks followed by strategy application to minimize, monitor and control the probability and the consequence of the undesired explosion events. The personnel safety and structural integrity will be the main objectives of the explosion risk management. During the lifespan of an oil and gas facility, the potential factors affecting the explosion risks are listed in Table 13.1. Based on the aforementioned process of general risk management and SIM, a lifecycle explosion risk management method is proposed, and four main steps are implemented in the proposed explosion risk management: • • • •

risk identification: data collection and hazard identification, risk evaluation: risk screening and detailed assessment, risk reduction: barriers and structure strengthening, risk management implementing and monitoring: implementation and monitoring of risk management process.

13.2 Data Collection The risk management database is constituted of all the information from the original designs, fabrication and installation process, inspections, evaluations, modifications, strengthening and repairing works of the target facility, whereas only data that are relevant to the specific facility under examination should be applied for risk evaluation. The risk management process requires the data to be up to date and as detailed as possible. Generally, the data for explosion risk analysis are divided into three categories, i.e. facility characteristic data, structural condition data and explosion risk-related data.

13.2.1 Facility Characteristic Data The facility characteristic data are the basic data from the construction of the facility. It includes: • • • •

general facility data, design data, fabrication data, installation data. Table 13.2 provides a summary of critical characteristic data.

294

13 Standard-Based Lifecycle Risk Management …

Table 13.2 Summary of facility characteristic data General

Original and current owner/operator Original and current facility use and function Location Facility type—refinery factory, petrol station, offshore platform, etc. Number of equipment Other site-specific information, manning level, etc.

Original design

Design contractor and date of design Design drawings and material specifications Design code Weather criteria—wind, seismic, ice, etc. Equipment arrangement

Construction

Construction contractors and date of construction Approved construction drawings or as-built drawing Fabrication, welding and construction specifications Material documents, such as construction specifications and/or mill certificates and material traceability

13.2.2 Structural Condition Data The structural condition data contain the data variations during the overall lifespan of the facility. The condition data include: • • • • • • •

in-service inspection data, damage evaluation data, corrosion protection data, strengthening/modification/repair (SMR) data, structural modifications, condition monitoring data, operational incident data. Table 13.3 gives the summary of basic condition data.

13.2.3 Explosion-Related Data The explosion-related data refer to those data that is indispensable for assessing the likelihood and the consequence of an explosion event, which includes: • facility conditions, • human and operational factors, • explosion barriers.

13.2 Data Collection

295

Table 13.3 Summary of structural condition data Facility history

Weather loading history—storms, earthquakes, etc. Operational loading history—accidental loads Performance during past accidents Survey and maintenance records Repairs—descriptions, analyses, drawings and dates Modifications—descriptions, analyses, drawings and dates

Present condition

All structures—actual size, location, etc. All structures—existing loading and equipment arrangement Production and storage inventory Appurtenances—current list, sizes and locations Wells—number, size and location of existing conductors

Table 13.4 is a summary of basic explosion-related data.

13.2.4 Data Management In the risk management process, inaccurate and/or insufficient data is usually the main issue due to the operators’ unawareness of the importance of the data. Most operators are not clear about what data they have and where the data are being kept. Therefore, a data management system is indispensable for the oil and gas projects throughout the whole lifespan of the facilities. All data should be filed in a timely manner and archived systematically as permanent record. A data tracking system should be constructed for the engineers to retrieve the data.

13.3 Risk Assessment 13.3.1 Risk Assessment Process To conduct the risk analysis for explosion events, different safety standards usually ensue different corresponding processes (API, 2006; HSE, 2003; ISO 13702, 2015). Most of the assessment processes include three main factors, which are risk screening, detailed assessment and risk reduction, as shown in Fig. 13.2. The simple risk-based screening process is the first-level assessment for explosion events. It is intended to identify the facilities at low risk so that the explosion events to be excluded from a detailed structural assessment. A detailed assessment is generally event-based, which is required when the risk level is derived as medium or high. Risk

296

13 Standard-Based Lifecycle Risk Management …

Table 13.4 Summary of explosion-related data Facility conditions

Equipment numbers and layout information Leak sources count Ignition sources count Process material and pressure Confinement conditions and venting conditions Congestion conditions and blockage ratio (e.g. pipeline length, etc.)

