Abstract: Recently, deep-learning-based fault diagnosis methods have been widely studied for rolling bearings. However, these neural networks are lack of interpretability for fault diagnosis tasks.