Job offer

Within the scope of the ERC Consolidator project AUTOMATIX (see details below), we are seeking a PhD candidate to develop machine learning approaches for constitutive modeling.

Context

With the advent of machine-learning (ML) techniques, numerous studies have explored replacing traditional constitutive models with black-box neural networks or other data-driven approaches. However, it has been shown that such black-box models may perform poorly outside their training domain if no physics-based constraints are imposed on the learning architecture. Current research therefore focuses on designing physics-informed or physics-constrained learning strategies for various classes of materials.

This PhD project will focus on dissipative material behaviors such as elastoplasticity, viscoelasticity, and related phenomena. Learning dissipative behaviors is particularly challenging due to inherent path dependence and the evolution of unobservable internal state variables. The objective of this PhD is to propose novel hybrid modeling architectures that combine classical phenomenological constitutive models with neural-based components.

Training data will initially rely on synthetic datasets generated from high-fidelity microstructural simulations at the scale of a Representative Volume Element (RVE). In a second stage, learning at the structural scale based on full-field experimental images will also be addressed within the project.

This full-time PhD position is fully funded for at least 3 years within the ERC project. The PhD candidate will be supervised by Jeremy Bleyer and will be a core member of the AUTOMATIX research team., Master Degree or equivalent, The PhD candidate should:

• have a strong background in solid mechanics;

• demonstrate a good understanding of dissipative behaviors such as plasticity or viscoelasticity;

• have experience in programming and scientific computing (Python, Julia, C++, or similar);

• be able to work collaboratively in a research team and communicate scientific results clearly.

Previous experience with finite element software such as FEniCSx and/or machine-learning frameworks (JAX, PyTorch, etc.) is a plus but not required.

Languages ENGLISH

The AUTOMATIX project aims to improve the modeling of material behavior in solid mechanics. Accurately capturing complex phenomena (such as plasticity, damage, or environmental effects) remains a major challenge in many applications. AUTOMATIX leverages advances in machine learning to automatically build models from experimental data while directly embedding physical and mathematical knowledge within the learning architecture. This hybrid approach produces more reliable models, consistent with mechanical laws and less dependent on large datasets. A key outcome will be an open-source, modular, and high-performance library accessible to both academia and industry. AUTOMATIX will be applied in particular to the modeling of 3D-printed concrete at the Navier laboratory, to better predict complex phenomena such as material curing and crack formation.