NeuralDEM -- Real-time Simulation of Industrial Particulate Flows

Abstract: Advancements in computing power have made it possible to numerically simulate large-scale fluid-mechanical and/or particulate systems, many of which are integral to core industrial processes. Among the different numerical methods available, the discrete element method (DEM) provides one of the most accurate representations of a wide range of physical systems involving granular and discontinuous materials. Consequently, DEM has become a widely accepted approach for tackling engineering problems connected to granular flows and powder mechanics. Additionally, DEM can be integrated with grid-based computational fluid dynamics (CFD) methods, enabling the simulation of chemical processes taking place, e.g., in fluidized beds. However, DEM is computationally intensive because of the intrinsic multiscale nature of particulate systems, restricting simulation duration or number of particles. Towards this end, NeuralDEM presents an end-to-end approach to replace slow numerical DEM routines with fast, adaptable deep learning surrogates. NeuralDEM is capable of picturing long-term transport processes across different regimes using macroscopic observables without any reference to microscopic model parameters. First, NeuralDEM treats the Lagrangian discretization of DEM as an underlying continuous field, while simultaneously modeling macroscopic behavior directly as additional auxiliary fields. Second, NeuralDEM introduces multi-branch neural operators scalable to real-time modeling of industrially-sized scenarios - from slow and pseudo-steady to fast and transient. Such scenarios have previously posed insurmountable challenges for deep learning models. Notably, NeuralDEM faithfully models coupled CFD-DEM fluidized bed reactors of 160k CFD cells and 500k DEM particles for trajectories of 28s. NeuralDEM will open many new doors to advanced engineering and much faster process cycles.

• The paper introduces NeuralDEM, a novel deep learning framework that replaces conventional DEM for simulating complex industrial particulate flows.
• It employs a field-based representation and multi-branch neural operators to capture both macroscopic behavior and rapid transient dynamics.
• Numerical experiments on hoppers and fluidized bed reactors validate NeuralDEM’s scalability and accuracy, simulating up to 500,000 particles in real time.

Overview of "NeuralDEM -- Real-time Simulation of Industrial Particulate Flows"

The research paper "NeuralDEM -- Real-time Simulation of Industrial Particulate Flows" presents an advanced methodological framework, NeuralDEM, designed to facilitate real-time simulations of particulate systems integral to various industrial processes. The discrete element method (DEM) has been widely acknowledged for its capability to accurately represent granular and discontinuous materials. Despite its utility, DEM's computational intensity, derived from the multiscale nature of particulate systems, presents significant challenges. This paper addresses these issues by introducing NeuralDEM, an innovative deep learning surrogate capable of replacing traditional DEM and coupled CFD-DEM simulations.

Key Contributions and Methodology

The authors propose NeuralDEM as a novel end-to-end approach to streamline DEM computations. This methodology is built upon two principal innovations:

1. Field-based Representation of Particle Systems: NeuralDEM conceptualizes the Lagrangian discretization of DEM within a continuous field framework. By doing so, it models macroscopic behaviors through auxiliary fields, omitting the reliance on microscopic model parameters. This abstraction allows the simulation of long-term transport processes across different flow regimes effectively.
2. Multi-branch Neural Operators: The system introduces scalable, multi-branch neural operators that extend traditional modeling capabilities to encompass a variety of multi-physics scenarios, from slow and pseudo-steady operations to rapid, transient dynamics. The multi-branch structure allows for the differential treatment of primary and ancillary quantities, managing both core physics in main branches and supplementary properties in auxiliary off-branches.

The NeuralDEM framework is built to manage complex industrial scenarios, processing them in real-time, which previously posed insurmountable obstacles for conventional deep learning approaches.

Numerical Experiments and Results

The paper demonstrates NeuralDEM's capabilities through two primary industrial applications: hoppers and fluidized bed reactors.

1. Hopper Flows: These simulations focused on slow, pseudo-steady behaviors typical of granular materials in storage and dispensing systems. The NeuralDEM accurately predicts phenomena such as flow regimes, outflow rates, drainage times, and residual materials without requiring traditional DEM's extensive calibration procedures.
2. Fluidized Bed Reactors: Characterized by rapid, transient dynamics, the fluidized bed experiments underscore NeuralDEM's ability to simulate coupled CFD-DEM systems with up to 500,000 particles and 160,000 CFD cells over extended trajectories. The system notably delivers reliable long-term simulations, maintaining stability with correct physics extrapolation for significant durations.

Implications and Future Directions

NeuralDEM's advancement suggests profound implications for engineering applications, as it reduces the computational burden and accelerates process cycles significantly. The ability to condition on macroscopic properties like internal friction angles provides ease of integration into engineering workflows, sidestepping laborious parameter calibrations typical of traditional DEM.

Future development efforts could explore NeuralDEM's application scaling to more expansive particulate systems and incorporating additional physical phenomena such as thermal effects and chemical reactions. Facilitating further adaptability in varying industrial contexts could enhance its utility, driving broader adoption in industrial process simulations.

NeuralDEM represents a pivotal step towards efficient, scalable, and comprehensible simulations of particulate systems, anticipating new avenues in digital twin technology and real-time operational feedback within industrial contexts.