Research On Bearing Fault Diagnosis Of Different Working Condition Based On Balanced Distribution Adaptation

Bearing,as one of the indispensable parts in mechanical equipment,plays a dual role of supporting shaft and bearing load.Bearing are prone to fatigue damage in longterm and high-speed working environment.If the bearing is not replaced or repaired in time before it is seriously damaged,it will easily lead to production accident.Therefore,it is very important to identify the running state of bearing in advance.Traditional machine learning methods need to meet the requirement that the training set and the test set have the same distribution,that is,when picking up vibration signal,the operating conditions of equipment must me basically stable.However,except for a few scenes such as laboratories,which can meet this condition,under normal conditions,the working conditions of mechanical equipment will change according to the actual production requirement during the actual operation.Therefore,the traditional machine learning methods have great limitations for bearing fault diagnosis under different working conditions.In order to solve the above problems,this paper takes the rolling ball bearing as the experimental object,and adopts the Balanced Distribution Adaptation Method(BDA)to realize the fault diagnosis of the bearing under different working conditions.The BDA method first maps the feature samples of two different working conditions into the Hilbert feature space through the kernel method,and uses the MMD distance as the criterion for the distribution distance of the two types of feature samples,and then reduces the distribution distance corresponding to each bearing fault state in the feature space.To achieve the adaptation and closeness of the sample distribution distance under different working conditions.Finally,the KNN classifier is used to classify and identify the samples after distribution adaptation,thus solving the problem of bearing fault diagnosis under different working conditions.Firstly,considering that the fault state of bearing should be described comprehensively,the time domain,frequency domain and time-frequency domain features of bearing signal are extracted as the input samples of BDA method.The feasibility of BDA method is verified by simulation samples and bearing data of Case Western Reserve University(CWRU),and the value of iteration times T is determined.Then,in order to obtain the bearing vibration data under different working conditions,with the help of ABLT-1A bearing test-bed in MCVN105 laboratory,the bearing fault data acquisition experiment under different working conditions was designed.By building a data collection system,the vibration signal of bearings with different fault size,different load and different rotating speed are picked up.At last,using MCVN105 bearing data to carry out the different working bearing transfer diagnosis on the same test bed,and using CWRU bearing data as the source domain and MCVN105 bearing data as the target domain,the effect of BDA method is verified.At the same time,the diagnostic results were compared with PCA,KPCA,TCA and JDA.The results show that BDA method has better effect on bearing fault diagnosis under different working conditions.