/
WG1 - Passing through time and space scales

WG1 - Passing through time and space scales

This working group will focus on solutions to gain orders of magnitude for tackling simulations at scales which are relevant for real life applications. Below are the objectives to achieve this

Driver

@nicolin @Hongyang Cheng @Deepak Tunuguntla

Approver

@Daniel Barreto

Contributors

@nicolin @Hongyang Cheng @Deepak Tunuguntla

Informed

Objectives

  1. High-Performance Computing: Leveraging new hardware architectures and parallelisation techniques (CPU and GPU) to overcome challenges in simulating non-spherical particles, load balancing, and contact algorithms.

  2. Model-Based Solutions:

    1. Developing enriched contact models and utilizing spatial and temporal periodicities to optimize particle usage and time steps.

    2. Exploring Alternative Techniques: Investigating other innovative methods, including machine-learning and model order reduction approaches, to further improve simulation efficiency.

Due date

Key outcomes

D1.1. (M9)–Short list of academic reference problems (including their full definition in terms of initial configuration, boundary condition, time step, expected results, and/or analytical solutions, etc.). The first list relates to problems that may be tackled by considering various hardware architectures, the second one to model-based challenges. Lists will be compiled as part of a short technical report and made available via the wiki and repositories.

D1.2. (M12, M24, M36 and M48) – Implementation of hardware- and model-based short-list problems using at least two open-source DEM codes to give a comprehensive comparison of the different investigated solutions. Two individual problems will be considered each year (i.e. one hardware-based, another one model-based) and summarised in individual technical reports (M12, M24, M36 and M48). These will also be made available as individual datasets fully documented and shared via repositories.

D1.3. (M48) – Two journal papers considering the conclusions and findings related to (i) hardware-based solutions and (ii) model-based solutions, respectively.

Status

in progress

 

 Problem Statement

This working group will focus on solutions to gain orders of magnitude for tackling simulations at scales which are relevant for real life applications. Below are three approaches to achieve this:

  1. Using high-performance computing and taking benefits of new hardware architectures.

    • Well-known solutions such as parallelisation with both CPU and GPU will be investigated.

    • Efficient use of these solutions in particular contexts is still challenging. For instance, simulating non-spherical particles with GPU-based algorithm, handling load balancing on CPU parallel architecture for simulations or parallelising contact algorithms such as Voronoi based searching remains difficult and are still scientific barriers.

    • Potential participation of hardware manufacturers will enable new insights for the optimisation of DEM algorithms.

  2. Using model-based solutions

    These could be

    1. enriched contact models;

    2. space periodicity; and/or time extrapolation strategies.

    3. Alternative methods investigating other innovative methods, including machine-learning and model order reduction approaches, to further improve simulation efficiency.

The first approach consists in enriching the contact model in order to do more with fewer particles. The second approach takes advantage of temporal and spatial periodicities to use less particles or time steps. Whereas the third focuses on dimensionality reduction and learning the hidden patterns in the provided data.

In the first grant period, the WP1-UP track focuses on model-based solutions within the framework of discrete element modeling, namely by reducing the number of particles in the simulation. This technique (also known as mesoparticles) is straightforward. However, obtaining equivalent, reproducible macroscopic responses remains a challenge, and one of the biggest problems is the scaling of contact laws (relating to their spatial and temporal representation of the bulk system), in complex settings, for example, where particle sizes vary and multiphysics (liquids) are involved. This position paper extends the work of two PhD students funded by MSCA DN scheme with the support of COST Action ON-DEM to broadly disseminate the knowledge of numerical upscaling. The paper draft is in its final form and expected to be handed in by December. With this deliverable, we have a better grasp on computational cost versus accuracy for granular flow problems. This knowledge allows us to move forward with multi-scale and data-driven models, providing reference data and a better evaluation of the tradeoff between CG-DEM and other advanced approaches.

 Scope

Must have:

  • Create a detailed list of reference problems related to both hardware and model-based challenges, including their full definitions and available via wiki and repositories.

  • Execute two open-source DEM code implementations each year, focusing on one hardware-based and one model-based problem, culminating in documented technical reports and datasets.

  • Produce two journal papers by the end of the project, summarizing the conclusions from hardware-based and model-based solutions.

Must have:

  • Create a detailed list of reference problems related to both hardware and model-based challenges, including their full definitions and available via wiki and repositories.

  • Execute two open-source DEM code implementations each year, focusing on one hardware-based and one model-based problem, culminating in documented technical reports and datasets.

  • Produce two journal papers by the end of the project, summarizing the conclusions from hardware-based and model-based solutions.

Nice to have:

  • Engage with hardware manufacturers to gain insights into optimizing DEM algorithms.

  • Investigate additional promising solutions, such as machine-learning techniques, to enhance simulation capabilities.

Not in scope:

 

How to contribute?

 

Milestones and deliverables

Milestone

Deliverable

Milestone

Deliverable

D1.1 (M9)

D1.2 (M18)

  • Report on upscaling with CG-DEM

Important Pages

 

Meet the team

 

Govender1.png
Nicolin Govender

WG1 & DEM-HPC sub-group lead

hongyang.jpeg
Hongyang Cheng

WG1 vice-lead & DEM-Model-based UP sub-group lead

Deepak4.png
Deepak Tunuguntla

WG1 co-lead & DEM-Model-based Alternative Methods sub-group lead