Unsupervised Learning for Anomaly Detection of Electric Motors

This paper presents a novel approach for discriminating abnormal electric motors from normal motors using real sound data. The proportion of abnormal data among all training data is extremely low. Feature extraction techniques are applied to the sound signals. After the extraction, those features are used in unsupervised learning algorithms, including the convolutional autoencoder and one-class support vector machine. Results show that the proposed methods successfully distinguished normal data from abnormal data. The techniques can be applied to determine whether an electric motor is faulty during quality inspection.