Constructing and navigating the design-performance space

It is generally believed that the utilization of computational building performance simulation tools can contribute to the improvement of building designs. Accordingly, many such tools have been developed. Yet, their application in (and thus their impact on) the building delivery process has been rather limited. This thesis focuses on one possible contributing factor, i.e., insufficient support for navigation in the design-performance space. While many efforts have been invested in algorithms and models that help generate building performance data, much less has been done to support the process of organizing, exploring, and evaluating such data. To address this shortcoming, this thesis presents, for the building design domain, an approach to generation and exploration of the design-performance space. In this approach, an initial design is used to generate a set of alternative designs that collectively constitute the design space. One way of doing this relies on the "scalarization" of design variables. The scalarization leads to the representation of a building as a point in a d-dimensional design space. Each coordinate of such a space accommodates a salient (semantic or geometric) design variable. Subsequently, the entire corpus of design alternatives is subjected to performance modeling. Based on the modeling results, an n-dimensional design-performance space is constructed, where n = d + p (d: number of design variables; p: number of performance indicators). Once constructed, this space can be visualized and used by the designer to explore the relationship between design variables and corresponding performance attributes.