Human and operational factors

Explosion barrier installation

Personal characteristics

• Competence • Working load/stress fatigue • Work environment

Task characteristics

• • • • • •

Methodology Task supervision Task complexity Time pressure Tools Spares

Characteristics of the technical system

• • • • • • •

Equipment design Material properties Process complexity Human machine interface (HMI) Maintainability/accessibility System feedback Technical condition

Administrative control

• Procedures • Work permit • Disposable work descriptions

Organizational factors/operational philosophy

• • • • • • •

Detection device Electricity isolation Emergency system shutdown Artificial vent Water deluge Inert gas Blast walls Blast relief panel Soft barriers Safety gap Alarms

Programs Work practice Supervision Communication Acceptance criteria Simultaneous activities Management of changes

13.3 Risk Assessment

297

Risk screening

Satisfy screening criteria

No

Yes

Detailed assessment

Assessment failed

Risk reduction

Assessment passed

Assessment complete

Fig. 13.2 Risk assessment process

reduction and mitigation methods should be deployed until the detailed assessment satisfies the risk criteria. HSE risk assessment process Figure 13.3 presents a flowchart of the risk assessment process from HSE (2006) structural strengthening of offshore topside structures as part of explosion risk reduction methods. This chart illustrates the application of the risk control and mitigation methods from both structure design and strengthening aspects.

13.3.2 Detailed Assessment In general, detailed assessment of explosion risks implements quantitative analyses for frequency estimates and consequence modelling. Some popular QRA methods, such as event tree, fault tree and Bayesian network have been introduced in Sect. 7.3. In the present section, other methods of frequency analysis and consequence evaluation will be introduced. Frequency analysis Frequency analysis estimates the likelihood of each failure case that has been specified in the hazard identification stage (Spouge, 1999). Frequency is the expected number of occurrences of the event per unit time, usually a year, but sometimes an hour or a project time. A frequency has the unit of 1/time and may take any positive value.

298

13 Standard-Based Lifecycle Risk Management …

New or existing facility assessment

Inputs to assessment

• • • • •

Safety case New structural data New loading data Life extension Change in use

Risk Screen

Risk acceptable?

Yes

OK

No

• Explosion inventory and development control • Equipment and structural layout • Deluge application

Consider control and mitigation measure

• Strengthening & ductility measures for blast walls • Strengthening & ductility measures for topsides • Strengthening & ductility measures for connections

Re-assess risk

Revise control & mitigation strategy

No

Yes

• Compare alternatives using CBA & ALARP against same base case • Risk acceptable?

Fig. 13.3 HSE risk assessment process (HSE, 2006, Permission from HSE)

Historical frequencies are estimated from the record of actual events and the associated exposure. An event frequency is usually calculated as follows, Event frequency =

Number of events Associated exposure

(13.1)

The associated exposure is a measure of the population size, from which the events have been recorded. This is usually a number of items or a number of years. Both the accidental experience and the exposure should be comprehensive collections from the same population. For example, if there have been 10 gas explosions on production platforms over a 20-year period, the frequency can be expressed as: Frequency =

10 explosions = 0.5 explosions per year 20 years

(13.2)

Supposing there is an average of 100 production platforms in service simultaneously, the frequency can be expressed as: 10 explosions 20 years × 100 platforms = 5 × 10−3 explosions per platform year

Frequency =

(13.3)

