Research funded by the Croatian Science Foundation – project: IP-2026-4668
Principal Investigator: Prof. Marko Čanađija, Ph.D.
Research Group: Assoc. Prof. Sanjin Kršćanski, Ph.D., Assoc. Prof. Ante Skoblar, Ph.D., Postdoctoral researcher Martin Zlatić, Ph.D.
Abstract:
The proposed research aims to extend the practical size limits of conventional finite element analysis (FEA) solvers for coupled multiphysics problems. Particular emphasis will be placed on thermo-mechanical systems, with additional applications to electro-mechanical and electro-magneto-mechanical phenomena. The initial phase of the project will focus on linear problems, followed by an extension to material nonlinearities in the second phase. To achieve these objectives, the research will build on recently proposed hierarchical deep-learning neural networks (HiDeNN/C-HiDENN), a computational framework that has demonstrated exceptional efficiency in single-physics simulations. HiDeNN reformulates classical FEA within a machine-learning framework that relies on massively parallel graphics processing units (GPUs) and is implemented in the JAX and JAX-FEM environments. All developed codes will be built using open-source Python libraries. The expected outcome is a unified C-HiDeNN-based computational framework capable of solving large-scale linear and nonlinear multiphysics problems beyond the limits of current FEA approaches. The framework will be validated against established benchmarks and analytical solutions, and new large-scale benchmarks relevant to real engineering applications will be developed. The resulting methodology will enable more detailed and realistic simulations than those currently achievable with standard FEA tools. Consequently, the proposed framework will support improved design of advanced materials, energy systems, and engineering structures operating in multiphysics environments. In addition, the results will benefit developers of FEA software and engineering teams working on complex multiphysics systems, while strengthening collaboration between academia and industry in computational mechanics.