Multi-fidelity load reconstruction in drone composite wings using in-service strain measurements
File(s)
Published version
Author(s)
Giannakeas, Ilias N
Pan, Yuhang
Naderi, Sadjad
Khodaei, Zahra Sharif
Aliabadi, Ferri MH
Type
Journal Article
Abstract
Accurate reconstruction of aerodynamic loads from sparse strain measurements is a longstanding challenge in structural health monitoring (SHM), particularly for lightweight composite wings in unmanned aerial vehicles (UAVs) where space and weight constraints limit instrumentation. This work presents a data-efficient pipeline for UAV wing load reconstruction under sparse sensing, explicitly designed to operate with limited experimental data. The proposed framework uses a two-stage approach consisting of multi-objective optimization for sensor placement and a hybrid load reconstruction strategy that combines physics-based (low fidelity) with data-driven (high-fidelity) observations. In the first stage, the optimal strain gauge locations are identified numerically through a Pareto-based optimization that balances the number of sensors against reconstruction accuracy. Then, in the second stage, a multi-fidelity approach is developed for the reconstruction of the aerodynamic loads where a calibration matrix (CM), derived from finite element simulations, is combined with an artificial neural network (ANN) trained on limited experimental data to account for modelling discrepancies and improve prediction fidelity. The framework is validated both numerically and experimentally, using wind tunnel experiments on a composite UAV wing. The results show that the multi-fidelity framework improves load reconstruction relative to the CM-only baseline and is particularly effective when experimental data are limited. With only three wind-tunnel runs used for training, the multi-fidelity model reduced the mean MAE from 8.72 to 4.58 for lift and from 8.94 to 4.08 for drag. For lift, the corresponding mean absolute percentage error (MAPE) decreased from 49.23% to 18.44%. When ten training runs were used, the mean MAE was further reduced from 8.73 to 3.85 for lift and from 8.44 to 3.91 for drag, while the lift MAPE decreased from 47.51% to 12.92%. The proposed framework is also compared against a fully data-driven ANN trained on the same experimental datasets. Across the full data-availability study, the multi-fidelity approach reached accurate and stable predictions with fewer training runs than the ANN, which exhibited higher variability under data-scarce conditions. These results highlight the potential of the proposed framework as a promising basis for SHM and digital twin (DT) applications in aerospace structures.
Date Issued
2026-11-01
Date Acceptance
2026-07-07
Citation
Aerospace Science and Technology, 2026, 178 (Part D)
ISSN
1270-9638
Publisher
Elsevier
Journal / Book Title
Aerospace Science and Technology
Volume
178
Issue
Part D
Copyright Statement
© 2026 Published by Elsevier Masson SAS. This work is licensed under a Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/).
License URL
Publication Status
Published
Article Number
113171
Date Publish Online
2026-07-08