Method of Classifying Motor Failure Conditions Using an Auto-Encoder-Based Ensemble Technique

In the field, unsupervised learning-based fault diagnosis is necessary due to the lack of fault data. However, conventional auto-encoder models face challenges in fault classification. To address this issue, this study proposes an unsupervised learning-based auto-encoder model enhanced with an ensemble approach. The proposed model performs fault classification using gear, bearing, and eccentricity fault data, and employs the Layer-wise Relevance Propagation (LRP) technique to improve fault classification. Through the LRP technique, the key frequency bands relevant to each fault are identified and incorporated into the learning process, allowing for frequency band-based training. As a result, the proposed auto-encoder model improves classification performance and effectively distinguishes between different types of faults.