Abnormal Behavior Detection of Needle Selector's Knife heads Based on Improved Generative Adversarial Networks

The needle selector is a key component of the circular weft knitting machine to realize the pattern knitting, to ensure that the jacquard pattern on the fabric is complete and smooth. Due to its fast movement frequency, conventional detection methods need to consume a lot of time and energy to process the data, and it is difficult to detect the movement of the knife heads effectively by human identification. In this study, GAN and convolutional neural network are mainly used to realize the detection of abnormal motion of the knife heads of the needle selector. The model not only improves the robustness and accelerates the convergence speed of the GAN model, but also realizes the data augmentation and solves the problem of insufficient abnormal data. The improved model achieves an average correct detection rate of 98.11% for three types of knife heads motion anomalies.