Version française
Multi-Physics-Informed Machine Learning for Non-Destructive Evaluation of Material Gradients
GERS - GeoEND - ABRAHAM Odile - odile.abraham@univ-eiffel.fr - +33 240845918
DEROBERT Xavier - xavier.derobert@univ-eiffel.fr - +33 240845911
PALMA-LOPES Sergio - sergio.palma-lopes@univ-eiffel.fr - +33 240845912
GERS - GeoEND - DEROBERT Xavier - xavier.derobert@univ-eiffel.fr - +33 240845911
Génie Civil ?
Nantes
Sciences de l’Ingénierie et des Systèmes (SIS)
ABRAHAM Odile - Université Gustave Eiffel - GERS - GeoEND
Contrat doctoral sur dotation des EPSCP

Thesis context

According to the European Environment Agency (EEA), buildings in the European Union account for approximately 42% of total energy consumption and 35% of greenhouse gas (GHG) emissions. As of 2020, nearly 75% of the EU building stock was considered energy inefficient. While the renovation of existing buildings could reduce the EU’s total energy consumption by 5–6% and decrease carbon dioxide emissions by around 5%, the current renovation rate remains below 1% per year.

In this context, the HORIZON-CL5-2024-D4-02-02 funded project RADIANCE aims to transform construction and renovation practices through the integration of advanced robotics, automation, and digitalization technologies. One of the project’s key expected outcomes is the development of innovative robotic solutions for façade renovation. Within this framework, Gustave Eiffel University is developing a multi-physics non-destructive evaluation (NDE) probe, designed to be integrated onto drone platforms engineered by project partners, with the objective of assessing the structural integrity of concrete façades.

Objectives

The objective of this PhD thesis is to develop a Machine Leaning interpretation methodology that exploits the complementary information provided by the three physical modalities embedded in the NDE probe (namely electrical resistivity, capacitive sensing, and ultrasonic testing [1]).
 The proposed methodology will enable a
quantitative reconstruction of property gradients in concrete, such as degree of saturation, elastic modulus, and porosity.

Main Tasks

The Doctoral Candidate will:

  • Perform numerical modelling of the three NDE techniques to evaluate the influence of relevant material property gradients on each NDE observable generating a sizable synthetic training datasets;

  • Design and carry out laboratory experiments to produce representative experimental training data;

  • Develop physics-informed machine learning algorithms, trained on both numerical simulations and experimental measurements (using a transfer learning approach), to recover quantitative material property gradients [2],[3],[4];

  • Demonstrate the utility of the algorithms in a field-scale experiment on a metric-scale concrete wall, using the multi-physics NDE probe integrated onto drone platforms.

Expected Results

  • A Python-based multi-physics Maching Learning framework for NDE characterization of property gradients, integrating multi-physics simulations and experimental measurements;

  • Experimental validation of the proposed methodology using the multi-physics NDE probe in both laboratory (handheld) and on-site (drone-mounted) configurations;

  • Open-access dissemination of the results, including peer-reviewed journal publications, publicly available source code (e.g. via GitHub), and curated experimental and numerical datasets.

References

[1] M. Fengal, P. Mora, P. Shokouhi, O. Durand, X. Dérobert, S. Palma-Lopes, M. Lehujeur, G. Villain, E. Gennesseaux, O. Abraham, Coherent and incoherent Rayleigh wave attenuation for discriminating microstructural effects of thermal damage from moisture conditions in concrete, NDT & E International, 156, 2025.  https://doi.org/10.1016/j.ndteint.2025.103473

[2] Y. Hu, X. Wei, X. Wu, J. Sun, J. Chen, Y. Huang, J. Chen, A deep learning-enhanced framework for multiphysics joint inversion, Geophysics, 88(1), K13-K26, 2023. https://doi.org/10.1190/geo2021-0589.1 

[3] Y. Ren, B. Liu, B. Liu, Z. Liu, P. Jiang, Joint Inversion of Seismic and Resistivity Data Powered by Deep Learning, IEEE Transactions on Geoscience and Remote Sensing, 62, 5929614, 2024. https://doi.org/10.1109/TGRS.2024.3458402 

[4] M. Skiadopoulos, D. Kifer, P. Shokouhi, A transfer learning approach to the prediction of porosity in additively manufactured metallic components, NDT & E International, 157, 2026. https://doi.org/10.1016/j.ndteint.2025.103531

Physics-informed Machine Learning, Ultrasonics, Capacitive measurements, Electrical resistivity, Non-destructive evaluation, Property gradients, Concrete
En cas de problème, contacter : Support thèses