This paper describes a specific computational framework for performance-guided design explorations. It relies on the concept of the "design-performance space", which denotes a virtual space defined by multiple design and performance dimensions. Each design dimension accommodates the range of possible values of a discretized design variable. Each performance dimension accommodates the range of the values of a specific performance indicator. We present a method for the rapid generation of the design-performance space around an initial design. This method relies on the observation that specialized neural network copies of detailed simulation programs may be derived that could yield dependable results for a typologically and contextually well-defined set of buildings. Once constructed, the design-performance space can be visualized and used by the designer to explore the relationship between a specific constellation of design variables and the resulting performance attributes. The paper concludes with a case study, which demonstrate the functionality of the system via illustrative design exploration trajectories in the design-performance space.
展开▼
