Study On Objective Evaluation Of Sewing Quality On Appearance And Sewing Parameters Design System

In the garment production process, the sewing is one of the important processes which affect the quality of clothing. Study on sewing process and sewing quality is helpful to technical improvement and equipment adjustment. As an important garment production process, sewing quality control is relatively complex, which is affected by many factors. Thus to achieve optimal quality of sewing must be relied on fully understanding the fabric performance, reasonable choosing the equipment and optimizing sewing parameters setting. Now, in garment processing, sewing parameters are empirically set, so the result is fuzzy. And the sewing quality is mainly evaluated by subjective compare so that the result is easy to be affected by environment and evaluators' subjective attitudes factors etc and lack of authority and certainty. In response to the status quo, the objective evaluation method of sewing quality on appearance is studied in this paper, and its goal is to replace the traditional subjective evaluation by use of quantitative analysis. On this basis, the rapid design method of sewing parameters is researched. Based on the actual work of garment processing, the method of fast matching sewing parameters is studied in the light of classification of fabrics. Finally, sewing parameters generation system is developed by means of hybrid programming of VB and Matlab.First of all, Research Status on evaluation of sewing quality and sewing parameters generation at home and abroad are comprehensive overviewed in this paper. The current studies, methodology and results achieved by domestic and foreign scholars in this area are analyzed. Then significance of study on objective quantitative evaluation of the sewing quality and sewing parameters generation is discussed.Then, this paper focuses on the method to achieve objective evaluation of sewing quality on appearance. The feasibility of objective evaluation of seam pucker by means of image processing and wavelet analysis discussed. Being a current common standard in the world, AATCC 88B standard pictures are shot into the computer and transformed gray images. According to the histogram and gray standard deviation of images, it is clear that uneven distribution of gray on images is significantly larger with the deterioration of seam pucker (from level5 to level 1).Wavelet transform possesses signal amplification ability of local mutations in time-frequency analysis, and can be used in the objective evaluation of seam pucker. The image of seam pucker can be regarded as a superposition signal composed by Low-frequency signal (main body of image), high-frequency signal (noise and edge of image) and middle and high frequency signal in this study. Therefore, the time-frequency localization characteristics of wavelet analysis are particularly suitable for extracting the features information of this signal.After wavelet transform of image, the horizontal detail coefficients mainly reflect the horizontal fold caused by bad seam pucker information, the vertical detail coefficients mainly reflect the vertical retails and the diagonal detail coefficients shows the whole image details. Therefore the horizontal direction is a main direction to research the seam pucker. Analysis of standard deviation of the horizontal detail coefficients shows that it is a constantly increasing trend with the deterioration of seam pucker (from level5 to level 1), and the span between adjacent grades is relatively uniform, with a larger total span, and therefore wavelet analysis used in determination of seam pucker is perfect. On the basis of definition of SPDI, analysis shows that the haar wavelet and the 5th dimension can be regarded as default wavelet and default analysis dimension used in analysis of the single or double seam images.Parameters used to objectively evaluate seam pucker are discussed in Chapter III. A group of one-dimensional Characteristic parameters are used to represent the main information of two-dimensional image so as to be more intuitive, efficient and easy to achieve signal extraction, detection or identification. The 160 random standard images are obtained respectively on AATCC 88B single and double seam pucker grade standard pictures. On the basis of gray-scale information and energy distribution on image itself, two parameters of histogram and image entropy are extracted, and analysis shows that the correlation coefficient between the above parameters and seam pucker grades reach to above 0.93.After wavelet transform, the six parameters that are standard deviation of horizontal, vertical and diagonal detail coefficients on 5th and 4th dimension are extracted. And the correlation coefficient between standard deviation of horizontal detail coefficient and seam pucker grades reaches more than 0.96 which is significantly higher than that between the standard deviation of other two directional detail coefficient and seam pucker grades. It is proved that the details on image are indeed enlarged. Finally, the image energy is extracted to be as another characteristic parameter according to the energy distribution, and analysis shows that the correlation coefficient between the total energy and seam pucker grads also reaches more than 0.96.Chapter IV discusses how to construct and test the objective evaluation model of seam pucker on appearance. According to the feature of characteristic