Research On Bearing Fault Diagnosis Based On Vibration Signal Analysis

Bearing failures account for 30%of the total failures of industrial machinery.To avoid property damage and personal injury caused by bearing damage,fault diagnosis of bearings is of great significance.the initial fault features of bearing are often submerged by background noise,making it difficult for most methods to obtain effective fault information.Sparse signal recovery（SSR）methods have good noise resistance.However,most existing fault diagnosis methods based on sparse signal recovery technology use ₁l-norm or its variants to approximate impulse sparsity,but they always have the following shortcomings:（1）In different scenarios,the existing methods need to properly adjust the hyperparameter to improve performance.How to automatically adjust the hyperparameter to achieve the best performance is an open problem,and the ₁l-norm and its variants are only approximate values of the ₀l-norm,inevitably resulting in performance losses;（2）The existing methods are usually divided into two steps.The first step is to recover the time-domain fault impulse,and the second step is to obtain the envelope spectrum of the recovered fault impulse through frequency transformation.However,if the time-domain impulse cannot be accurately recovered in the first step,the frequency transformation in the second step will cause error accumulation problems,and traditional frequency transformation still has low-resolution issues when processing short-term sampling data.To address the above issues,this article conducted the following research work:Aiming at the problem that hyperparameters cannot be automatically adjusted and the performance loss caused by ₁l-norms and their variants,this paper proposes a bearing fault diagnosis method based on sparse Bayesian learning（SBL）.Because the fault signal of the bearing exhibits a brief impact signal with a short duration,and the energy of the signal is usually concentrated at some sampling points,the impulse signal in the time domain has a block sparse characteristic.By utilizing the block sparsity characteristics of fault impulses and combining sparse Bayesian learning methods to model vibration signals in the time domain,fault impulses in the time domain can be recovered.However,in the time domain,the amplitude attenuation of fault impulses is fast,their sparsity is high,and there is strong interference noise,which makes traditional prior modeling methods unable to accurately recover fault impulse signals and has performance degradation issues.To improve the accuracy of this method for time-domain fault impulse recovery and reduce performance loss,we further introduced the generalized double Pareto（GDP）prior,which significantly improves sparsity compared to traditional priors such as gamma and Laplace,thereby improving the performance of fault pulse recovery.Experimental verification using both public and private data shows that the proposed method has stronger anti-interference ability and higher accuracy in recovering time-domain fault impulses compared to other methods.Aiming at the low-resolution problem of traditional time-frequency transform and error accumulation caused by inaccurate time domain signal recovery,we present a bearing fault diagnosis method based on the estimation of signal parameters via rotational invariance techniques（ESPRIT）.Due to the influence of various interference harmonics in the actual working environment,the fault impulse signal of fault bearings contains much high-frequency clutter.To prevent the impact of high-frequency clutter on the detection results,we extracted the envelope of the detection signal and removed the interference of high-frequency clutter.However,ESPRIT can only deal with complex signals,so we introduce Hilbert transform（HT）to transform real envelope signals into complex signals,and then use ESPRIT to estimate the envelope spectrum of fault impulse.On the other hand,there is a problem of information redundancy in the real and imaginary parts of complex signals,which greatly increases computational complexity.To solve the above problems,we further propose a bearing fault diagnosis method based on the rotation invariance of real value space.This method converts complex signals into real value signals by introducing a specific unitary matrix,removing the information redundancy of complex signals,and reducing the algorithm’s computational complexity.The results of simulation and actual experiments indicate that both proposed methods can accurately estimate the envelope spectrum of fault impulses.In addition,the method based on real value ESPRIT significantly reduces the computational complexity compared to ESPRIT.