Towards Sustainable Architecture: 3D Convolutional Neural Networks for Computational Fluid Dynamics Simulation and Reverse DesignWorkflow

TL;DR

This paper introduces a flexible 3D CNN-based model for real-time turbulent flow prediction and reverse design workflow, enabling architects to simulate wind flow and generate building designs efficiently.

Contribution

It presents a novel residual CNN model for near real-time CFD simulation and a reverse workflow for architectural design based on wind flow targets.

Findings

Achieves near real-time turbulent flow prediction.

Enables reverse design of building volumes from wind flow targets.

Provides a flexible tool for architectural wind flow analysis.

Abstract

We present a general and flexible approximation model for near real-time prediction of steady turbulent flow in a 3D domain based on residual Convolutional Neural Networks (CNNs). This approach can provide immediate feedback for real-time iterations at the early stage of architectural design. This work-flow is then reversed and offers a designer a tool that generates building volumes based on target wind flow.