Building exterior design system by hierarchical combination fuzzy model

We propose a new "Kansei" design method for a building exterior by using a decomposed fuzzy model and its inverse problem solution. The Kansei of the human to CG image of the building exterior is measured by the SD technique. The decomposed fuzzy model is used to perform the human reactions. The model is divided into sub-models with a single output variable. Input variables are common in these sub-models. In one sub-model, each input variable is fed into the fuzzy inference unit separately, and the output signals of these fuzzy inference units are combined two by two, by combination unit. Parameters of the model are tuned by an error backpropagation algorithm using the date measured by the SD technique. For the design of the building exterior to satisfy a customer's Kansei request, the inverse problem of the model is solved. Firstly, inverse problems for all sub-models are solved separately, under a constraint of noninferior property to the solution. These solutions are unified by taking the intersection for all sub-models. In general, the unified solution is still a set of intervals, then the unique solution is required to draw a CG image of the building exterior, and is obtained by a numerical optimization method to satisfy the customer's requests as much as possible. An experimental result is reported, and the method is implemented on the WWW system using Java and CGI.