Research On Fault Diagnosis Of Rolling Bearing Based On NLM-VMD And Metric Learning

As an important part of rotating machinery,rolling bearings are widely used in industrial production.Bearings often work in high-load,high-speed environments,which are prone to failures and cause equipment downtime,and may even pose a threat to life safety.This thesis focuses on the practical significance of rolling bearing fault diagnosis.Based on the vibration signal,the diagnosis is divided into two parts: feature extraction and model training.Firstly,a fault feature extraction method combining Nonlocal Mean Denoising(NLM)and Variational Mode Decomposition(VMD)is proposed.Then,metric learning is introduced to change the sample distribution to improve the performance of the classifier.Based on the analysis of the research results at home and abroad,this thesis presents a bearing diagnosis method with good noise resistance and high diagnostic accuracy.In this thesis,NLM is deeply analyzed,and it is used for bearing vibration signal denoising in one-dimensional.In view of the fact that NLM denoising effect is greatly affected by the parameters,a bayesian method is proposed to optimize NLM parameters.In reality,due to various factors,the real bearing fault vibration signal may contain strong environmental noise.Aiming at the rare problem of feature extraction of fault signal with low signal-to-noise ratio,this thesis proposes a fault feature extraction technology combining NLM denoising and VMD.The original signal is decomposed by VMD,and the main modal component representing fault feature information is selected.Furthermore,the Bayesian optimized NLM is used to denoise the main modal component signal,which solves the disadvantage that VMD can not completely eliminate the noise in the signal,and effectively strips out the fault information which is buried in the noise.For the training of classifiers,this thesis successfully introduces the metric learning into the research of bearing fault diagnosis.Metric learning can adaptively learn metrics from raw data,change the original sample distribution,and greatly improve the accuracy of the classifier.In this thesis,based on the Large Margin Nearest Neighbor algorithm(LMNN),the K-Nearest Neighbor Classifier(KNN)is improved,and multiple sets of controlled experiments are designed to illustrate the effectiveness of metric learning in bearing fault diagnosis.