Soft-margin hyperdisk tensor machine and its application to the intelligent fault diagnosis of rotating machinery

Abstract

Abstract Compared with a support vector machine, a hyperdisk (HD) classification model is a more effective model for intelligent fault diagnosis. But due to its defect of a hard margin, the formed category area sometimes does not approximate the real category area well, which means that the HD model has poor anti-interference ability, generalization ability and stability. Therefore, in order to overcome the above defects, a soft-margin HD tensor machine (SHDTM) is proposed. Firstly, by introducing the adaptive weight, the HD margin can be adaptively adjusted, that is, the details of the HD margin are added to obtain a soft margin so that it can better approximate the real category region and improve its anti-interference ability for outliers and samples with noise. Secondly, the model input is extended from vector data to tensor data. This can further improve the generalization ability and stability of the model by increasing the richness of the input information. The results of the rotating machinery fault diagnosis experiments fully prove the effectiveness of the proposed model. The SHDTM model has excellent resistance to outliers and noise interference, and also obtains good diagnostic results when diagnosing unbalanced datasets.