A Study of Motor Fault Condition Classification Using an Unsupervised Learning-Based Ensemble Model

In motor fault diagnosis, auto-encoder based methods are effective in detecting abnormal patterns by utilizing only normal data. However, this approach has limitations in classifying various fault types, as it is primarily trained with normal data. To solve this problem, a voting technique using an ensemble model based on an auto-encoder is proposed in this study. The proposed method combines several auto-encoder models to synthesize the prediction results of each model, showing higher performance in fault type classification than traditional single model approaches. This method has the advantage of more clearly distinguishing the boundaries between failure types and maintaining stable performance even in various failure scenarios. The experimental results confirm that the proposed method can classify various types of faults with high accuracy, even though it was trained only on normal data.