Rolling Bearing Fault Diagnosis Based On Time And Frequency Analysis Of Vibration Signal

Rolling bearing is one of the important components of rotating machinery transmission.Due to its complex structure,heavy load,harsh working environment,and long-term high-speed working condition,it will inevitably cause fault.Once the rolling bearing fails,it will affect the working condition of the entire mechanical system.In order to avoid unnecessary property losses and personnel injury,it is very necessary to use effective bearing fault diagnosis methods to detect the health states of the rolling bearing.In general,rolling bearings cannot be inspected directly due to disassembly,machine size,or working environment limitations.Vibration signal of rolling bearing can reflect the working condition of rolling bearing,so vibration signal analysis is one of the main methods for fault diagnosis of rolling bearing.However,due to the small amplitude and strong environmental noise of the fault signal,the vibration signal is a non-stationary signal with strong time-varying and frequency-varying,which makes it more challenging to extract fault features.To solve this problem,this paper first proposes a rolling bearing fault diagnosis method based on Robust local mean decomposition(RLMD)and Kmeans++ that can suppress the end effect and modal mixing.Then,since time-reassigned Multisynchrosqueezing Transform(TMSST)based on time-frequency masking(TFM)can better extract the features of strong time-varying and frequency-varying signals,a fault diagnosis method for rolling bearings based on TFM-TMSST and Extreme Learning Machine(ELM)is proposed.The main research contents are as follows:(1)A fault diagnosis method for rolling bearings based on RLMD and Kmeans++is proposed.In view of the characteristics of rolling bearing vibration signals that are usually non-stationary and non-linear Frequency Modulation(FM)and Amplitude Modulation(AM)signals,the rolling bearing vibration signals are decomposed into several product functions(PF)using RLMD,the sensitive PF components are selected by the cross-correlation function,and the signal is reconstructed using the selected sensitive PF components;Calculate the time and frequency domain statistical features of the original vibration signal and the reconstructed signal as the initial feature;Linear discriminant analysis(LDA)was used to extract the Fisher featur of rolling bearing.The training sample set of rolling bearing Fisher features is used to train the Kmeans++clustering model and obtain the corresponding clustering center.By calculating the hamming closeness degree between the test sample and the cluster center,the fault condition of the test sample is diagnosed.The experimental results show that the proposed method can effectively diagnose the fault type and damage degree of the rolling bearing by using the simulated vibration data of the rolling bearing with different signal to noise ratio and the vibration data of the rolling bearing with different types and damage degrees.(2)A fault diagnosis method for rolling bearings based on TFM-TMSST and ELM is proposed.In view of the strong time-varying and strong frequency-varying characteristics of rolling bearing vibration signals,the TMSST is used to process the rolling bearing vibration signals.The TFM technique is used to TMSST result for noise reduction,and get the reconstructed signal.The time and frequency domain statistical characteristics of the original vibration signal and the reconstructed signal are calculated as the initial features.The Principal component analysis(PCA)is used to extract the principal component features containing the main rolling bearing fault information,and the LDA method is further used to increase the distance between classes and reduce the distance within classes.The principal component Fisher features training sample is used to train the constructed ELM neural network model;Finally,the test sample is used for fault diagnosis of rolling bearing.Based on the simulation fault rolling bearing vibration data and the test bench rolling bearing vibration data are used for experimental analysis,the experimental results show that the method can diagnose rolling bearing faults more effectively than other traditional methods.(3)The experimental results of two proposed methods in(1)and(2)are compared and the data of the rolling bearing data center at the university of Paderborn experiment is used to analyze too,the results show that the method diagnosis performance of both methods are very good,even though the number of training set is less and the SNR of signal.The recognition rate of the method in(1)is slightly higher,but the diagnosis time is longer.