Tapered Roller Bearing Failure Diagnosis Based on Improved Probability Box Model

Hong Tang,Hongliang Dai,Yi Du

doi:10.1109/access.2020.3013875

Hong Tang, Hongliang Dai + Show 1 more

Open Access

https://doi.org/10.1109/access.2020.3013875

Copy DOI

Abstract

Bearing failure often occurs in rotating machinery. The fault diagnosis method based on the vibration signals has been studied for many years. This paper proposed an improved probability box (ip-box) modeling method for diagnosing bearing faults. The major theoretical principles involved with the probability box (p-box) modeling methods and a projection method. Since a larger aggregated width results in the p-box not being conducive to a fault identification and diagnosis, the mean of the focal element interval and the amount of data fluctuation between the adjacent focal elements were used as additional information. Then, the additional information was added to the ip-box model by the cooperative optimization method. Finally, the experimental results showed that the classification performance of a support vector machine (SVM) trained with eight measured values from the ip-box was significantly improved.

Highlights

The fault diagnosis technology of bearings has become an important means and key technology to ensure the safety and stability of production systems in the development of a modern industry [1]
The mean value of the interval and data fluctuation of the adjacent focal element interval is added as additional information to the p-box, which can further reduce the aggregated width to avoid the overlap between p-box models
This study presents a procedure for obtaining tighter p-boxes by reducing coincidence intervals using the cooperative optimization method and adding additional information

Summary

INTRODUCTION

The fault diagnosis technology of bearings has become an important means and key technology to ensure the safety and stability of production systems in the development of a modern industry [1]. An interval model is one of the common methods to describe the uncertainty of bearing signals, but it can obtain only the range of the bearing data, not its probability distribution, and the statistical information of the bearing data cannot be perfectly used. The method optimizes all the focal element interval sets in a collaborative way, avoiding the repeated calculation of an internal search, reducing the aggregated width of the p-box, and improving the overall calculation efficiency. The mean value of the interval and data fluctuation of the adjacent focal element interval is added as additional information to the p-box, which can further reduce the aggregated width to avoid the overlap between p-box models. It should be noted that the current method is used to optimize the width of the focal element interval for the p-box model, i.e., ip-box model, using the nonlinear programming process of the projection method. The difference between Eqs. (1a) and (1b) is the area of y and z, which is not equivalent

P-BOX MODELING METHOD OF BEARING TIME-DOMAIN SIGNAL

ANALYZING P-BOX AND IP-BOX

FEATURE-BASED DEMPSTER-SHAFER STRUCTURE

PATTERN RECOGNITION WITH DIFFERENT METHODS

Findings

CONCLUSION