Research On Typical Failure Prediction Of Bearing Based On Vibration Data Analysis

With the rapid development of the Chinese industrial technology,the safety and reliability of industrial system have become more and more important.Bearing is an important part of rotating machinery,and is widely used in the connection between rotating parts and fixed parts.Due to the extremely complicated working environment of bearing,it is easy to be damaged,then which will affect the operation of the entire rotating machinery.For example,the bearing of aero engine is one of the main sources of flight accidents.Bearing fault prediction technology can provide theoretical guidance for preventive maintenance and make a good maintenance plan in advance,which has important practical significance and economic value for improving the entire rotating machinery operating life.This paper proposes a new hybrid data-driven(GM-RVM-CEEMD)algorithm and an improved relevance vector machine(KPCA-ACS-RVM)algorithm firstly.The specific work content is as follows:Firstly,the basic knowledge which is involved in the research of the Relevance Vector Machine(RVM)is briefly described,two problems that need to be solved by the current Relevance Vector Machine algorithm are pointed out: the rapid accuracy decline of long-period prediction and the poor generalization ability of the prediction model..Subsequently,in order to meet the needs of engineering applications,the study about the long-period prediction of the relevance vector machine is deeply studied.The Grey Model(GM)algorithm with good long-period prediction performance was proposed to fuse RVM.Because the fusion of RVM and GM is based on the principle of error correction,in order to make the coupling closer,the complementary ensemble empirical mode decomposition(CEEMD)is used to reconstruct the prediction error of the GM algorithm,and the reconstructed error is applied to RVM modeling,finally the GM prediction error is corrected by the RVM prediction output.In view of the problem that the prediction model cannot be updated online,this paper uses dynamic sliding window technology to continuously update the model,so that the model always maintains a good matching performance during the prediction process,thereby improving the applicability of the GMRVM-CEEMD algorithm on long-period prediction problems.Then,through in-depth analysis of the characteristics of bearing vibration data,it is difficult to fully reflect the bearing degradation problem by a single time domain or frequency domain characteristic parameter.The Kernel Principal Component Analysis(KPCA)is applied to bearing vibration signals process.The time domain and frequency domain features are fused by KPCA to construct a new fault feature extraction method that characterizes the degradation state of bearing.At the same time,through in-depth research on the mechanism of the relevance vector machine algorithm,two optimizations and improvements are made to solve the problem of the low accuracy(poor generalization performance)of the standard relevance vector machine algorithm.First,for the problem that different types of kernel functions have significant differences in the mapping effects,the global and local kernel functions are fused to form a new combined kernel function.Then,for the problem that the kernel parameter selection has a significant effect on the regression effect,an adaptive cuckoo search(ACS)algorithm is used.Finally,a combined kernel relevance vector machine algorithm based on adaptive cuckoo search algorithm is formed.Finally,online verification is performed through a real bearing accelerated degradation experimental platform.Under the same experimental conditions,by comparing and analyzing the prediction results of different algorithms,the prediction performance of the proposed algorithm is improved compared with the comparison algorithm.That is to say,the experimental results prove the effectiveness of the proposed improved algorithm.