Intelligent advisory systems for fabric selection

Abstract

In general, the selection of fabric relies on the user's preferences. Such preferences are expressed in terms of psychological perceptions of fabric hand. Fabric hand is commonly adopted for the assessment of fabric quality and prospective performance in a particular end use. The fabric hand is primarily assessed by subjective judgments from sensory experts. Subjective judgment treats fabric hand as the outcome of the psychological chain reactions initiated from the sense of touch and combined with the sensitivity and the experience of judges. Conventional fabric hand evaluation system utilizes an old-fashioned Total Hand Value (THV) as a metric to model the subjective fabric hands based on fabric properties. The Kawabata's evaluation system (KES) developed by Kato Tech Company is an example that has a systematic approach on modeling fabric hand. This method has become a standard evaluation system in the industries nowadays. However, these specialized systems could not advise the users which type of fabric is most likely matched his/her psychological perceptions of fabric hand. In this thesis, we have built and characterized an intelligent advisory system for individuals in fabric selection based on his/her psychological perceptions of fabric hand. Firstly, we modeled the relationship between the sensory perceptions of fabric hand and fabric properties using a neural networks approach. In the experiments, we found that 12 fabric properties significantly correlated with 14 sensory fabric hand descriptors that are highly reliable over extended periods of time. Such correlation was validated by reliable performance of prediction of fabric hand based on fabric properties using neural networks. Thus, we implemented an intelligent advisory system for individual fabric selection incorporated with the proposed model. This advisory system was constructed using a fuzzy rule-based expert system and was validated by individuals. It was found that the prediction error is statistically insignificant. It was proved that the proposed advisory system is able to advise on the desired fabrics for individuals based on inputs of his/her preferred sensory ratings of 14 fabric hand descriptors. In addition, experimental results verified the robustness of the proposed hybrid fuzzy-neural advisory system that could provide satisfactory performance. The convergence speed increased with the knowledge accumulated from individuals' selection behavior that proved the adaptability of the proposed system. Moreover, the qualities of the rules constructed by the hybrid systems were measured and were found to be at very low mean square error after fine-tuning. Five extra fabrics were applied to the proposed system without training, and the performance of the proposed system was in close agreement with reality, this shows the generalities of the hybrid system. Taking our approach, industry could benefit from developing an appropriate fabric to match with individuals' expectations within a reasonable period of time.