Expanding architectural design automation

Abstract

Design automation is on the rise in architecture and urban design, integrating performance evaluation into the design process, exploring an array of potential designs, and assisting from the earliest stages of the design process. Multi-objective optimization enables efficient design automation by balancing trade-off objectives, such as real estate profitability and energy efficiency. This research applies multi-objective optimization to the design of district-scale mixed-use buildings, considering various requirements including land use, transportation networks, and infrastructure systems. The objective is to explore multi-objective optimization algorithms for automating the design of mixed-use residential complexes, requiring the setting of appropriate parameters and constraints, and defining an objective function based on the design goals. In architectural design, architects consistently modify floor plans to meet various requirements and regulations. Form and location of spaces for different uses are considered in a geometrically integrated manner. Previous studies focused on pre-typed floor plans, and algorithms have failed to accommodate the variability of floor plans that constantly change during the design process. Studies incorporating simulations into algorithms to find optimal energy-efficient designs have been confined to a single-use. Though differentiated studies exist that incorporate the location and orientation of floor plans, legal and regulatory review into the algorithm for the design of actual apartment complexes, these are confined to the automation of the design of single-use buildings, namely residential facilities. This study extends the differentiation by adding a mixed-use characteristic. Genetic algorithms excel at solving non-differentiable optimization problems where the values to be optimized are nonlinear and discrete. Among genetic algorithms, NSGA-II for multi-objective optimization addresses multiple trade-off objectives simultaneously to produce a set of optimal solutions. Therefore, NSGA-II is well-suited for automating the design of mixed-use residential complexes with multiple geometric parameters and objectives and non-differentiable optimization problems. In this study, algorithmic exploration was conducted based on NSGA-II. Rhino 3D, Grasshopper, and Wallacei X plug-ins were utilized to execute the algorithm. The experiment site was chosen as the Garak Hyundai 5th apartment small-scale reconstruction project area, which is a Class III general residential area with a plan for 179 households. The algorithm exploration was conducted within a confined scope of three apartments and one podium on this site. This was due to the difficulty in converging to the global optimum as the number of potential solutions increases. Parameters were set such that the shape, location, and orientation of the apartments and podium could vary, while still maintaining geometric integration. Performance evaluation factors included maximizing the floor area ratio for profitability and minimizing the surface to volume ratio for energy efficiency. To ensure the generation of a feasible design, constraints were established through a regulatory review. The algorithm minimizes the number of intersection points or the intersection area between polygons, offering the advantage that the feasibility of the design is also subject to performance evaluation during optimization, which provides information on convergence. Two experiments were carried out to enhance the algorithm in terms of feasibility and performance. The initial implementation indicated that in terms of feasibility, a metric indicating how close the design is to the planned number of households best signifies optimal convergence. Regarding performance, a negative correlation was found between floor area ratio and surface area to volume ratio, suggesting a rediscovery of the trade-off between profitability and energy efficiency. This research also discovered that convergence to the optimal surface area to volume ratio enlarges the podium floor plan area, which increases the building coverage ratio, and decreases feasibility by extending the podium floor plan beyond the site boundary. A second algorithm was implemented with the enhanced direction of adding objective-based constraints. This research refrained from considering the weighting of the objective-based constraints as this would skew the trade-offs and decrease the optimization performance. For the seven constraints defined as intersections between polygons, this research introduced seven additional constraints with different formulas but the same implications. These new constraints switched the measured elements: from the number of intersection points to the intersection area, and vice versa. Compared to the initial implementation, the second one demonstrated an improvement in feasibility, better aligning with the planned number of households. The negative correlation between floor area ratio and surface area to volume ratio, and the decreased feasibility of convergence to the surface area to volume ratio optimum, were reaffirmed in the second implementation. The hypothesis for the second algorithm proposal involved adding objective-based constraints, which were designed to mirror the original ones. However, out of the seven constraints added, four demonstrated significant positive correlation with the pre-existing constraints. This indicates that incorporating objective-based constraints into an algorithm can provide information about convergence, but constraint control remains challenging. This study deepens the understanding of the complexity of automating the design of mixed-use residential complexes. It notably rediscovered the trade-off between profitability and energy efficiency, even when using metrics of floor area ratio and surface to volume ratio, and found an inverse correlation between surface to volume ratio and feasibility for the podium. Meanwhile, a feasibility improvement method known as additional objective-based constraints was examined over two algorithm implementations, but it did not result in significant improvement. Another limitation is that a confined scenario of three apartments on a small site was used in the experiment, and only part of the energy efficiency, represented by the surface area to volume ratio, was reflected in the algorithm. Future research will continue to explore various optimization algorithms like integer programming beyond genetic algorithms to enhance the design's feasibility. Furthermore, to develop a universal and generalized design automation algorithm, additional performance metrics such as energy efficiency will be considered, and experiments will be conducted in various contexts and scales.