Dynamic probability analysis on accident chain of atmospheric tank farm based on Bayesian network
Introduction
Massive flammable and volatile liquid materials are stored in chemical atmospheric tank, the maximum volume has reached 150,000 m3 in China, which has significant fire and explosion risk (Ning, 2019). Due to the dense layout, once a fire or explosion occurs, it may cause the domino effect accident of tanks. On March 18, 2019, a naphtha tank fire broke out at ITC tank farm in Houston, USA, triggered domino effect involving 11 tanks in the tank farm to be burned down, the whole accident chain lasted 70 h (CSB, 2020). On June 2, 2013, the explosion and fire of toluene tank in the triphenyl tank farm of Sinopec Dalian Petrochemical Company initialized domino effect, and three other tanks in the tank farm exploded and burst into flame, resulting in four deaths and RMB 6.97 million economic loss. From the above accident cases, domino accidents in the chemical and process industry may cause serious economic losses and casualties. Therefore, assessing the accident chain risk has extremely important theoretical significance and application value for the prevention and control of the domino effect.
Due to the low probability of domino accident, it is difficult to adopt the experimental method to simulate the accident scenario. Summary and analysis of characteristics through accident cases is the main research direction at present. Darbra et al. (2010) made statistics on the accident types, causes, consequences of 225 domino accidents, and found that Explosion-Fire, Fire-Explosion and Fire-Fire are the most common accident chain modes. Abdolhamidzadeh et al. (2011) summarized 224 domino accidents of chemical process equipment, explosion is main primary accident, accounting for 57%.
Analyzing the escalation probability and evaluating the severity of the accident with the help of numerical simulation software or the empirical model is an important part of the risk assessment of the domino effect. Khan and Abbasi (1998a) built a model to evaluate the scale and possible consequences of the domino effect in chemical process industries. And they developed the DOMIFFECT software, which can evaluate the domino effect and can be used for the possibility of domino accidents in chemical process industries (Khan and Abbasi, 1998b). Cozzani and Salzano (2004) modified probit model under overpressure damage and proposed a simplified escalation probability model (Salzano and Cozzani, 2005a) and damage threshold of atmospheric tank (Cozzani et al., 2005b, Cozzani et al., 2006). Mingguang and Juncheng (2008) improved the accuracy of the overpressure failure probability model. Abdolhamidzadeh et al. (2010) proposed a Monte Carlo-based algorithm to evaluate the domino effect of chemical process accidents. Landucci et al., 2009a, Landucci et al., 2009b analyzed the failure time of storage tanks under the influence of fire and improved the probit model under the effect of heat radiation. After that, Landucci et al. (2015) researched the accident case data with safety barrier function and improved the calculation model of device failure time. Zhang et al., 2019a, Zhang et al., 2019b analyzed 165 domino accidents, established the propagation probability model of the accident chain, and verified the accuracy of the model.
In recent years, some scholars have begun to use Bayesian network to assess the risk of the domino effect. Khakzad et al. (2013) applied Bayesian network to the study of domino accident and determine the most likely escalation path of domino effect in tank farm. Khakzad, 2015, Khakzad et al., 2018 introduced dynamic Bayesian (DBN) method to simulate the accident escalation in domino scenario, which made up for the shortage of static Bayesian in time escalation. Zeng et al. (2020) combined with dynamic Bayesian network, explained the changes in the escalation probability of domino accidents under the protection of safety barriers. WenHu established an accident probability model of hybrid storage tank area. The paper analyzed and calculated the damage probability of explosion and pool fire to adjacent equipment in liquefied gas storage tanks by considered that the occurrence of a pool fire accident in a single tank is a result of the incident (WenHu et al., 2019). Zerouali and Hamaidi (2019) combined bow tie and fuzzy logic with Bayesian networks to propose a fire and explosion risk prediction model for storage tanks, and verified the effectiveness of the proposed method. Santana combines fuzzy mathematics with Bayesian networks to propose a new method to determine the failure probability of thermal radiation.
Although many scholars have studied the characteristics of domino accidents in the tank farm, specific research of domino accidents on the types of tanks is lacking, as atmospheric tanks and pressurized tanks are different in terms of containing chemicals, internal pressure, and tank structure. In previous literature, the escalation probability model does not consider the real extinguishing process of fire in the propagation process of accident chain, which may last for hours before the fire burnt out on. Therefore, the dynamic probability analysis method of domino accident chain in atmospheric tank farm is proposed in this paper. Event trees are constructed to achieve prior probability through 136 domino accidents. Then the equivalent pool fire radius is proposed to analysis the influence of fire-fighting factors on the escalation probability. Finally, the method is applied to a case study.
