Adaptive Dynamic Mode Decomposition Theory And Its Application In Mechanical Fault Diagnosis

With the rapid development of society,industrial production is gradually developing towards intelligent automation,and mechanical equipment become the main body and power generator of industrial production.As rolling bearing is one of the most important parts in machineries,it is particularly important to diagnose and recognize rolling bearing faults.It is the most common method in rolling bearing fault diagnosis to monitor and diagnose the state of bearings,which their components by using random vibration signals.In the research of fault diagnosis technology,Dynamic mode decomposition(DMD)theory,as a data-driven mode decomposition technology,can accurately extract the spatio-temporal characteristics of complex dynamic systems.Meanwhile.It has been widely used in the state assessment and feature extraction of time series.In the process of rolling bearing fault diagnosis,fault feature extraction and fault type recognition become the main concerns of fault diagnosis.Therefore,aiming at rolling bearing fault vibration signals,this thesis takes DMD as the research method,and takes feature extraction and fault diagnosis as the research objectives.The main research contents of this thesis are as follows:1)In order to solve that the truncated rank is difficult to be accurately selected by traditional DMD,an Adaptive dynamic mode decomposition(ADMD)algorithm is proposed.The Improved particle swarm optimization(IPSO)algorithm is used to realize the adaptive selection of truncated rank and threshold in DMD,which can avoid human intervention.2)To solve the problem that fault feature signals are difficult to be extracted effectively under strong background noise,a fault feature extraction method of rolling bearing based on ADMD is proposed.The method is verified by using the bearing fault data of the experimental bench and the field measured rolling bearing fault data of 1.5MW direct drive permanent magnet fan.By comparing with traditional Empirical mode decomposition(EMD),Singular value decomposition(SVD)and Proper orthogonal decomposition(POD)methods,it is verified that the proposed method can extract fault features effectively and have good results.3)To solve the problem that bearing faults in planetary gearbox are difficult to be identified,a fault classification method for planetary bearings based on Adaptive multiresolution dynamic mode decomposition(AMDMD)is proposed.The method is verified by the fault data of planetary gear box bearing experimental bench.Compared with traditional methods such as EMD and Convolutional neural network(CNN),the classification accuracy of the proposed method is 96.43%,which has obvious classification effect.