Tomáš Karásek

Tomáš Karásek

Tomáš Karásek is currently the Head of the National Competence Centre for High Performance Computing in the Czech Republic and the head of the Parallel Algorithm Research Lab at IT4Innovations National Supercomputing Centre. Previously, he spent seven years in several positions at the Institute of High-Performance Computing in Singapore. In the past, he participated in several EC-funded projects focusing on SMEs, such as SESAME Net, InnoHPC, and the PRACE SHAPE program. He has received funding for more than 30 contractual research projects with industrial partners over the past 15 years, and he's the author or co-author of more than 20 scientific papers.

CFD-Based Digital Twin of a PEM Fuel Cell with AI-Driven Surrogate Modelling

Presentation abstract: This talk presents the development of a digital twin of a polymer electrolyte membrane (PEM) fuel cell aimed at efficient performance analysis and operational optimisation. The proposed framework integrates a high-fidelity CFD model with a data-driven surrogate model based on artificial neural networks. The CFD model captures the coupled electrochemical, fluid flow, and transport phenomena within the fuel cell and is used to generate training data, including polarisation curves, across a range of operating conditions.

A surrogate model is trained using normalised CFD data, with careful selection of network architecture, activation functions, regularisation, and optimisation strategy to ensure robustness and accuracy. The trained surrogate enables rapid inference and gradient-based optimisation of selected operating parameters. Promising configurations identified by the surrogate model are subsequently validated using full CFD simulations to assess accuracy and physical consistency.

The results demonstrate that the surrogate-assisted optimisation successfully identifies operational configurations with prediction errors below 3% compared to the reference CFD solutions. The proposed digital twin significantly reduces computational cost while maintaining high accuracy, making it suitable for design studies, sensitivity analyses, and real-time optimisation of PEM fuel cells. This approach provides a scalable foundation for advanced control strategies and future integration into hydrogen energy systems.