Prediction Models For Worsted Finishing Processing Based On Artificial Neural Network Technology

Textile processing is a very complicated and long procedure, especially for worsted finishing. It is well known that finishing processing is the most complicated and changeable one and has a significant effect on the quality and end-use of the worsted fabric, besides the fiber properties, spinning and weaving. However, the studies on the theories and the relationships of the procedures have been done slowly and little that many problems as well as the corresponding relationships are still described in quality. One of main reasons is that the relationships between the quality of worsted fabric and the parameters of fiber, yarn and processing are fuzzy, complex and nonlinear relations. Hence, it is very significant to find out an accurate and effective approach to characterize these relationships and to realize quick response processing in textile industry. At present, the worldwide development of economy makes domestic textile industry faced with tough marketing competition. If want to win the market, how to obtain low cost, high quality and added-value textiles is a very urgent task for China textile industry. The project, named the intelligent technological designing and smart quality forecasting of worsted processing in textile is the key technology and the actual method to solve the problems mentioned above.The current evaluation of fabric quality mainly focuses on the style, handle and physical properties based on measurements of fabric properties and performance. In this paper, a great deal of work was done to filtrate and confirm the parameters of the material and processing before establishing the predicting models. The importance of inputted parameters affecting the fabric quality was estimated and confirmed by gray correlative analyzing. The virtual processing of whole finishing was realized by establishing artificial neural network (ANN) models. For simplification, the two ends for the worsted fabric processing and key procedures were paid the most attention during the research, mainlyfocusing on the fiber properties, parameters of yarn, finishing and fabric quality. The appropriate ANN models have been selected for the prediction and deduction to the actual processing for a China worsted mill. The essential relationships between input and output values have been found accurately and automatically after the enough training of independent data arrays. Comparing to traditional statistic methods, the hypotheses or predigestions are always made to set up a mathematical model, which results in bad adaptability and forecasting, so that multivariate linear regression models are inferior to ANN models in the prediction of fabric quality.It has been found in experiments that the predicted results obtained from the analysis for 78 groups of measured data were not better than those of the 30-groups data selected from the 78group ones. Though the analysis based on residuals analysis, a data-selecting method was put forward to improve the efficiency and prediction effect of the ANN, The calculated results indicate that the selection is effective to enhance the effectivity and accuracy of the ANN models. At the same time, ANN technique has been found to have a good capability of solving the linear and nonlinear problems.In the paper, several ANN models have been set up to deduce the characteristics of fiber, yarn, fabric and the parameters of worsted finishing. The expected results were obtained. So the methods are very practical and the suggestions from the analysis can be used for references in technological control, quality monitoring and product design.The prediction models based on ANN technique developed in this paper were achieved through the Matlab programming and Web technique, i.e. .NET framework technology, data mining and SQL database to realize the storage, maintenance, transfer and analysis of the data in this system. It can be accessed if inputted correct address of the server. The prediction and deduction can be achieved after collocating the data and the parameters for ANN models. Meanwhile, the friendline...