13.3 Risk Assessment

299

As mentioned in Sect. 2.3, insufficient and inaccurate data can always be the knottiest problems for quantitative risk analysis. Some of the data sources are listed below. • Accident databases: several regulatory authorities and other organizations have their own accident databases with the linked exposure data, from which frequencies can be determined. • Published accident frequency analyses: several organizations have published accident data in the form of frequencies that are suitable for risk analysis. • Published risk assessments: this involves previous risk assessments, conference papers, magazines, etc. • Reliability data: for individual equipment item, there exist several reliability databases that provide failure rates. However, those data usually should be appropriately addressed in a reliability analysis, such as fault tree analysis, before they can be used as frequencies in a QRA. • Accident and exposure data collections: all historical frequency data originates from collections of accident data and the associated experience. In many cases, the best practice to derive frequency estimates is to make one’s own collection and analysis of the data. Consequence modelling To model the consequence of an explosion event, the first step is to estimate the explosion overpressures. It is impractical and unreliable to predict overpressures in detailed assessment using simple qualitative methods. Some more complicated methods including empirical models, phenomenological models and numerical simulations are available to predict explosion overpressures on oil and gas structures. The following step is to model the escalation of an explosion event. Escalation may occur as a result of the failure of the key items of structure or equipment subject to blast load. In order to prevent the escalation, oil and gas facilities are deployed with a wide variety of active and passive fire and blast protection systems. The potential failure of these systems due to fire or explosion is one of the key factors that may lead to escalation. It should be noted that the modelling of the consequences of escalation is highly subjective and specific case oriented. If structural failure is predicted to occur, consequent effects are usually evaluated based on personnel judgement. To consider the total losses (C T ) of a structure collapse, El-Reedy (2012) proposed a method to convert the three main consequences, i.e. environmental losses (C E ), business losses (C B ), and injuries and safety-related losses (C S ) to monetary value (USD). This method was proposed for offshore installation and will be briefed herein as an example. Accordingly, total loss can be calculated as CT = CE + CB + CS

(13.4)

300

13 Standard-Based Lifecycle Risk Management …

• Environmental Losses Environmental losses are calculated according to the amount of liquid that is spilled. This amount is usually the daily amount, or the amount of liquid released. There are two types of costs in the environmental losses, the fixed cost and the variable cost. The mobilization of personnel and equipment to perform the clean-up service and the regulatory cost constitute the fixed cost. The variable costs will be determined by the volume of the oil spill. Both fixed and variable costs vary with different global regions. The differences are described in Table 13.5. It takes the mathematical form of CE = f (d) × {FC + Vc × min(DP , R)} f (d) = 1; d > dm   ds − d 2 f (d) = 1 + ; d ≤ dm ds

(13.5)

where C E is the environmental loss in USD; V C denotes the variable cost in USD/bbl; F C represents the fixed environmental cost in USD; R stands for the minimum released oil volume in bbl; DP denotes the daily production in barrel of liquid bpl; d represents the distance offshore in km s and d s is the maximum significant distance offshore, which is the distance from the shoreline in km. • Business losses The business losses come from the replacement cost and the deferred production losses due to the complete or partial platform failure. The monthly net production can be calculated as   PM = DPoil × Poil + DPgas × Pgas × (30 days)

(13.6)

where DP is the daily production in bpd or MSCFD; Poil is the oil price/bbl in USD/bbl; Pgas denotes gas price/Mscf in USD/Mscf and PM represents the net monthly production. The deferred production loss is equal to the difference between the present value and the reserved value discounted for the time of stoppage at a given interest rate. So that PV = PM /rm CDP = PM

n  k=1

1 k 1 + rm

(13.7) (13.8)

where PV is the investment present value, which is used to model the oil and gas reserve herein; r m stands for the monthly return on the investment that is equal to

