Early-stage shading design evaluation for buildings: A workflow accessing energy performance on AI-generated 3D geometry

¹ Architecture, The Design School, Arizona State University, Tempe, AZ, United States
² School of Sustainable Engineering and the Built Environment, Arizona State University, Tempe, AZ, United States

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Abstract. This study presents a simplified and controllable workflow for the early-stage evaluation of self-shading building façade designs. Early design decisions strongly influence long-term building energy performance; however, façade shading strategies are often defined through qualitative guidelines rather than quantitative assessment, largely due to the time-intensive nature of energy modeling and the limited simulation expertise among architects. The proposed workflow addresses this limitation by enabling a rapid transition from conceptual design ideas to quantitative feedback on building operational performance, while preserving architectural intent. Using text-based design inputs derived from initial abstract ideas, AI-based tools are employed to generate biomimetic façade shading geometries inspired by the saguaro cactus, tailored to a hot-arid climate context in Tucson, Arizona. The AI-generated façade geometries are then integrated into a standardized DOE Medium Office reference building using a Grasshopper-Honeybee-EnergyPlus workflow. This workflow enables consistent comparison across multiple shading design alternatives, supporting rapid decision-making in early-stage design optimization. Three self-shading scenarios with shading depths of 1, 2, and 3 meters are evaluated against an unshaded baseline. The results indicate that the extension of shading depth reduces annual cooling demand but may increase annual heating demand, revealing diminishing returns of overall building performance beyond certain shading depths. The early-stage quantitative performance evaluation feedback enabled in this study informed early-stage design decision-making without significantly increasing modeling complexity. While Large Language Models (LLMs) have been used to support model integration and result interpretation in this work, future efforts will focus on further simplifying modeling complexity empowered by LLMs.