Research On Defect Detection And Quantitative Method Of Cotton Spinning Yarn

As the textile industry as a whole develops towards high quality,intelligence and digitalization,the yarn quality evaluation standard has evolved from the initial simple judgment of the presence of yarn defects to the direction of yarn defect classification and grading.Yarn defect detection based on analog circuits technology can no longer meet the testing needs of modern spinning production.In the detection and quantitative analysis of cotton yarn defects,it is found that the mixed noise factor in yarn signal increases the difficulty of extracting the yarn defects with high accuracy and identification.Due to the coupling effect of multiple factors,the yarn signal has a complex nonlinear relationship with the geometric dimension of yarn defects,and it is difficult to establish the quantitative analysis model of yarn defects.Therefore,the following research work was carried out in this paper:Firstly,based on the yarn signal sensing principle,the theoretical analysis model and simulation model of yarn signal under the influence of yarn defects were established,which laid a theoretical foundation for the subsequent yarn defect detection and quantitative analysis;An experimental platform for yarn defect detection based on capacitance sensor was built,and about 450 yarn signal samples were obtained.Secondly,a stepwise yarn defect classification method based on CWT-CNN was proposed.Two-dimensional time-frequency features with higher identification were extracted from yarn signals by continuous wavelet transform.Based on the depth separable convolution yarn defect classification network,a step yarn defect detection method was proposed based on"recognition"and"classification".And,the cotton experiments were carried out with 21 yarns as the detection object,the results show the recognition accuracy of the stepwise yarn defect detection method based on CWT-CNN can reach 94.5%on the test set,and the elapsed time is only 1/5 of the average time of the single step yarn defect detection method.Furthermore,a method for quantifying and grading yarn defects based on stacking learning was proposed.The time domain characteristic parameters of yarn defects quantitative analysis were extracted,and the algorithm model of yarn defects quantitative analysis based on stack learning was built.In combination with the classification standard of yarn defects,the quantitative analysis and classification of yarn defects were realized.At the same time,the yarn defects signal data set of methods to carry out the experiment,the experimental results show that the stack based on the quantitative analysis model of fitting to learn several R~2 tend to 1,and on the fitting effect is superior to other single model algorithm,and yarn defects classification results consistent with the sample real level,shows the yarn defects of stack based on learning,the model has better stability and the accuracy of quantitative analysis.Finally,a yarn defect detection and quantitative analysis software was developed based on the above yarn defect classification algorithm and quantitative analysis and classification algorithm.The performance of the software was tested with test sample data.The test results show that there is an acceptable deviation of 1%-5%between the quantitative analysis results of yarn defect diameter and yarn defect length and the real sample data,but the yarn defect classification results are basically consistent with the real sample data,which verifies the feasibility and validity of the research method in this paper.