Data parallel shape optimization : application to the generation of energy efficient building shapes / by Elias Georges Gemayel.

Abstract

Energy consumption in buildings has always been an important concern. Constructing energy-efficient buildings is becoming a trend in architecture and engineering, either to contribute to the protection of the environment by reducing energy consumption and CO2 emissions, or for sheer nancial purposes. Previous works demonstrate that applying sustainable strategies during the early phases of the building's design highly contributes to reducing the energy cost of the building. In particular, the design of the building's shape during the conceptual phase is an important factor for reducing this cost. In this thesis, we build an engine that optimizes a building's shape to minimize the solar insolation on its envelope, greatly reducing the building's cooling energy cost, and delivering more pro table real-estate investment projects. The problem is formulated as an optimization problem and solved using state-of-the-art algorithms. Moreover, we provide various parameterizations for the building's shape along with their respective variants of the objective function. Furthermore, to ensure real-time output of the results while optimizing the building's shape, methods exploiting data parallelism on CUDA are proposed and implemented. Finally, for realistic rendering and e cient shading computation, we use the OptiX engine for general-purpose ray tracing on GPUs. We demonstrate our system on prismatic and free-form building shapes. Our preliminary results show reductions in insolation of up to 35percent relative to standard designs.