A framework to model contractors’ hazard and risk exposure at process plants using unsupervised text mining

Abstract

Increasing outsourcing of operations in process industries is leading to contractors facing higher exposure to complex processes and machinery. This necessitates development of safety management systems by performing fault detection and diagnosis, accident/failure modelling and risk assessment of contractors’ operations and associated hazards. Unfortunately, limited work has been done in this area. Moreover, in broad process safety domain, using text mining to perform accident modelling and risk assessment has not been fully exploited. We address these research gaps by proposing an unsupervised text mining framework to generate accident/failure models like chain of events and fault trees from investigation reports of contractor incidents. A case study is used to demonstrate the framework where incident investigation reports of contract personnel working at a steel plant in India are analysed using a combination of topic modelling and association rule mining. Twenty sequences of basic events are identified as major chains of events. These chains of events are then combined using logic rules to generate fault trees depicting all possible propagation paths of contractor safety failures leading to top events. Our framework enables generation of chain of events and fault trees from unstructured incident reports with minimal human supervision with accuracy of 81%.