Machine Tool Fault Diagnosis And Application Based On Unbalanced Data Classification Method

As the mother of manufacturing,CNC machine tools play a vital role in modern manufacturing industry.However,in the production process,they will malfunction from time to time.Failure to accurately diagnose faults in a timely manner will result in a decline in product quality.Depending on the weight,it can cause unimaginable losses.Therefore,it is necessary to discover and eliminate the failure of CNC machine tools in a timely manner.Fault diagnosis of machine tools and other mechanical equipment is essentially a pattern recognition process.Conventional SVM fault diagnosis methods are based on the premise of sample equilibrium.However,in real life,failure data is extremely rare and difficult to collect which leads to an imbalance in the data set.In view of the disadvantages of unbalanced data sets,severe overlap of samples,and poor SVM performance in noise interference,a sampling algorithm is used to study sampling algorithms that are more suitable for unbalanced data.Based on this,a new type of fault diagnosis is proposed.The method was used to verify the fault diagnosis of the spindle of the machine tool.The verification results show that the proposed sampling algorithm and fault diagnosis method are feasible and effective.The main research contents are as follows:(1)Firstly,taking an artificial data set TD containing three special points as an example,it is trained by SVM,RAMO-SVM,SMOTE-SVM,BSMOTE-SVM,and ADASYN-SVM.Then,based on the training results,the advantages and disadvantages of the traditional oversampling algorithms RAMO,SMOTE,BSMOTE,and ADASYN are analyzed in detail.Finally,by taking advantage of the four types of traditional oversampling methods,an over-sampling(OCAI)algorithm considering the amount of information is proposed,paving the way for the oversampling algorithm(OSSC)based on sample characteristics.(2)According to the OCAI idea,an oversampling method based on sample characteristics(OSSC)is proposed.The algorithm considers the sample distance,the density of the nearest neighbors,and the amount of information of the samples to sort the positive samples.The noise points can be ranked in the normal sample.Then,the sorted samples are finally synthesized based on the SMOTE algorithm of SVM's(9)K-information near-neighborhood reduction synthesis ratio.The classification testproves that the OSSC has a higher positive and negative sample identification rate than the existing over(under)-sampling sampling method,at the same time,the algorithm has stronger robustness,which islaid the groundwork for the follow-up of new fault diagnosis methods.(3)Based on ICEEMDAN modal decomposition algorithm,Shannon energy entropy and OSSC algorithm,a fault diagnosis method based on sample characteristics(ICEEMDAN-Shannon-OSSC-SVM)was proposed.The experimental verification of the inner ring of the rolling bearing of the machine tool is carried out as a fault diagnosis object.The experimental results show that:the new fault diagnosis method of ICEEM-DAN-Shannon-OSSC-SVM compares the different fault diagnosis methods combinations of CEEMDAN,EEMD,EMD modal decomposition algorithm,Shannon energy entropy and SMOTE,BSMOTE,SC oversampling algorithm,which have some improvements in three aspects: fault recognition rate,algorithm stability and robustness.