Chain of events model for safety management: Data analytics approach

Abstract

In chain of events model, an accident path is defined by a series of sequential events from initiating event to accident realization. In this study, we define accident path as a 5-tuple set comprising hazardous element (HE), initiating event (IE), pivotal event (PV), accident scenario (AS), and consequence (C). We propose a new methodology to identify the path components (i.e., HE, IE, PV, AS and C) and quantify the path using data analytics approach. We have used both proactive (workplace observation (WO) and high risk control program (HRCP)) and reactive data (incident records (IR)) for this quantification. The HEs and their corresponding IEs are extracted from WO and HRCP, and then, IR is used to identify PVs, ASs and Cs. In the process, we have used K-modes categorical clustering and Expectation-Maximization (EM) algorithm based text clustering. We could identify 39 accident paths using proposed approach and after quantification, nine paths are found to be frequently recurring. Preventive strategies are identified for accident paths of high concern. The quantification of accident paths will assist safety management in decision making by prioritizing accident paths and assessing performance of preventive barriers. By the use of data mining techniques, experts will not have to read the huge safety data for accident path analysis. This analysis can be done by experts by identifying safety keywords from the clusters generated by clustering algorithms. This will reduce the labour of safety experts in accident analysis with respect to time and resource.