in collaboration with Alvise Simondetti of Arup’s Foresight team
Design is the process of balancing competing concerns – aesthetically and in performance across many objectives, e.g. reducing carbon emissions within a cost budget. Analysing design performance often relies upon computationally intensive analysis models (ray tracing for lighting, computational fluid dynamics for acoustics and pollution dispersal). Such expensive models make exploring the vast design space intractable and frequently only a couple of designs are quantitatively analysed. This means opportunities to shape a better world are missed.
The goal of this project is to develop novel tool support for tackling architectural design problems, focusing initially on urban masterplanning. The key challenge is not outright automation or optimisation, but rather supporting humans in deriving insight, particularly to understand the most profitable and flexible parts of the design space. We propose to extend our prior work with with Arup in this field with techniques from statistical machine learning, in particular the idea of Kriging (also known as Gaussian process regression) to derive a simplified “proxy” model that can be evaluated quickly. Such models should help offer instant, interactive design feedback, as well as enabling us to identify the statistically most likely-profitable directions for further exploration, with respect to multiple design objectives.
This project is funded under the EPSRC’s Industrial CASE scheme, which supports collaborative research which is based in the university but benefits from regular contact and guidance from our industry partner, Arup.