Work Packages

The REMODEL programme is structured around a set of interconnected work packages, each addressing a critical component of next-generation mesh generation and geometry handling for high-performance computing. Together, these work packages tackle one of the key bottlenecks in modern simulation: the ability to efficiently generate, adapt, and utilise high-quality meshes for complex geometries at scale. By combining advances in geometry, meshing, adaptivity, machine learning, and multi-physics integration, the programme aims to unlock the full potential of Exascale computing for industrial and scientific applications.

WP1: Geometry Handling

Lead: Queen’s University Belfast — Prof. Trevor Robinson
Supporting: University of Edinburgh, Swansea University

WP1 addresses a critical limitation in modern simulation workflows: the difficulty of scaling complex CAD-based geometries across high-performance computing systems. Current approaches are typically designed for single-processor environments and struggle with the size, complexity, and distribution required for Exascale computing.

The objective is to develop scalable geometry handling strategies that enable efficient distribution of large models across heterogeneous CPU–GPU architectures. This includes evaluating alternative geometric representations such as NURBS, implicit, and faceted models, alongside strategies for geometry partitioning and data locality.

Key developments focus on geometry complexity reduction and physics-driven de-featuring, as well as the creation of a plug-and-play geometry handling library integrated with simulation workflows. These advances will enable efficient, high-fidelity simulations of complex engineering systems at scale.

WP2: Hybrid Meshing Strategies

Lead: King’s College London — Prof. David Moxey
Supporting: University of Edinburgh, Imperial College London

WP2 is driven by the need to overcome limitations in existing meshing approaches, which struggle to balance accuracy, flexibility, and scalability for complex geometries and multi-physics simulations. Current tools are often restricted in their support for high-order elements and parallel performance.

The objective is to develop hybrid meshing strategies that combine multiple element types and varying orders of approximation, enabling meshes to adapt locally to both geometry and physical behaviour. This includes the development of automatic hybrid mesh generation tools and improved handling of transitions between element types.

Key developments include new mesh quality metrics for hybrid and high-order meshes, tools to ensure high-order geometric fidelity, and a parallel hybrid mesh generation framework tightly integrated with geometry. These capabilities are essential for accurate and scalable simulation on next-generation HPC systems.

WP3: Adaptive Mesh Generation

Lead: Swansea University — Prof. Oubay Hassan
Supporting: University of Edinburgh, Imperial College London

WP3 addresses the need for dynamic, solution-driven mesh adaptation in large-scale simulations, where static meshes are inefficient for capturing evolving physical phenomena. Current adaptive approaches are limited in their scalability and integration with parallel workflows.

The objective is to develop parallel mesh and degree adaptation algorithms that enable refinement, coarsening, and variation in approximation order based on the evolving solution. This includes robust strategies for load balancing and efficient operation on heterogeneous HPC architectures.

Key developments focus on preserving mesh quality during adaptation, enabling efficient geometry and topology access, and supporting time-dependent simulations. A central outcome is the integration of adaptivity through a plug-and-play framework compatible with existing simulation codes.

WP4: Machine Learning for Mesh Generation

Lead: Swansea University — Prof. Rubén Sevilla
Supporting: University of Edinburgh, Queen’s University Belfast

WP4 explores the use of machine learning to transform mesh generation and adaptation, addressing the limitations of manual and heuristic-based workflows. Current approaches are time-consuming and do not fully utilise the wealth of available simulation data.

The objective is to develop AI-driven methods that support geometry processing and mesh generation by learning from historical simulations. This includes geometry feature detection and classification and the prediction of mesh requirements for complex, multi-scale problems.

Key developments include AI-driven mesh adaptation for multi-scale simulations, dynamic mesh prediction for time-dependent systems, and a scalable AI framework designed for HPC environments. These advances aim to significantly accelerate simulation workflows while maintaining accuracy and robustness.

WP5: Multi-Physics Integration

Lead: Imperial College London — Prof. Spencer Sherwin
Supporting: University of Edinburgh, King’s College London

WP5 addresses the challenge of integrating mesh generation and adaptation within multi-physics simulation frameworks, where different physical models impose competing requirements on geometry, resolution, and computational performance.

The objective is to develop scalable multi-physics integration strategies that couple meshing and simulation processes efficiently in parallel HPC environments. This includes understanding the sensitivity of simulations to geometry and mesh resolution and ensuring reliable solver performance.

Key developments include verification of multi-physics solver accuracy, the development of tightly and loosely coupled integration strategies, and performance assessment for transient simulations with moving boundaries. These advances will enable robust, high-fidelity simulation workflows for complex engineering applications.