Explainable bearing fault diagnosis based on class activation mapping with physical domain knowledge

초록

Rolling element bearings are critical components in rotating machinery, and their faults can lead to downtime and product quality degradation. While deep learning models such as convolutional neural networks (CNNs) have achieved high accuracy in fault diagnosis, their decision-making processes are often uninterpretable, limiting their reliability in industrial applications. To address this issue, this study proposes an explainable fault diagnosis framework that combines CNN with class activation mapping (CAM). Vibration signals from the MFPT bearing dataset were transformed into time-frequency representations using the short-time Fourier transform (STFT). The trained CNN model achieved high classification accuracy. CAM was applied to visualize the model's attention, revealing strong alignment with known bearing fault frequencies such as BPFI and BPFO. These results demonstrate that the model learned diagnostically meaningful patterns, enhancing its interpretability and supporting its deployment in industrial fault diagnosis systems.

키워드