Considering that the defect of four kinds of mathematical models can not reflect the complex relationship between the features in variable predictive model based class discriminate method

a variable predictive model based radial basis function (VPMRBF) method was put forward.

Firstly

the abstracted features were input into VPMRBF classifier

and then the training samples were used to establish radial basis function models

which could reflect the complex relationship between features; finally

the established radial basis function prediction models were used to predict the features of those test samples

and the sum of squares prediction error could be employed as a basis for necessary classification. Experimental results of roller bearing fault diagnosis showed that the recognition rate of VPMRBF increased by 4.75%

1.75% and 5.25% respectively

compared with the radial basis function neural network

the support vector machine and variable predictive model based class discrimination method. By making full use of and effectively combining the advantages of radial basis function and variable predictive model based class discrimination method

this realized entire diagnosis from fault feature extraction to fault identification.