Research On Rolling Bearing Fault Diagnosis Based On Multi-class Mahalanobis Taguchi Model

Rolling bearing as the key components of complex equipment in industrial production,often is in a state of fatigue run and carry a heavier load,so it is necessary to establish a set of perfect fault diagnosis system to judge its fault mode and damage degree timely and effectively,assist related personnel to make maintenance or replacement decision,so as to reduce maintenance costs and accident rate,improve production quality and ensure man-machine safety.Mahalanobis Taguchi system is a pattern recognition method has been widely applied in recent years,based on the data analysis of the ideas of multivariate data for classification prediction.It has significant advantages in selecting effective feature variables and realizing feature dimension reduction.This paper improves the traditional two-class Mahalanobis Taguchi system and proposes a multi-classification method of Mahalanobis Taguchi system based on spectral clustering algorithm.Spectral clustering is combined with Mahalanobis Taguchi system,and two kinds of multi-classification models are proposed: one is to construct multiple two-class classifiers and combine them with a certain structure to form multi-class classifiers,a SC-MTS model of Mahalanobis Taguchi system based on spectral clustering is constructed,which indirectly solves the multi-class classification problem;another is to change the Mahalanobis Taguchi system discriminant rules,establish Multiple Mahalanobis Space,and construct SC-MMS model of Multiple Mahalanobis Space based on spectral clustering,which directly solves the problem of multi-class classification.In this paper,the rolling bearing fault vibration signal data collected by the bearing data center of Case Western Reserve University in the United States using the fault simulation experiment platform are used.The data samples are divided into 10 categories according to the rolling body fault,inner ring fault,outer ring fault,normal state fault type and the degree of mild,moderate,and severe damage of the bearing.In this paper,the Empirical Mode Decomposition method is used to decompose and extract the original vibration signal of the bearing,construct the initial feature space of the bearing data,and realize the multi-class identification of the rolling bearing fault mode by using the Mahalanobis Taguchi system multi-classification models based on the spectral clustering algorithm.The results show that both models have good classification effect for fault diagnosis of rolling bearings.The SC-MTS classification process is simpler than the SC-MMS model without too much calculation,but in terms of threshold,SC-MMS model is superior to SC-MTS model because it only needs to compare the measurement scale of samples to be tested and does not involve threshold determination.