Feature optimization for bearing fault diagnosis

This paper presents methods of feature optimization for bearing fault diagnosis. These methods optimize statistical features in time domain and frequency domain. These optimization methods mainly consist of dimensionless processing and evaluation. Dimensionless processing method is used to avoid the influence of dimension and magnitude to the sensitivity. Fault sensitivity and discrete degree of features are evaluated. And features are selected according to the evaluation results. Analysis results of vibration signals of normal bearings, bearings with outer ring fault, bearings with inner ring fault and bearings with rolling element fault are presented. The results show that these methods are efficient to improve the separability of features.