EUVIMEDEUVIMED entdecken
Crossref

Full-Field Strain and Stress Reconstruction Using a Hybrid ML–iFEM Approach

Ilias N. giannakeas, Yuhang Pan, Maria Loizidou, Zahra Sharif khodei, Ferri M.H. Aliabadi

e-Journal of Nondestructive Testing · 2026-08 · 31 (8)

DOI: 10.58286/33831 ↗

Kurz gefasst

Fachartikel in der Zeitschrift e-Journal of Nondestructive Testing, 2026-08. Verfasst von Ilias N. giannakeas und 4 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate full-field reconstruction of structural stress and strain remains a central challenge in structural health monitoring, particularly when only sparse sensing data are available during operation. While the inverse finite element method (iFEM) is widely recognised for its robustness in displacement and stress estimation, its practical implementation is limited by the requirement for dense and well-distributed in-situ strain measurements. This can make the method infeasible in complex and large-scale structures where sensor layouts are limited by cost, accessibility, or weight constraints. Recent advances in operator-learning methods offer promising avenues to overcome this limitation by inferring missing field information directly from sparse observations. Approaches such as Deep Operator Networks (DeepONets) can learn continuous strain operators and enrich discrete strain data, enabling full-field reconstruction from a reduced sensing configuration. This work introduces a two-step reconstruction framework that combines an operator-learning–based strain augmentation module with iFEM to recover the full stress and strain fields from limited strain measurements. First, the measured strains are augmented using DeepONet to estimate the strain values at the iFEM elements that do not have direct strain measurements. These augmented strains are then used within iFEM to compute displacements and stresses, ensuring the reconstruction remains grounded in physical constraints. The framework is adaptable to diverse sensor layouts and explicitly incorporates model parameters to account for uncertainties and potential sensor loss. The reconstructed fields enable the subsequent extraction of stresses at critical locations, supporting traditional integrity assessments and remaining useful life analyses. The methodology is demonstrated on a numerical cantilever composite plate benchmark. A dataset of 400 laminate configurations is generated by varying the number of plies and ply orientations while keeping the geometry, loading, sensor layout, ply thickness, and material properties fixed. The generated layups include symmetric, unsymmetric, balanced, and unbalanced configurations, enabling the model to learn across a mechanically diverse laminate family. Results show that conditioning the DeepONet on the laminate stiffness matrices enables strain reconstruction across unseen layups, providing a compact and physically meaningful route for incorporating laminate variability into data-driven SHM workflows. The study demonstrates the feasibility of combining sparse sensing, laminate mechanics, and operator learning as a preliminary step towards more generalizable iFEM-based reconstruction frameworks for composite structures.

Zitieren

APA

giannakeas, I. N., Pan, Y., Loizidou, M., khodei, Z. S., & Aliabadi, F. M. (2026). Full-Field Strain and Stress Reconstruction Using a Hybrid ML–iFEM Approach. e-Journal of Nondestructive Testing, 31(8). https://doi.org/10.58286/33831

Vancouver

giannakeas IN, Pan Y, Loizidou M, khodei ZS, Aliabadi FM. Full-Field Strain and Stress Reconstruction Using a Hybrid ML–iFEM Approach. e-Journal of Nondestructive Testing. 2026;31(8). doi:10.58286/33831

BibTeX
@article{giannakeas2026,
  author = {giannakeas, Ilias N. and Pan, Yuhang and Loizidou, Maria and khodei, Zahra Sharif and Aliabadi, Ferri M.H.},
  title = {Full-Field Strain and Stress Reconstruction Using a Hybrid ML–iFEM Approach},
  journal = {e-Journal of Nondestructive Testing},
  year = {2026},
  volume = {31},
  number = {8},
  doi = {10.58286/33831}
}
Kostenlos starten →