Towards scientific machine learning for granular material simulations - challenges and opportunities

Marc Fransen *, Andreas Fürst *, Deepak Tunuguntla *, Daniel N. Wilke *, Benedikt Alkin, Daniel Barreto, Johannes Brandstetter, Miguel Angel Cabrera, Xinyan Fan, Mengwu Guo, Bram Kieskamp, Krishna Kumar, John Morrissey, Jonathan Nuttall, Jin Ooi, Luisa Orozco, Stefanos-Aldo Papanicolopulos, Tongming · 2025

Read Towards scientific machine learning for granular material simulations - challenges and opportunities by Marc Fransen *, Andreas Fürst *, Deepak Tunuguntla *, Daniel N. Wilke *, Benedikt Alkin, Daniel Barreto, Johannes Brandstetter, Miguel Angel Cabrera, Xinyan Fan, Mengwu Guo, Bram Kieskamp, Krishna Kumar, John Morrissey, Jonathan Nuttall, Jin Ooi, Luisa Orozco, Stefanos-Aldo Papanicolopulos, Tongming (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-08766.txt.

Introduction

Marc Fransen and a large group of co-authors wrote a position paper in twenty twenty-five on machine learning for granular materials — sand, grain, powders. Particle-scale physics governs how these materials flow or jam, but simulating every grain is prohibitively costly. The paper grew out of a Lorentz Center workshop joining the granular and learning communities.

The authors define the terrain and name seven key challenges spanning gas-like, fluid-like, and solid-like regimes. They then survey learning tools fit for the job: sequence models for path-dependent behavior, encoder-decoder models for high-dimensional states, graph networks, neural operators, and model reduction. The aim is robust digital twins for industrial granular processes, from silos to pharmaceutical powders.

Catalog record

CREATORMarc Fransen *, Andreas Fürst *, Deepak Tunuguntla *, Daniel N. Wilke *, Benedikt Alkin, Daniel Barreto, Johannes Brandstetter, Miguel Angel Cabrera, Xinyan Fan, Mengwu Guo, Bram Kieskamp, Krishna Kumar, John Morrissey, Jonathan Nuttall, Jin Ooi, Luisa Orozco, Stefanos-Aldo Papanicolopulos, Tongming
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE93 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-08766