13.3 Risk Assessment

301

Table 13.5 Differences in default values for consequence costs Variable

Location

Units

Description

10%

%

Discount rate

$110.00

$/bbl

Oil revenue per bbl

$1.50

$/Mscf

Gas

$1 MM

$

Personnel marginal exposure cost

400

Days

Default deferred production period

$350 M

$500 M

USD

Fixed environmental cost in open sea

$500

$500

$500

$/bbl

Variable environmental cost in open sea

65

25

25

50

Miles

Default distance offshore

100

100

100

100

100

Miles

Distance offshore to open sea

–

$5 M

$5 M

$5 M

$5 M

$5 M

$/ton

Default replacement value

M

$150 M

$150 M

$150 M

$150 M

$150 M

$

Default daily production value

R

250

250

250

250

250

bbl

Default spill volume

N crew

20

20

20

20

20

Persons

Default crew size: service unknown

N crew

30

30

30

30

30

Persons

Default crew size: production platform

N crew

20

20

20

20

20

Persons

Default crew size: drilling platform

N crew

50

180

50

50

50

Persons

Default crew size: quarters platform

N crew

10%

10%

10%

10%

10%

%

Percent of crew exposed if evacuated

G

N

C

M

E

I

10%

10%

10%

10%

Poil

$120.00

$130.00

$110.00

$130.00

Pgas

$1.50

$3.00

$1.50

$1.50

C ex

$1 MM

$1 MM

$1 MM

$1 MM

N

200

545

180

180

FC

$500 M

$750 M

$400 M

VC

$500

$500

D

50

dm

(continued)

302

13 Standard-Based Lifecycle Risk Management …

Table 13.5 (continued) Variable

Location

Units

Description

2

Persons

Default crew size: unmanned platform

20%

%

The percent increase in the expected safety losses in gas production platforms

G

N

C

M

E

N crew

2

2

2

2

–

20%

20%

20%

20%

G = Gulf of Mexico, N = North Sea, C = Canada, M = Malaysia and E = elsewhere

approximately (r/12); r represents the annual rate of return on investment; n = N/30, where N is the total number of downtime days. The business loss C B has the form of CB = CDP + CR

(13.9)

where C R is the replacement cost in USD; C DP is the deferred production loss in USD. • Safety losses Safety loss is equal to the crew number time the marginal safety cost per crew member. Adjustments for safety loss as follows arise from the considerations of the specific conditions of the target. • The consequence value is suggested to be increased around 50% when the platform processes H2 S. • The value is decreased by 10% when the platform is evacuated during the severe environmental events. • The penalty is increased by 20% for gas platforms. So that, there are CS = Cex × N × G

(13.10)

G = 1, if DPgas = 0; G = 1.2, if DPgas = 0 herein C S is the safety-related losses; C ex represents the marginal safety cost per crew member; N stands for the number of crew on a platform; G denotes the penalty factor for gas platforms; DPgas = 0 implies non-gas platform. Some of the factors from the above equations are different for different regions around the world as listed in Table 13.5.

13.3 Risk Assessment

303

13.3.3 Risk Acceptance Criteria Risk criteria are usually applied to convert numerical risk estimates, e.g. 10−7 per year, derived from a risk analysis into qualitative dissertations, such as negligible risk to be commensurate with other judgements, or high economic benefits in the decision-making process. Risk criteria comprise the technical aspects of the decision-making process that is one of the key threads leading quantitative risk analysis into risk management as a whole. The simplest framework for risk criteria is a single risk level that dissertates tolerable risk from intolerable ones. The attraction of such criteria is the simplicity, whereas discretion should be exerted for its application because no uncertainty is reflected in estimating risks and in assessing what is tolerable. Therefore, a more flexible framework, as low as reasonably practical (ALARP) is introduced by HSE (2017a). It divides risks into three bands as shown in Fig. 13.4, which is explained as below. • An intolerable region: risks are intolerable in this region expect in extraordinary circumstances, and thus, additional risk reduction measures are indispensable. • A tolerable region: it is usually specified as an ALARP region, which indicates that the risks are considered acceptable providing that they have been made as low

Fig. 13.4 Illustration of ALARP (HSE, 2017b, Permission from HSE)

304

13 Standard-Based Lifecycle Risk Management …

as reasonably practicable. It implies that risk reduction measures are desirable but may not necessarily be implemented if a cost-benefit analysis shows that their cost is disproportionate to the benefit to be achieved. • A broadly acceptable region: risks in this region are tolerable and no risk reduction measures are required. Usually, the process industry level of risk for a particular facility is based on one of the two parameters: the average risk to the individual, i.e. fatality accident rate (FAR) or potential loss of life (PLL) and the risk of a catastrophic event at the facility. It has been generally acknowledged in the petroleum and chemical industries that the average risk for an individual at a facility should not exceed a value in the order of 10−3 per year. The facility risk is the total frequency of each main type of incident and it should not usually exceed a value in the order of 10−4 per year for most petroleum and chemical facilities.

