Photogrammetry-based computational fluid dynamics

Abstract

Computational fluid dynamics (CFD) is the cornerstone of the design and analysis process in many engineering applications. Not only has it been applied in the design phase, but it has also been employed for analyzing the fluid flow phenomena during the operation phase for many in-use structures, such as vehicles, buildings, and landscapes. However, creating a 3D mesh-based model of in-use structures that can be used by conventional boundary-fitted CFD methods is labor-intensive, time-consuming, and sometimes impossible. Due to the challenges introduced by geometry complexity and lack of design information, it is often difficult to perform an accurate and efficient CFD analysis of these objects. This paper aims to overcome such challenges by proposing a novel photogrammetry-based CFD framework for simulating in-use structures whose design models and analysis meshes are hard to obtain. The proposed framework integrates machine learning-based 3D point cloud reconstruction of structures from 2D images obtained from portable devices (e.g., cell phones and drones) and an immersogeometric approach that can carry out flow analysis directly on reconstructed point clouds. We first present the technical details of point cloud reconstruction techniques and the immersogeometric analysis method. We then simulate the flow past a standard 12 oz soda can reconstructed using photogrammetry and compare the results with reference solutions to assess the accuracy of the approach. Finally, the proposed photogrammetry-based CFD is applied to simulations of a bell tower and the Kavita and Lalit Bahl Smart Bridge on the University of Illinois Urbana-Champaign (UIUC) campus to demonstrate the robustness of the framework and its applicability to real-world in-use civil structures.