Oil and Gas Processing Equipment

Risk Assessment with Bayesian Networks

G. Unnikrishnan

• Éditeur: CRC Press
• Année: 2020
• ISBN:
  9781000174236
• Édition: 1
• Sujet: Technology & Engineering
• Sous-sujet: Chemical & Biochemical Engineering

Oil and gas industries apply several techniques for assessing and mitigating the risks that are inherent in its operations. In this context, the application of Bayesian Networks (BNs) to risk assessment offers a different probabilistic version of causal reasoning. Introducing probabilistic nature of hazards, conditional probability and Bayesian thinking, it discusses how cause and effect of process hazards can be modelled using BNs and development of large BNs from basic building blocks. Focus is on development of BNs for typical equipment in industry including accident case studies and its usage along with other conventional risk assessment methods. Aimed at professionals in oil and gas industry, safety engineering, risk assessment, this book

• 
  
  

• Brings together basics of Bayesian theory, Bayesian Networks and applications of the same to process safety hazards and risk assessment in the oil and gas industry

• 
  
  

• Presents sequence of steps for setting up the model, populating the model with data and simulating the model for practical cases in a systematic manner

• 
  
  

• Includes a comprehensive list on sources of failure data and tips on modelling and simulation of large and complex networks

• 
  
  

• Presents modelling and simulation of loss of containment of actual equipment in oil and gas industry such as Separator, Storage tanks, Pipeline, Compressor and risk assessments

• 
  
  

• Discusses case studies to demonstrate the practicability of use of Bayesian Network in routine risk assessments

1

Introduction

The oil and gas industry handle highly inflammable and toxic fluids under severe operating conditions and have high inventories of the same. Given the hazardous nature of its operations, it is important that the industry ensures such facilities are designed, maintained and operated in a safe manner. Different methods have evolved over time to analyze and mitigate the risks involved. However, major accidents continue to occur and at that time issues on safety and risk assessment come up. For example, risk assessment came into sharp focus during incident investigations of major accidents in British Petroleum’s Texas City Refinery, Buncefield fuel storage and Jaipur tank farm, India. Oil and gas industry typically use Quantitative Risk Assessment (QRA) methodology to analyze and understand risks in its facilities. The method started in nuclear industry and was later adopted in process as well as oil and gas industries. Based on practice of more than 25 years, the industry is aware of the limitations of the method.

Bayesian Network (BN) or Bayesian Belief Network (BBN) is being applied productively as probabilistic risk assessment method in several areas like medicine, computer science, ecology and chemical industry. The method offers certain advantages over QRAs and reveals a fuller risk profile with causes and effects. This book focuses on the application of BN methods to assess risk of major hazards in oil and gas industry. Specifically, the book presents BN models for the major hazards for oil and gas separator, atmospheric hydrocarbon storage tanks, hydrocarbon pipelines and centrifugal compressors and pumps. Examples of the BN models’ simulation with generic data are also given.

1.1 Application of BNs for Risk Assessment

BN have been applied with good results in the areas such as computer science, ecology, medicine and chemical industry (Fenton and Neil, 2012; Pouret, Naim, and Marcot, 2008; Kjaerulff and Madson, 2008; Korb and Nicholson, 2010; Hubbard, 2009). However, applications of BN to oil and gas facilities have been very limited specifically with regard to the risks due to major hazards in oil and gas facilities. Loss Of Containment (LOC) scenarios constitute major hazards in the oil and gas facilities. Therefore, causal mechanisms and BN have been developed for such scenarios for the more common equipment, namely oil and gas separator, hydrocarbon pipelines, atmospheric hydrocarbon storage tanks and centrifugal compressor and pumps. The BNs are then simulated and analyzed with generic data.

1.2 The Readership

The book is primarily meant for practicing engineers and researchers in the oil and gas industry as well for graduate students in Process Safety who want to explore the risk assessment area. Rather than describing the theoretical aspects of BN in detail, they cover only the extent necessary for the continuity of the topic. Adequate references have been provided for the interested reader to pursue the relevant topics. The practical aspects are emphasized and given in more detail since that will be what the practicing engineers would want to know.

1.3 Major Limitations of QRA

The main motivation for this book arose out of the author’s own experience in the oil and gas industry for over three decades. QRA is well embedded in the process safety studies as a tool for land use planning and spacing of critical units in oil and gas facilities CCPS (2001) and Tweeddale, (2003). However, the majority of the QRAs done during the design stage of a facility usually end up in the records section or library shelves. During operational phase there is very little or no attempts to update these QRAs. When changes are made to the facilities, most of the time QRAs are done only for that portion that undergoes change, which has proven to be fundamentally wrong.

Industry and academia are aware of the limitations of QRA. Please see the ASSURANCE project Lauridsen, Kozine, and Markert (2002) and also Pasman and Rogers (2013); Pasman, Jung, Prem, Rogers, and Yang (2009). The author’s personal experience in conducting QRAs also provided the first-hand proof of these limitations. In summary they are as follows:

1. Uncertainties in data for failure frequencies, lack of precision in models and difficulties in identifying common cause failures.
2. Assumptions are not visible to all concerned.
3. Models are static, difficulties in capturing variations/changes to the facility
4. Requires considerable specialist efforts and time
5. Software is costly, calculations are not transparent and limit flexibility
6. The causes of loss of containment are not investigated in detail

Further comparisons between QRA and BN methodologies are given in Chapter 9.

