Work Packages

WP1 addresses a fundamental limitation in modern simulation workflows: the difficulty of scaling complex geometric models across high-performance computing systems. Existing CAD-based approaches are typically designed for serial environments, requiring entire models to be loaded and processed on a single machine. This creates major challenges when dealing with large assemblies, moving geometries, and repeated geometric interrogation within simulation pipelines.

As simulation fidelity and model complexity increase, geometry handling becomes a critical bottleneck, particularly in HPC environments where efficient execution depends on parallel scalability, data locality, and memory efficiency. Current approaches struggle to distribute geometry effectively across processors, leading to excessive memory usage, poor performance, and limited scalability on heterogeneous CPU–GPU architectures.

WP1 tackles these challenges by rethinking how geometry is represented, partitioned, and accessed in parallel, enabling scalable, distributed geometry handling that integrates seamlessly with meshing and simulation workflows.

The objective is to develop scalable geometry handling strategies that enable efficient distribution and interrogation of geometry across heterogeneous CPU–GPU architectures. This includes evaluating alternative geometric representations such as NURBS, implicit, faceted, and hybrid models, alongside methods for geometry partitioning and data locality. A key aim is to ensure geometry is fully integrated within simulation workflows, rather than acting as a preprocessing constraint.

WP1 will deliver parallel-ready geometry handling tools that support distributed models, dynamic updates, and efficient interaction with simulation workflows, removing geometry as a bottleneck in high-fidelity simulation.

WP2 addresses the challenge of generating meshes that can simultaneously capture complex geometric detail, multi-physics behaviour, and large-scale computational efficiency. Existing meshing approaches are often limited in their ability to combine element types and approximation orders effectively, leading to either over-refined meshes or loss of accuracy in critical regions.

This challenge becomes significantly more acute in high-performance computing environments, where hybrid meshes must also support parallel generation, efficient data distribution, and consistent quality across partitions. Achieving smooth transitions between element types, maintaining geometric fidelity, and ensuring numerical stability across heterogeneous regions introduces substantial complexity, particularly for high-order methods.

WP2 tackles these issues by developing strategies that enable flexible, high-quality hybrid meshes that are both physically accurate and computationally efficient, forming a critical link between geometry representation and scalable simulation.

The objective is to develop robust hybrid meshing strategies that combine multiple element types and approximation orders within a unified framework. This includes improving mesh resolution control, transition handling, and geometric fidelity, while ensuring compatibility with parallel workflows and high-order methods. These approaches aim to generate meshes that are both computationally efficient and physically accurate.

WP2 will deliver automated hybrid mesh generation capabilities that operate in parallel, enabling consistent, high-quality meshes for complex geometries and large-scale simulations.

WP3 focuses on enabling efficient and scalable adaptive mesh generation for high-fidelity simulations involving multi-scale and time-dependent phenomena. While adaptivity is essential for resolving complex physical behaviour, many existing approaches struggle to operate effectively in parallel and heterogeneous HPC environments, where dynamic mesh modification introduces challenges in load balancing, data consistency, and communication overhead.

In large-scale simulations, the ability to refine and coarsen the mesh dynamically must be balanced against the need to maintain mesh quality, numerical stability, and computational efficiency. This is particularly challenging for distributed systems, where mesh updates can lead to data migration, partitioning constraints, and degraded performance if not carefully managed.

WP3 addresses these challenges by developing scalable, solver-integrated adaptive meshing strategies that enable meshes to evolve efficiently alongside the physics, ensuring accuracy is achieved only where needed while maintaining performance at scale.

The aim is to develop parallel adaptive meshing strategies that allow mesh resolution to evolve dynamically based on the underlying physics. This includes refinement, coarsening, and degree adaptation, as well as support for moving geometries and transient simulations. A key goal is to maintain mesh quality and numerical stability while ensuring efficient execution on heterogeneous HPC platforms.

WP3 will deliver scalable adaptive meshing tools that integrate directly with simulation codes, enabling dynamic, efficient, and solver-driven mesh evolution at scale.

WP4 explores the use of machine learning and data-driven methods to transform mesh generation from a largely manual, iterative process into a more predictive and automated workflow. Traditional approaches rely heavily on user expertise and repeated simulation cycles, making it difficult to achieve consistent performance across different geometries and physical scenarios.

This challenge is amplified in high-performance computing environments, where inefficient mesh design can lead to significant increases in computational cost. Leveraging the vast amounts of available simulation data presents an opportunity to guide decisions such as mesh density, anisotropy, and feature resolution, but integrating these approaches in a reliable and scalable way remains an open problem.

WP4 addresses this by developing AI-driven meshing strategies that learn from historical data, assist in geometry preparation, and guide mesh adaptation, reducing manual effort while improving consistency, efficiency, and scalability across complex simulation workflows.

The objective is to enable AI-driven mesh generation strategies that can predict optimal mesh configurations based on geometry and physics. This includes learning from historical simulation data to improve mesh density, anisotropy, and feature resolution, while ensuring compatibility with high-fidelity and multi-physics simulation requirements.

WP4 will deliver data-driven meshing tools that automate key decisions in geometry preparation and mesh generation, reducing manual effort while improving consistency and efficiency.

WP5 addresses the challenge of enabling efficient and scalable multi-physics simulation by integrating advances in geometry handling, meshing, and adaptivity into coherent workflows. Real-world engineering problems involve tightly coupled physical processes, yet combining these within HPC environments remains difficult due to challenges in data exchange, synchronisation, and computational efficiency.

In particular, inconsistencies between geometry, mesh resolution, and solver behaviour can lead to reduced accuracy or unnecessary computational cost. Understanding how different physical models interact, and how sensitive they are to geometric and discretisation choices, is essential for developing robust simulation pipelines.

WP5 addresses these challenges by developing integrated multi-physics simulation strategies that ensure consistency, scalability, and efficiency, enabling high-fidelity simulations suitable for complex industrial applications.

The objective is to establish integrated multi-physics simulation strategies that ensure consistency between geometry, mesh, and solver behaviour. This includes understanding the sensitivity of simulations to geometric features and mesh resolution, and improving how different physical models are coupled in parallel environments.

WP5 will deliver scalable multi-physics simulation workflows that bring together geometry, meshing, and solver technologies into coherent, high-performance pipelines suitable for industrial applications.

WP-Parallel underpins the entire programme by addressing the fundamental challenge of achieving scalable, efficient execution on heterogeneous high-performance computing systems. While advances in geometry handling, meshing, and simulation are essential, their impact depends on the ability to execute efficiently across distributed CPU–GPU architectures.

At Exascale, performance is increasingly limited not by computation, but by data movement, communication overhead, and memory access patterns. Poor data locality, excessive communication, and imbalanced workloads can significantly reduce efficiency, making it essential to design algorithms that are inherently parallel rather than adapted from serial workflows.

WP-Parallel addresses these challenges by developing communication-aware, data-efficient algorithmic strategies that enable all components of the simulation pipeline to scale effectively, ensuring the REMODEL framework can fully exploit modern HPC systems.

The objective is to develop parallel algorithmic strategies that enable all components of the simulation pipeline to scale efficiently on modern HPC systems. This includes designing methods that minimise communication, optimise data locality and memory access, and exploit heterogeneous hardware through balanced CPU–GPU utilisation. A central aim is to ensure that all workflows are built around scalable, communication-aware designs, rather than adapting serial algorithms for parallel execution.

WP-Parallel will deliver high-performance computational frameworks that support geometry handling, mesh generation, adaptivity, and multi-physics simulation at scale, ensuring the entire REMODEL pipeline is Exascale-ready.