Jumeaux numériques d'ordre réduit intégrant des modèles physiques pour la prévision en temps réel du risque de lésions des tissus mous chez les utilisateurs d'emboîtures prothétiques // Physics-Enhanced reduced-order digital twins for real-time prediction

.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

The core objective of this PhD is to transition from high-fidelity, computationally intensive multiscale finite element models to low-fidelity, real-time digital twins. By integrating physics-based equations with machine learning, this research aims to create scalable models that maintain high predictive accuracy for clinical deployment. This task builds on strong prior work, including real-time strain localization using dictionary-based ROM-nets (Rohan et al., 2023) developed by an expert in reduced-order modeling and expert in biomechanical modeling.
The main contribution of this PhD is the reduced order modeling of patient specific multiscale models.

A Hybrid computational Workflow will be developed to merge physics-based equations and machine learning for real-time predictions. The dictionary involved in a ROM-net is composed of local PCA (Principal component analyses) related to groups of similar patients, where the similarity criteria aim to facilitate dimensionality reduction of biomechanical data. The reduced digital twin will be implemented onto a demonstrative platform:
A functional prototype of the digital twin showcasing clinical utility. A real-Time evaluation against Embedded Sensor Data will be performed on a functional prototype: The digital twin will be developed based on the scientific data collected in TWIN-IT and evaluated against partial data collected from embedded sensors that will be collected by the functional prototype. Goal-oriented error estimators and AI techniques will be employed
to enhance computational efficiency and manage prediction uncertainties.
The dimensionality reduction of mechanical data will be supplemented by a dimensionality reduction of the geometrical data related to the patient morphology (Ferhat et al., 2026). The latter reduction will be performed using a spectral shape analysis (ShapeDNA) of specific patient meshes.
Upon successful validation, the research will extend to the design of prosthetic sockets, specifically testing the feasibility of real-time monitoring for transfemoral amputees. Clinically, this work aims to identify and monitor mechanical factors that contribute to soft tissue lesions and pain, thereby supporting the development of preventive strategies and personalized patient care.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Début de la thèse : 01/10/2026
WEB : https://www.cemef.minesparis.psl.eu/wp-content/uploads/2026/04/PhD-PEPR-TWIN-IT.pdf

Other public funding

ANR Financement d'Agences de financement de la recherche

Mines Paris-PSL

364 SFA - Sciences Fondamentales et Appliquées

.
We are looking for a highly motivated researcher with a strong background in computational mechanics. The ideal candidate should possess the following qualifications: • Core competencies: Strong skills in finite element simulation and a background in statistics or machine learning, particularly regarding biomechanical solid models. • Preferred experience: Prior experience in finite element modeling and the analysis of multiscale data will be highly valued. • Soft Skills & Mindset: The candidate must demonstrate independence, scientific rigor, and a genuine curiosity regarding the development of thermomechanical digital twins..