1.4 BN and Its Advantages

BN is seen as a viable alternative to and/or complementary QRA methodology (Weber, Oliva, Simon, and Iung, 2012; Roy, Srivastava, and Sinha, 2014). As noted earlier, BN is being widely applied to computer science, ecology, finance and chemical industries. Bayesian approaches have also been the subject of a popular book (McGrayne, 2011).

However, oil and gas applications are generally limited to (Kalantarnia, Khan, and Hawboldt, 2010; Kujath, Amyotte, and Khan, 2010; Khakzad, Khan, and Amyotte, 2011; Pasman, and Rogers, 2012; Rathnayaka, Khan, Amyotte, 2012; Khakzad, 2012; Khakzad, Khan, and Amyotte, 2013a, 2013b, 2013c; Cai, Liu, Zhang, Fan, Liu, and Tian, 2013; Abimbola, Khan, and Khakzad, 2014; Ale, Gulijk, Hanea, Hudson, Lin, and Sillem, 2014; Tan, Chen, Zhang, Fu, and Li, 2014).

Main advantages of BN are as follows:

1. It presents the risk in a visually and easily understandable manner
2. The methodology is transparent.
3. Failure data and thereby the risk profile can be easily updated in line with changes/updates of the facility
4. Site-specific data (even if it is sparse) and experts’ opinion can be incorporated.
5. Layers of interconnecting causes of loss of containment can be fully explored.
6. BN can be simulated in predictive and diagnostic mode since causes and effects with their relationships are represented in a transparent model.

In summary, while QRA has its place in Land Use Planning (LUP) and safe distances, Bayesian techniques offer models that can represent a version of cause and effect.

1.5 Scope of the Book

The scope covered in the book is given below:

1. Identification of major hazards and Layers of Protection provided to equipment units in a typical Oil & Gas facility. These are obtained by review of several designs, Piping & Instrument diagrams, Hazard and Operability Study (HAZOP) study & Layers Of Protection Analysis (LOPA) reports from industry.
2. Development of causal relationship networks for critical equipment/systems failures & its causes, hazards & consequences using the above data
3. Conversion of these causal relationships to BN.
4. Simulation of the networks using a suitable software. Testing of the networks with data.

The sources and information required for building the BN is given in Figure 1.1:

FIGURE 1.1
Sources and information required for building the BN.

In order to develop cause and effect relationships, relevant process safety documents, namely HAZOP, LOPA and Safety Integrity Level (SIL) study reports were studied in detail. These are actual reports from the industry and due to confidentiality are not listed here. Several accident investigation reports were also studied in detail to analyze the root causes that led to such accidents (Lal Committee, 2011; Buncefield Major Incident Investigation Board, 2007, 2008; Herbert, 2010). In parallel, the techniques of developing BN were reviewed to select the right approach to model the cause and effects/influence diagrams (Fenton and Neil, 2012; Kjaerulff and Madson, 2008; Korb and Nicholson, 2010).

BN requires parameterization with failure/incident data. Failure data from several data sources were analyzed to parameterize the BN and the same are given in Table 3.4. In certain cases, expert opinions were also sought. An inter-disciplinary approach was required, and materials from many sources (Gulvanessian, and Holický, 2001; Lannoy and Cojazzi, 2002; Bayraktarli, Ulfkjaer, Yazgan, and Faber, 2005; Straub, 2005; Twardy, Nicholson and Korb, 2005; ANU Enterprise-The Murray Darling Basin Authority, 2016) were used to understand the application of BN to risk assessments.

1.6 Structure of the Book

The book is written with nine Chapters. The contents of each chapter are summarized below:

Chapter 1 Introduction provides an overview of the topic as well as the background and motivation for writing this book. Purpose, objectives and scope are given in this chapter.

Chapter 2 Bayes Theorem, Causality and Building Blocks for BN contains basics of probability, description of Bayes theorem and how it can be applied to model cause and effect. A summary of use of discrete and continuous distributions in representing the nature of causes is provided. Further, it describes how complex cause and effect mechanisms can be visually represented as BN using simple building blocks and directed graphs and presents two examples to illustrate the flexibility and power of the BN.

Chapter 3 Bayesian Network for Loss of Containment from Oil and Gas Separator presents the immediate and root (parent) causes for an LOC scenario in a typical oil and gas separator. Causes for LOC as well as the post event scenario are modeled in BN. Application of BN to SIL calculations is also given. Sensitivity feature of BN and how it can be used to find out the sensitivity of other nodes to a target node are illustrated in this chapter.

Chapter 4 Bayesian Network for Loss of Containment from Hydrocarbon Pipeline gives the application of BN to an LOC scenario from a hydrocarbon pipeline. The immediate and root causes as well as the post LOC event scenarios are modeled as BN. Predictive and diagnostic modes of simulating the BN are described. Sensitivities of parent nodes to target ...

Table des matières

• Cover
• Half Title
• Title Page
• Copyright Page
• Dedication
• Table of Contents
• Preface
• Author
• 1. Introduction
• 2. Bayes Theorem, Causality and Building Blocks for Bayesian Networks
• 3. Bayesian Network for Loss of Containment from Oil and Gas Separator
• 4. Bayesian Network for Loss of Containment from Hydrocarbon Pipeline
• 5. Bayesian Network for Loss of Containment from Hydrocarbon Storage Tank
• 6. The Jaipur Tank Farm Accident
• 7. Bayesian Network for Centrifugal Compressor Damage
• 8. Bayesian Network for Loss of Containment from a Centrifugal Pump
• 9. Other Related Topics
• References
• Index