Application of data mining to minimize fire-induced domino effect risks

Long Ding,Faisal Khan,Jie Ji

DOI: https://doi.org/10.1111/risa.13925

Abstract:Data mining (DM) has been applied in many advanced science and technology fields, but it has still not been used for domino effect risk management to explore minimum risk scenarios. This work investigates the feasibility of DM in minimizing the risk of fire-induced domino effects in chemical processing facilities. Based on DM, an evidential failure mode and effects analysis (E-FMEA), which could bridge chemical facilities' operational reliability and domino effect risk, is combined with fault tree analysis (FTA) for the occurrence risk modeling of loss of containment (LOC) event of chemical facilities, which is often the triggering point of fire-induced domino effects. Industry specific data such as reliability data, inspection records, and maintenance records are of great value to model the potential occurrence criticality of LOC. The data are used to characterize the LOC risk priority number (RPN) of chemical facilities through FTA and E-FMEA, search and statistics rules are proposed to mine inspection records to assess LOC risk factors. According to the RPN scores of facilities, inherent safety strategies to minimize risk via inventory control are proposed, and their effectiveness is tested using a well-known probit model. In this way, the approach proposes a unit-specific evidence-based risk minimization strategy for fire-induced domino effects. A case study demonstrates the capability of DM in the risk minimization of fire-induced domino effects.