Diagnosis of Rolling Bearing Based on Classification for High Dimensional Unbalanced Data

Qi Hang,Jinghui Yang,Lining Xing

doi:10.1109/access.2019.2919406

Qi Hang, Jinghui Yang + Show 1 more

Open Access

https://doi.org/10.1109/access.2019.2919406

Copy DOI

Journal: IEEE Access	Publication Date: Jan 1, 2019
Citations: 29	License type: CC BY 4.0

Affiliation: National University of Defense Technology

Abstract

Motor systems are becoming more and more vital in modern manufacturing and bearings play an important role in the performance of a motor system. Many problems that arise in motor operation are related to bearing faults. In many cases, the accuracy of the devices for monitoring or controlling a motor system highly depends on the dynamic properties of motor bearings. Thus, fault diagnosis of a motor system is inseparably related to the diagnosis of the bearing assembly. The fault diagnosis of rolling bearings is substantially a classification problem. The traditional application of random forest (RF) to fault diagnosis methods is based on balanced data. However, in a practical situation, it is difficult to collect the fault data that are usually unbalanced. In order to solve this problem, in the first step, we propose a two-step (TS) clustering algorithm to enhance the original synthetic minority oversampling technique (SMOTE) algorithm for the unbalanced data classification. Then, based on the improvement of the SMOTE algorithm, we propose the principal component analysis (PCA) and apply it in the field of high-dimensional unbalanced fault diagnosis data. In this paper, we apply this new method to the fault diagnosis of rolling bearings, and the experiments conducted in the end show that the improved algorithm has a better classification performance.

Highlights

Nowadays, it generates a large amount of data in the field of finance, Internet and intelligent manufacturing
The main work is divided into three parts: firstly, we propose a two-step clustering algorithm (TS) to enhances the original synthetic minority oversampling technique (SMOTE) algorithm for the unbalanced data classification which can solve the shortcomings of using SMOTE algorithm alone, and we call this combined algorithm as TS-SMOTE algorithm
The study indicates that this new fault diagnosis method, principal component analysis (PCA)-TS-SMOTE-RF, has better performance in setting evaluating indicator like recall, specificity, accuracy, AUC and G-mean than directly categorizing the original data by random forest algorithm or classifying after applying TS-SMOTE or PCA

Summary

INTRODUCTION

It generates a large amount of data in the field of finance, Internet and intelligent manufacturing. The study indicates that this new fault diagnosis method, PCA-TS-SMOTE-RF, has better performance in setting evaluating indicator like recall, specificity, accuracy, AUC and G-mean than directly categorizing the original data by random forest algorithm or classifying after applying TS-SMOTE or PCA. 2) MACHINE TOOL FAULT DIAGNOSIS BASED ON RANDOM FOREST ALGORITHM UNDER MACHINE LEARNING support vector machine has better generalization ability in small sample, nonlinear dataspace, etc., its classification performance is poor under multi-dimensional and large sample setting. The original random forest algorithm is under the condition of balanced data set, but in actual production process the fault samples are unbalanced Due to these reasons, when using the original random forest algorithm for rolling bearing fault diagnosis, the recognition rate of fault samples will drop which directly leads to poor performance of the classification.

EXPERIMENT AND ANALYSIS

EXPERIMENTAL DATA

FEATURE EXTRACTION AND PCA DIMENSIONALITY REDUCTION

VIII. CONCLUSION