Fault Prediction in Induction Motor Using Artificial Neural Network Algorithms

The paper reflects on the investigation of current signals and vibration signals monitoring for induction motor (IM) effective fault prediction using artificial neural network (ANN) algorithms. Failures in induction motor may occur due to the propagation of various mechanical and electrical faults. In this study, vibration and current signals were acquired after multiple experiments of varying rotational speeds from the experimental test rig and converted into the frequency domain. In this study, ten different fault conditions which are frequently encountered in IM, i.e., four mechanical fault conditions, five electrical fault conditions, and one no defect condition, have been considered. Fault prediction has been done using vibration and current signals concurrently. Three statistical parameters, i.e., standard deviation, skewness, and kurtosis have been considered for extracting features from the signal dataset. F1-score, which is a combination of precision and recall of the classifier into a single metric which is done by taking the harmonic mean, has been used as a metric for evaluating the prediction accuracies of each fault. On the other hand, accuracy has been used for evaluating the overall performance of the ANN model. The results showed that the proposed methodology achieves a very good accuracy and F1-score for the standard deviation data at all speeds while for skewness and kurtosis extracted data, it gives an average prediction accuracy. Owing to this, standard deviation turns out to be the best statistical feature for feature extraction, followed by skewness and kurtosis for the fault prediction of IM.