A novel data mining approach for analysis of accident paths and performance assessment of risk control systems

The data mining researches to facilitate the process of safety management is fairly new, compared to other industrial management domains. The implementation of appropriate, effective, and safe risk control systems (RCSs) is vital to ensure zero-accident and zero-harm vision of industrial work-systems. In this work, we propose a data mining based tool to analyze accident paths from incident data and assess the performance of RCSs. Our work upgrades the existing pattern analysis methods through three new types of analyses (i) temporal frequent itemset generation (T-FIG) for studying the time effect on patterns, (ii) elevated severity itemset generation (ESIG) for examining the risk reduction due to RCSs, and (iii) High impact itemset generation (High_impact_IG) to identify accident paths with high risk. T-FIG and ESIG assist in performance assessment of preventive and mitigating RCSs, respectively. The results from each of the analyses are compared and eight types of inferences regarding the performance of RCSs are drawn. The proposed methodology is applied to 612 incident records reported during steel making process in a steel manufacturing plant. It was found that there are four accident paths which have ineffective preventive and mitigating RCSs, have high risk and are probable to recur in future. Two among four of these paths include hot metal/steel/slag as the hazardous element and three of them are due to damaged/degraded/poorly maintained equipment. Moreover, the case study also demonstrates that proposed data mining approach is an effective and easy to use tool for performance assessment of RCSs and accident path analysis.