A Novel Method to Classify Rolling Element Bearing Faults Using K-Nearest Neighbor Machine Learning Algorithm

Abstract

A novel method is proposed in this work for the classification of fault in the ball bearings. Applications of K-nearest neighbor (KNN) techniques are increasing, which redefines the state-of-the-art technology for defect diagnosis and classification. Vibration characteristics of deep groove ball bearing with different defects are studied in this paper. Experimentation is conducted at different loads and speeds with artificially created defects, and vibration data are processed using kurtosis to find frequency band of interest and amplitude demodulation (Envelope spectrum analysis). Bearing fault amplitudes are extracted from the filtered signal spectrum at bearing characteristic frequency. The decision of fault classification is made using a KNN machine learning classifier by training feature data. The training features are created using characteristics amplitude at different fault and bearing conditions. The results showed that the KNN's accuracies are 100% and 97.3% when applied to two different experimental databases. The quantitative results of the KNN classifier are applied as the guidance for investigating the type of defects of bearing. The KNN Classifier method proved to be an effective method to quantify defects and significantly improve classification efficiency.