To design comfortable clothing, a quantitative assessment of clothing comfort is important. However, no effective assessment methods have been developed. In this study we carried out wearing tests and evaluated clothing comfort quantitatively using a three-layered neural network (NN) by back-propagating errors. In the wearing tests, cotton, polyester, and modified polyester apparel were used. Three subjects were kept at rest during 120 minutes' exposure to three consecutive ambient humidity changes (40→75→40%) at 33.0°C. Skin temperatures at 6 sites on the body, temperature and humidity inside the clothing, weight loss, and wearing sensation were measured. In the NN, the number of input factors was considered in two different cases; one with 9 physical factors, and the other with 12 factors adding three subjective factors. Clothing comfort was used as an output of the NN. An increase in the generalizing ability of the NN was observed when the subjective factors were added. Similar results were obtained using quadratic-multivariate analysis with twelve input factors and it was confirmed that the sensations in clothing were important factors when evaluating clothing comfort. A difference between hydrophilic and hydrophobic fibers was observed as follows as a difference in generalizing ability and weight coefficients in this model and the effects of clothing material on clothing comfort was made evident. The generalizing ability of clothing comfort in cotton clothing was smaller than that in polyester clothing and modified polyester had similar weight coefficients of the networks to cotton in the learning of the NN, both ascribable to the fact that water sorption ability in cotton and modified polyester was larger than that in polyester. It was also found that evaluation by NN was very useful for even an object which was difficult to model exactly, such as in an experiment with physiological changes.
View full abstract