A critical study on the importance of feature extraction and selection for diagnosing bearing defects
Midwest Symposium on Circuits and Systems
This paper presents a general data-driven diagnostic scheme to classify bearing faults in induction motors. Case western reserve university bearing data center are used to create two scenarios with different fault diameters of 0.007 and 0.014 that are induced in the inner race, the ball and the outer race. The diagnostic system could successfully conduct signal processing and classification steps to achieve an accurate condition assessment of the motor. In this work, the vibration signal is decomposed into several number of components by means of five different state-of-the-arts signal processing techniques. The extracted features which belong to the Time-domain, the Frequency-domain and the Time-Frequency domain are employed to create a pool of diverse features. Moreover, a feature selection strategy based on the correlation of the features to motor operating conditions is assessed. The obtained result shows that the combination of the most correlated features could provide an informative feature set for the fault classification and improve the diagnostic accuracy. In addition, feature selection can reduce the model complexity and facilitate the learning process of the fault classifiers.
Farajzadeh-Zanjani, Maryam; Razavi-Far, Roozbeh; and Saif, Mehrdad. (2019). A critical study on the importance of feature extraction and selection for diagnosing bearing defects. Midwest Symposium on Circuits and Systems, 2018-August, 803-808.