Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AB-UPT for Automotive and Aerospace Applications

Benedikt Alkin, Richard Kurle, Louis Serrano, Dennis Just, Johannes Brandstetter

TL;DR

AB-UPT addresses the need for accurate, fast neural surrogates for external CFD in automotive and aerospace design. It uses an anchored-branched transformer architecture to map CFD meshes to surface and volume fields with efficient cross-attention anchored at a subset of points, enabling full-field predictions from limited inputs. The study introduces SHIFT-SUV and SHIFT-Wing datasets and demonstrates that AB-UPT outperforms state-of-the-art baselines, achieving near-perfect predictions of drag and lift and enabling rapid inference even with isotropic CAD geometries. The results suggest AB-UPT is well-suited for industry-scale CFD surrogacy, offering fast training (about a day on a single GPU) and fast evaluation (as low as 0.6 seconds for CAD-based queries).

Abstract

The recently proposed Anchored-Branched Universal Physics Transformers (AB-UPT) shows strong capabilities to replicate automotive computational fluid dynamics simulations requiring orders of magnitudes less compute than traditional numerical solvers. In this technical report, we add two new datasets to the body of empirically evaluated use-cases of AB-UPT, combining high-quality data generation with state-of-the-art neural surrogates. Both datasets were generated with the Luminary Cloud platform containing automotives (SHIFT-SUV) and aircrafts (SHIFT-Wing). We start by detailing the data generation. Next, we show favorable performances of AB-UPT against previous state-of-the-art transformer-based baselines on both datasets, followed by extensive qualitative and quantitative evaluations of our best AB-UPT model. AB-UPT shows strong performances across the board. Notably, it obtains near perfect prediction of integrated aerodynamic forces within seconds from a simple isotopically tesselate geometry representation and is trainable within a day on a single GPU, paving the way for industry-scale applications.

AB-UPT for Automotive and Aerospace Applications

TL;DR

AB-UPT addresses the need for accurate, fast neural surrogates for external CFD in automotive and aerospace design. It uses an anchored-branched transformer architecture to map CFD meshes to surface and volume fields with efficient cross-attention anchored at a subset of points, enabling full-field predictions from limited inputs. The study introduces SHIFT-SUV and SHIFT-Wing datasets and demonstrates that AB-UPT outperforms state-of-the-art baselines, achieving near-perfect predictions of drag and lift and enabling rapid inference even with isotropic CAD geometries. The results suggest AB-UPT is well-suited for industry-scale CFD surrogacy, offering fast training (about a day on a single GPU) and fast evaluation (as low as 0.6 seconds for CAD-based queries).

Abstract

The recently proposed Anchored-Branched Universal Physics Transformers (AB-UPT) shows strong capabilities to replicate automotive computational fluid dynamics simulations requiring orders of magnitudes less compute than traditional numerical solvers. In this technical report, we add two new datasets to the body of empirically evaluated use-cases of AB-UPT, combining high-quality data generation with state-of-the-art neural surrogates. Both datasets were generated with the Luminary Cloud platform containing automotives (SHIFT-SUV) and aircrafts (SHIFT-Wing). We start by detailing the data generation. Next, we show favorable performances of AB-UPT against previous state-of-the-art transformer-based baselines on both datasets, followed by extensive qualitative and quantitative evaluations of our best AB-UPT model. AB-UPT shows strong performances across the board. Notably, it obtains near perfect prediction of integrated aerodynamic forces within seconds from a simple isotopically tesselate geometry representation and is trainable within a day on a single GPU, paving the way for industry-scale applications.
Paper Structure (28 sections, 4 equations, 19 figures, 11 tables)

This paper contains 28 sections, 4 equations, 19 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. AI surrogate modeling for external aerodynamics
  4. Anchored-Branched Universal Physics Transformer (AB-UPT)
  5. Aerodynamic forces
  6. Experiments
  7. Model configuration and baseline models
  8. Luminary SHIFT-SUV
  9. Dataset description
  10. Training setup
  11. Training data
  12. Comparison against state-of-the-art baselines
  13. Relative errors
  14. Aerodynamic forces
  15. Qualitative visualization
  16. ...and 13 more sections

Figures (19)

  • Figure 1: Estate (left) and Fastback (right) configurations of the reference AeroSUV platform.
  • Figure 2: The deformation cage used to create the geometry variants constructed around the Estate configuration.
  • Figure 3: Diagrams visualizing the deformation parameters described in Table \ref{['table:suv_morph_params']}. The parameters in the right image modify the vehicle symmetrically about the centerline (left and right sides equally).
  • Figure 4: Images depicting the mesh resolution refinement regions defined around the vehicle to capture the marge near-body and wake dynamics.
  • Figure 5: Instantaneous velocity field (center plane) and wall shear stress contours (vehicle surface) for a full-scale Estate variant.
  • ...and 14 more figures