Increasing Feasibility of Neural Network-Based Early Fault Detection in Induction Motor Drives

Modern industrial plants are complex and very sensitive to costs to the business of unscheduled downtime when a motor fails. This is the case of broken bars in induction motor (IM) drives, which still represents a large share of the market. In principle, an early defect detection is made possible by advanced artificial intelligence (AI)-based techniques, but their complexity clashes with the essential nature of IMs. This article aims to bridge the gap by using motor current signature already available in standard drives and proposing a mix of simulations and data augmentation to train efficiently the neural network (NN) without the need of many broken prototypes, which is the major flaw for the industrial feasibility.