Reducing biomass to a stable particle size is a key step towards biofuels and biobased products. In practice, this leads to blockages, variable particle distribution, and machine wear. The American bioenergy hub BETO has cited material handling as a barrier to large-scale production for years. New computational models from Idaho National Laboratory address this bottleneck.
Researchers combine measurements of stems with a particle model that mimics the behavior of thousands of small fragments of material, and with a learning computational model that extracts patterns from the data. This allows them to predict the behavior of material in the blade grinder and the evolution of particle size. Two factors are more important than often realized: the mesh size of the exhaust screen and the moisture content of the supplied material. By explicitly including moisture as input, predictions become more accurate and repeatable. INL reports that a deep neural operator model, an advanced neural network that learns relationships in the process, achieves high accuracy, as long as the source data is representative. This shortens the step from design to field testing and reduces the number of expensive full-scale tests.
Application in European installations
Wood chips and agricultural residues in European chains show similar variations in shape and moisture content. Models that optimize screen selection and drying reduce the risk of downtime due to clogging and limit the energy consumption of the milling step. The message is pragmatic. It's not a more powerful motor that makes the difference, but better data and settings regarding moisture and sieving. This aligns with the work of the Feedstock Conversion Interface Consortium, which aims to control variation in biofeedstock from field to reactor. INL refers to facilities where industry and researchers conduct joint tests, such as the Biomass Feedstock National User Facility. There, virtual and physical tests are combined, accelerating the design of pre-processing.
This step is relevant because grinding consumes a significant portion of power consumption in many installations and causes disruptions further down the chain. A more stable particle size results in more consistent drying, gasification, or fermentation, with fewer emergency shutdowns and less wear and tear on transport and screening. In projects where every percent of throughput counts, a better-tuned grinding process can make the difference between a pilot program and a profitable operation.
In practice, it works like this. The researchers feed a physical model that tracks individual particles with measurements of actual material. This provides insight into how stems buckle, tear, and break. On top of that layer, a fast computational model runs as a surrogate. This allows engineers to quickly calculate numerous scenarios, such as different screen sizes or higher moisture content, without the need for a full-scale test run each time. The result is a shorter iteration time between design and reality and a lower test load on the installation.
The test campaign focused on corn residue, a commonly used waste stream in the United States. The setup in the Process Development Unit varied with different sieves and moisture contents, so the model learned not only the average but also the particle size distribution. This approach aligns with the idea that material behavior is not constant but depends on season, storage, and logistics.
For European installations, the core remains the same. Wood chips for district heating networks, roadside grass in digesters, or straw as a feed additive all vary. By measuring moisture upon inlet and using the model as a decision support, an operator can pre-adjust the sieve selection and speed. In wet periods, this means drying first or avoiding a finer sieve; in drier months, a finer sieve can be used. The model thus forms a digital twin of the milling step, useful for training and daily operations.
A logical next step is calibration for each raw material. Poplar, willow, or pine wood behave differently than corn residue. Adding measurement data from European streams increases usability. Collaboration with machine builders is a logical next step, ensuring that moisture and particle size sensors are included as standard in new lines. This transforms a weak link into a predictable process step.
Source, INL feature story, November 17, 2025
https://inl.gov/feature-story/modeling-better-biomass-milling/
Image created with AI.
