Research On Fault Feature Extraction And State Identification Of Rolling Bearings

Rolling bearings are the most commonly used and most easily damaged components in rotating machinery.If they fail,they will directly affect the operating status of the entire mechanical equipment.Therefore,in-depth research on rolling bearing fault diagnosis technology is of great significance to ensure the safe and stable operation of mechanical equipment and reduce accidents and economic losses.Traditionally,the main research content of rolling bearing fault diagnosis has four parts: signal preprocessing,fault feature extraction,fault pattern recognition,diagnosis and decision.Among them,fault feature extraction and fault pattern recognition are the focus and difficulty of the whole process,and they play a key role in rolling bearing fault diagnosis and condition monitoring.This research takes rolling bearings as the research object,starting with vibration signals,and for the purpose of accurately extracting rolling bearing fault characteristic information and accurately identifying its fault types.So,it mainly solves two problems: 1)accurate extraction of fault features of rolling bearing;2)accurate identification of fault state of rolling bearing.The detailed research contents are as follows:(1)The key to identify the local fault of the rolling bearing is to extract the weak periodic fault characteristics accurately from the noisy fault vibration signal.Aiming at this problem,a method for extracting weak fault features of rolling bearings based on double clustering and Teager energy spectrum is proposed.In this method,the frequency spectrum of the original vibration signal is clustered twice,the frequency band corresponding to the maximum steepness is selected as the passband filtered signal,and the time-domain signal is demodulated and analyzed using the Teager energy operator.Experimental results show that this method can accurately and effectively extract the weak fault features of rolling bearings.2(2)Aiming at the problem that rolling bearing data sets contain high-dimensional non-sensitive feature information,a low-dimensional sensitive feature extraction method based on structural information fusion of kernel principal component analysis(KPCA)and t-distributed stochastic neighbor embedding(t-SNE)is proposed.The quantized features of the signal are extracted from multiple angles in the time domain,frequency domain and time-frequency domain to construct an initial high-dimensional feature set.Then KPCA and t-SNE are used to extract low-dimensional sensitive features with high discriminative power.Finally,the classification recognition rate and clustering analysis result input into the K-nearest neighbor classifier with the low-dimensional sensitive feature subsets are used as measurement indicators to analyze the results.Through comparison with other typical feature extraction methods and experimental analysis under variable speed,the results show the effectiveness of the proposed method.(3)In order to accurately identify the fault type of rolling bearing,a fault identification method based on quantum-behaved particle swarm optimization multi-scale permutation entropy(QPSO-MPE)is proposed.First,use the EEMD method to decompose the original signal,and use the kurtosis as a metric to filter out the IMF component containing the main fault feature information to reconstruct the vibration signal;then,use the QPSO algorithm to optimize the key parameters of MPE,and obtain MPE through optimization The model calculates the multi-scale permutation entropy of the reconstructed signal to construct the multi-scale permutation entropy feature set of bearing faults.Finally,the fault feature set is input to the GG fuzzy clustering algorithm for cluster recognition.The results show that the QPSO-MPE based rolling bearing fault identification method can achieve accurate identification of typical rolling bearing faults.