September 9, 2014

Generating Art from Simulation

While working on a series of daylighting studies for an Atelier Jones designed chapel, the role of simulation in the creation of art came under scrutiny. The studies, performed by Michael Gilbride and graduate students Nicole Peterson and Justin Schwartzhoff, focused primarily on the size, orientation, and transparency of the skylights and sidelights bordering a cross laminated timber (CLT) sculptural piece on the north wall of the chapel. The CLT wall’s design was intended to mimic a fabric curtain and was therefore open to a variable shape. This suggested that, instead of altering the window conditions, dramatic lighting conditions could be incorporated into the piece itself. To test this, the lab pursued a mixture of daylighting simulations, parametric scripting, and genetic algorithms in an effort to generate a more perfect form for the wall.
The project was an attempt to explore what potential the Rhinoceros plugin DIVA offered over traditional Ecotect studies. Both Ecotect and DIVA make use of the same Radiance lighting analysis system, but where they differ is their ability to connect with other systems. Typically, Ecotect models must be created within the program, or imported and adjusted within Ecotect before running simulations. DIVA exists within the modeling too Rhinoceros, allowing for a quick workflow from modeling to testing, as models can be tested immediately without export. Furthermore, the model geometry in Rhino can be adjusted through parametric scripting software such as Grasshopper, allowing DIVA to run a series of tests over a wide range of options without constantly requiring updates from the user.

To test the CLT wall, a grasshopper definition was developed to define the geometry of each CLT panel. The total number of panels remains consistent with the original design intent, but every vertex of the panel’s points is adjustable. The panels are designed to remain planer and have a maximum width of 8′, ensuring their constructability with CLT. In total, the system contains 27 variables, each containing at least 9 unique positions. Even on simplified Radiance settings, testing all of these options would take a prohibitively long time. Thankfully, Grasshopper includes a system to help work towards optimized solutions.

Galapagos allows for an evolutionary solution of the problem, given a specific “fitness parameter.” The system runs simulations on a set of random starting points, then checks each point, or genome, against the fitness parameter. After each point is tested, the top genomes are combined to form a new generation of genomes, where the process repeats. Overtime, the random variations begin to die out, and a “most fit” geometry emerges. The process takes time, with each solution in this test requiring about 20 generations to stabilize. The corresponds to an average of around 48 hours of constant simulation, so the process is not entirely efficient. Worse yet, given the low power draw of the IDL as a whole, the process makes up to 40% of the total energy use during the test period. Still, the results are intriguing, producing a design that meets the fitness criteria, but may not be what is originally assumed. The Grasshopper-Galapagos-DIVA combination is simply one more tool that can be used to explore design.