Research on rolling bearing state health monitoring and life prediction based on PCA and Internet of things with multi-sensor

Abstract

It is an important means to improve the utilization rate and reliability of the equipment to realize the state health monitoring of the mechanical equipment by building the Internet of things with multi-sensor. Studying the correlation of multi-source and similar sensor signals can improve the comprehensive utilization of information. In this paper, the similarity measure is used to describe and analyze the correlation between the multi-sensor monitoring signals of rolling bearing, and the failure bearing data is used to realize the comprehensive utilization of information prediction of service bearing life. To overcome the disadvantage of inconsistent prediction results and low reliability of rolling bearing single feature characterization and similarity life prediction algorithm, a comprehensive similarity life prediction method of rolling bearing based on multi-dimensional feature fusion is proposed in this paper. Nine degradation features of bearing vibration signal, such as kurtosis and root mean square, are extracted. Based on principal component analysis (PCA), multi-dimensional features are fused to fully characterize the operation state of rolling bearings. The maximum and minimum life values of rolling bearings given by different features under normal conditions are obtained. By calculating the comprehensive similarity, the corresponding life proportional adjustment functions are constructed respectively. The life results predicted by PCA fusion features are corrected in real time, and the predicted life intervals of the monitoring bearings are given. The experimental data of rolling bearings at the University of Cincinnati are used to carry out the applied research. Compared with the single time domain feature prediction results, the multi-dimensional feature prediction algorithm describes the life information of rolling bearings from various angles, and the prediction results are more accurate and reliable. The method presented in this paper provides a theoretical basis for predictive maintenance and health management of rolling bearings.