Development of a Surrogate Model for Reducing Experimental Data in Wing Surface Pressure Mapping

• Grant number: 25/06575-2
• Support Opportunities: Scholarships in Brazil - Master
• Start date: December 01, 2025
• End date: June 30, 2027
• Field of knowledge: Engineering - Aerospace Engineering - Aerodynamics
• Principal Investigator: João Paulo Eguea
• Grantee: Caio Augusto Mascarenhas Dias Filho
• Host Institution: Escola de Engenharia de São Carlos (EESC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

Aircraft design is a complex and costly process in terms of financial and human resources. Throughout the stages of aircraft development, the acquisition of aerodynamic data is fundamental to assess performance, with pressure distribution over lifting surfaces being an essential parameter for aerodynamic characterization. Wind tunnel tests represent the main source of high-fidelity data for mapping this distribution, but they involve high costs and limitations regarding the number of measurements. Computational fluid dynamics (CFD) simulations, on the other hand, allow the generation of large volumes of data, although with varying degrees of fidelity. In this context, the application of machine learning techniques emerges as a promising alternative to integrate these different sources of information, enabling the identification of complex patterns and the extraction of knowledge from a reduced number of measurements or simulations, without significant losses in accuracy. This project proposes the development of a metamodel based on Physics-Informed Neural Networks (PINNs), trained with multi-fidelity data from experimental tests and numerical simulations, with the objective of predicting surface pressure distribution on wings from a reduced set of experimental measurements. The model will be progressively developed, assessing the reduction in the number of experimental measurements and observations under different flow conditions, with its predictive and generalization capabilities verified using established statistical metrics such as RMSE, MAE, and R². The loss function of the metamodel will include a physics-informed term associated with the error between the predicted and experimentally measured lift coefficient. The expected outcome is that the developed metamodel will be able to accurately predict pressure distribution on wings, significantly reducing the need for high-fidelity experimental measurements while maintaining generalization capability for different wing geometries, thereby contributing to cost and time reduction in the aircraft development process.