Risk propagation analysis of domino effect in chemical accident: An integrated approach with data mining and Bayesian networks

Abstract

Domino effects in chemical process industries can lead to catastrophic consequences due to their complex, multi-stage escalation mechanisms. Existing approaches to domino accident analysis often lack integration between qualitative causal insights and quantitative modeling of risk propagation. To address this gap, this study proposes an integrated methodology that combines the grounded theory, association rule mining, and Bayesian network modeling to systematically identify and evaluate risk pathways in hazardous chemical accidents. The grounded theory analysis of historical accident reports led to the identification of five core dimensions: accident type, human error, material properties, environmental conditions, and systemic management deficiencies. Using the Apriori algorithm, 418 high-confidence association rules were extracted from leakage- and explosion-initiated disaster chains, with the sequence ‘equipment defect → leakage → explosion’ occurring in 78 % of cases. A dual-layer Bayesian network model comprising 106 nodes was constructed to quantify the interactions among causative factors. Sensitivity analysis using the expectation–maximization algorithm revealed that shockwaves (sensitivity = 0.275) and debris dispersion (0.258) are dominant contributors to secondary escalation. This study proposes an integrated approach combining data mining and Bayesian networks for analyzing risk propagation patterns of the domino effect in hazardous chemical incidents, providing insights to enhance safety resilience in high-risk chemical industries.