📄 agriculture

Establishment and parameter calibration of double-layer bonded discrete element model for alfalfa stems at optimal harvest stage

This study establishes and calibrates a double-layer bonded discrete element model that accurately captures the structural heterogeneity of alfalfa stems, demonstrating its effectiveness in simulating mechanical behaviors like shear force and repose angle to optimize harvesting and drying equipment.

Original authors: Wei Zhang, Haitang Cen, Wang Guo, Penghui She

Published 2026-07-09

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Wei Zhang, Haitang Cen, Wang Guo, Penghui She

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you're trying to build a perfect digital twin of an alfalfa stem for a video game, but instead of a cartoon, this game is actually a high-tech simulation used to design farm machines. The goal? To figure out exactly how to cut, move, and dry these stems without breaking them or wasting energy.

For a long time, scientists tried to model these stems by treating them like a single, uniform stick—kind of like a solid piece of chalk. But the authors of this study say, "Hold on a minute!" They argue that this "one-size-fits-all" approach is wrong because real alfalfa stems are actually more like a straw with a tough, crunchy shell and a soft, squishy center. If you ignore that difference, your simulation is like trying to predict how a chocolate-covered marshmallow breaks by pretending it's just a block of chocolate. It just doesn't work right.

So, the team at Inner Mongolia University of Technology decided to build a "double-layer" model. Think of it as constructing a digital stem out of thousands of tiny, virtual marbles. They used slightly larger marbles to build the tough outer skin (the epidermis) and smaller, softer marbles to fill the inside (the core). They then glued these marbles together with a special digital "glue" that mimics how real plant cells stick to each other.

To make sure their digital glue was the right strength, they didn't just guess. They went into the lab with real alfalfa stems harvested at their perfect moment (the "optimal harvest stage"). They measured everything:

How wet the stems were: 0.723 ± 0.044 kg/kg.
How thick they were: 3.41 mm on average.
How heavy they were: 994.46 kg/m³.
How much force it took to slice through one: 56.36 N.
How they pile up when dumped in a heap (the "angle of repose"): 31.62°.

With these real-world numbers in hand, they ran a massive series of computer tests. They used a clever "treasure hunt" strategy to find the right settings for their simulation. First, they screened a bunch of variables to see which ones mattered most (like a detective ruling out suspects). Then, they took "steepest ascent" steps—basically walking up a hill of data to find the peak where the simulation matched reality best. Finally, they used a complex mathematical map (called a Box–Behnken response surface) to pinpoint the exact settings.

They even taught a computer brain (a Feedforward Neural Network) to predict the cutting force, and it did a great job, matching the simulation data with an accuracy score (R²) of 0.9077.

The result? Their new double-layer model was incredibly accurate. When they simulated the stems piling up, the digital heap formed an angle of 31.46°. Compare that to the real-life measurement of 31.62°, and you get a tiny error of just 0.51%. That's like measuring a football field and being off by less than the width of a human hair.

However, the authors are careful not to claim they've solved everything. They admit their model only works for stems at this specific harvest stage with this specific moisture level. They haven't tested it on stems that are wetter, drier, or at different growth stages. Also, while their simulation is great for static tests (like cutting or piling), they haven't yet proven it works perfectly for high-speed, chaotic situations like stems flying through a machine at full speed.

In short, the paper suggests that by treating alfalfa stems as a two-layered structure rather than a single block, we can create a much more reliable digital blueprint. This blueprint is now ready to help engineers design better farm equipment that cuts and handles alfalfa with less waste and more efficiency, but it's a tool for specific conditions, not a magic wand for every farming scenario.

Technical Summary: Establishment and Parameter Calibration of a Double-Layer Bonded DEM Model for Alfalfa Stems

Problem Statement
Accurate simulation of alfalfa stems is critical for optimizing mechanized harvesting, conveying, and drying operations. However, conventional Discrete Element Method (DEM) models typically treat crop stems as homogeneous, isotropic materials. This simplification neglects the inherent structural heterogeneity between the outer epidermis (mechanical tissue) and the inner core (parenchyma and vascular tissue), leading to significant errors in simulating anisotropic mechanical behaviors such as shearing and stacking. Furthermore, existing models often fail to accurately capture the separation processes between the epidermis and inner core during cutting. There is a lack of high-fidelity models specifically calibrated for alfalfa stems at the optimal harvest stage that account for this double-layer structure.