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Towards scientific machine learning for granular material simulations - challenges and opportunities Marc Fransen *, Andreas Fürst *, Deepak Tunuguntla *, Daniel N. Wilke *, Benedikt Alkin, Daniel Barreto, Johannes Brandstetter, Miguel Angel Cabrera, Xinyan Fan, Mengwu Guo, Bram Kieskamp, Krishna Kumar, John Morrissey, Jonathan Nuttall, Jin Ooi, Luisa Orozco, Stefanos-Aldo Papanicolopulos, Tongming Qu, Dingena Schott, Takayuki Shuku, WaiChing Sun, Thomas Weinhart, Dongwei Ye, Hongyang Cheng † Abstract Micro-scale mechanisms, such as inter-particle and particle-fluid interactions, govern the behaviour of granular systems. While particle-scale simulations provide detailed insights into these interactions, their computational cost is often prohibitive. At a recent Lorentz Center Workshop on “ Machine Learning for Discrete Granular Media ”, researchers explored how machine learning approaches can aid the development of constitutive laws and efficient data-driven surrogates for granular materials while also addressing uncertainty quantification. Attended by researchers from both the granular materials (GM) and machine learning (ML) communities, the workshop brought the ML community up to date with GM challenges. This position paper emerged from the workshop discussions. In this position paper, we define granular materials and identify seven key challenges that characterise their distinctive behaviour across various scales and regimes–ranging from gas-like to fluid-like and solid-like. Addressing these challenges is essential for developing robust and efficient models for the digital twinning of granular systems in various industrial applications. To showcase the potential of ML to the GM community, we present classical and emerging machine/deep learning techniques that have been, or could be, applied to granular materials. We reviewed sequence-based learning models for path-dependent constitutive behaviour, followed by encoder-decoder type models for representing high-dimensional data in reduced spaces. We then explore graph neural networks and recent advances in neural operator learning. The latter captures the emerging field evolution of interacting particles via efficient latent space representation. Lastly, we discuss model-order reduction and probabilistic learning techniques for high-dimensional parameterised systems, both of which are crucial for quantifying and incorporating uncertainties arising from physics-based and data-driven models. We present a typical workflow aimed at unifying data structures and modelling pipelines and guiding readers through the selection, training, and deployment of ML surrogates for granular material simulations. Finally, we illustrate the workflow’s practical use with two representative examples, focusing on granular materials in solid-like and fluid-like regimes. A Preprint † † footnotetext: * Co-first authors. † Corresponding author. All other co-authors are listed alphabetically. Keywords Granular materials, multi-scale methods, machine learning, sequence modelling, model order reduction, neural operator learning, probabilistic learning, uncertainty quantification 1 Introduction Granular materials (GMs), ranging from beach sand to raw materials such as iron ore, are integral to various industrial processes. They play an essential role across many engineering disciplines, including geotechnical [ 128 ] , coastal [ 121 ] , hydraulic engineering [ 2 ] , pharmaceutical [ 2 ] , additive manufacturing [ 129 , 148 , 5 ] , agriculture [ 80 , 159 ] , bulk handling [ 154 , 50 , 107 ] and robotics [ 98 , 115 ] . Although simple in appearance, GMs exhibit solid-, fluid-, and gas-like behaviour [ 47 ] , making it one of the most complex materials to handle. A first-principle approach for simulating GM behaviour involves modelling individual grains’ interactions. However, this is computationally expensive. Alternatively, continuum approaches can be adopted, where constitutive laws are required to describe the material’s bulk behaviour that emerges from grain-scale interactions by relating stress and strain and/or their rates under specific conditions. For certain solid-like behaviour, existing theories from computational solid mechanics, e.g., Mohr-Coulomb, Drucker-Prager and more advanced models incorporating micro-structural information like fabric tensors [ 99 , 55 ] can be utilised. Similarly, for fluid-like GM processes, such as dense granular flows, theories from computational fluid mechanics [ 170 ] provide useful frameworks. However, these models are typically crafted by “domain experts” within specialised fields and often limited to specific conditions. As a result, they often fail to fully capture granular behaviour across the broad range of regimes and conditions encountered in practice. It is important to recognise that these conventional approaches were developed by combining theoretical frameworks with sparse experimental data, typically obtained under highly controlled laboratory settings (e.g. Fig. 1 ). While they perform adequately within these limited domains, extending them beyond their calibration range becomes increasingly challenging. Unlike fluid and solid mechanics, granular mechanics lack a unified continuum theory, necessitating expensive, phenomenon-specific, multi-scale approaches to develop effective continuum models. Figure 1: Examples of laboratory devices for granular materials in (a) triaxial compression and (b) continuous ring-shear conditions. Copyright: Dietmar Schulze . Current research interests towards modelling granular materials Current research focuses on (1) developing a unified theory capable of capturing GM phenomena across solid-like, fluid-like and gas-like regimes [ 47 , 86 ] ; (2) upscaling particle-level methods to handle the size and complexity of industrial-scale problems; and (3) incorporating real-world variability, including boundary condition uncertainties, measurement noise and the inherent stochastic nature of granular systems [ 36 , 49 ] . From (1)–(3), a central question emerges: can we derive or learn unified constitutive theories and governing equation solvers for granular materials directly from particle-scale information, resulting in models that are not only more efficient but also provide deeper insights than contemporary, phenomenological approaches? To address this, researchers increasingly rely on grain-scale numerical simulations, primarily employing the Discrete Element Method (DEM), in conjunction with micro-scale observations, e.g., using the micro-Computed Tomography [ 175 ] . DEM [ 44 ] remains the most widely used technique for modelling granular systems and has demonstrated success across multiple fields, including process engineering [ 89 ] , geotechnical engineering [ 88 ] , bulk handling [ 155 , 51 ] , mining [ 1 ] , chemical engineering [ 60 ] and additive manufacturing [ 5 , 157 ] . From physics-based solvers to data-driven alternatives DEM simulations are information-rich, capturing particle positions, velocities and orientations, and interaction forces with other particles and geometries. This data can be subsequently mapped to macroscopic stress and strain fields, ultimately enabling the construction of closed continuum-scale models. While effective, this multi-step approach remains highly phenomenon- and regime-specific, relying heavily on specialised domain knowledge and ad hoc upscaling [ 173 ] or multi-scale coupling [ 67 , 38 ] algorithms. Recall that, unlike solids and fluids, granular materials do not have a unified theory. As a result, the entire experiments–micro–macro–continuum–model pipeline becomes an iterative and costly endeavour. Addressing this challenge raises a new set of research questions: 1. Can data-driven approaches expedite the multi-step process of realising phenomenon-specific granular theories? 2. Can data-driven alternatives fully replace contemporary physics-based models by overcoming their inherent limitations? 3. Can data-driven methods help realise the unified theory that the GM community has long strived for? 4. More broadly, how can we best utilise the available GM data to complement our current approaches to modelling discrete media? Over the last two decades, machine learning techniques have evolved into powerful alternatives for computational mechanics, enabling researchers to process vast amounts of data and extract deeper insights from detailed simulations (see the recent review [ 186 ] ). As a result, ML has enabled researchers to perform tasks like constitutive model development [ 7 ] and surrogate

Read the complete text