Fault diagnosis method of rolling bearings Based on SPA-FE-IFSVM

Hui Han,Lina Hao

doi:10.1177/1687814020969470

Abstract

Rolling bearings are the most frequently failed components in rotating machinery. Once a failure occurs, the entire system will be shut down or even cause catastrophic consequences. Therefore, a fault detection of rolling bearings is of great significance. Due to the complexity of the mechanical system, the randomness of the vibration signal appears on different scales. Based on the multi-scale fuzzy entropy (FE) analysis of the vibration signal, a rolling bearing fault diagnosis method based on smoothness priors approach (SPA) -FE-IFSVM is proposed. The SPA method was used to adaptively decompose the vibration signal and obtain the trend item and de-trend item of the vibration signal. Then the fuzzy entropy of the trend item and de-trend item was calculated respectively. Meanwhile, aiming at the problem that the support vector machine (SVM) cannot process the data set containing fuzzy messages and was sensitive to noise, the fuzzy support vector machine (FSVM) was introduced and improved, and then the FE as the feature vector was input into the improved fuzzy support vector machine (IFSVM) to identify the failure. The method was applied to the rolling bearing experimental data. The analysis results show that: this method can achieve 100% fault diagnosis accuracy when only two component features are extracted, which can effectively realize the fault diagnosis of rolling bearings.

Highlights

Rolling bearing is one of the most basic parts in mechanical equipment, known as ‘‘Industrial Joint,’’ and have the most extensive application in aerospace, electric power, metallurgy and other industries
According to the purpose of introducing membership function (MBSF), the fuzzy MBSF can be divided into two categories: one is by introducing the fuzzy MBSF, the independent variable of the function is each training sample, considering the spatial distribution characteristics of the sample in the sample set, it is assigned to membership value (MBS), so that all training samples will get the MBS, according to the difference of MBS, the importance of the sample can be determined, so as to solve the characteristics of support vector machine (SVM)’s sensitivity to noise and outliers to a certain extent; another is to indicate the subordinate probability of the sample to this category
This paper presents a fault diagnosis method based on smoothness priors approach (SPA)-fuzzy entropy (FE) and improved fuzzy support vector machine (IFSVM), and it is applied for the fault diagnosis of rolling bearing

Summary

Introduction

Rolling bearing is one of the most basic parts in mechanical equipment, known as ‘‘Industrial Joint,’’ and have the most extensive application in aerospace, electric power, metallurgy and other industries. The main idea of these multiscale analysis methods is to decompose the original signal by using the multiscale decomposition method, and solve the entropy value of the components obtained by the decomposition, and constitute the feature vector of rolling bearing fault diagnosis. In order to solve the problem of the multiscale decomposition, a fault diagnosis method for rolling bearings based on SPA-FE-IFSVM is proposed in this paper. Compared with the traditional feature extraction method, the method proposed in this paper selects two components with large difference between trend items and de-trend items to calculate the FE value, which can better reflect the essential characteristics of signal fault. Smoothness priors approach (SPA) is a nonlinear de-trend method for signals proposed by Dr Karjalainen[21] from Kuopio University in Finland This algorithm assumes that the original data signal, namely the time series Z, consists of two parts:. For the matrix Dd, the aperiodic trend item in the signal can be well estimated when the order is 2, so the order of Dd is 2, D2 2 R(NÀ2) 3 N can be expressed as:

À2 1 0 Á Á Á 0

Xn À Á

X nÀ X nÀ À

Procedure of the fault diagnosis method

Conclusion