Vulnerability analysis of process plants subject to domino effects

doi:10.1016/j.ress.2016.06.004

Reliability Engineering & System Safety

Volume 154, October 2016, Pages 127-136

https://doi.org/10.1016/j.ress.2016.06.004 Get rights and content

Highlights

•
Graph theory is a reliable tool for vulnerability analysis of chemical plants as to domino effects.
•
All-closeness centrality score can be used to identify most vulnerable installations.
•
As for complex chemical plants, the methodology outperforms Bayesian network.

Abstract

In the context of domino effects, vulnerability analysis of chemical and process plants aims to identify and protect installations which are relatively more susceptible to damage and thus contribute more to the initiation or propagation of domino effects. In the present study, we have developed a methodology based on graph theory for domino vulnerability analysis of hazardous installations within process plants, where owning to the large number of installations or complex interdependencies, the application of sophisticated reasoning approaches such as Bayesian network is limited. We have taken advantage of a hypothetical chemical storage plant to develop the methodology and validated the results using a dynamic Bayesian network approach. The efficacy and out-performance of the developed methodology have been demonstrated via a real-life complex case study.

Introduction

Severity of consequences of potential hazards in a system is a function of both the magnitude of hazards and the vulnerability of the system. In other words, exposed to the same level of hazard a system with higher vulnerability is susceptible to larger damages and thus suffering higher levels of risk. Johansson et al. [1] use the term vulnerability to address the inability of a system to withstand the failures. In the context of domino effect modeling, Khakzad and Reniers [2] defined the domino effect vulnerability as the capability of a plant to escalate a primary accident (fire or explosion) to higher order accidents (e.g., secondary fires and explosions). For the purpose of the present study, we adopt the definition by Khakzad and Reniers [2] and define domino vulnerability as the susceptability of a plant which allows a primary accident to spread throughout the plant via cascading effects, triggering secondary accidents, and so on. Compared to traditional risk analysis which is aimed at identification of hazards and estimation of the system failure probabilities, vulnerability analysis is usually performed to pinpoint the system components the failures of which contribute most to the cascading of failures not only within the system of interest but also across other interdependent systems. As such, the emphasis of vulnerability analysis is more on the extent of failures rather than the probabilities thereof although the incorporation of the vulnerability analysis in the probabilistic risk analysis can address both.

Over the past two decades, the issue of cascading effects in complex and interdependent systems such as water distribution networks [3], power grids [4], [5], and process plants [6], [7], [8], [9], [10], [11], [12], [13], [14], [15] has drawn much attention. The increasing trend in modeling and risk analysis of cascading effects – better known as domino effects in hazardous industries such as process plants – mainly lies in the fact that such cascading failures, although rare, can result in catastrophic consequences. Compared to other infrastructures, however, the potential consequences of domino effects in process plants¹ can be much more severe owing to the presence of hazardous materials such as flammable, explosive, and toxic substances. For example, a series of explosions in a LPG² storage plant in Mexico in November 1984 left around 600 deaths and 7000 severe injuries; similarly, a series of fires and explosions in Buncefield oil storage depot in the U.K in December 2005 led to the largest fire in peace time Europe, leaving 43 injuries and incredible property damages.

Compared to long-established methods for accident modeling and risk analysis of domino effects in process plants, relevant work in the field of vulnerability analysis has been relatively few [13], [15], [16]. In most previous work, however, either a full simulation of potential domino effects has been performed to identify the units contributing the most to the vulnerability of the plant or an iterative deterministic analysis has been carried out with one failure at a time to evaluate the extent of the failure cascade. Regardless of which aforementioned approaches is used, simulations can turn out too time-consuming and even intractable in case of large process plants containing many process installations and equipment.

Many infrastructures such as water distribution networks, power grids, process plants, and communication networks can be displayed as graphs in an abstract form, where the components of the infrastructure are represented as nodes, and the flows of materials, energy, or information among the nodes are denoted as edges. Accordingly, some graph metrics have been suggested to infer about the attributes of the graph (infrastructure) under consideration. In this regard, graph metrics have been used to investigate the robustness of (single) communication networks with regard to technological errors or man-made attacks [17], vulnerability of (joint) interdependent networks (power network and Internet communication network) to the cascading effect of random node failures [18], vulnerability analysis of water distribution networks [19], and vulnerability analysis of process plants in the context of domino effects triggered by terrorist attacks [20] or random failures [2].

