The issue of this paper is the use of perceptual evaluations of products as a simulation platform for improving decision making in the design process of a new product. In previous work, we proposed bayesian kansei models using unsupervised learning of relationship between perceptual evaluation of a family of similar products and technical characteristics of these products. We showed the main advantages of such bayesian models compared to other techniques in that they are not opaque to the designer and they can be easily used in compound analysis/synthesis scenarios. The objective of this paper is to show the limits of such unsupervised bayesian kansei models and to propose the use of complementary supervised bayesian kansei models. We show that these models are more accurate to carry out a local optimization of a design.

This content is only available via PDF.
You do not currently have access to this content.