TL;DR
This paper introduces a novel, fully virtual workflow for predicting the hydrodynamics of packed beds filled with non-spherical, polydisperse particles. By combining Discrete Element Method (DEM) for particle packing and Computational Fluid Dynamics (CFD) for flow simulation, the approach accurately predicts bed permeability, offering a significant improvement over traditional semi-empirical methods.
What Problem Does This Paper Address?
Packed bed reactors are ubiquitous in chemical engineering, but accurately characterizing their hydrodynamic properties, such as permeability, is crucial yet challenging. Traditional methods include using semi-empirical correlations (like Ergun or Kozeny-Carman), which often rely on assumptions of spherical particles and can lead to significant overestimations. Experimental measurements, while more accurate, can be time-consuming and resource-intensive, especially for complex materials like wood chips. The authors aim to bridge this gap by developing a robust, fully numerical workflow that can predict these properties with high fidelity, using only particle size distribution as input. This is particularly relevant for biomass conversion processes where wood chip packed beds are common, but their hydrodynamic data is scarce.
Methods
The core of this work is a multi-stage numerical workflow:
- Particle Generation: Wood chips, used as a case study, were manually measured to obtain their dimensions. These measurements were then used to create numerical equivalents of the chips, assuming a parallelepipedic shape.
- Packed Bed Simulation (DEM): The DEM code LMGC90 was employed to generate a virtual packed bed of these non-spherical particles. This step accounts for particle-particle interactions and gravity to create a realistic packing.
- Meshing (CFD): The OpenFOAM CFD toolbox, specifically the
snappyHexMeshutility, was used to automatically mesh the fluid domain (the void spaces between particles) within the generated packed bed. This process involves background meshing, refinement based on particle surfaces, snapping the mesh to the geometry, and adding boundary layers. - Fluid Flow Simulation (CFD): Navier-Stokes equations were solved using the
simpleFoamsolver in OpenFOAM to simulate fluid flow through the meshed domain. The Launder-Reece-Rodi turbulence model was used to capture fluid-wall interactions accurately. - Permeability Calculation: The pressure drop across the simulated bed was extracted from the CFD results, and permeability was calculated using Darcy's law. The workflow was also extended to simulate flow in the inertial regime and determine the Forshheimer equation constants.
Datasets
The primary "dataset" used in this study was a collection of 536 manually measured wood chips (Fagus sylvatica). These measurements provided the dimensions (length, height, width) for creating the numerical particle models. No pre-existing datasets were used for training or evaluation; instead, the numerical results were validated against experimental measurements performed by the authors on a packed bed of the same wood chips.
Key Insights and Ideas
- Virtualization of Complex Geometries: The workflow demonstrates the feasibility of creating accurate virtual representations of complex, non-spherical, polydisperse granular media, moving beyond simplified spherical particle models.
- DEM-CFD Synergy: The seamless integration of DEM for particle packing and CFD for fluid flow simulation provides a powerful tool for predicting macroscopic properties from microscopic particle behavior.
- Meshing Challenges and Solutions: The paper highlights the importance of meshing techniques like
snappyHexMeshfor accurately capturing the intricate pore geometry, and it rigorously assesses mesh quality to ensure reliable results. - Impact of Particle Shape: The study implicitly shows that particle shape significantly influences packing and, consequently, bed hydrodynamics, necessitating approaches that go beyond simple sphericity corrections.
- Validation Against Experiment: The direct comparison of numerical predictions with experimental measurements provides strong validation for the proposed virtual workflow.
Key Results
- The developed workflow predicted a bed permeability that was 16.0% higher than the experimental value.
- This result represents a significant improvement over the Kozeny-Carman equation, which, even with a sphericity correction, overestimated the bed permeability by 115%.
- The numerical approach successfully captured the transition from Stokes to inertial flow regimes and allowed for the fitting of the Forshheimer equation constant (C = 7865) for the wood chip bed.
- Mesh quality assessment indicated that a refinement level of 4 was sufficient to achieve a pressure drop within 1.60% of the reference value, balancing accuracy and computational cost.
- Computational domain size analysis showed that predictions stabilize for domains larger than 11 $d_{eq}$, with an averaged permeability of 1.89 ± 0.10 × 10⁻⁸ m².
Limitations and Future Work
- Particle Shape Simplification: The use of parallelepipeds to represent wood chips is a simplification. The authors acknowledge that a more accurate representation (e.g., using trapezoidal shapes) could further improve predictions, particularly for inertial effects.
- Computational Cost: While the workflow is robust, DEM simulations can be computationally intensive, requiring significant time and resources, especially for large particle numbers or long simulation times.
- Experimental Uncertainty: The authors note that the experimental permeability has an associated uncertainty (± 2.50%), which should be considered when comparing numerical and experimental results.
Future work could involve exploring more sophisticated particle shape representations, investigating the impact of particle surface roughness, and extending the workflow to predict other important parameters like tortuosity, dispersion coefficients, and volumetric heat exchange coefficients. The authors envision this approach as a valuable tool for optimizing reactor design and troubleshooting existing systems.
Summary updated