The existing rolling bearing intelligent diagnosis algorithm is mainly based on a single vibration acceleration signal

and the feature extraction process requires human selection

which cannot guarantee the stability of feature quality. In view of the above problems

a multiclass relevance vector machine (MRVM) bearing fault diagnosis method is proposed using the multicore parameter optimization of acoustic and vibration fusion features of bearing faults. Firstly

the concept of combined silhouette coefficient (CSC) is proposed; secondly

the CSC is utilized as an adaptation function to adaptively determine the symbol number ε \varepsilon and the embedding dimension m m of the refined composite multiscale symbolic dynamic entropy (RCMSDE). Finally

the MRVM with multicore parameter optimization is used to realize the fault diagnosis based on bearing acoustic vibration signal fusion features. The effectiveness of CSC for quantifying the class distinctiveness among sample feature vectors is verified using the public dataset of Case Western Reserve University; the effectiveness of the proposed CSC-RCMSDE-MRVM method is verified by the data from the test bed of traction motor bearings of a high-speed train. The accuracy of fault identification reaches 99.91%

which is higher than that of the traditional RCMSE-MRVM method of 99.39%

the RCMFE-MRVM method of 99.55%

and single-channel CSC-RCMSDE-MRVM method (96.01% for vibration signal and 98.95% for acoustic signal).