Enhancing fault diagnosis of undesirable events in oil & gas systems: A machine learning approach with new criteria for stability analysis and classification accuracy

Abstract

Petroleum serves as a cornerstone of global energy supply, underpinning economic development. Consequently, the effective detection of faults in oil and gas (O&G) wells is of paramount importance. In response to the limitations observed in prior research, this study presents an innovative fault diagnosis system, rooted in machine learning techniques. Our approach encompasses a comprehensive analysis, incorporating stability assessment via standard deviation (STD), and a meticulous evaluation of accuracy and stability for distinct fault scenarios. By integrating data preprocessing, feature selection methods, and deploying a robust random forest classifier, our model achieves a substantial enhancement in fault classification accuracy and stability. Extensive experimentation substantiates the superiority of our approach, surpassing the performance of previous studies that predominantly emphasized overall accuracy while disregarding stability analysis. Notably, our model attains remarkable accuracies, notably achieving a flawless 100% accuracy for scenario 3 faults. Detailed examination of mean accuracies and STDs further reinforces the precision and consistency of our model's predictive capabilities. Additionally, a qualitative assessment underscores the practical utility and reliability of our model in accurately identifying critical fault types. This research significantly advances fault detection methodologies within the O&G industry, providing valuable insights for decision-making systems in oil well operations.