Intelligent Bearing Fault Diagnosis Method Based on HNR Envelope and Classification Using Supervised Machine Learning Algorithms

Research on data-driven bearing fault diagnosis techniques has recently drawn more and more attention due to the availability of massive condition monitoring data. The research work presented in this paper aims to develop an architecture for the detection and diagnosis of bearing faults in the induction machines. The developed data-oriented architecture uses vibration signals collected by sensors placed on the machine, which is based, in the first place, on the extraction of fault indicators based on the harmonics-to-noise ratio envelope. Normalisation is then applied to the extracted indicators to create a well-processed data set. The evolution of these indicators will be studied afterwards according to the type and severity of defects using sequential backward selection technique. Supervised machine learning classification methods are developed to classify the measurements described by the feature vector with respect to the known modes of operation. In the last phase concerning decision making, ten classifiers are tested and applied based on the selected and combined indicators. The developed classification methods allow classifying the observations, with respect to the different modes of bearing condition (outer race, inner race fault or healthy condition). The proposed method is validated on data collected using an experimental bearing test bench. The experimental results indicate that the proposed architecture achieves high accuracy in bearing fault detection under all operational conditions. The results show that, compared to some proposed approaches, our proposed architecture can achieve better performance overall in terms of the number of optimal features and the accuracy of the tests.