Failure prediction in the refinery piping system using machine learning algorithms: classification and comparison

Abstract

Pipelines play a pivotal role in transporting large volumes of oil and gas within refineries. However, over time, they are susceptible to deterioration, leading to potential failures. Effective monitoring is imperative to maintain their optimal performance and safety. This research introduces a machine learning (ML) approach to pinpoint failure sources in oil and gas pipelines. Analysing an industrial dataset, we compared six ML models to predict failures in refinery pipelines. Leakage sources are predicted based on three operational parameters: transported fluid, temperature, and pressure. The models are evaluated and compared in terms of precision, recall, F1-score, accuracy, and the ROC-AUC. Remarkably, the XGBoost classifier exhibited a 99.7% accuracy, outperforming other algorithms in predicting the failure source. Emphasizing the value of Industry 4.0 solutions, this study underscores the potential of advanced ML in enhancing pipeline monitoring. Such predictions empower operators to pre-empt failures, reinforcing industry safety and sustainability.