Supporting Exploration of Design Alternatives using Multivariate Analysis Algorithms |

Supporting Exploration of Design Alternatives using Multivariate Analysis Algorithms

Rusne Sileryte, Antonio D’Aquilio, Danilo Di Stefano, Michela Turrin, Ding Yang (Delft University of Technology, ESTECO)

CHALLENGE - Parametric modeling allows quick generation of a large number of design alternatives. Ultimately, it can be combined with optimization algorithms for obtaining optimal performance-driven design. However, setup of design space for optimization is a very complex task requiring prior knowledge and experience. Therefore, this paper focuses on the process that happens before the optimization. It proposes to use multivariate analysis algorithms for exploring and understanding the relations between various design parameters, after sampling the design space. 

SOLUTION - This paper presents a design environment that integrates guidance-based support for exploration of the design space, combing efficiency, user-friendliness and flexibility. It uses multivariate analysis algorithms of modeFRONTIER optimization software in tandem with Grasshopper-based geometry visualization dashboard. The case study is the swimming pool of the Jiangmen Sports Centre in China. Four performance values are set up: maximizing average UDI (Useful Daylight Illuminance) value, minimizing energy need, maximizing energy production value, and minimizing the area of roof surface. Grasshopper plugins Honeybee and Ladybug together with simple mathematical functionsare used as external solvers for the evaluation of performance values. Design space sampling has been performed by modeFRONTIER using Uniform Latin Hypercube algorithm with 340 alternative designs, out of which 334 were evaluated as feasible. 

BENEFITS - The proposed method is computationally efficient and integrated into an environment familiar to architects. It relies on algorithms available in modeFRONTIER software together with database querying capabilities available in PostgreSQL and a developed dashboard. Clusters based on output variables suggest that the most desired behavior can be found in clusters 2.5 and 2.6, where many designs with expressive roof curvature can be found that is also preferred by the designer.