Following the work of Khakzad and Reniers [2], in the present study we examine the reliability and efficacy of graph theory (graph metrics) in domino vulnerability analysis of process plants under multiple accident scenarios and varying environmental conditions. We validate the results obtained from graph theory using a dynamic Bayesian network (DBN) methodology developed by Khakzad [15]. The efficacy of the developed methodology is demonstrated via a real case study. To this end, a short review of Bayesian networks is given in Section 2. The graph theory and graph metrics are presented in Section 3, while the development, validation, and application of the methodology are demonstrated in Section 4. The main conclusions drawn from this work are presented in Section 5.

Section snippets

Ordinary Bayesian networks

Bayesian network (BN) is a probabilistic tool for reasoning under uncertainty, where the nodes represent random variables and directed arcs imply local conditional dependencies between parent and child nodes [21], [22]. The type and strength of such conditional dependencies are defined by means of conditional probabilities assigned to the nodes. Those parent nodes which are not children of other parent nodes – so-called root nodes – are assigned marginal probabilities. satisfying the Markov

Graph theory

A mathematical graph is an ordered pair $G = (V, E)$ where $V = {v_{1}, v_{2}, \dots, v_{n}}$ and $E = {e_{1}, e_{2}, \dots, e_{m}}$ denote sets of n vertices (nodes) and m edges (directed or undirected), respectively. In a weighted graph, a set of numerical values can also be assigned to either the nodes or edges of the graph. In this case, the weighted graph can be presented as $G = (V, E, W_{V}, W_{E})$ where $W_{V}$ and $W_{E}$ are weight vectors allocated to the vertices and edges, respectively.

In a directed graph, a walk from the node $v_{i}$ to $v_{j}$ is a

An example

In this section, we take advantage of a simple example to develop and verify the methodology. To analyze the domino vulnerability of process installations, consider a hypothetical chemical storage plant as shown in Fig. 4, comprising four atmospheric storage tanks T1–T4. The characteristics of the storage tanks are listed in Table 1. Considering a major leak with a probability of 1.0 E−04 for the atmospheric storage tanks [25], probable accident scenarios can be determined using an event tree

Conclusion

In the present study we demonstrated the applicability and efficacy of graph theory in domino vulnerability analysis of chemical plants. We illustrated that graph centrality metrics, in particular all-closeness score, can reliably and effectively be employed to identify the most vulnerbale installations within a chemical plant during domino effects and under multiple accident scenarios and varying atmospheric conditions. In this study a dynamic Bayesian network methodology was employed to

References (31)

J. Johansson et al.
Reliability and vulnerability analyses of critical infrastructures: comparing two approaches in the context of power systems
Reliab Eng Syst Saf
(2013)
N. Khakzad et al.
Using graph theory to analyze the vulnerability of process plants in the context of cascading effects
Reliab Eng Syst Saf
(2015)
I. Bentes et al.
A new tool to assess water pipe networks vulnerability and robustness
Eng Fail Anal
(2011)
D. Codetta-Raiteri et al.
A dynamic Bayesian network based framework to evaluate cascading effects in a power grid
Eng Appl Artif Intell
(2012)
Y. Koç et al.
An entropy-based metric to quantify the robustness of power grids against cascading failures
Saf Sci
(2013)
D.F. Bagster et al.
The estimation of domino incident frequencies – an approach
Process Saf Environ Prot
(1991)
J. Gledhill et al.
Development of methods to assess the significance of domino effects from major hazard sites
(1998)
F. Khan et al.
Models for domino analysis in chemical process industries
Process Saf Progress
(1998)
F. Khan et al.
Studies on the probabilities and likely impacts of chain of accidents (domino effect) in fertilizer industry
Process Saf Progress
(2000)
V. Cozzani et al.
The assessment of risk caused by domino effect in quantitative area risk analysis
J Hazard Mater
(2005)

G. Reniers et al.

DomPrev planning: user-friendly software for planning domino effects prevention

Saf Sci

(2007)

G. Reniers et al.

Domino effects within a chemical cluster: a game-theoretical modeling approach by using Nash-equilibrium

J Hazard Mater

(2009)

