Bearing Fault Classification of Induction Motor Using Statistical Features and Machine Learning Algorithms

Abstract

Condition monitoring can avoid sudden breakdown and ensure the reliable and safe operation of rotating machinery used in the industry. The early detection of fault signatures and accurately classifying them in time will ensure efficient maintenance operation and reduce the possibility of losses due to uncertain breakdown. In the fault diagnosis mechanism feature extraction from the raw signal is important to reduce the dimensionality of the original signal, which also carries the most important features and plays a vital role to further the fault classification process. In our analysis, a simple and novel diagnosis method is carried out with a vibration signal, where eight different bearing conditions are considered. The time and frequency domain statistical features are considered and a total of 20 statistical features are extracted and later, 4 classification algorithms (SVM, RF, KNN, ANN) are applied to estimate the overall accuracy in fault classification. It is found that, after hypertuning the model parameters, all the algorithms show more than 99% accuracy. The result possesses that the discussed procedure can classify complex bearing faults and thus, can be used for practical applications.