I’ve started a series of articles for sztuka-architektury.pl about harnessing the generative design methods for optimization purposes rather than generation of forms uninformed by  their performance in any sense. “Generative optimization depending on sun exposure” is my first article from the series (full text in polish here) – written thanks to Fred Labbe from Expedition Engineering. Expedition Engineering acted as structural engineers of a new metro station canopy in Napoli. Rogers, Stirk, Marbour + Partners are the architects of the project and Metropolitana Di Napoli is the client.

Expedition Engineering decided to use genetic algorithms (GA) to optimize the canopy structure, because it has to fulfill overlapping conditions. The structure has to both provide shading for the elevator platforms (red on the diagram underneath) and allow sunlight to reach the lower level of the train platforms (yellow).

GA operates across a pre-set number of individuals and through a pre-set number of populations. Each individual is represented by a genotype and a phenotype. First generation of individuals is a random one. Then the process of optimization begins to take place. First step of the process is decoding the genotypes into phenotypes. This means that the strings of 0s and 1s drive the generation of the actual canopy structures in Rhino. The “fitness value” is then assigned to each of the genotypes depending on how well does it’s phenotype fulfill the “fitness conditions”. The fitness condition in the case of this project is the amount of shading mentioned above. Each of the phenotypes is then tested against the position of the sun during few days in the year. Next step is the usage of the genotypes to create next generation. This step of the process is called “mating” – the “parents” of each of the next genotypes are selected randomly, but before this selection each of the genotypes is assigned a particular amount of space in the “wheel of lottery” depending on how fit it is. The most fit individuals become the parents of the next generation much more often than the less fit ones. The process is then repeated starting from the first step. The diagram below shows the phenomena of convergence – with time GA tends to produce phenotypes with similar fitness value.

The usage of artificial selection in GA for purpose of optimization is quite close to Darwin’s original model of evolution. There is however another kind of optimization taking place outside of the GA in the design process devised by Expedition Engineering. It is the structural optimization of each of the member. Initially each of the member in every individual was structurally optimized using finite elements methods. This process took place within the loop of members’ distribution optimization using GA. This approach was then changed due to time constraint. Finally, the thickness of each of the members was approximated depending on it’s position, based on the thickness of the optimized members in an evenly distributed grid. Final outcome of the GA optimization was then structurally optimized again, using proper finite elements methods.

Expedition Engineering was optimizing across the spectrum of 400 individuals in 70 generations. This means 28 000 canopies generated using Rhino script.
The process was conceived as a parametric tool. The designer is able to change the importance of every fitness condition. Presumably, assigning 80% of importance to shading and 20% to sunlight would result in quite a different outcome than assigning 50 / 50% or 20 / 80%. In reality the designer needs to wait considerable amount of time for the each of the outcomes, so it doesn’t feel like interacting with a parametric device. This however does not diminish the virtues of the approach which leads towards aesthetically satisfying spatial intervention, whose form is actually informed by it’s performance on the site. To some extend, this approach could be called a bottom-up approach to design.