Methodology
This study established a double-layer bonded DEM model for alfalfa stems and employed a systematic parameter calibration strategy.

Physical Characterization: Alfalfa stems were collected at the optimal harvest stage in Hohhot, Inner Mongolia. Key physical properties were measured, including moisture content (0.723 ± 0.044 kg/kg), diameter (3.41 mm), density (994.46 kg/m³), Poisson's ratio (0.4), and shear modulus (18.26 MPa). Contact parameters, including static friction coefficients and coefficients of restitution between stems and steel plates, as well as between stems themselves, were determined via inclined plane and free-fall tests.
Model Construction: A double-layer bonded model was developed using the Hertz–Mindlin with Bonding (HMB) contact model. The stem was approximated as a cylinder composed of two distinct regions: an outer epidermis and an inner core.
- Geometry: Spherical particles (0.4 mm diameter) formed the epidermis, while smaller particles (0.30–0.35 mm) filled the inner core.
- Arrangement: Particles were arranged using a spatial coordinate positioning method to ensure no overlap, resulting in a model with 7,946 particles per stem.
- Bonding: Parallel bonding was applied at contact points to simulate the interaction forces and fracture behaviors between the epidermis and core.
Parameter Calibration: A stepwise calibration strategy was used to identify and optimize key parameters:
- Screening: Plackett–Burman screening tests identified significant bonding parameters (affecting shear force) and contact parameters (affecting repose angle) from a larger set of candidates.
- Optimization: Steepest ascent tests were conducted to approach the optimal parameter ranges.
- Response Surface Methodology (RSM): Box–Behnken designs were used to establish second-order regression models relating the significant parameters to the shear force and repose angle.
- Machine Learning: A Feedforward Neural Network (FNN) was trained to predict shear force based on the significant parameters identified by RSM.
Validation: The calibrated model was validated by comparing simulated shear forces and repose angles against physical experimental data.

Key Contributions

Double-Layer Modeling: The study successfully developed a DEM model that explicitly distinguishes between the epidermis and inner core of alfalfa stems, addressing the limitations of homogeneous models in capturing structural anisotropy.
Systematic Calibration: A robust workflow combining Plackett–Burman screening, steepest ascent, and Box–Behnken RSM was applied to efficiently calibrate complex bonding and contact parameters for a multi-layered biological material.
Hybrid Prediction: The integration of an FNN model with RSM provided a comparative analysis for predicting shear forces, demonstrating the capability of neural networks to capture nonlinear relationships in DEM parameter calibration.
High-Fidelity Data: The study provided a comprehensive dataset of physical and contact parameters for alfalfa stems at the optimal harvest stage, including specific values for the double-layer structure.

Results

Shear Force: The Box–Behnken response surface model for shear force achieved a coefficient of determination ( $R^2$ ) of 0.9036. The FNN model yielded an $R^2$ of 0.9077 and a Root Mean Square Error (RMSE) of 0.9407, indicating comparable and high predictive accuracy.
Repose Angle: The response surface model for the repose angle showed an $R^2$ of 0.9773.
Validation: Using the calibrated parameters, the simulated angle of repose was 31.46°, which deviated from the experimental average of 31.62° by only 0.51%. This low relative error confirms the accuracy of the double-layer bonded model.
Significant Factors: The calibration identified specific bonding parameters (e.g., epidermis-pith critical tangential stress) and contact parameters (e.g., epidermis-epidermis static friction coefficient) as the primary drivers of shear force and repose angle, respectively.

Significance
The paper claims that the proposed double-layer bonded DEM model accurately represents the anisotropic mechanical behavior of alfalfa stems, overcoming the errors introduced by treating stems as homogeneous materials. By providing a reliable simulation basis with a relative error of less than 1% for the repose angle, the study offers a technical reference for the structural design and parameter optimization of alfalfa harvesting and drying equipment. The modeling framework and calibration strategy presented are intended to serve as a reference for developing DEM models for other herbaceous crops with similar structural heterogeneity. The authors acknowledge limitations regarding the single moisture content and growth stage tested, noting that future work will address diverse conditions and dynamic working scenarios.

Technical Summary: Establishment and Parameter Calibration of a Double-Layer Bonded DEM Model for Alfalfa Stems

More like this