Over the last three decades researchers have applied artificial intelligence (AI) techniques to assist engineers in exploring design alternatives over a vast array of possible solutions under constraints. Relevant techniques and examples are the following: expert systems (Maher 1988, Bedard and Ravi 1991), formal logic (Jain, Krawinkler and Law 1991, Eisfeld and Scherer 2003), grammars (Meyer 1995, Shea and Cagan 1998), case-based reasoning (CBR) systems (Bailey and Smith 1994, Kumar and Raphael 1997), evolutionary algorithms (Grierson and Khajehpour 2002, Sisk, Miles and Moore 2003, Rafiq, and Mathews and Bullock 2003) and hybrid systems that combine AI techniques such as a CBR system with a genetic algorithm (Soibelman and Pena-Mora 2000).
The impact of Al-based methods in design practice is negligible mainly because many of the proposed systems are standalone with no interactions with design representations currently employed in practice, such as building information models (BIM). In fact, only three of the above research projects (Meyer 1995, Bailey and Smith 1994 and Kumar and Raphael 1997) use architectural models with 3D geometry as input for structural synthesis. In the absence of such models, global gravity and lateral load transfer solutions can be explored to satisfy overall building characteristics and requirements. However, these solutions need actual architectural models to be substantiated and validated.
Another disadvantage of the above research systems that hinders their practical use is that the support provided is mainly automatic and the reasoning supported is monotonic (i. e. based on some inputs, these systems produce outputs that meet specified requirements). By contrast, a hierarchical decomposition/refinement (i. e. top-down) approach to conceptual design is followed in this research. This approach enables knowledge-based feedback to the engineer and engineer’s interactions with an architectural-structural model at various decomposition/refinement levels. A similar approach has been proposed by Sacks et al. (2000), however, their approach automates design tasks that are exclusive to architects and engineers, such as positioning spaces and proposing structural layouts. In addition, it provides no interactions with building architectural models. The approach proposed in this paper is described in the next section.