Computational intelligence to detect bearing faults using optimal features from motor current signals

In recent times, there has been a notable growth in research investigations into the fault diagnosis of electrical machines. The effective detection of permanent magnet synchronous motor bearing faults is a significant challenge; however, it is crucial for ensuring safety and cost-effectiveness in industries. The data referring to faults needs to be studied under distinct operating conditions, with effective features. Consequently, a meticulous choice of features is required before fault identification. The study aims to find the fewest and most reliable features in the dataset from Paderborn University so that a simple and accurate way can be found to diagnose faults using current signals. The selection of optimal features is initially performed using three algorithms: the equilibrium optimizer, the emperor penguin optimizer (EPO), and the butterfly optimization method. The k nearest neighbour (kNN) and random forest classifiers are used for classification. The results are performed based on the metrics of sensitivity, specificity, accuracy, precision, F1-score, and Matthew's Correlation Coefficient. The results indicate that machine learning models employing different feature selection techniques exhibit superior performance across different feature dimensions. Specifically, the model utilizing the kNN classifier and features selected through the EPO method achieved the maximum accuracy of 100%. The model's efficacy was also compared to the similar work presented in the literature. The efficacy of the optimal features is experimentally confirmed by analyzing current data from squirrel cage induction motors and has shown a high accuracy of 95.2%.