A fault diagnosis approach for roller bearing based on improved intrinsic timescale decomposition de-noising and kriging-variable predictive model-based class discriminate

Abstract

The measured vibration signal of roller bearings has noise signals, which will largely influence the accuracy of roller bearing fault diagnosis. This paper puts forward a vibration signal de-noising method based upon improved intrinsic timescale decomposition (ITD); fuzzy entropy is then extracted as the fault feature of the roller bearing. Essentially the fault diagnosis of roller bearings is a process of pattern recognition. Targeting the limitation of existing pattern recognition methods, a new pattern recognition method – variable predictive model-based class discriminate (VPMCD) – is introduced into roller bearing fault identification. In the original VPMCD classifier, however, only the regression model could be used to predict, which will reduce the prediction’s accuracy when the relation between features is complicated. Aimed at this defect, a kriging-variable predictive model-based class discriminate (KVPMCD) pattern recognition method is presented in this paper. The kriging model is composed of a regression model and a correlation model, of which the correlation model is a local deviation that is created on the basis of a global model, to make up for the shortcomings of the simple regression model in the original VPMCD. Therefore, a fault diagnosis approach for roller bearing based on improved ITD de-nosing and KVPMCD is proposed in this paper. The analysis results from experimental signals with normal and defective roller bearings indicate that the proposed fault diagnosis approach can accurately and effectively distinguish the status of roller bearings with or without fault and fault patterns.