Failure Causality Diagnostic in Industrial Systems through Automated Machine Learning
Résumé
In the context of modern industrial systems, efficient failure management is crucial for maintaining operational integrity, minimizing downtime, and maintenance optimization. This paper explores the application of Automated Machine Learning (AutoML) to enhance both failure causality diagnostic and failure causality prognostic in industrial systems. Different failure causes are detected by failure causality diagnostics, and the upcoming failure could be prevented by failure causality prognostics. Indeed, upcoming failures could be avoided by preventing their causalities. Traditional machine learning approaches require significant manual intervention for model selection, hyperparameter tuning, and feature engineering, which can be time and cost-consuming. AutoML, on the other hand, automates these processes, enabling quicker and more accurate predictions while reducing the need for extensive domain expertise. Our approach integrates AutoML into real-time failure diagnostics, identifying the root causes of system malfunctions using historical and sensor data. Simultaneously, it applies AutoML for prognostics, predicting Remaining Useful Life (RUL) of components and foreseeing future failures. By leveraging both data-driven models and physics-based insights, our approach improves the reliability of diagnostics and prognostics in various industries, including manufacturing, aerospace, and energy. The competing risks can be considered an application of this approach. High probable competing risks to cause are detected based on historical data in diagnostic phase and by handling them in the prognostic phase, the anticipated failures could be prevented. Finally, the experiments are conducted using real-world industrial datasets, demonstrating the superior performance of AutoML compared to traditional machine learning methods in both diagnostic accuracy and prognostic precision. This study shows that AutoML can significantly enhance decision-making processes in maintenance planning and risk mitigation, ultimately reducing operational costs and improving system reliability.