Letter to the Editor

Citation

Stylios, G.K. and Cheng, L. (1998), "Letter to the Editor", International Journal of Clothing Science and Technology, Vol. 10 No. 2. https://doi.org/10.1108/ijcst.1998.05810baf.001

Letter to the Editor

G.K. Stylios and L. ChengUniversity of Bradford, Bradford, UK

"Fabric handle" is one of the most important factors in the textile and garment manufacturing and retailing industries for the assessment of fabric and garment qualities. Traditionally, fabric handle is judged by experts, imposing many problems, owing to disagreements, which can be costly in time and quality. The lack of a unified standardisation of textile handle, which is linked with the design and manufacture of the final product and retailing, has prompted a study for searching an appropriate model for prediction of fabric handle values.

In the present study a neural-fuzzy model has been proposed and applied to predict fabric primary and total hand values (Stylios, 1996). The results show that the neural-fuzzy technique adopted in the present study is a more efficient and effective method compared to the conventional linear equation approaches (see Figure 1).

Figure 1. A neuro-fuzzy model for engineering fabric handle

The combination of the feed-back and back-propagation algorithm of neural networks, which is the first-time tested in this way, is a very convenient and flexible technology for quality control and has potential applications in the textile manufacture industries. By using this algorithm, a trained neural network could calculate the variation of each group of input parameters according to the designed change of the outputs (primary handle value for the present study). The algorithm is used to reverse engineer the desirable handle of fabrics successfully.

When a new set of measured fabric properties is presented as new input into a trained neural network, the network can provide the predicted primary handle value for those fabrics in terms of stiffness, smoothness, etc. A question that will be naturally asked may be what should be done if different primary handle values are demanded, e.g. an increase in the value of stiffness by 80 per cent; how should the input (mechanical properties of the fabric) be changed to meet this requirement? Can a neural network be trained to provide that kind of feed-back information? In this case re-engineering or reverse engineering of the fabric would be possible. In the present study, a modified back-propagation algorithm is developed and tested to facilitate engineering, re-engineering or reverse engineering of textile materials, in the general textile genetic engineering domain described and discussed first by the author.

In this feedback process, the network error is defined as the summer-squared-difference between the original network output, A2 t_lj(handle values) and the new desired output, t_lj; the following expression is adopted:

The changing of networks input is set to be proportional to the error derivative; the momentum term and adaptive learning rate are also used. The change of the input is:

Similar to the training process of the back-propagation algorithm, the changing of input is repeated until the value of the output is close enough to the desired value.

Reference

Stylios, G.K. (1996), "Textile aesthetics; material/human interaction in the determination of intelligent tactile and visual data for the design, manufacture and retailing of textile products", 10th Symposium of Human Sensation and Textile and Clothing Performance, Proceedings of the Fibre Society of Japan, 9-10 June.