Evolutionary Computation for the Topological Design of Framed Buildings

Building design is a multi-disciplinary process. The range of choices faced by the design team is vast and, given the time constraints that usually apply in the commercial world and the cognitive limitations of the human brain, the solutions reached by designers can be considered as no more than satisfying. Several research teams have been working on the development of techniques to assist building designers and help them locate good areas of the search space. The most promising techniques to date have used genetic algorithms as the search engine with parametric representations of the building and its properties. This approach has limited the designs that can be considered to buildings with regular column grids and rectangular floor plates. While regular column grids are commonplace, very few buildings are rectangular, especially in the Europe where construction sites are often based on ancient, irregular boundaries. This paper therefore describes a development of the above work using genetic programming rather than genetic algorithms as the search engine. This allows the use of a graph based form of representation, rather than parameters which coupled to additional features using a plane sweep algorithm allows the algorithm to produce building topologies for non-rectangular floor plates.