Machine learning-aided multiscale design of porous materials tailored to application-specific, hydro-mechanical performance requirements

Through continuous interaction between computational fluid dynamics, mechanics of solids, material engineering, and machine learning, with my host, I will develop a novel and computationally efficient method, implemented in open-source software, for the multi-scale design of engineered porous materials (EPMs) that meet user-specified hydro-mechanical functional requirements.

This computer-aided approach will accelerate the discovery of EPMs and shorten the time for technology development, and is aimed at EPM design for additive Manufacturing (i.e. 3D-printing).

The basic notion of the proposed approach is: (1) to employ a dimensionality reduction techniques to obtain a low-dimensional proxy for the high-dimensional problem of characterizing a porous micro-structure, (2) to develop physics-informed neural networks (PINNs) for scale-specific hydro-mechanical simulation of porous media at the micro (pore) scale, the meso (pore-network) scale, and the macro (Darcy) scale, (3) to employ a physics-based coupling mechanism for scale-specific PINNs, allowing them to form a chain of neural networks for hydro-mechanical structure-property-performance (S-P-P) linkage, and (4) to incorporate a topology optimization algorithm for the multi-scale design of porous media.

The focus is on fluid-saturated, poroelastic materials, with special emphasis on biomedical applications that require a defined porous structure, such as meniscus implants and bone scaffolds.

I will work on the project at the University of Luxembourg (host institute), in collaboration with the University of Strasbourg (secondment institute).