| Rolling bearing is the key transmission component in rotating machinery and its working status has an important impact on the normal and stable operation of the mechanical devices and the whole system.However,due to the mechanical wear,contact fatigue and other reasons caused by long-term operation,rolling bearing is easily partially damaged.Therefore,the study of fault diagnosis methods for rolling bearings is of great significance for ensuring the safe operation of equipment,reducing economic losses and avoiding potential security accidents.In this thesis,rolling bearing is taken as the research object and vibration signal of rolling bearing is taken as the research carrier.The diagnosis techniques of three single faults of the rolling bearing,i.e.,inner ring defect,ball defect and outer ring defect,are studied under different loads and different fault depths.The feature extraction technology of vibration signal and intelligent classification technology for bearing failure mode are mainly discussed and studied respectively.The main work of the thesis is as follows:(1)The background and significance of bearing fault diagnosis are introduced,and current technical schemes at home and abroad are investigated and discussed.Finally,the principles,advantages and disadvantages of feature extraction technology and intelligent bearing fault recognition technology for vibration signal are summarized.(2)Aiming at the difficulty of distinguishing single fault type of rolling bearing effectively under various fault depths and loads,a feature extraction method based on wavelets analysis called Hermitian scale-energy spectrum under continuous wavelet transform is proposed.The method can analyze the distribution and amplitude of vibration signal from the viewpoint of energy,and further extract the energy characteristics.After forming joint features under combing the time domain statistical features,it can be successfully applied to bearing fault diagnosis with small samples.Experiments show that the scheme can effectively diagnose single fault type of bearing even under different classifiers,and has good robustness and reliability when some candidate wavelets are selected.(3)In order to realize the automation of bearing fault diagnosis and improve the diagnostic accuracy,this thesis customizes the parameter optimization model of support vector machine based on genetic algorithm,and successfully applies it to the intelligent recognition of inner ring defect,ball defect and outer ring defect.Experiments show that the model can automatically optimize the parameters,and the optimized support vector machine can effectively identify the single fault type of the rolling bearing.At the same time,the whole diagnostic model can reliably identify the fault type under three different data sets.(4)In this thesis,after summarizing the dictionary learning technology and representation classification technology,an improved collaborative representation based classification(CRC)model based on dictionary learning for the multi-classification problem of identification of single bearing fault with different depths.For the proposed dictionary learning optimization model,a solution algorithm based on alternating iterative strategy of coding and dictionary updating is presented.This algorithm is very efficient because it can obtain analytical solution at each iteration.According to the characteristics of the representation classification model under learned dictionary and vibration signal,two simple feature extraction schemes combined with the improved CRC model are proposed for fault identification of bearings.Experiments show that the proposed method can diagnose three single fault locations under four different fault depths.High diagnostic accuracy of the three single fault locations under three different data sets also indicates the method has certain reliability. |
