Research On Yarn Detection System Of Joint Machine Based On Machine Vision

With the rapid development of the textile industry,it is the general trend to promote the intelligent textile workshop.The core competitiveness of textile enterprises will be enhanced by realizing the replacement of manual textile machinery automation.In the automatic production line of circular weft knitting,the joint machine is used instead of manual to realize the automatic joint process of yarn.The joint machine absorbs the end and end yarns on different yarn tubes through the yarn guide tube,and sends the absorbed yarns to the knotting device to complete the joint.In this process,to ensure that the yarn guide tube can successfully absorb the required yarn is the premise of realizing the automatic knotting process.If there is no yarn in the yarn guide tube or only a single yarn is sucked,the yarn joint process will fail.In this paper,to solve the problem that multi-state and multi-type yarn absorption in the yarn guide tube of the joint machine is difficult to detect effectively in real time,a yarn detection system of the joint machine based on machine vision is proposed to realize the real-time monitoring of yarn number and color in the yarn guide tube of the joint machine,so as to ensure the reliability of the joint.The main contents of the paper are as follows:(1)On the basis of traditional image processing,this paper proposes the effective detection of yarn number based on frame difference method and Hough line algorithm.Gaussian and median filter is used for denoising and binary segmentation.Finally,frame difference and edge detection of pixels are used as Hough lines to judge the number of yarn in the yarn guide.In order to solve the difficulty of threshold segmentation of light-colored yarns by frame difference method,HSV operation was added to extract color targets.In view of the yarn breakage in the fitting of Hough line,a 1×13 operator closure operation is added to smooth the fracture.The accuracy of yarn detection based on frame difference method and Hough line method is 97.83%and 98.17% respectively,and the yarn detection effect is generally good.(2)In yarn detection by traditional image processing algorithm,it is easy to be affected by the environment and different types of yarn need to set the algorithm parameters separately.Later,the type of yarn detection by the splicer will increase,and its algorithm still needs to be improved.Therefore,an image classification method based on deep learning is proposed to identify the number and color of different types of yarns in the yarn guide tube.A lightweight small student network is constructed by using superimposed depth detachable convolutional module.In order to accelerate the training speed and improve the accuracy of the model,a combination of transfer learning and knowledge distillation was used for training.After testing,the accuracy of top-1 on the verification set was 99.57%,and the accuracy of the student network was improved to 99.28% after distillation of the original accuracy of 96.00%on the test set.The robustness,stability and accuracy of the whole algorithm have been improved to some extent.(3)In order to realize the practical application of yarn detection algorithm,the platform of yarn detection system is designed.Including the detection system device structure,hardware circuit and PCB layout.Then,the circuit board is modularized and tested.Finally,the student network model obtained by combination training on the PC side is deployed on the embedded platform after parameter quantization and format conversion.After testing,the actual yarn detection accuracy is 99.16%,and the detection speed after NPU acceleration is about 10.75 fps,which can meet the requirements of yarn detection in practical application.