Research On Rolling Bearing Fault Diagnosis Methods Based On Vibration Signal Analysis

With the rapid development of science and technology,mechanical equipment is growing larger and more precise.The structure of mechanical equipment is becoming more complex.As the "joint" of mechanical equipment,the rolling bearing is one of the primary and essential parts commonly used in mechanical equipment.However,due to long-term operation of harsh conditions,such as variable speed,variable load,high-temperature,and high-pressure conditions,rolling bearings are vulnerable to damage,which may accelerate the performance degradation of mechanical equipment and cause nonscheduled shutdown or even catastrophic casualty.Therefore,the research on rolling bearing fault diagnosis technologies has important scientific significance and engineering application value.To realize the fault diagnosis of rolling bearing with high accuracies,this thesis nominates rolling bearing as the research object and takes signal processing methods and machine learning models as the core techniques to carry out experiments based on rolling bearing vibration signals.There are some critical issues,including periodic fault impulse enhancement,feature extraction,and fault diagnosis.The research contents of this thesis include the following four aspects:(1)The vibration mechanism and signal characteristics of rolling bearings are analyzed.Different characteristics of rolling bearing vibration are identified by studying the rolling bearing vibration caused by stiffness change,bearing movement,and early fault defects.A rolling bearing fault simulation platform was built.Under the speeds of 1750 revolutions per minute(RPM),2000 RPM,and 2250 RPM,twenty rolling bearings with different health states were tested,and vibration signals were sampled from them,thus forming the Shandong University rolling bearing dataset with 3720 pieces of vibration signals,which is one of the dataset sources for the experiments in this thesis.(2)Aiming at the problem of periodic fault impulse enhancement of rolling bearing,a method named maximum average impulse energy ratio deconvolution(MAIERD)is proposed.In order to adaptively design the finite impulse response filter,the MAIERD method maximizes the index,average impulse energy ratio,and the objective function method is applied to solve the filter coefficients.In the MAIERD method,the Morlet wavelet is appointed as the initial filter,and the fault period is detected by the autocorrelation function.The results on simulated and measured signals show that compared with other deconvolution methods,the MAIERD method can overcome different kinds of noise,and fault characteristic frequency and its multiples can be examined from the envelope spectrums of deconvolution signals.Also,the performance of the MAIERD methods is greatly improved in terms of feature energy ratio and computation time.Therefore,it can be declared that the MAIERD method is robust,practical,and effective.(3)Aiming at the problem of rolling bearing fault diagnosis,a rolling bearing fault diagnosis method based on adaptive variational mode decomposition and extreme learning model optimized by particle swarm optimization(AVMD-PSOELM)is proposed.First,rolling bearing vibration signals are collected,and an adaptive variational mode decomposition method is utilized to process rolling bearing vibration signals,thus obtaining four different modes.Then,oscillation energy values are extracted from all decomposed modes.Subsequently,an extreme learning machine model,whose weight and bias values are optimized by particle swarm optimization,is established and applied to identify the fault types of rolling bearings.The fault diagnosis testing results from two rolling bearing datasets indicate that the proposed AVMD-PSOELM method can realize the rolling bearing fault diagnosis goal with higher accuracy.(4)Aiming at the problem of rolling bearing fault diagnosis under noise conditions with different signal-to-noise ratios(SNRs),a fault diagnosis method based on synchrosqueezing wavelet transform and kernel extreme learning machine(SWT-KELM)is proposed.First,rolling bearing vibration signals are collected,synchrosqueezing wavelet transform is employed for high-resolution signal representation in the time-frequency domain,and inverse synchrosqueezing wavelet transform is applied for sub-signal reconstruction.Sub-signals with high correlation coefficients are selected to form a two-dimension matrix for further feature extraction.Furthermore,singular value decomposition is implemented on the two-dimension matrix to obtain singular values as the fault feature.Subsequently,a kernel extreme learning machine model,whose parameters(kernel function parameter and regularization parameter)are optimized by the beetle antennae search algorithm,is established and applied to identify the fault types of rolling bearings.The fault diagnosis testing results from two rolling bearing datasets indicate that under noise conditions with different SNRs,the proposed SWT-KELM method can still realize the goal of rolling bearing fault diagnosis with higher accuracies.Thus,the proposed SWT-KELM method is robust.What’s more,compared with the modes obtained by empirical mode decomposition and variational mode decomposition,the sub-signals obtained by the synchrosqueezing wavelet transform method contain more fault-related information,which is revealed by envelope spectrum analysis.The proposed methods in this thesis can effectively improve the fault diagnosis accuracy of the rolling bearing in dealing with issues on periodic fault impulse enhancement,feature extraction,and fault diagnosis.Additionally,the proposed methods in this thesis extend the application scope of existing methods,providing new solutions for the fault diagnosis of rolling bearings.