Feature Selection for Bearing Fault Detection Based on Mutual Information
This paper deals with the important task of feature selection for the detection of faulty bearings in a rotor-bearing system. Various time, frequency and time-frequency based features are obtained from signals measured from bearings with and without outer race defect. The features are divided into a training set, a validation set and a test set. The task is to develop an optimal subset of features for a pattern classification algorithm which can efficiently and accurately classify the state of the machine as healthy or faulty. The features are ranked based on the mutual information content between the feature subset and the state of the machine. A validation set from the measured data is then used to obtain the optimal subset for classification. The performance of the method is evaluated using the test set.
KeywordsBearing defect Feature selection Mutual information Optimal feature set
- 2.Barkova, A., Barkov, N.: Condition assesment and life prediction of rolling element bearings. Sound Vibr. 28, 10–17 (1995)Google Scholar
- 9.Mori, K., Kasashmi, N., Yoshioka, T., Ueno, Y.: Prediction of spalling on ball bearings by applying discrete wavelet transform to vibration signals. Wear 8, 195–162 (1996)Google Scholar
- 10.Nataraj, C., Pietrusko, R.G.: Dynamic response of rigid rotors supported on rolling element bearings with an outer raceway defect. In: ASME (ed.) Proceedings of IDETC/CIE 2005. Long Beach, CA, USA (2005)Google Scholar
- 15.Spectra Quest: http://www.spectraquest.com (2009)
- 18.Ypma, A.: Learning methods of machine vibration analysis and health monitoring. Ph.D. thesis, Delft University (2001)Google Scholar