13.4 Risk Reduction For oil and gas facilities, risk reduction can be achieved by either reducing explosion risks or/and increasing structural resistance to overpressures.

13.4.1 Explosion Risk Reduction Measures Risk reduction includes consequence mitigation and likelihood reduction at all stages of the explosion management. Measures to reduce explosion risks are listed and briefly explained in Tables 13.6 and 13.7 based on facility design and barrier installations.

13.4.2 Structural Strengthening Structural strengthening methods are usually adopted to reduce both the likelihood and consequence for an explosion event by increasing the system strength capacity. API RP2SIM (2014) introduced some structural strengthening methods for offshore installations to be implemented in the strengthening, modification and repairing techniques (SMR). These methods are briefly explained in Table 13.8. For other types of oil and gas facilities, different structural strengthening methods may be applied based on the specific conditions and the problems detected.

13.4 Risk Reduction

305

Table 13.6 Explosion risk reduction methods from design aspects Measures

Descriptions

Equipment minimization

Leak frequency is proportional to the number of process equipment of facilities. Therefore, the simplest possible process systems are desirable

Inventory minimization

The inventory in the process system may be related to the duration of any leak and the time required for blowdown

Inventory Pressure

Flammable cloud size is determined by the leak dimension and the pressure of the inventory. Reduced inventory pressure will reduce explosive cloud dimensions, and thus, the severity of the explosion event. It will also result in a lower inventory mass within the system which will give the potential for a more rapid blowdown and reduced escalation consequence

Operations and maintenance procedures

Errors in maintenance and operating procedures are important causes of leaks. The potential effect of improvements in these areas on the leak frequency is mainly judgmental at present, although human reliability modelling may give some hints

Ventilation

The ignition probability depends on the gas concentration and the ignition sources in this area. Free or forced ventilation is able to reduce the gas concentration

Installation orientation

If, as is often the case, one side forms a solid partition, then, orientation will affect the ventilation air change rate within the module

Ignition source minimization

In general, the main ignition sources are welding/hot work, compressors, electrical equipment and engines/exhausts. Removing or minimizing some of those sources are possible. For instance, lights can be switched off when not needed

Ignition source location

Highest overpressures in congested modules tend to arise when the ignition point is at the furthest point from a main vent. Although there is potential for ignition to occur practically at any point within the module, removing ignition sources away from such extremities will, to some extent, lower the potential for high explosion overpressure to occur

Minimization of congestion

Explosion events are most likely to occur in congested areas, and therefore, reducing congestion in the modules can decrease both the probability of explosion and the overpressure if an explosion does occur

Emergency procedures

Local fatalities may be prevented from exposure of explosion if the personnel in the area is alerted to a leak by alarms or their own observation, and thus, can escape from the area before the ignition occurs. It can be implemented in the emergency procedures

Integrity of living areas

The integrity of the living areas or any other area used as a temporary refuge during accidental events should be secured against smoke, gas, fire and blast impacts

306

13 Standard-Based Lifecycle Risk Management …

Table 13.7 Explosion risk reduction method based on barriers Blast relief panels

The overpressure can be diverted away from potential escalation sources by blast relief panels. Blast relief panels will open quickly during an explosion to reduce peak overpressures

Emergency shutdown systems (ESD)

An effective ESD system will restrain the inventory released in an incident, and therefore, the size and duration of any resulted cloud size. The location of the ESD valves will be determined according to the areas where each particular inventory could be released

Isolation and blowdown

A leak may be reduced by isolating it manually or using the ESD system, and depressurising the leaking section using the blowdown system. Damage or fatality risk in escalation can be reduced by isolation and blowdown, and sometimes evacuation may be avoided

