Rolling bearing defect severity assessment under varying operating conditions

This paper investigates a multilayer feed-forward neural network structure for automated bearing defect severity assessment under varying operating conditions. A bearing health index was proposed, based on the Weibull theory. Defect-related frequency features were used as inputs to the neural network. The neural network was trained to establish relationship between defect characteristic and bearing conditions, and outputs specific a health index value corresponding the defect severity. Experiments have shown that the neural network was effective in differentiating faulty bearings from a 'healthy' bearing, with a 99% and 97% success rate for classifying defects in the inner and outer raceways, respectively.