{"slug": "a-data-fusion-framework-for-grounding-aerospace-surrogate-model-via-experimental", "title": "A Data Fusion Framework for Grounding Aerospace Surrogate Model via Experimental Wind-Tunnel Observations", "summary": "Researchers at an undisclosed institution presented a data fusion framework that adapts a CFD-trained deep learning surrogate using wind-tunnel PSP measurements to correct systematic discrepancies between CFD and experimental data. The Geotransolver surrogate, trained on 2,300 high-fidelity CFD simulations of the NASA CRM wing-body configuration, reproduced CFD forces with R2 > 0.99 but failed to match experimental data; the correction network, trained on PSP measurements at Mach 0.70 and 0.85, improved agreement at Mach 0.85 and outperformed direct interpolation at all held-out angles of attack, reducing prediction errors without retraining the surrogate.", "body_md": "arXiv:2609.04267v1 Announce Type: new \nAbstract: Aerodynamic surrogate models trained on high-fidelity CFD data reproduce numerical predictions of both scalar outputs and entire fields accurately, yet their predictive fidelity is limited by systematic discrepancies between CFD and experimental observations. We present an experimentally grounded correction framework that adapts a CFD-trained deep learning surrogate using wind-tunnel PSP measurements. A Geotransolver surrogate trained on 2,300 high-fidelity CFD simulations of the NASA CRM wing-body configuration, spanning geometric variation, Mach 0.70-0.85, and angles of attack 0 to 4 degrees, reproduces the CFD integrated aerodynamic forces and pitching moment to R2 > 0.99 but does not match the experimental data. To incorporate experimental information without retraining the surrogate, a correction network is trained on spatially registered PSP measurements at two freestream Mach numbers (0.70 and 0.85) across the same angle-of-attack range, learning the discrepancy between the surrogate-predicted and experimentally measured surface-pressure distributions. At Mach 0.85 the correction substantially improves agreement with PSP, particularly at the wing suction peak, shock location, and subsequent pressure recovery, reducing both the magnitude of the prediction error and the fraction of wetted surface on which it exceeds 0.05 in Cp, and it does so from a limited experimental dataset without modifying the pretrained surrogate parameters. On held-out angles of attack the grounded surrogate agrees with measurement to within 2.3-2.7% of the measured Cp range, and outperforms direct interpolation between the measured conditions at every state tested. Experimental measurements can therefore ground a large-scale simulation-trained surrogate by learning systematic CFD-to-experiment discrepancies while preserving its generalization capability and computational efficiency.", "url": "https://wpnews.pro/news/a-data-fusion-framework-for-grounding-aerospace-surrogate-model-via-experimental", "canonical_source": "https://arxiv.org/abs/2609.04267", "published_at": "2026-09-07 04:00:00+00:00", "updated_at": "2026-09-07 04:26:51.977865+00:00", "lang": "en", "topics": ["machine-learning", "ai-research"], "entities": ["NASA CRM", "Geotransolver"], "alternates": {"html": "https://wpnews.pro/news/a-data-fusion-framework-for-grounding-aerospace-surrogate-model-via-experimental", "markdown": "https://wpnews.pro/news/a-data-fusion-framework-for-grounding-aerospace-surrogate-model-via-experimental.md", "text": "https://wpnews.pro/news/a-data-fusion-framework-for-grounding-aerospace-surrogate-model-via-experimental.txt", "jsonld": "https://wpnews.pro/news/a-data-fusion-framework-for-grounding-aerospace-surrogate-model-via-experimental.jsonld"}}