A Virtual Worsted Woven Production System Based On Artificial Neural Network Technology

According to the total technology requirement of textile and the yarn quality, weaving is an important procedure in order to manufacture accepted fabrics. During weaving, there exist nonlinear relationships among yarn quality and properties, weaving parameters and loom efficiency and gray goods quality. Under the same quality of yarns, the optimization of weaving parameters can reduce the load during spinning, and improve fabric quality, and benefits to realize low cost and high efficiency.At present, there are many ways to predict textile quality, such as traditional statistic method, fuzzy algorithm and gray system theory. In this paper, artificial neural network (ANN) are adopted in order to predict loom efficiency, fabric defects, yarn count, yarn strength, weaving warp tension and weaving machine speed respectively.Loom efficiency and fabric defects are woven quality forecast models, they predict performance of loom and fabric quality based on index of materials and technical parameters. The average value of absolute error is 2.05% and the biggest error is 5.12% in the model of loom efficiency. The results indicate that ANN model has high forecast precision . The average value of error for the model in fabric-defect prediction is 7.0% and the biggest error for that is 13.90% that means there is a certain error in this model so that needs to adopt new parameters or renew quantity indexes to improve the accuracy of the model.Yarn count and yarn strength are predicted by using the feedback models. The purpose is to help to choose appropriated materials. The main method is to input target variable, such as weaving faults and efficiency etc, and technological parameters to forecast the requirements of yarn quality in terms of the feedback model. The average value of forecast error for yarn counts is 4.02% and the biggest error is 10.4%. Weaving warp tension and the speed of loom machine are also the technological parameters predicted though the feedback models in the same principle. The method can help technicians to adjust and design the weaving technology. The average value of feedback error is 5.46% and the biggest error is -8.97% using the model of the weaving warp tension; the average value of the feedback error is 3.49% and the biggest error is-7.23% by means of the model of the loom speed. The precision meets the request of the factory, and The ANN models have become into the foundation of weaving design parameter, selection and quality control in the processing.According to the features of ANN, the comparison between ANN model and the methods of multi-linear regress, fuzzy algorithm and gray theory has been conducted. The corresponding six regress models are built for prediction and compared with the results attained from the ANN models. There is available but limited precision only for the models of loom efficiency and weaving machine speed. The precision of the other models is unacceptable. It is verified that the multi-linear regress method is not suitable for the forecast. Based on fuzzy algorithm and gray theory that can solve some nonlinear and dispersing problems, the predicting models of loom efficiency built on fuzzy and gray theory have been compared with the corresponding ANN models. The predicted results from these models show that ANN is superior to fuzzy and gray theory in the prediction. The reasons is that the traditional mathematics models have lower accuracy than ANN method because some conditions are often assumed and simplied when they are established and used. On the contrary, ANN, especially for the improved BP algorithm, can achieved a high accurate prediction if the input nodes, hidden layer's nodes and training times are enough. Meanwhile, the capacity that ANN solves linear problem has also been proved.In this research, all the neural network models have been compiled in Matlab toolbox that possesses excellent calculation function and can realize the design of core algorithm. The system adopts SQLServer2000 as the base database that realizes inputting data , saving and maintenance. The...