A novel approach for domino effects modeling and risk analysis based on synergistic effect and accident evidence

Highlights

• A novel approach for domino effects modeling and risk analysis.

• Fire accident evidence and synergistic effects are considered.

• Energy accumulation effect is used to assess the domino effect risk.

• A visible image of domino evolution and associated risk are presented.

Abstract

Modeling the spatial-temporal evolution of domino effects and analyzing its risk could guide risk prevention and control. In the present work, a novel approach is proposed to model the spatial-temporal evolution and perform a risk analysis of fire-induced domino effects based on synergistic effect and accident evidence. The approach takes fire accident evidence into account to model the synergistic effect of burning units. The receivable thermal dose is adopted as the failure criterion of endangered units to consider heat energy accumulation. The time to failure and escalation probability are used to assess domino effect risk. The proposed approach is effective to model the dynamic evolution of accident propagation, and the results provide a visible image of domino effect evolution and associated risk, and also the results are helpful to identify the most vulnerable unit and the most dangerous units. A case study is illustrated to demonstrate the credibility and applicability of the proposed approach.

Introduction

Domino effect is a low-frequency high consequence chain of accidents, where a primary accident occurs and spreads to adjacent units, causing secondary accidents and the total consequences much severer than the consequence of the primary accident [1]. Chemical industrial parks have infrastructures, such as fuel tanks and pipelines, high complexity and growing interdependencies make infrastructures increasingly vulnerable to domino effects [2]. The need for risk analysis of domino effects in chemical industrial parks has been recognized [3], [4], [5], [6]. In the past, the studies on risk analysis of domino effects focused on four main tasks: the identification of domino scenarios, the assessment of escalation frequencies, the calculation of loss of life in complex domino scenarios and the assessment of equipment vulnerability [7].

In most of the early studies, such as [3,4,6], simple "domino chains" were considered and secondary accidents were considered as independent accidents. The over-simplification of studies loses a fundamental characteristic of real-world domino effects, which is the spatial-temporal evolution of domino effects. As a result, the early studies considered the risk of a particular independent accident rather than the risk of domino effects. The spatial evolution of domino effect refers to the escalation route in which the primary accident propagates to adjacent units. For example, if a storage tank is on fire, spatial evolution means which adjacent tank could be the secondary accident unit, the tertiary accident unit and so on. To address the over-simplification problem, some studies aimed to model the spatial evolution of domino effects, such as determination of the most likely propagation route given a primary accident. Cozzani et al. [8] introduced a methodology based on the binomial distribution, the methodology considered all possible accident scenarios triggered by a primary accident to handle the uncertainty arisen from the lack of knowledge about the likely units involved in a domino effect. Khakzad et al. [9] applied ordinary Bayesian network (BN) to model the spatial evolution of domino effects and to determine the most likely propagation route given a primary accident. There are similar studies based on different approaches, such as game theory [10], Monte Carlo simulation [11], and event tree [12], [13], [14]. However, these studies did not take the temporal evolution of domino effects into account.

Since domino effects are quite complicated due to probabilistic propagation, synergistic effects (synergistic effect refers to the collaboration of concurrent primary and secondary accidents to trigger another accident in a tertiary unit and so forth, making an already started domino effect continues [2]), dynamic evolution and behavior of target unit (adjacent/neighboring unit) [15], therefore, a domino effect is a time-dependent process, not only spatial evolution of involved units but also temporal evolution of involved units should be identified. The temporal evolution of domino effect refers to the escalation time at which the primary accident propagates to adjacent units. For example, if a storage tank is on fire, temporal evolution means when the adjacent tanks will be destroyed and be the secondary accident unit, the tertiary accident unit and so on. More importantly, nearly half domino effects are caused by fire [16], [17], [18], and since the escalation caused by fire is delayed concerning the start of fire, foreseeing the temporal evolution of domino effects and predicting the most likely sequence of accidents with timeline could guide the allocation of preventive and mitigative safety measures [2].

To address the drawbacks of previous studies, modeling the spatial-temporal evolution of domino effects has been studied recently. The spatial-temporal evolution of domino effect refers to the escalation route and escalation time the primary accident propagates to adjacent units. Khakzad [2] proposed a methodology based on dynamic Bayesian network (DBN) to model the spatial-temporal evolution of domino effects and also to obtain the most likely sequence of accidents in a potential domino effect. However, the approach is unable to model the temporal dependency quantitatively and obtain accident propagation time considering the synergistic effect. Later, Khakzad et al. [19] introduced a methodology based on graph theory which models a domino effect as a directed graph, graph centrality measures such as out-closeness and betweenness scores were used to identify the units playing a pivotal role in initiating and propagating domino effect. Zhang et al. [15] proposed an agent-based modeling and simulation approach to study the spatial-temporal evolution of domino effects, which explains a domino effect from a bottom-up perspective. Zhou and Reniers [20] proposed a matrix-based approach to model the spatial-temporal evolution of fire-induced domino effects, the approach uses a simulation-based algorithm to calculate accident probability. Chen et al. [21] proposed a methodology involving a Domino Evolution Graph (DEG) model and a Minimum Evolution Time (MET) algorithm to model the spatial-temporal evolution of domino effects.

In these works, fire-induced domino effects were studied, and the synergistic effect [9] of fires was considered by superimposing heat radiations. However, the synergistic effect was dealt with statically, and the superimposed escalation vector on a target unit was compared with an empirical threshold value to determine whether a target unit is damaged or not, as a result, consideration of synergistic effect lacks detailed modeling of time dependency. In studies [2,19,20], if the intensity of the escalation vector is below the threshold value, the likelihood of accident propagation is neglected. Thus, accumulated heat energy or strength loss of tank wall due to the escalation vectors which are below the threshold value is overlooked. For example, consider two fuel storage tanks T1 and T2 as shown in Fig. 1, and a threshold value of 15 kW/m2. If the thermal radiation T2 received from T1 is 14 kW/m2, since it is below the threshold value, therefore it is deemed the primary accident in T1 could not propagate to T2 no matter how long the fire in T1 lasts. From the viewpoint of heat energy accumulation, the radiation intensity threshold value method is inappropriate. However, it should be noted that there are different radiation thresholds commonly used (e.g. 12 kW/m2 or even 7.5 kW/m2) because of the theoretical possibility that after a sufficient amount of time, if the radiation received is constant and below a given threshold, the target unit cannot fail because a sort of steady-state can be reached in which heat received through radiation is balanced by heat dispersed through convection with ambient air.

More accurately, whether a target unit loses its structural integrity in a fire depends on the received thermal dose which decides the thermal stress temperature that can cause failure [22,23]. In this work, the fire synergistic effect model (FSEM) [24] is used to model the spatial-temporal evolution of domino effects considering accident evidence and synergistic effect quantitatively. The FSEM based approach models the escalation route of accidents and predicts escalation time and probability through modeling time-dependent synergistic effect. Besides, from the viewpoint of energy accumulation, the receivable thermal dose is adapted as the failure criterion to consider heat energy accumulation of escalation vectors which are below the threshold value. The FSEM based approach could deliver a macroscopic image of the domino effect evolution route with a microscopic evolution time.

The present work aims to examine the capability of FSEM in spatial-temporal evolution modeling and risk analysis of domino effects in chemical plants. The remainder of this paper is organized as follows: Preliminaries of FSEM are firstly recapitulated in Sections 2. The proposed methodology is presented in Section 3. In Section 4, the application of the methodology to a hypothetic chemical storage plant is illustrated. Finally, the conclusions of the present work are drawn in Section 5.