Topgene: an artificial Intelligence Approach to a Design Process

This work deals with two architectural design (AD) problems at the topological level and in presence of the social norms community, privacy, circulation-cost, and intervening opportunity. The first problem concerns generating a design with respect to a set of above mentioned norms, and the second problem requires evaluation of existing designs with respect to the same set of norms. Both problems are based on the structural-behavioural relationship in buildings. This work has challenged above problems in the following senses: (1) A working system, called TOPGENE (The TOpological Pattern GENErator) has been developed. (2) Both problems may be vague and may lack enough information in their statement. For example, an AD in the presence of the social norms requires the degrees of interactions between the location pairs in the building. This information is not always implicitly available, and must be explicated from the design data. (3) An AD problem at topological level is intractable with no fast and efficient algorithm for its solution. To reduce the search efforts in the process of design generation, TOPGENE uses a heuristic hill climbing strategy that takes advantage of domain specific rules of thumbs to choose a path in the search space of a design. (4) TOPGENE uses the Q-analysis method for explication of hidden information, also hierarchical clustering of location-pairs with respect to their flow generation potential as a prerequisite information for the heuristic reasoning process. (5) To deal with a design of a building at topological level TOPGENE takes advantage of existing graph algorithms such as path-finding and planarity testing during its reasoning process. This work also presents a new efficient algorithm for keeping track of distances in a growing graph. (6) This work also presents a neural net implementation of a special case of the design generation problem. This approach is based on the Hopfield model of neural networks. The result of this approach has been used test TOPGENE approach in generating designs. A comparison of these two approaches shows that the neural network provides mathematically more optimal designs, while TOPGENE produces more realistic designs. These two systems may be integrated to create a hybrid system.