The Tool(s) Versus The Toolkit

Abstract

In this paper we present an approach to enable inference when coupling computational design systems and engineering simulation, in order to narrow the ambiguity of a design space to a space that is meaningful for a designer's goals. Inference is a statistical technique to draw judgement about data. The emergence of computational design systems in architecture has enabled the utilization of engineering simulation to evaluate and drive exploration of the design space. However, we argue that designers find it challenging to infer an thorough understanding of the design space when considering many variables because coupled systems are limited to one directional operations (input {\textrightarrow} output). Consequently, the qualitative control over the quality of design comes into question. In response, we present a probabilistic representation of the design-analysis system whereby, input and output variables are represented as probability distributions to enable bi-directional inference between input and output. Subsequently, the capability to infer cause from effect provides a deeper understanding about the relationships between design variables and physical behaviour. Furthermore, we discuss Bayesian networks as a statistical technique to handle inference over complex design spaces.