Building codes are established to safeguard the health and safety of a city. It is however a universal matrix which constrains each site in the same manner. The thesis proposes a new tool to generate and evaluate how by multiple parameters and quantitative analysis building massing would impact context. By using a fully parametric approach, layers of spatial data are generated to test for its performance. Iterative designs are evolved by means of a machine learning algorithm to reach the optimal solution by defined weighting of parameters. This would be customizable depending on the user of the tool. The outlined methodology will formulate a criticism on Hong Kong’s building code as its need to adapt for digital data in the future.
The multi-objective criteria explored in this thesis includes building code constraints, environmental performance, view orientations
building performance and feasibility tests etc. Such tool would redefine concept design with
data informed designs and ultimately optimize time and sustainability for building massings.
Do we optimize for maximum ‘Visibility’ or ‘Structural’ efficiency? What about ‘Connectivity’ or ‘Sustainability?’ With multi-objective optimization we develop a scoring matrix to nest each variable together. Our algorithm generates design options and scores them using a multi-point radar diagram. This allows for further evaluations by the architect using quantified data.
A machine learning approach is used to refine results over time. For each generation, the computer generates massings randomly and performs an evaluation matrix with nested parameters. The best scoring designs are know as ‘Elites’, these are carried forward into the next generation. Another 95% is randomized and the process is repeated for eight generations. With more start designs per generation, the data would be more diverse whilst increased number of generations would refine for one optimal solution. The machine learning algorithm is conducted through ‘Discover’ a grasshopper plugin.
The setup of the parametric plan uses 3 control points to form a polyline. A building profile is shaped by an offset relationship to such polyline. A top and bottom profile is used to create a lofted massing. The machine algorithm will test different location of the 3 control points to make variations and fine tuning adjustments.
