关键词: 经编机, 断纱检测, 小波变换, 图像处理, 深度学习

Abstract:

Objective For warp knitting machines, yarn breakage is inevitable in the working process. When a yarn breakage occurs, the warp knitting machine should be stopped immediately for yarn repair so as to avoid causing fabric defects. However, because the yarn diameter is only tens of microns, and tens of thousands of yarns are knitted at high speed at the same time in the knitting process, it brings great difficulties to the online detection of yarns in warp knitting machines.

Method Aiming at the above problem, this paper proposes a composite detection algorithm. Firstly, the defect feature enhancement method based on dimension transformation was adopted to enhance the stability of the algorithm by data processing. Secondly, on the basis of dimensional transformation data, a yarn breakage detection algorithm based on wavelet transformation was proposed, and the gray level transformation rule of the image was analyzed from the perspective of time domain to realize defect detection. Finally, deep learning was adopted to further improve the robustness of the algorithm.

Results To verify the effectiveness of the proposed algorithm, the detection system was built in KARL MAYER RD7/2-12 warp knitting machine (Fig.8). Ten cameras were adopted to shoot at a distance of 1 m. Five groups of experiments were carried out to verify the feasibility of the algorithm under different processing conditings (Tab.1). The first group of data was taken as an example and the detection processes (Fig.10). Five sets of yarn breakage experiments were conducted on three different warp knitting machines, and the methods proposed in this paper can effectively identify yarn breakage. It is shown that the algorithm proposed in this paper can effectively detect yarn breakage. The algorithm proposed in this paper was comared with STL (time series decomposition algorithm) and wavelet decomposition algorithm (Tab.2). The timeliness of different algorithms was analysed, indicating that the conventional STL decomposition algorithm had the worst timeliness (Tab.2). The proposed method and wavelet decomposition showed better timeliness, and it also proved that the introduction of deep learning had no impact on the timeliness of the algorithm. About 120 h of continuous experiment was set to verify the stability of the algorithm proposed in this paper, compared with other algorithms, the missed detection rate and the false detection rate were both decreased (Fig.11).

Conclusion Aiming at the problem of online detection of broken yarn defects in warp knitting machines, this paper proposes a robustness detection algorithm, which is proven to be feasible by a large number of experiments. An innovative method of weak defect feature enhancement based on dimension transformation is proposed to overcome the degradation of detection stability caused by various weaving processes and serious external noise interference.The problem of restricting the accuracy and timeliness of broken yarn detection on warp knitting machines has been solved. The proposed algorithm has important implications for textile defect detection. At the same time, the proposed algorithm is expected to promote the further development of textile industry towards automation and intelligence.

Key words: warp knitting machine, yarn breakage detection, wavelet transformation, image processing, deep learning