Trends in Generative Design

Originally published on Archinect, March 31, 2012.

I have been researching optimization methods in architecture and I was delighted to learn about a few high-profile projects that have utilized some fantastic algorithms to assist in the realization of complex forms.

Multi-Objective Optimization, or Multi-Criteria Design Optimization (MCDA), software has been developed primarily for and by industrial engineers to design airplanes, automobiles, high-speed trains, and appliances. Process Integrated Design Optimization is another name for the field. The idea of optimization in architecture has been around since nearly the beginning of CAD in the 1950s and '60s, but had been limited by the power of computers and the huge number of variables, so it has not yet achieved a critical mass of proponents. The drive towards BIM technology has been justified in part because of the benefits that result from having a digital representation of the building to develop performance simulations.

Designers who work with MCDA craft search spaces, and guide searches with goal sets. In Galapagos — the in-house evolutionary solver for Grasshopper (the parametric modeling plugin for Rhino) — there are two inputs: genomes and a fitness function. Genomes represent variables that can be changed in order to manipulate a model. A fitness parameter is a "score" that the solver wants to target.

It is essentially a matter of measuring something, changing a variable in the experiment, measuring again, and guiding the changes in the model towards the desired score. Of course this process is automated and guided by optimization algorithms; genetic algorithms are frequently used. Richard Dawkins's seminal book The Selfish Gene outlines many of the principles still used in this process, including the mutations, crossovers, and population-size parameters that I will describe later.

Recently, projects at the Stanford Center for Facilities Engineering have advanced the interest in architectural application through integration with these new process-engineering software systems. Digital Project (CATIA) can be integrated directly to the Phoenix Integration Model Center interface; there does not yet exist a publicly available wrapper for Grasshopper/Rhino. Skidmore, Owings and Merrill used the Digital Project → ModelCenter workflow for fine-tuning the structural system of their Transbay Terminal Tower proposal.

Other projects that have used optimization in the recent past include OMA for their new Hong Kong transit project, PTW's Water Cube (the structure was rationalized using optimization), and many others consulted on by ARUP, who have developed a promising integration software, DesignLink.

A video by engineers at Ramboll shows how the möbius-strip form of the Astana Library was engineered to maintain the ideal form and minimize deflections, panel variations, cost, and material usage to make the project a reality. A similar but less sophisticated process was used in the Morphosis Phare Tower in Paris.

With more design-phase models representing a realistic picture of what the built construction will be, simulations are increasing in popularity as benchmarks for making design decisions. The legal issues concerning liability in the use of these figures are bound to become a hot topic as the tools become seamlessly incorporated into design software environments. Autodesk Revit incorporated an Analyze tab in Revit 2010, providing solar radiation and energy analysis through Green Building Studio. Vasari now has computational fluid dynamics (CFD) simulation, though it is based on a video game engine and is not as accurate as programs like CFdesign or Fluent, which can provide good analysis of passive ventilation strategies.

When I worked with Autodesk last summer there was a big push for optimization procedures, with three projects in the IDEA Studio (the university-tied research wing of Autodesk) directly supporting this study. The projects were Local Code by Nicholas de Monchaux; a project by David Benjamin of The Living — the structural optimization of the Living Light pavilion in Korea (Benjamin's studio at GSAPP has produced a great deal of experimentation with modeFRONTIER, a multi-objective optimization tool); and a third team of PhD researchers from USC adapting the Revit API for volumetric optimization.

Additionally, Zach Kron of the great blog buildz has developed a Goal Finder for the Vasari interface, enabled through the Python plug-in for Vasari. Vasari, if you don't know, is an Autodesk Labs product (free, for now!) that allows you to perform conceptual design studies and load them directly into Revit.

I have been experimenting with Grasshopper, Python, Rhino, and Phoenix Integration to create "search space" models that allow a designer to provide a framework for manipulation that is guided by goal sets. I have been working on goal sets — algorithms inspired by shape grammar and architectural manifestos to generate many design schemes, evaluating each and providing a catalog of design options for the architect and client to choose from. A quick project that I worked on a few weeks ago codified Corbusier's design aesthetic from Towards a New Architecture — The Problem of the House — into a goal set. I then created a grid-based model and generated a massive family of Villa Savoyes.

Will post much more as my thesis comes to a close (37 days).

The original post linked video demonstrations: the Astana Library optimization, the Living Light structural optimization, and my own Villa Savoye Generator.