Xiao Song

Induction motors are important equipment in industrial production processes. To solve the problem of characteristic harmonics overlapping when there is deterioration in different parts of the induction motor and diagnose the deterioration degree of internal components such as bearings, stator winding insulation, and air gap balance, a diagnostic method based on kernel principal component analysis (KPCA) and radial basis function neural network (RBF) is proposed. Firstly, through the experimental analysis, it is found that the 2nd, 3rd, 4th, and 5th order characteristic frequency harmonics in the current can reflect the deterioration of the motor. Secondly, KPCA is used to determine the correlation degrees between characteristic frequency harmonic contents and the deterioration of the corresponding parts of the motor. Finally, the products of characteristic frequency harmonic contents and corresponding correlation degrees are taken as the input vectors of radial basis neural network, and the deterioration degrees are taken as the output vectors to diagnose the deterioration of the motor. Through the diagnostic analysis of the experimental unit and the comparison of the actual deterioration degree of the motor after disassembly, it is proved that the proposed method can accurately diagnose the deterioration of the motor. By the proposed method, the correlation between motor deterioration and characteristic frequency harmonics has been identified, and the degree of deterioration of each part of the motor has been quantified. It can monitor the degree of motor deterioration in real-time, grasp the trend of motor deterioration, and detect early signs of motor deterioration.

Fine tuning self supervised pretrained models using pseudo labels can effectively improve speech recognition performance. But, low quality pseudo labels can misguide decision boundaries and degrade performance. We propose a simple yet effective strategy to filter low quality pseudo labels to alleviate this problem. Specifically, pseudo-labels are produced over the entire training set and filtered via average probability scores calculated from the model output. Subsequently, an optimal percentage of utterances with high probability scores are considered reliable training data with trustworthy labels. The model is iteratively updated to correct the unreliable pseudo labels to minimize the effect of noisy labels. The process above is repeated until unreliable pseudo abels have been adequately corrected. Extensive experiments on LibriSpeech show that these filtered samples enable the refined model to yield more correct predictions, leading to better ASR performances under various experimental settings.