A Data Fusion Framework for Grounding Aerospace Surrogate Model via Experimental Wind-Tunnel Observations

SummaryAI generated

The researchers proposed an experimental-based wind tunnel correction framework, using wind tunnel PSP measurement data to calibrate the aerodynamic proxy model trained with high-precision CFD. This Geotransolver proxy model was trained on NASA CRM wing configurations with 2,300 sets of CFD simulation data (Mach numbers 0.70–0.85, angle of attack from 0 to 4 degrees). Although it could reproduce the CFD integrated aerodynamic forces and pitching moments (R2 > 0.99), it still showed systematic deviations from experimental data. By training a correction network to learn the differences between the proxy predictions and experimental surface pressure distributions, calibration was achieved without re-training the main proxy model. At Mach number 0.85, this correction significantly improved the agreement of the peak suction forces on the wings, the position of shock waves, and the pressure recovery region, reducing the wet surface area ratio by more than 0.05 Cp. At the remaining angle of attack, the prediction error of the corrected proxy model was within 2.3%–2.7% of the measured Cp range, which is better than direct interpolation methods.

Related eventsRELATED EVENTS

2026-09-09 05:29OpenAI claims huge maths breakthrough on a famed ‘Millennium Problem’
2026-09-09 06:11OpenAI claims to have solved the Navier-Stokes equation problem, but its originality is questioned by the academic community
2026-09-08 22:00Google DeepMind Maps 9 Billion Possible DNA Variants

Key entitiesKEY ENTITIES

NASA CRM

SignalsSIGNALS

Keyword heat

NASA CRM1

All reports (1)SOURCES

↗

A arXiv cs.LG en 2026-09-07 12:00

The researchers proposed an experimental-based wind tunnel correction framework, using wind tunnel PSP measurement data to calibrate the aerodynamic surrogate model trained with high-precision CFD. This Geotransolver surrogate model was trained on NASA CRM wing configurations with 2,300 sets of CFD simulation data (Mach numbers 0.70-0.85, angle of attack from 0 to 4 degrees). Although it could reproduce CFD integrated aerodynamic forces and pitching moments (R2 > 0.99), it still showed systematic deviations from experimental data. By training a correction network to learn the differences between surrogate predictions and experimental surface pressure distributions, calibration was achieved without re-training the main surrogate model. At Mach number 0.85, this correction significantly improved the agreement of wing suction peak values, shock wave positions, and pressure recovery regions, reducing the wet surface area ratio by more than 0.05 Cp; at retained angles of attack, the prediction error of the corrected surrogate model was within 2.3%-2.7% of the measured Cp range, which is better than direct interpolation methods. This framework…