Fault diagnosis of needle selector drive parts based on adaptive stochastic resonance with improved generative adversarial networks

Piezoelectric needle selectors, as key weaving components in the jacquard knitting process of knitting machinery, are widely used in textile equipment. Accurate diagnostic procedures for needle selector drive faults are crucial to ensure the normal operation of the equipment. However, traditional vibration diagnosis methods cannot detect weak periodic signals, resulting in low accuracy of monitoring results. To address this issue, this paper proposes an adaptive stochastic resonance (SR) method based on an improved generative adversarial network (IGAN). Firstly, in order to solve the problem of difficulty in obtaining stochastic resonance parameters, SR is combined with IGAN, and IGAN provides the optimal SR parameters. Secondly, a soft threshold residual attention mechanism and residual network were introduced in the GAN network, and multiple generators were utilized to alleviate the problem of model collapse, in order to adapt to the actual working environment. In addition, due to the large amount of data, it is recommended to use feature parameters for training to improve the efficiency of model training. Finally, through a typical vibration data, the influence of different parameter quantities on the training accuracy of the model was studied, and the superiority of the proposed model compared to other models and the stability of the model under different environmental influences were explored. The results show that this method can effectively detect weak periodic signals of the needle selection driver, effectively alleviate the problem of model collapse in different environments, and is superior to existing methods in terms of accuracy and stability.