Magenetic Tile Fault Detecion of High Voltage Electitric Machine: A consistent soft-label-based Multi-view feature selection Method

The detection of high voltage permanent magnet motors has always been a big problem due to the interference of high voltage and magnetic field on the diagnosis. Especially the magenetic tile of the motor, the failure of the magenetic tile will directly lead to the operation failure of the motor. We propose Multi-view Unsupervised Consistent Soft-label Feature Selection(MUCSFS). This method constructed consistent pseudo-labels through soft labels of clustering affinity of each view sample and constructed the model by integrating selection constraints into the mapping model. This model is used to filter the fault data set to get the feature subset, and the feature subset is used to cluster. We verify the effectiveness of the method by simulating multi-view data and through the fault clustering experiment of the magnetic tile fault data set of high voltage motor, it is confirmed that our method can effectively cluster the fault categories.