Blast wall

Blast walls have long been used to protect adjacent areas from the effects of overpressure. These walls are designed to absorb blast energy by displacement

Water deluge

Deluge has been found suitable for reducing overpressure in congestion-generated explosion. If explosion mitigation is considered critical, a deluge flow-rate of at least 13–15 l/min/m2 is recommended for general area coverage

Artificial vent

Artificial vent is the ventilation, which is not governed by the environmental wind alone. Upon detection of flammable gas, the stand-by fan(s) is switched on to give maximum possible ventilation to dilute the leak. Thus, the generation of an explosive cloud is prevented or restrained

Inert gas

Inert gas, e.g. CO2 , N2 , can be used to dilute the flammable mixture by flooding the volume, within which the gas has been detected. The density of the explosive gas can then be reduced below its lower explosive limit

Detection device

Detection measures can be applied to identify hazardous conditions on the plant, such as excessive process pressure, an unignited release of flammable gas or a fire. Detection devices trigger controlling or mitigation measures as emergency response to accidental events

Alarm

The alarm system may alert operators to mitigate leaks before they ignite, or at least to evacuate the area

Soft barriers

Progress has been made in the manufacture of soft barriers, such as the micro-mist device consisting of a cylinder of superheated water, which is released quickly as a fine mist in response to pressure or flame sensors during an explosion. This device could suppress the explosion and significantly reduces overpressures

Safety gap

In the process industry, the safety gap of an open space without congestion is purposely placed in between the congested process areas. The absence of obstacles in a safety gap eliminates the fluid–obstacle interaction, thereby preventing the generation of turbulence. It can be very effective in reducing pressures prior to the onset of detonation

13.4 Risk Reduction

307

Table 13.8 Brief explanation of structural strengthening methods Structural strengthening methods Damage removal Member removal

Only applicable when the member is identified to be no longer required for the structure during the assessment process

Crack removal

This may be achieved by remedial grouting

Localized strengthening or repair Member grouting

Including completely member grouting or grouting only the dented part of the members

Joint grouting

Improving the static strength of the joint and also increasing the fatigue life of the joint

Structural clamps

An effective way to repair brace members or joints of jacket structures. Also, it is useful for connecting external brace to additional piles in a global strengthening scheme, adding new members into a structure to reinforce existing members and/or joints

Welding

Welding is usually considered as the best method of strengthening or repairing technique and it includes: – dry welding at or below sea surface at an atmosphere with a cofferdam or a pressure-resisting chamber – hyperbaric welding using habitats – underwater wet welding

Bolting

Bolts are useful for topside repair because the platform does not need to be shutdown when a bolted joint is installed on the topside in a hazardous area. Tension and corrosion problems should be considered for configuration and material selection of bolts

Member removal

This may be a staged measure in a comprehensive repair scheme or may constitute in its own right

Member flooding

The intentional flooding of structural members may be used as a method to increase the load carrying capacity of the member. The impact of the increased gravity loads and dynamic mass, and the influence on decommissioning should be considered

Adhesives and epoxy grouting

Resins are able to be used offshore as adhesives, grout and the matrix in composite materials

Cold forming

Cold forming technique involves two main categories: mechanical connectors and swaging

Global strengthening or repair Leg-pile annulus grouting

Grouting of the annulus between the jacket legs and piles provides a reliable and economical way to increase the global capacity of the platform. There are three issues need to be addressed before grouting the main piles. They are the impact on platform decommissioning, the influence of the increased weight and the increase in dynamic mass

External bracing

Additional external bracing to additional piles can globally strengthen the structures of small platforms. Either welded or clamped connections are suitable for connecting the external braces to the structure

308

13 Standard-Based Lifecycle Risk Management …

The most appropriate structural strengthening methods should be selected based on the comprehensive consideration of technics, costs and safety aspects. Some basic considerations for selection of structural strengthening methods are listed below: • • • • • • • • • • •

safety of construction, operation, diving and diving support personnel, potential use of diver-free techniques, difficulty in fabrication, handling and installation, rigging complexity and layout, support vessel type, availability and access, fit-up tolerance of clamps and members, interference with conductors, jacket members and appurtenances, possibility of collision with existing risers and control bundles, requirements for predesign inspection, field measurements and material samples, outfitting with well-designed installation aids, required weather window.