parameters, the Probabilistic neural network model and multiple regression model used to objectively evaluate the seam pucker are constructed. It is tested that Multivariate regression model is simple, easy to understand, but its requirements to its distribution characteristics of the data are very strict, and prediction accuracy relatively is low, the correlation coefficient between the predictive value and the expected value is above 0.97. The probabilistic neural network model hasn't requirement to data distribution and has higher precision, so the effect can be better judged. Probabilistic neural network is a non-linear pattern classification technology, whose essence is a parallel algorithm developed on the basis of Bayesian minimum risk criteria. It is a fully forward calculation process, and does not like the traditional multi-layer feed-forward network such as BP algorithm which needs error back-propagation calculation. Therefore, compared to BP algorithm, the PNN network has higher degree of stability on the occasions of limited samples, which has more simple structure, shorter training time, and is not easy to converge to the local optimal advantage. The forecast results are compared when six optimal parameters are regarded as input and all nine input parameters as input of PNN model. Analysis showed that the prediction accuracy of the former is significantly higher than that of the later. However, when a larger sample volume is used, the difference becomes less obvious. And the correlation coefficient between the forecast value and the expectations value can reach to above 0.99, which proved that the model is effective and has higher precision. The PNN model can be used to predict seam pucker grade of unknown samples and the larger number of training samples benefits the accuracy of the network, allowed under the sample size.The typical fabrics are selected to test above PNN and the multiple regression models. Comparing the objective analysis result to the subjective analysis result shows that these two models have good precision, and the forecast accuracy of PNN model is above 90%, the correlation coefficient between objective and subjective evaluation reach more than 0.95. While the forecast accuracy of multi regression model is about 80%, and poorer than PNN model, the correlation coefficient between objective and subjective evaluation is above 0.90.Chapter V focuses on the realization of sewing parameters generation methods. Now, in the actual garment processing, sewing process is empirically configured on the basis of the different characteristics of the fabric. So the study on the sewing parameters generation is also based on the classification of fabric properties in this paper. 69 kinds of fabrics, which can cover the common clothing fabric from thickness, weight and performance, are selected to be tested by FAST style instrument. Then 16 performance indicators are obtained and analyzed. To simplify the follow-up processing and reduce errors, factor analysis is done to these performance parameters so as to find four common factors to replace the original 16 indicators. And then K-means clustering method is used to classify the 69 kinds of fabrics to 7 types which is the optimum type amount determined by means of mixed-f statistics. After clustering, the common performances of the various types are analyzed, which acts as a basis of follow-up configuration process.Then one or two samples are selected from various types to do orthogonal experimental design under the different sewing process. Test results are analyzed by the objective evaluation PNN model of seam pucker, and significant impact of various sewing parameters to test result is obtained and the optimal sewing parameters configure is found. The analysis result shows that sewing tensile are the most important parameter to affect the seam pucker, stitch density and thread count are second factors, and the needle number is the slightest factor to seam pucker. For unknown fabric samples, their respective categories are determined by discriminant analysis method, in order to quickly match its optimal sewing parameters. With the expansion of sample cases, classification of fabrics can be dynamically adjusted so as to make more rational classification. Finally, a sewing parameters design system is developed by using VB and MATLAB mixed programming based on all research achievements in this paper. This system not only carries out data management of fabric specifications, fabric properties, and sewing parameters, cluster Calculation of fabrics and sewing parameter generation, but also can achieve objective and subjective evaluation of seam pucker and automatically compute the errors. This system is with the characteristics of simple operation, convenient data-processing and friendly user interface, and can be dynamically expanded in order to optimize data classification and sewing parameters.In this paper, a more comprehensive study on the main factors in sewing process is done, an objective evaluation system of seam pucker is constructed and sewing parameters design system is programmed based on classification of fabrics. This approach is more close to the garment manufacture process in reality, so it has better practicality and popularization. The study achievement in this paper is helpful to more rapidly and accurately select apparel sewing parameters, control sewing quality, and finally promote rapid response in garment processing enterprises...