A methodical approach for knowledge-based fire and explosion accident likelihood analysis

Abstract

An accident database is an excellent data source if appropriately leveraged along with domain expertise. However, a proper framework and tools are required to extract data from a database. The current work aims to develop such a framework by systematically introducing a unique approach to integrate three techniques. First, Natural Language Processing (NLP) is used to extract causal and contributing factors from an accident database. Second, an Interpretive Structural Model (ISM) establishes the interrelationship and hierarchy of the extracted factors. Third, a probabilistic method for quantitative reasoning and accident analysis is employed. This integrated approach is applied to the US Chemical Safety and Hazard Investigation Board (CSB) oil and refining (downstream) incident database to develop a generalized accident causation model. The model provides insight into the factors responsible for accidents (i.e., commonalities among casualties), interactions, and accident pathways. It can also be used to develop strategies for preventing accidents. The model is tested on ten scenarios from the CSB and verified on six incidents from the IChemE database. The results are promising in establishing the model's efficacy in predicting adverse events. Sensitivity analysis shows that management of change and lack of procedure and training have the highest sensitivity towards fire and explosion, and therefore need proper attention. This approach will be an essential tool for Safety 4.0, enabling process safety in the digitalization process.