Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ONERA's CRM WBPN database for machine learning activities, related regression challenge and first results

Jacques Peter, Quentin Bennehard, Sébastien Heib, Jean-Luc Hantrais-Gervois, Frédéric Moëns

TL;DR

This work introduces ONERA's CRM WBPN CFD database to enable and benchmark machine-learning approaches for predicting wall-field distributions Cp and Cf under unseen flight conditions. By compiling 468 RANS simulations on the CRM WBPN geometry and providing both pointwise and structured-wall data, the authors define a Codabench regression challenge and evaluate a range of pointwise and global regressors, including MLP, λ-DNN, decision trees, kNN, POD+RBF, and IsoMap+RBF. The results show that modewise MLP yields the strongest global performance, while several pointwise methods achieve high $R^2$ yet vary in worst-case errors, underscoring the task's intermediate difficulty and the value of the dataset for future development. The dataset and benchmark are designed to foster community participation and accelerate ML-assisted aero-field prediction across the flight envelope, with potential benefits for aeroelastic and aero-structure computations.

Abstract

This paper presents a new Computational Fluid Dynamics database, developed at ONERA, to support the advancement of machine learning techniques for aerodynamic field prediction. It contains 468 Reynolds-Averaged Navier-Stokes simulations using the Spalart-Allmaras turbulence model, performed on the NASA/Boeing Common Research Model wing-body-pylon-nacelle configuration. The database spans a wide range of flow conditions, varying Mach number (including transonic regimes), angle of attack (capturing flow separation), and Reynolds number (based on three stagnation pressures, with one setting matching wind tunnel experiments). The quality of the database is assessed, through checking the convergence level of each computation. Based on these data, a regression challenge is defined. It consists in predicting the wall distributions of pressure and friction coefficients for unseen aerodynamic conditions. The 468 simulations are split into training and testing sets, with the training data made available publicly on the Codabench platform. The paper further evaluates several classical machine learning regressors on this task. Tested pointwise methods include Multi-Layer Perceptrons, $λ$-DNNs, and Decision Trees, while global methods include Multi-Layer Perceptron, k-Nearest Neighbors, Proper Orthogonal Decomposition and IsoMap. Initial performance results, using $R^2$ scores and worst relative mean absolute error metrics, are presented, offering insights into the capabilities of these techniques for the challenge and references for future work.

ONERA's CRM WBPN database for machine learning activities, related regression challenge and first results

TL;DR

This work introduces ONERA's CRM WBPN CFD database to enable and benchmark machine-learning approaches for predicting wall-field distributions Cp and Cf under unseen flight conditions. By compiling 468 RANS simulations on the CRM WBPN geometry and providing both pointwise and structured-wall data, the authors define a Codabench regression challenge and evaluate a range of pointwise and global regressors, including MLP, λ-DNN, decision trees, kNN, POD+RBF, and IsoMap+RBF. The results show that modewise MLP yields the strongest global performance, while several pointwise methods achieve high yet vary in worst-case errors, underscoring the task's intermediate difficulty and the value of the dataset for future development. The dataset and benchmark are designed to foster community participation and accelerate ML-assisted aero-field prediction across the flight envelope, with potential benefits for aeroelastic and aero-structure computations.

Abstract

This paper presents a new Computational Fluid Dynamics database, developed at ONERA, to support the advancement of machine learning techniques for aerodynamic field prediction. It contains 468 Reynolds-Averaged Navier-Stokes simulations using the Spalart-Allmaras turbulence model, performed on the NASA/Boeing Common Research Model wing-body-pylon-nacelle configuration. The database spans a wide range of flow conditions, varying Mach number (including transonic regimes), angle of attack (capturing flow separation), and Reynolds number (based on three stagnation pressures, with one setting matching wind tunnel experiments). The quality of the database is assessed, through checking the convergence level of each computation. Based on these data, a regression challenge is defined. It consists in predicting the wall distributions of pressure and friction coefficients for unseen aerodynamic conditions. The 468 simulations are split into training and testing sets, with the training data made available publicly on the Codabench platform. The paper further evaluates several classical machine learning regressors on this task. Tested pointwise methods include Multi-Layer Perceptrons, -DNNs, and Decision Trees, while global methods include Multi-Layer Perceptron, k-Nearest Neighbors, Proper Orthogonal Decomposition and IsoMap. Initial performance results, using scores and worst relative mean absolute error metrics, are presented, offering insights into the capabilities of these techniques for the challenge and references for future work.
Paper Structure (19 sections, 11 equations, 9 figures, 2 tables)

This paper contains 19 sections, 11 equations, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Wing-body-pylone-nacelle CRM data base
  3. Geometry
  4. Selected flow conditions
  5. CFD simulations
  6. Numerical improvement and verification of the database
  7. Physical verification of the database
  8. Scattered and structured data
  9. Regression challenge
  10. Performance of classical pointwise regressors
  11. Pointwise MultiLayer Perceptron
  12. $\lambda$-DNN neural network
  13. Decision Tree
  14. Performance of classical global regressors
  15. Modewise MultiLayer Perceptron
  16. ...and 4 more sections

Figures (9)

  • Figure 1: Numerical Design Of Experiment
  • Figure 2: CRM WBPN. Left: wall mesh. Right: imposed laminar zones
  • Figure 3: Left: Numerical verification, standard deviation of the drag over the last 20% iterations. Right: For drag and lift, relative importance of change in flow conditions and standard deviation
  • Figure 4: Example of polar plots from the database (at various Mach and Reynolds numbers)
  • Figure 5: isolines of pressure and friction lines for $p_i = 4 \times 10^5$ (all) and $M_\infty = 0.85$, AoA = 1.5$^o$ (left), AoA = 5$^o$ (center), AoA = -5$^o$ (right).
  • ...and 4 more figures