N. Khakzad et al.

Domino effect analysis using Bayesian networks

Risk Anal

(2013)

V. Cozzani et al.

Quantitative assessment of risk caused by domino accidents

N. Khakzad

Application of dynamic Bayesian network to risk analysis of domino effects in chemical infrastructures

Reliab Eng Syst Saf

(2015)

Cited by (84)

Optimizing safety barrier allocation to prevent domino effects in large-scale chemical clusters using graph theory and optimization algorithms
2024, Process Safety and Environmental Protection
Domino effects are high-impact low-probability events that can have catastrophic consequences. To prevent and to reduce risks related to such events, safety barriers (SBs) are crucial. However, the initiation, propagation, and stopping processes of domino effects are characterized with complexity and uncertainties and hence they are unpredictable. This makes it challenging to allocate SBs based on predicted probabilities. In this study, a multi-objective optimization model which integrates graph theory with Non-dominated Sorting Genetic Algorithm II (NSGA-II) was proposed to allocate add-on SBs effectively. Graph metrics were used to quantify the escalation risks related to storage tanks and to optimize the allocation of add-on SBs, thereby minimizing the consequences of a domino effect under a budget constraint. The results of the case study demonstrate great efficiency in finding globally optimal strategies with a largest reduction of 94.3% in the out-closeness score due to the implementation of add-on SBs, allowing decision-makers to choose the most preferable investment strategy in face of domino effect risk. Our study therefore provides a novel approach to achieve an optimal allocation of add-on SBs globally and can be useful in preventing domino effects in large-scale chemical clusters equipped with a large number of storage tanks.
Dynamic Domino Effect Assessment (D2EA) in tank farms using a machine learning-based approach
2024, Computers and Chemical Engineering
The current work presents a Dynamic Domino Effect Assessment (D2EA) methodology for chemical storage tank farms. While the application of the proposed approach is focused on atmospheric tanks, it applies as well to pressurized tanks. It utilizes the temperature and time-dependent material strength property (yield strength) as a structural health indicator. The D2EA methodology uses random forest (RF) and feed-forward neural network (FFNN) to predict yield strength during fire exposure and applies the predicted yield strength to dynamic failure assessment. A lumped parameter model generates datasets to train the dynamic failure prediction model. Two case studies have been used to demonstrate how the method can be used. The results suggest that RF and FFNN can predict gamma distribution-aided dynamic failure probability assessment. The RF is a better tool than FFNN due to its lower computational cost and good performance. The current work will be helpful in the design and operation of tank farms – an essential and critical infrastructure.
Improved pool fire-initiated domino effect assessment in atmospheric tank farms using structural response
2024, Reliability Engineering and System Safety
Domino effects are severe accident scenarios affecting storage tanks and are often initiated by pool fires. Flame engulfment and heat radiation are the two major sources triggering domino effect. Threshold-based and probit-based methods are widely used to assess the possibility and probability of a secondary accident. These methods are also a part of advanced methods devoted to examining synergic or coupling effects. The current work examines (i) how effective the threshold-based methods are and (ii) how accurate the current time to failure (TTF) estimation models are, which are the basis of probit-based methods. The results suggest that threshold-based methods are not pertinent for the quantitative assessment of domino effect and that significant improvement can be made in the existing TTF prediction models using site-specific structural response data. A new set of equations for TTF estimation using data analytics is proposed. Application to 4,080 pool fire scenarios demonstrates that the newly developed model can improve the TTF prediction performance compared to the existing models (around 22% in terms of R²). In addition, a method has been proposed and validated to correlate time with the failure probability for time-dependent domino effect assessment, which is a limitation of probit-based methods.
Domino index: A rapid quantification tool for the domino effect in chemical plants
2023, Heliyon
The severity of industrial accidents involving domino effects is widely acknowledged in chemical and process industries. The interdependence of installations and complexity of layouts pose significant challenges for the rapid quantitative assessment of domino effects in large chemical plants. In this study, a set of domino indices was introduced to measure the extent to which a given installation triggered and propagated domino effects, as well as to assess the overall domino effect in a specified area. An accelerated algorithm for domino accident modelling was developed based on Monte Carlo simulations to calculate the domino index. This algorithm can simulate all potential domino accident propagation pathways and the failure frequencies of installations. Two case studies, derived for a hypothetical chemical plant and actual oil-storage facilities, were examined to evaluate the applicability of the method. Furthermore, the method was validated using conditional probability calculations and vertex metrics. The results demonstrated that the proposed domino index is a useful tool for rapidly quantifying domino effects and that it can assist in identifying critical installations, designing plant layouts, and screening hazardous areas. The method and indices can provide guidance for the prevention of severe domino accidents.
A goal programming approach to multi-objective optimization of firefighting strategies in the event of domino effects
2023, Reliability Engineering and System Safety
In the event of fire in tank terminals, an ideal firefighting strategy would include simultaneous suppression of all burning tanks and cooling of all exposed tanks. This strategy, if effective, would confine the fire and prevent its escalation throughout the plant. However, limited firefighting resources, which is the case at most tank terminals, do not usually allow for conducting an ideal firefighting particularly if fire propagates from an originally burning tank to adjacent tanks, creating a domino effect. As such, an optimal strategy would be needed to determine which burning tanks to suppress and which exposed tanks to cool so as to best satisfy the safety objectives. For tank terminals that are near communities or offsite assets, the situation can become more challenging as a multi-objective firefighting strategy would be needed not only to limit the possibility and extent of domino effects within the plant but also to protect people and assets located outside the premises of the plant. For this purpose, in the present study, a methodology is developed based on goal programming – a multi-objective optimization technique – for identifying optimal firefighting strategies while considering both onsite and offsite risks of domino effects. The resulting firefighting strategies are demonstrated to be more effective and consistent than the ones identified using single-objective optimization techniques or general guidelines.
A comparison of DBN model performance in SIPPRA health monitoring based on different data stream discretization methods
2023, Reliability Engineering and System Safety
The energy and industry sectors depend upon the reliability of complex engineering systems (CESes), such as nuclear power plants or manufacturing plants; it is important, therefore, to monitor system health and make informed decisions on maintenance and risk management practices. One proposed approach is to use causal-based models such as Dynamic Bayesian Networks (DBN), which contain the structural logic of and provide graphical representations of the causal relationships within engineering systems. A current challenge in CES modeling is fully understanding how different data stream discretizations used in developing underlying conditional probability tables (CPTs) impact the DBN’s system health estimates.
This paper demonstrates the impact that different time discretization strategies have on the performance of DBN models built for CES health assessments. Using simulated nuclear data of a sodium fast reactor (SFR) experiencing a transient overpower (TOP), different strategies for discretizing CES data streams are used to construct the CPTs for a health-based DBN model. This study finds that these strategies generate different models with varying levels of performance for determining different assessments of overall system health. By understanding how these design factors impact the model’s health assessments, future risk models can be developed to provide a more meaningful assessment of a system’s health, resulting in more informed decisions.

