Classification method of planetary gear box sun gear faults based on multi-class relevance vector machine

The invention relates to a classification method of planetary gear box sun gear faults based on a multi-class relevance vector machine. The classification method comprises the steps of acquiring original vibration signals of a planetary gear box sun gear by using an acceleration sensor, segmenting different fault types of the sun gear vibration signals, acquiring a characteristic value for each segment of signals, normalizing the extracted characteristic indexes, using the normalized characteristic values as an input variable, numbering characteristics corresponding to different faults of the sun gear, using the characteristics as a target value, dividing a training sample and a test sample, selecting a Gauss radial basis function as a kernel function of the multi-class relevance vector machine, substituting the training sample into the multi-class relevance vector machine to carry out classification training, carrying out classification test on the test sample by using an acquired classification model to obtain a predication target value, and comparing a predication class with an actual class to obtain the effectiveness of the classification model. The classification precision of the classification method is not lower than a support vector machine. Meanwhile, compared with an adaptive neural-fuzzy inference system, the classification method provided by the invention is obviously better in classification effect.