Context :Deep learning methods based on neural networks are universal approximators and have been applied for many years to identify the parameters of systems represented by mathematical models. However, their performance depends on the data used for training. The objective of the internship is to develop a deep learning algorithm in which the convergence would be driven by constraints from the physics of the system.
These constraints can be expressed, for example, in the form of PDEs or a finite element model.[...]

Objective : It is to create a coupling between machine learning and a model based on the mechanical behavior of a structure (beam or plate) to identify properties of structures that are not always easy to access without destructive testing (elastic modulus, loss factor, etc.), in particular for composite structures. It will combine numerical developments in Python (learning and identification algorithms), finite element simulations and experimental characterizations.

Duration : last 5 to 6 months, starting in February 2025, and a PhD grant on this topic may be available from October 2025.