The method proposed to assess the risk of domino effect in tank farms and determine the most likely accident chain and key accident tank, which can provide guidance for the layout design and emergency response of atmospheric tank farm.
Section snippets
Accident node escalation probability
For domino accidents of atmospheric tanks, 136 cases from 1970 to 2020 were counted, and the escalation law of accident chain is summarized. These database sites include:
- 1)
Emergency Management Department of the People's Republic of China and the official website of the local provincial and municipal emergency management departments.
- 2)
Safety Management Network (http://www.safehoo.com/),
- 3)
Yi'an Network (http://www.esafety.cn/shiguanli/List_105_29.html),
- 4)
China Safety Production Association website (//www.china-safety.org.cn/
Dynamic probability analysis method of accident chain
Based on the analysis of the escalation of the accident chain, a dynamic probability analysis method of the accident chain in the atmospheric tank farm under the condition of fire extinguishing is presented. This method can be used to analyze the most likely accident tank and accident chain, which has certain guiding significance for emergency firefighting. The flow chart of the analysis method is shown in Fig. 2.
The following assumptions are put forward when using this method: ① The failure
Case study
In this part, the method of Section 3 is used to assess the risk of the accident chain and verify the feasibility of the method. Since there is no information about storage tank parameters and layout in the accident report, the atmospheric storage tank area of a chemical storage company is used as the research object to verify the research in this paper. The layout diagram and scene image of the tank area follows (Fig. 5):
The materials and tank parameters stored in the tank farm are shown in
Conclusion
In this paper, taking the accident chain of atmospheric tank farm as the analysis object, the initial accident scenario and escalation mode of 136 domino effect accidents of atmospheric storage tank were statistically analyzed. From the results, the most likely escalate segment of the initial accident storage tank is "Leakage-Explosion". The most likely escalate segment of the secondary accident tanks is "Explosion-Fire", the most frequent scenario of accident escalation between adjacent tanks
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
This work is supported by National Natural Science Foundation of China (No. 71971110), and Postgraduate Research & Practice Innovation Program of Jiangsu Province (SJCX20 0389) for its financial support.
References (27)
- et al.
Domino effect in process-industry accidents–an inventory of past events and identification of some patterns
J. Loss Prev. Process Ind.
(2011) - et al.
A new method for assessing domino effect in chemical process industry
J. Hazard. Mater.
(2010) - et al.
The assessment of risk caused by domino effect in quantitative area risk analysis
J. Hazard. Mater.
(2005) - et al.
Escalation thresholds in the assessment of domino accidental events
J. Hazard. Mater.
(2006) - et al.
The quantitative assessment of domino effects caused by overpressure: Part I. Probit models
J. Hazard. Mater.
(2004) - et al.
Domino effect in chemical accidents: Main features and accident sequences
J. Hazard. Mater.
(2010) - et al.
How to address model uncertainty in the escalation of domino effects?
J. Loss Prev. Process Ind.
(2018) Application of dynamic Bayesian network to risk analysis of domino effects in chemical infrastructures
J. Reliab. Eng. Syst. Saf.
(2015)- et al.
DOMIFFECT (DOMIno eFFECT): user-friendly software for domino effect analysis
J. Environ. Model. Softw.
(1998) - et al.
Quantitative assessment of safety barrier performance in the prevention of domino scenarios triggered by fire
J. Reliab. Eng. Syst. Saf.
(2015)
The assessment of the damage probability of storage tanks in domino events triggered by fire
J. Accid. Anal. Prev.
The assessment of the damage probability of storage tanks in domino events triggered by fire
J. Accid. Anal. Prev.
An improved probit method for assessment of domino effect to chemical process equipment caused by overpressure
J. Hazard. Mater.
Cited by (11)
A neural network approach to predict the time-to-failure of atmospheric tanks exposed to external fire
2024, Reliability Engineering and System SafetyDynamic Domino Effect Assessment (D2EA) in tank farms using a machine learning-based approach
2024, Computers and Chemical EngineeringCurrent status, challenges, and future pathways of chemical industrial park safety in China
2024, Journal of Loss Prevention in the Process IndustriesImproved pool fire-initiated domino effect assessment in atmospheric tank farms using structural response
2024, Reliability Engineering and System SafetyModeling the evolution of industrial accidents triggered by natural disasters using dynamic graphs: A case study of typhoon-induced domino accidents in storage tank areas
2024, Reliability Engineering and System Safety