13.5 Risk Management Implementation and Monitoring 13.5.1 Implementation Risk management implementation carries out the detailed work scope to accomplish the activities identified in the previous risk management processes. The risk management implementation should fulfil the following works. • • • •

Define the appropriate timing and strategy for implementing the framework. Apply the risk management policy and process to the organizational processes. Comply with legal and regulatory requirements. Ensure that decision making, including the development and setting of objectives, is aligning with the outcomes of risk management processes. • Communicate and consult with stakeholders to ensure that its risk management framework remains appropriate and up to date. Risk management should be implemented to ensure that the risk management process complies a risk management plan at all relevant levels and functions of the organization as part of its practices and processes.

13.5.2 Monitoring To ensure that risk management is effective and continues to support organizational performance, both monitoring and reviewing should be planned and integrated into the risk management process to maintain regular checking or surveillance. The monitoring and reviewing process includes:

13.5 Risk Management Implementation and Monitoring

309

• measuring risk management performance against indicators, which are periodically reviewed for appropriateness, • periodically checking progress against and deviation from the risk management plan, • periodically reviewing whether the risk management framework, policy and plan are still appropriate given the organizations’ external and internal context, • reporting on progress with the risk management plan and how well the risk management policy is followed, • reviewing the effectiveness of the risk management framework. The monitoring and reviewing process should encompass all aspects of the risk management process for the purposes of: • ensuring that controls are effective and efficient in both design and operation procedures, • deriving further information to improve risk assessment, • analysing and learning lessons from events, changes, trends, successes and failures, • identifying changes in the external and internal context, including changes to risk criteria and the risk itself, which may demand revision of risk treatments and priorities, • Identifying emerging risks. The results of monitoring and reviewing should be recorded and externally and internally reported as appropriate. Based on the results of monitoring and reviewing, decisions should be made on how the risk management framework, policy and plan can be improved. These decisions should lead to improvements in the organization’s management of risk and the risk management culture.

References API. (2005). Recommended practice for planning, designing and constructing fixed offshore platforms—Working stress design. Section 17: Assessment of existing platforms. Washington, D.C.: American Petroleum Institute. API. (2006). Recommended practice for the design of offshore facilities against fire and blast loading (1st ed.). Washington, D.C.: American Petroleum Institute. API, API RP 2SIM. (2014). Structural integrity management of fixed offshore structures (1st ed.). Washington, D.C.: American Petroleum Institute. El-Reedy, M. A. (2012). Offshore structures design, construction and maintenance. Waltham, MA, U.S.A.: Gulf Professional Publishing, imprint of Elsevier. HSE. (2003). Fire, explosion and risk assessment topic guidance, Issue no. 1. HSE. (2006). Structural strengthening of offshore topsides structures as part of explosion risk reduction methods. The steel Construction Institute. HSE. (2017a). ‘ALARP “at a glance”’. Available at http://www.hse.gov.uk/risk/theory/alarpglance. htm. HSE. (2017b). ‘Guidance on ALARP Decisions in COMAH’. Available at http://www.hse.gov.uk/ foi/internalops/hid_circs/permissioning/spc_perm_37/.

310

13 Standard-Based Lifecycle Risk Management …

ISO 13702. (2015). Petroleum and natural gas industries—Control and mitigation of fires and explosions on offshore production installations—Requirements and guidelines. ISO 31000. (2009). Risk management—Principles and guidelines (1st ed.). Switzerland. ISO 6184/4. (1985). Determination of the efficacy of explosion suppression systems. Spouge, J. (1999). A guide to quantitative risk assessment for offshore installations. CMPT.