Bearing fault diagnosis of induction motor based on mRMR-SOM method
Aiming at the bearing fault diagnosis problem of induction motor,a fault diagnosis method combining feature selection based on maximum-relevance and minimum-redundancy(mRMR),and self-organizing map(SOM)was proposed.These two methods were used under different stages when bearing fault diagnosis was carried out.Firstly,a fault diagnosis test platform of induction motor was built in the laboratory environment.During experimental stage,vibration,current and voltage signals were respectively collected under different motor states.The high-dimensional hybrid feature set including time domain and frequency domain features was obtained by statistical method.Then,with mutual information as the background,mRMR was used to screen features with strong distinguishing ability according to the correlation and redundancy between features and status tags,so as to avoid computational redundancy and posteriori diagnostic performance degradation.Finally,SOM was used to classify the healthy state and bearing faulty state of induction motor.During this stage,the effectiveness of SOM for bearing fault diagnosis and the influence of mRMR on fault diagnosis results were verified.The experimental results show that SOM was able to accurately distinguish healthy state and bearing fault states.It shows good clustering effect and clear classification boundary,the classification accuracy reaches 89％.Using mRMR feature extraction can reduce the 154-dimensional feature to 17 dimensions,shorten the network convergence time by 23.5％,and improve the classification accuracy from 89％to 98％.The experiment results verify that the mRMR-SOM method is effective for the bearing fault diagnosis of induction motor and has good diagnostic effect.
self-organizing map(SOM)feature selection based on maximum-relevance and minimum-redundancy(mRMR)mutual informationfeature dimensionality reductionfeature selectionneural network algorithmunified distance matrix(U-Matrix)
万方数据
维普
