Treffer: Conceptual design of buildings with Artificial Intelligence

The conceptual design stage is the most crucial part for the development of functional and resource-efficient buildings. However, due to the open-ended nature of the design projects, it is a difficult task to recognize the most promising solutions out of the wide scope of possibilities. Artificial Intelligence (AI) can be applied to create support tools that facilitate design exploration to structural engineers. In this thesis research, an AI conceptual design assistant tool has been developed based on the Genetic Algorithm (GA) method to assist senior engineers in the decision-making process of building design. The tool has been developed in Python using NSGA II, which is a multi-objective evolutionary algorithm. In this work, the design scope focused on medium-rise buildings with a rectangular plan and equally distributed span distances. The design variables to be determined by the assistant include the structural material (reinforced concrete and steel), building and grid dimensions, and floor system type. A multi-objective optimization approach has been used including three major conceptual design objectives: minimizing structural cost, maximizing free space and minimizing the environmental impact. The NSGAII algorithm is used to compute the optimal Pareto front which enables the exploration of the design space with extended information about the impact of the design variables on the objective functions. Structural performance constraints have been estimated through a preliminary sizing of the structural members and a posterior penalization of disproportionate solutions. The most appropriate lateral load system (Moment-resisting frames, braced frames or shear walls) is also determined and sized by the program using a deterministic approach based on consolidated rules of thumb in engineering practice. The assistant tool has been validated using design examples found in the literature. It provides the user with a graphical representation of the best feasible struc