Automated learning of programmatic information from plans of existing buildings.

Programming is an integral part of the design of any building. A better program reduces the necessity for changes during planning, design, and construction, and results in a design solution which is better suited to current and future needs and less prone to malfunction. Computers are assisting architects in the performance of an increasing number of design and production tasks. Because of the nature of the tasks performed in the programming phase the use of computers has not been effective. Architects transform their perception of an architectural problem into a representation of a physical solution. One of their main concerns in the programming phase is to define the functional characteristics of spaces to be housed. They predict such characteristics based on their past experiences. The transformed physical solutions implicitly contain design information. A knowledge-based system is proposed in this research. This system acquires design information from plans of existing buildings, generates a building program from the acquired information, and evaluates a design solution based on the generated building programs. An automated knowledge acquisition method is developed to acquire design knowledge from plans of existing buildings. The proposed system uses frames to represent design knowledge and employes probability theory in making decisions. The proposed system is tested with elementary schools located in Ann Arbor and Detroit. The findings confirm that a computer can acquire some design knowledge through the proposed automated knowledge acquisition method, generate reasonable design guidelines with the properly defined design requirements generation variables, and evaluate a design solution against the accumulated design knowledge during a feedback process. This research contributes to ongoing explorations in the areas of architectural programming, functional design evaluation, and knowledge engineering.PhDArchitectureComputer scienceUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/162272/1/9001584.pd