An architectural design method using rank-based interactive evolutionary algorithm

Abstract

Evolutionary algorithms are a stochastic search methodology that has been widely researched and used in engineering design and has recently found applications in other design fields. In each context, their aim is to maximize the satisfaction of one or more goals, based on calculated satısfaction of the goals. There are tasks, such as architectural design, where calculating satisfaction is problematic, because goals involve experiential qualities, such pleasure, comfort, meaning, etc. Although this creates a bottleneck for computational treatment, a person can evaluate experiential qualities by aesthetic judgment or some reasoning. Interactive evolutionary algorithm (IEA) is a form of evolutionary computation designed to utilize information provided through human subjective assessments. Aesthetic judgment is subjective measurement of pleasure resulting from a perception. In this respect, aesthetic judgment takes place in real-time and without conceptual abstraction. Because there is no abstraction in this type of judgment, the knowledge a designer can exercise when he/she faces multiple alternative designs, is determining which design is more pleasant to perceive compared to another one. Determining the exact score of each design on some absolute scale is problematic due to the subjective nature of judgment, ı.e. the absence of a consensus about such a scale. Therefore, this study proposes a design method based on an interactive evolutionary algorithm using a non-dominated sorting method that is well-known in the context of multi-objective evolutionary algorithm. In this method, the fitness value is assigned based on people's subjective preferences, allowing to gradually approach to the best design solution based on one's aesthetic judgment. The method developed is applied to the case of a theater named Schauspielhaus designed by Jorn Utzon (1918-2008) in Zurich, Switzerland. The ceiling module, one of the conspicuous of Utzon's design, was modeled and parameterized in Grasshopper. The convergence behavior of the proposed algorithm during the design process was examined throughout 364 design generations by 25 participants. The results indicate that the proposed algorithm is able to integrate the aesthetic preferences of the designers. The study also yields hints about the richness of the resulting information produced by the interactivity, by applying non-parametric statistical tests as well as unsupervised machine learning for design knowledge elicitation.