Root cause analysis and well intervention risk prediction for typical failures in downhole safety valves using explainable AI

Understanding the causes of well integrity issues when they arise is a vital part of well integrity management and continuous improvement. Well barrier elements such as the Downhole Safety Valve (DHSV) are essential in preventing uncontrolled flow of hydrocarbons from wells to surface. However, they are susceptible to degradation, design and human induced failure, requiring repair in form of intervention and workover. With time, well barrier failure frequency increases, and so do the required intervention, workover and costs to fix DHSV failures. In this work the objective is to use an integrated data analytics approach that leverages explainable machine learning for failure root cause analysis and reliability analysis for well intervention/workover risk. The k-nearest neighbors (KNN) algorithm had the highest accuracy and micro average F1 score of 0.75 from the 785 DHSVs under study. We discovered correlation between service time and depth to the failure modes. Using reliability analysis, the workover risk was lower than intervention risk but contributed to the highest expected failure cost. This analysis can be applied to other repairable well barriers. Machine learning has potential for well barrier failure diagnosis and intervention risk prediction with reliability methods acts as a decision-support tool in well intervention planning.