Machine Learning-Based Online Multi-Fault Diagnosis for IMs Using Optimization Techniques With Stator Electrical and Vibration Data
Shih-Hsien Hsu, Chien-Hsing Lee, Wen-Fang Wu, Joe‐Air Jiang
Abstract
Induction motors (IMs) have been commonly applied to industrial fields since the past decades; thus, developing advanced fault diagnosis methods becomes vital for IM applications. This study proposed an online fault diagnosis system for IMs based on the Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) algorithms to reduce the additional repair costs and prevent unexpected downtime. It focused on detecting healthy three-phase IMs and five common fault conditions of the IMs, involving broken rotor bars, rotor unbalance, and composite faults with short-circuited stator windings that combined two or three types of the faults, for practical purposes. The experimental results show that the model performance improved by 15% over the default model when train-test split ratios, feature selection, and hyperparameter optimization, notably in XGBoost, are considered. The proposed XGBoost model enables a high accuracy of 96.06% for RF to perform a motor fault diagnosis under six different motor conditions. Furthermore, the execution time required by the proposed fault diagnosis system is 57% less than the time required by existing motor fault diagnosis methods. These results successfully demonstrate the effectiveness of the methods proposed in this study for online motor diagnosis.