Image feature extraction from wavelet scalogram based on kernel principle component analysis

The scalogram image features of unbalance,misalignment,rub-impact and oil whirl fault were analyzed,and a new feature extraction method from the wavelet scalogram of fault signals was put forward based on kernel principle component analysis(KPCA).By using ZT-3 multi-functional rotor test bed,32 samples for each type of fault were obtained,continuous wavelet transformation was carried out,and KPCA feature,scalogram texture feature and spectrum feature were extracted.Finally,the extracted features were tested and classified by using parameter self-adaptive support vector machine.Analysis result shows that the average recognition effect of features extracted by KPCA is up to 90%,and is higher than the classification results of scalogram texture feature and spectrum feature,so KPCA can effectively extract the features of scalogram and is helpful for the intelligent diagnosis of rotor faults.2 tabs,2 figs,10 refs.