Status Quo, Advances and Futures of Machine Learning in Fault Detection and Diagnosis for Energy: A Review

Fault Detection and Diagnosis (FDD) plays a crucial role in maintaining the integrity and efficient operation of modern industrial systems, from manufacturing sectors to process industries. FDD involves identifying and classifying abnormal conditions that could lead to equipment failure, production inefficiencies, or safety hazards. However, traditional FDD techniques face challenges in handling vast data and complex system dynamics and ensuring timely and accurate fault detection in dynamic environments. Manual inspections and heuristic approaches are inadequate, and statistical process control methods have limitations in capturing complex relationships and adapting to evolving process conditions. To overcome these challenges, advanced techniques such as deep learning-based approaches have emerged, leveraging the capabilities of neural networks for fault detection and diagnosis. These approaches have shown promising results in handling high-dimensional, nonlinear, and time-varying process data. This paper reviews the advancements, challenges, and prospects of deep learning in FDD in industrial systems. Firstly, it discusses the emergence and development of deep learning methods applied to FDD and their applications in relevant fields. Secondly, a new development path that combines deep learning with big data is proposed to address the increasing production data in modern industrial settings. Finally, the opportunities and limitations of deep learning in FDD are clarified, providing insights for future research and development in this area.