Unsupervised Learning Based Diagnosis Model for Anomaly Detection of Motor Bearing with Current Data

Abstract Lifecycle engineering is a key concept for promoting environmentally sustainable practices among manufacturing firms. A key aspect of life cycle management for pursuing sustainability is condition-based maintenance system that uses data analytics process such as anomaly detection to understand machine condition. Conventional machine learning technique applied to anomaly detection uses classifiers based on supervised learning to detect anomalies of motor bearings using the bearing specifications. However, supervised learning generally requires a large volume of data at the time of abnormal operation, which takes time to acquire. This paper therefore proposes a data analytics process to detect motor bearing failure using data during normal condition. In the data analytics process, we first make a power spectrum from current sensor signals. Then, after grouping a power spectrum into bins, we employ a Gaussian Mixture Model (GMM) to learn the normal condition of a motor bearing. Next, we calculate likelihood with the GMM to check for any difference from the normal condition. Experimental results demonstrate that the proposed method can show increasing anomaly of a bearing condition corresponding to insufficient grease.