Machine Learning Aided Modeling of Granular Materials: A Review (2410.14767v1)

Abstract: AI has become a buzz word since Google's AlphaGo beat a world champion in 2017. In the past five years, machine learning as a subset of the broader category of AI has obtained considerable attention in the research community of granular materials. This work offers a detailed review of the recent advances in machine learning-aided studies of granular materials from the particle-particle interaction at the grain level to the macroscopic simulations of granular flow. This work will start with the application of machine learning in the microscopic particle-particle interaction and associated contact models. Then, different neural networks for learning the constitutive behaviour of granular materials will be reviewed and compared. Finally, the macroscopic simulations of practical engineering or boundary value problems based on the combination of neural networks and numerical methods are discussed. We hope readers will have a clear idea of the development of machine learning-aided modelling of granular materials via this comprehensive review work.

• The paper demonstrates that ML techniques can reduce computational costs and improve accuracy in simulating granular materials.
• The paper details how ML-based microscopic modeling, using neural networks, accelerates DEM processes for particle interactions.
• The paper highlights that integrating ML with continuum methods via surrogate constitutive models offers efficient macroscopic simulations despite extrapolation challenges.

Machine Learning Aided Modeling of Granular Materials: A Comprehensive Review

The application of ML in modeling granular materials has gained significant traction in recent years, as detailed in this review by Wang, Kumar, et al. The paper meticulously surveys the recent advancements in integrating ML with the paper of granular materials, addressing both micro and macro scales of analysis. The discussion encompasses the use of ML for modeling particle-particle interactions, the development of constitutive models, and ML-aided macroscopic simulations.

Granular materials, by nature, present complex challenges in modeling due to their intrinsic discontinuity and range-dependent stress-strain responses. Traditional numerical approaches such as the Finite Element Method (FEM), Discrete Element Method (DEM), and Material Point Method (MPM) have been used extensively in this field. However, these methods often involve computational trade-offs, including high costs and limitations on large-deformation problems. This has paved the way for ML techniques, known for their ability to handle complex, high-dimensional problems, to offer transformative solutions in granular material modeling.

The paper presents a comprehensive review of ML algorithms applied in various contexts:

1. ML-based Microscopic Grain Modeling: ML approaches have been adopted to refine the DEM's contact detection and resolution processes. By employing classification and regression neural networks, researchers aim to speed up these computationally intensive steps. However, the completeness and robustness of training datasets remain challenges that limit the broad applicability of these ML-based models.
2. Constitutive Models of Granular Materials: Traditional constitutive models often involve complex assumptions. ML models, trained on either experimental or synthetic data, provide an opportunity to bypass some limitations by learning directly from data. Time-sequence networks like LSTM and GRU have shown promise in capturing history-dependent behaviors, while simpler models like MLPs require carefully selected history variables to perform adequately.
3. ML-Aided Macroscopic Simulations: For continuum-based approaches, such as the FEM and MPM, integrating ML models as surrogate constitutive models has demonstrated potential in reducing computational burdens significantly. Efforts in the FEM-ML and MPM-ML frameworks reflect this integration. GNN-based approaches have also been leveraged to simulate the kinematic features of macroscopic particles, though issues like error accumulation and extrapolation capabilities remain prevalent challenges.

The paper emphasizes that while ML models in granular materials can drastically improve computational efficiency and offer insights devoid of traditional assumptions, they are not free from limitations. The weaker extrapolation capability and the necessity for comprehensive datasets inhibit widespread application. Solutions such as active learning, transfer learning, and uncertainty quantification are discussed as future directions to augment the robustness and applicability of ML models in granular material studies.

The implications of these advancements are substantial for both theoretical understanding and practical engineering applications related to granular materials, like soil mechanics and geotechnical engineering. However, ensuring model generalization across varied conditions and addressing issues like computational memory requirements are essential steps forward.

In summary, this review highlights the progress and potential of ML in granular material modeling while critically assessing the current challenges and offering pathways for further research and development within this interdisciplinary field.