Application of Bayesian network and artificial intelligence to reduce accident/incident rates in oil & gas companies

Abstract

Process safety management (PSM) is a framework that demonstrates a company’s commitment to process safety, a better understanding of hazards and risks, a comprehensive assessment and management of risks, and enhanced learning from experience to improve overall safety and operational performance. Companies often use an incident data reporting system to execute PSM. While companies keep incident data in thousands of reports, rarely do they glean full value in learning from these to prevent and reduce future incidents. To overcome this challenge, this research applied machine learning and keyword analysis to label and classify 8199 incident reports from an oil and gas company into nine groups identified in the latest version of PSM guidelines published by the Center for Chemical Process Safety (CCPS). To converge on an optimal solution, two different Bayesian network techniques (Tabu and hill climbing) were applied. Both methods resulted in the same map, showing that the Total Number of Incidents has the maximum dependency (50%) on Asset Integrity & Reliability; this means focusing resources on this aspect could reduce the total number of incidents by half. Cross correlation analysis (CCA) was also applied, which validated and confirmed this result. This analysis identifies which measures enhance the company’s safety management strategy to reduce these latent causes, but also supports critical thinking, enhanced communication, and learning culture to improve organizational safety.