Research On Fault Diagnosis Method Of Rolling Bearing Using Genetic Programming

Rolling bearing is an indispensable part of rotating machinery,and its running status will affect the operation of the whole equipment.The research of rolling bearing fault diagnosis technology is of great significance to ensure the safe and stable operation of rotating machinery.The traditional rolling bearing fault diagnosis methods have strong design and specificity,but weak intelligence and adaptability.In the real engineering environment,the monitoring signals are complex and changeable will cause the misdiagnosis of traditional diagnosis methods.Therefore,exploring the intelligent diagnosis methods of rolling bearing that can adaptively achieve diagnosis according to the monitoring signal is conducive to guarantee the safe operation of rotating machinery.Genetic programming(GP)is a computational intelligence method that can automatically generate solutions according to problems.The research and development of GP can realize the intelligent fault diagnosis of the rolling bearing.In this paper,the vibration signal of rolling bearing is taken as the research object,GP is taken as the technical means,and GP is used to solve the problem of rolling bearing fault detection,fault type identification,and fault type identification using a small number of samples.The main contents and innovations of this paper are as follows.(1)Traditional rolling bearing fault detection methods typically focus on using signal processing algorithm to capture fault-related pulses from monitoring signals.However,these algorithms need a lot of prior knowledge,and it is hard to put forward an intelligent algorithm that can effectively process all kinds of signals.For this issue,this paper proposes a fault detection approach using GP to design the composite morphological filter(GPDCMF).In the GPDCMF approach,GP can automatically generate a composite morphological filter according to the characteristics of the monitoring signal to capture the fault-related pulse.The fault detection will be achieved by analyzing the envelope spectrum of the captured pulse.The GPDCMF approach is applied to the simulation and experimental fault signals of rolling bearing,and the results show that the proposed approach can effectively detect the bearing fault.(2)Traditional rolling bearing fault type recognition methods typically include many independent technique steps,such as signal detection,feature extraction,feature reduction,classifier optimization,and et al.However,the results of each step will affect the final diagnosis.An effective fault recognition method needs to design and combine these steps properly using the rich expert experience,and it may only be effective for a certain fault diagnosis task.Therefore,this paper proposes a fault type recognition approach using GP to automatic feature extraction and construction(GPAFEC).In the GPAFEC approach,GP can acquire an informative feature vector from the original vibration signals according to the characteristics of the existing bearing sample signals.Having the obtained feature vector,k-Nearest Neighbors(KNN)is employed to perform fault type recognition.The performance of the GPAFEC approach is evaluated on three rolling bearing fault data sets,and the results show that the proposed approach can accurately identify different fault types of rolling bearing.(3)Most of the existing fault type recognition methods are proposed by assuming that there are enough samples to establish the diagnosis model.However,in the actual engineering environment,collecting a large number of fault samples is not easy.It is necessary to achieve the rolling bearing fault type recognition using a small number of samples.To address this issue,a new approach,namely GP multi-view feature construction and ensemble(GPMFCE),is proposed for identifying rolling bearing faults using a small number of samples.In the GPAFEC approach,GP can construct the low-level features from three views of samples into high-level features.To further improve the generalization performance,an ensemble of classifiers based on KNN is created by using the constructed features from every single view.Three few-shot rolling bearing fault datasets are used to validate the effectiveness of the GPAFEC approach,and the results show that this new approach can accurately identify the fault types of rolling bearing using a small number of samples.