Explainable artificial intelligence based intelligent fault diagnosis: A systematic review from applications to insights

Deep learning (DL)-based intelligent fault diagnosis (IFD) has become a key enabler for condition-based maintenance, yet its “black-box” nature limits explainability and reliability. Explainability concerns understanding the model’s decision process, including the physical meaning and contribution of extracted features, while reliability addresses the rationality and stability of diagnostic results under varying conditions. As explainability analysis reveals potential biases and irrational dependencies, it serves as the foundation for improving reliability, spurring extensive research on explainable intelligent fault diagnosis (XIFD). This survey first reviews the core components of IFD to clarify how XIFD enhances model transparency, and then categorizes existing approaches into ante-hoc and post-hoc methods. Ante-hoc methods focus on building inherently interpretable models, whereas post-hoc methods explain results through external analysis. Then, we provide the usage and disadvantages of existing post-hoc XIFD methods and emphasize the significance of ante-hoc XIFD by giving insights into the most popular signal processing-informed IFD methods. Finally, we consolidate the pressing challenges that necessitate immediate attention within the realm of XIFD research, and offer insights into promising prospects for future research endeavors. The library is available at: https://github.com/HazeDT/Awesome-Explainable-IFD-Library .