Identification and Diagnosis of Asynchronous Motor Imbalance Faults Using Surrogate Models

Abstract

Asynchronous motors have a wide range of industrial applications due to their robust structure, low maintenance costs, and high reliability. However, these motors can be exposed to electrical and mechanical faults caused by environmental and operational conditions. Among the types of faults are problems such as bearing failures, stator winding faults, and rotor bar breakages, with mechanical imbalance faults standing out as a critical issue that adversely affects motor performance. This study aims to compare the performance of surrogate models (RBF and KRG) and deep learning models (RNN, GRU, LSTM), which represent a novel approach for diagnosing imbalance faults in asynchronous motors. For this purpose, experimentally collected current (Ia, Ib, Ic) and vibration (X, Y, Z) signals were analyzed in the frequency domain, and the features obtained via FFT were used in classification processes for three classes (Healthy, DA_1, DA_2). According to the results, the RBF model exhibited the best performance with an accuracy of 97.78% and a precision of 97.64%, while the KRG model showed remarkable success with an accuracy of 93.89% and a precision of 93.71%. In contrast, the deep learning models with the highest accuracy, RNN and LSTM, demonstrated lower performance with an accuracy of 87.22% and a precision of 87.23%. Compared to the RNN model, which is the most accurate deep learning model, the RBF model achieved an improvement of 12.11% in accuracy and 11.93% in precision, proving to be a superior tool in diagnosing imbalance faults. Notably, it achieved 100% accuracy in the DA_2 class and distinguished itself from other classes with its distinct features. These findings show that surrogate models offer an effective solution in asynchronous motor fault diagnosis by providing high accuracy and precision rates along with limited data requirements and low computational cost.