Machine Learning Based Incipient Fault Diagnosis of Induction Motor

An induction motor is the most important machine used in all industries, so the health of the machine is checked regularly by the machine learning method described in this paper. The objectives of this paper is to get high accuracy in predicting the fault of a machine and reduce the training time period of the model. Prediction of electrical and mechanical faults is achieved by monitoring the acoustic signal and analyzing it to identify deviations from expected values, which can indicate the presence of faults. The AI algorithms are trained using acoustic data acquired from electrical machines. A benchmark data set available online is used to test the proposed algorithm. The data is segmented into non-overlapping epochs. Features are extracted by the Kruskal-Wallis algorithm, and then the data is trained on various models to compare their accuracy. Machine learning-based diagnostic techniques such as weighted KNN, fine KNN, wide neural networks, and coarse Gaussian SVM are used and achieve 100% accuracy. The best algorithm is weighted KNN due to its fast processing time.