Area impact analysis of chemical installations and critical installations identification

Abstract

Identifying critical installations in an area containing many chemical installations is important for the safety management of the chemical area. In this study, the area impact degree of chemical installations is defined, which is determined by analyzing the probability of accidents at other installations in the area when an accident occurs at a given installation, and the losses caused by each installation in the event of corresponding accident. Critical installations for area safety can be easily identified by comparing the area impact degree of different installations in the area. An illustrative example demonstrates the proposed approach. Analysis of the area impact degree of 6 tanks in a storage area and identification of more important installations for the safety of the entire area illustrate the applicability of this approach. Furthermore, our novel area importance analyzing approach is also compared with the graph approach.

Introduction

In the chemical process industry, there are often many production or storage installations in a certain area to deal with or store a large amount of flammable and explosive dangerous chemicals. An accident at an installation may escalate to other nearby installations. This is called the domino effect. The formation of the domino effect must satisfy three conditions (Reniers and Cozzani, 2013): an initial accident that can cause accidents at nearby installations; escalation effect under the action of escalation vectors, such as thermal radiation and overpressure; and a secondary accident that is formed with worse consequences than the primary accident. In the petrochemical industry, many major accidents often involve domino effects, such as the accident which occurred at the Buncefield oil depot in northeast London in the early hours of December 11, 2005 due to excessive filling, the gasoline vapor was ignited and caused a violent vapor cloud explosion, which subsequently caused a series of explosions and fires, and the accident destroyed more than 20 storage tanks; In the evening of October 23, 2009, the Puerto Rican Caribbean Oil Company (CAPECO) spilled over a tank during the unloading of gasoline from the tanker Cape Bruny to the storage tanks. The gasoline volatilized to form a huge vapor cloud, which was subsequently ignited. An explosion then occurred and caused multiple storage tank fires. The explosions and fires severely damaged 17 of the 48 storage tanks. Domino effects often cause much greater losses than the initial accident in the chemical industry, so preventing domino effects in chemical installations is an important task of safety management.

So far, there have been many studies on domino effects of chemical installations, mainly involving the following aspects. The first is the condition under which the domino effect occurs. Many researchers have found that the escalation of accidents is mainly caused by the corresponding escalation vectors. For fires, the main escalation vector is heat radiation, and the escalation vector of explosions is overpressure. There are certain thresholds for the value of escalation vectors. When the value of an escalation vector received by the neighboring installations is less than the threshold, the domino effect will not occur. Only when this threshold is exceeded, the domino effect may occur (Cozzani and Salzano, 2004a; Cozzani et al., 2006; Alileche et al., 2015). There are differences in thresholds of escalation vectors in the literature. For heat radiation, the threshold ranges from a few kW/m² to tens of kW/m², and the threshold of overpressure also ranges from a few kPa to tens of kPa. For atmospheric vessels and pressure vessels, the thresholds of the same escalation vector are also usually different. The second is the probability of the domino effect. These studies focus on the damage probability of adjacent installations under the influence of escalation vectors when an accident occurs. In existing researches, the Probit model is commonly used to analyze the damage probability of neighboring installations, and the coefficients of the Probit model are determined according to the effect of accidental heat radiation or overpressure on the neighboring installations, so that the Probit model can be easily used to estimate the failure probability of the installation (Khan and Abbasi, 1998; Zhang and Jiang, 2008; Salzano and Cozzani, 2005; Landucci et al., 2009). In addition, there are methods such as event tree (Bernechea et al., 2013) and simulation analysis (Rad et al., 2014) to analyze the escalation probability of accidents. Another important aspect of domino effect research is how to prevent domino effects or reduce the loss of domino effects. This involves decision-making and management (Reniers et al., 2008), safety barriers (Janssens et al., 2015; Landucci et al., 2015), installation layout (de Lira-Flores et al., 2013; López-Molina et al., 2013), emergency response (Zhou et al., 2016; Hosseinnia et al., 2018), etc.

These studies on domino effects mainly focused on those between the initial accident and the second accident, and the higher-order domino effects are less studied. Darbra et al. (2010) and Hemmatian et al. (2014) studied hundreds of accidents involving domino effects and analyzed the sequences of domino effects. The results showed that some accidents formed a "four-step" domino sequence. Fire and explosion were the main accident types that caused domino effects. In recent years, there have been some studies involving high-order domino effect. For example, a Monte Carlo simulation based approach named FREEDOM was used by Abdolhamidzadeh et al. (2010) and Rad et al. (2014) to assess the frequency of domino accidents. A methodology of Bayesian network was proposed by Khakzad et al. (2013) to model domino effect propagation modes and to estimate the probability of domino effects at different levels. Yuan et al. (2016) developed a method based on Bayesian networks to estimate domino effect probability of dust explosions. Zhou and Reniers (2017, 2018) used probabilistic Petri-net (PPN) and matrix-based methodology to model and analyze the domino effect of VCEs and the domino effect caused by fire, respectively.

In an area with many installations of dangerous chemicals, an accident at one installation may escalate to other installations in the area through domino effects. Which installations are more important to the safety of the area? This question requires identifying critical installations in the area. Some researchers proposed novel graph theory based approaches and Bayesian networks to analyze high-order domino effects (Khakzad and Reniers, 2015; Khakzad et al., 2016, 2017; Chen et al., 2019). However, these approaches still have certain limitations in dealing with synergistic effects or probability interdependence in the case of multiple installation accidents. This paper studies the identification of critical installations in a chemical industry area, and an approach is proposed to evaluate the installation area impact which is used to measure the importance of installations. If primary accidents occurring at a given installation in an area are possible to cause greater losses to the area than primary accidents at other installations, it can be considered that this installation has a greater impact on the area safety. It should be noted that after a primary accident occurs, other installations in the area may also be damaged due to domino effects. Therefore, the area impact of an installation must consider not only the possible accident loss of the installation itself, but also the possible losses of other installations under domino effects.

The remaining parts of this paper are arranged as follows: Section 2 discusses the problem in analyzing the importance of chemical installations in an area, and proposes the definition of area impact degree. In Section 3, the process using the area impact of chemical installations to evaluate the importance of installations is presented. Section 4 illustrates the proposed approach through an example. Conclusions of this work are drawn in Section 5.