Early Fault Detection in Induction Motors Using AdaBoost With Imbalanced Small Data and Optimized Sampling

Intelligent fault detection in induction motors (IMs) is a widely studied research topic. Various artificial-intelligence-based approaches have been proposed to deal with a large amount of data obtained from destructive laboratory testing. However, in real applications, such volume of data is not always available due to the effort required in obtaining the predictors for classifying the faults. Therefore, in realistic scenarios, it is necessary to cope with the small-data problem, as it is known in the literature. Fault-related instances along with healthy state observations obtained from the IM compose datasets that are usually imbalanced, where the number of instances classified as the faulty class (minority) is much lower than those classified under the healthy class (majority). This paper presents a novel supervised classification approach for IM faults based on the adaptive boosting algorithm with an optimized sampling technique that deals with the imbalanced experimental dataset. The stator current signal is used to compose a dataset with features both from the time domain and from the frequency domain. The experimental results demonstrate that the proposed approach achieves higher performance metrics than others classifiers used in this field for the incipient detection and classification of faults in IM.