Benchmarking Convolutional Neural Network and Graph Neural Network based Surrogate Models on a Real-World Car External Aerodynamics Dataset

Sam Jacob Jacob, Markus Mrosek, Carsten Othmer, Harald Köstler · 2025

Read Benchmarking Convolutional Neural Network and Graph Neural Network based Surrogate Models on a Real-World Car External Aerodynamics Dataset by Sam Jacob Jacob, Markus Mrosek, Carsten Othmer, Harald Köstler (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-06699.txt.

Description

Sam Jacob Jacob, Markus Mrosek, Carsten Othmer, and Harald Köstler compare two fast stand-in models for car aerodynamics in this 2025 paper. Full airflow simulations are slow, which hampers joint work between engineers and stylists. The authors test a convolutional network that reads a signed distance field against a graph network tool that reads the car surface mesh directly.

Their test set is drawn from real car work: 343 shapes built from 32 base cars across five projects. The convolutional model missed the true drag by 2.3 counts on average, against 3.8 for the graph model. Both called the direction of a drag change right about 77 percent of the time. Each method gives an answer in under two minutes, over 600 times faster than a full simulation, though fine detail within one car family is still hard to capture.

Catalog record

CREATORSam Jacob Jacob, Markus Mrosek, Carsten Othmer, Harald Köstler
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-06699

Browse the full library · About the Tower of Babel