S(i)ma.rt : Leveraging Conditional GANs to Develop Real-Time Inverse-Simulation based Design Optimization Workflows

Building performance simulation is an approach to emulate specific aspects of real-world building performance through a computational model, built based on the fundamental principles of physics, mathematics and engineering. The primary objective of building performance simulation is to quantify the domains of building performance with respect to building design, construction, operation and control.[1] 

Architects and designers in the past relied on simulation engines such as DOE-2.2[3], EnergyPlus[45] and Radiance[33], which required a lot of detailed input information and time to compute and output the simulation results.[2] Also, given that the mentioned simulation engines were advanced, it required that the user have some specialist technical knowledge to use them, in order to output accurate simulation results.[4] 

Attempts have been made in recent times to make these high-end simulation engines more accessible, by bundling them as plugins into various standard 3D modelling design software like Rhinoceros[43] and Grasshopper[44]. This paves way for users to rely on simulation-based iterative workflows to make design decisions. Although this type of workflow has been successful and well received by both academia and the industry, improvements with respect to reducing the amount of time a simulation engine takes to compute the calculations and run the simulation (for every iteration) are highly desired.[2][39][40][41] Improvements with respect to making the tools more accessible and user-friendly (in order to promote sustainable design practices to a broader audience of users) are also highly sought after.[40][42] 

The research exploration hopes on leveraging machine learning algorithms to pave the way for a novel ‘top-to-bottom’ simulation-free design workflow, with a special focus on image grid-based simulation metrics in high performance building design (such as Solar Radiation and Daylight Autonomy), that primarily hopes to reduce the time spent on the design exploration and hopes to make building design analysis more accessible to a broader audience. If the results are compelling, this exploration would act as a proof of concept for the aforementioned metrics and could be expanded to other metrics in the domains of high-performance building design.