Research And Realization Of Visual Defect Detection System For Cord FabricYarn Of Air-jet Loom

Cord fabrics are framework materials of rubber products such as tires,which determine the strength of tires to a large extent.Therefore,the effective defect detection is of important significance for the cost control,quality management and competitiveness promotion of enterprises so as to improve the quality of cord fabrics.At present,the defect detection of cord fabrics by domestic enterprises is dependent on the operator's visual detection.The results tend to be affected by the inspector's experience,physical condition and sense of responsibility,so the frequent errors and omissions reduce the reliability of the detection and affect the quality of the product.In view of the above problems,this paper used machine vision technology to replace naked eye and achieve on-line automatic defect detection of cord fabric yarn.The system improves the detection efficiency and accuracy,liberates labors,and promotes the level of automatic production.The main work and contributions of this paper are as follows:1.The existing cord fabric defect detection systems are all carried out after the completion of the weaving.In fact,the defects of cord fabrics are caused by those in the yarn itself and not generated during the weaving process.The on-line defect detection and interception alarm before weaving was first proposed and realized in this paper,which avoids reworking after weaving and greatly improves the production efficiency and the quality of cord fabric products.2.A defect detection algorithm for cord fabric yarn based on Gabor feature was proposed.The Gabor response features of the image were extracted by Gabor filter energy values from different scales and directions.Then,the binary image analysis was used to realize yarn defect detection.By comparing the detection effects of Gabor features from different scales and directions on the defect image,the scale and direction parameters of the Gabor filter suitable for yarn defect detection during weaving were selected.The experimental results showed that the yarn defect detection accuracy by this detection algorithm reached 99.2%.3.In order to reduce detection errors caused by yarn cross,according to the characteristics of defective yarn and the difference in the direction of crossed yarn,this paper adopted the rectangular subwindow analysis with variable width,set an appropriate width to reduce the Gabor feature area of crossed yarn in the rectangular subwindow,and set the area threshold to reduce the system stop due to yarn cross and improve the production efficiency.4.The yarn has two types of joints(yarn joint and hand seam joint),and the Gabor feature response is close to yarn defect features,resulting in error stop in the production process.In order to solve this problem,through the statistical analysis of the length and width geometrical parameters of them,the distribution range of the geometric eigenvalue was obtained.Accordingly,the minimum distance classifier was designed based on the geometric parameters to achieve rough classification,that is,yam joint and hand seam joint seemingly.The HOG(Histogram of Oriented Gradient)features of the rough classification results were extracted further.Through the two SVM(Support Vector Machine)classifiers,the defect missing report rate was adjusted to promote the joint recognition rate.The results showed that the accuracy ratio of the minimum distance classifier for yarn joint and hand seam joint was 99.97%.The recognition rate of the two HOG-SVM classifiers for yarn joint and hand seam joint was 56.7%and 43.2%respectively in the case of 0%defect missing report rate.The yarn joint filter increased the production efficiency of the air-jet loom by 8.33%.5.Based on the algorithm in this paper,the on-line automatic defect detection system of yarn before weaving was realized first in China.14 systems were installed in the cord fabric weaving workshop of a high-performance fiber company in Guangdong to support its 14 machines,and after 3-shift operation for 4 months(August,2016-November,2016)the statistics showed that due to the system the customer complaints of the number of abnormal yarn rolls decreased from 59.3 to 8.5 per month.