View all citing articles on Scopus

View full text

Vulnerability analysis of process plants subject to domino effects

Highlights

Abstract

Introduction

Section snippets

Ordinary Bayesian networks

Graph theory

An example

Conclusion

Reliability and vulnerability analyses of critical infrastructures: comparing two approaches in the context of power systems

Reliab Eng Syst Saf

Using graph theory to analyze the vulnerability of process plants in the context of cascading effects

Reliab Eng Syst Saf

A new tool to assess water pipe networks vulnerability and robustness

Eng Fail Anal

A dynamic Bayesian network based framework to evaluate cascading effects in a power grid

Eng Appl Artif Intell

An entropy-based metric to quantify the robustness of power grids against cascading failures

Saf Sci

The estimation of domino incident frequencies – an approach

Process Saf Environ Prot

Development of methods to assess the significance of domino effects from major hazard sites

Models for domino analysis in chemical process industries

Process Saf Progress

Studies on the probabilities and likely impacts of chain of accidents (domino effect) in fertilizer industry

Process Saf Progress

The assessment of risk caused by domino effect in quantitative area risk analysis

J Hazard Mater

DomPrev planning: user-friendly software for planning domino effects prevention

Saf Sci

Domino effects within a chemical cluster: a game-theoretical modeling approach by using Nash-equilibrium

J Hazard Mater

Domino effect analysis using Bayesian networks

Risk Anal

Quantitative assessment of risk caused by domino accidents

Application of dynamic Bayesian network to risk analysis of domino effects in chemical infrastructures

Reliab Eng Syst Saf