Prior-weighted support matrix machine for mine gearbox fault diagnosis under mislabeled data

The majority of present data-driven fault diagnosis methods for mine gearboxes are constructed in vector space, which fails to properly benefit from the structural information of 2D fault features like wavelet time-frequency images. Besides, the presence of mislabeled data can greatly impact the capability of fault diagnosis. Hence this article develops a prior-weighted support matrix machine (PWSMM) for mine gearbox fault diagnosis, which has the following characteristics: Firstly, the proposed PWSMM model possesses strong matrix learning ability, allowing the direct exploitation of structural information within matrix-form fault features. Secondly, a prior weight assignment strategy is provided for PWSMM to eliminate the negative effect of mislabeled data. Finally, an efficacious solution procedure is developed for PWSMM using the alternating direction method of multipliers (ADMM) method. Experiments reveal that PWSMM is superior to other advanced fault diagnosis models, especially when dealing with mislabeled data.