UNIKPiML - Research project

In many physical systems, the governing partial differentiation equations (PDEs) are known with high confidence, but simulating a numerical solution can be prohibitively expensive. In other contexts, PDEs are unknown (or partly known) and unveiling them from experimental data is the central goal since they could shed some lights on the underlying physical process. Recently, physics-guided machine learning models have shown to be a promising tool in both abovementioned scenarios. They rely on neural networks (NNs) to simulate the physical quantities of interest at various temporal and spatial positions. Training such NNs entails to incorporate physical constraints, usually in the form of a PDE and boundary conditions, and/or to be able to generate plausible simulated data [Raissi et al., J. Comp. Physics, 2019]. The originality of this proposal is to embrace the hard setting encountered in surface engineering, that is limited prior physical knowledge and few experimental data [Brandao et al., Entropy 2022]. To overcome both limitations, the thesis will develop a unified end-to-end framework from the physics modeling to the algorithms used for training physics-guided models:
 

The advances made will be applied for the modeling of laser/radiationmatter interaction issues which can be impactful in many societal challenges such as in the fields of space, energy, health, ...