Using machine learning and keyword analysis to analyze incidents and reduce risk in oil sands operations

Abstract

Many companies maintain large databases of incident reports. Incidents that have severe consequences are analyzed in detail to prevent recurrence, while minor incidents are typically just stored without any further evaluation. Especially with common incidents and those with lesser consequences, details that are necessary to understand the cause of the incident might be missing or recorded inconsistently. We argue that incidents can be reported more accurately and analyzed to provide learning value to companies maintaining databases to better prevent and mitigate risks, lower the cost of losses, and improve safety culture. The aim of this research is to apply machine learning and keyword analysis to create a digitalized system for efficiently reporting incidents that can be used to generate a risk matrix, trend report, prevention and mitigation strategies, and leading indicators for every incident report that is inputted.During this research project, 15,000 incident reports were analyzed to build a customized library. The customized library included the labels used in machine learning, the keywords from the incident database, and a list of statements used to accurately describe incidents. The labels and keywords were matched to the statements in a logical manner and output results were also programmed to match the statements using a company’s safety guidelines, standard operating procedures, and asset management systems. The basic structure for generating outputs was demonstrated using a large incident database provided by collaborators of the project and anonymized sample inputs. Three incident report case studies are also processed and presented using the proposed methodology, delivering risk matrix, trend analysis, prevention and mitigation strategies, leading indicators that can be used by workers and companies to increase hazard awareness and improve safety performance.