Shaking Moon Dust: Apollo Soil and Simulants Reveal Secrets of Lunar Size Separation

When engineers dream of building a lasting human presence on the Moon, they do not imagine hauling every kilogram of raw material from Earth. Instead, they look to the fine grey dust that blankets the lunar surface, known as regolith, as a feedstock for oxygen, metals, and water. Turning that dust into usable resources, a field called in situ resource utilisation, or ISRU, demands a reliable supply of consistently sized particles. Now, a team led by researchers at Imperial College London has taken a close look at one of the simplest possible ways to sort lunar soil by size: shaking it. Their study, published in the journal Space and Planetary Resources, compares the behaviour of genuine Apollo 15 regolith with three widely used terrestrial simulants, and the results carry important lessons for anyone hoping to process Moon dust at industrial scale.

The problem the researchers set out to solve is deceptively simple. On Earth, mining operations classify crushed ore by size using water or air, relying on cyclones, tables, and other fluid-driven devices. The Moon offers neither. Its hard vacuum and extreme temperature swings make any process that depends on a gas or liquid impractical, a challenge earlier NASA-era studies described as a formidable technical problem. Removing coarse fragments larger than a millimetre and fine particles below roughly 90 micrometres would dramatically improve the efficiency of oxygen production reactors, reducing parasitic heating, blockages, fouling, and power draw. Because mineralogy varies across size fractions, size classification can even serve as a first step in enriching valuable minerals. Vibrational segregation, the phenomenon popularly known as the Brazil Nut Effect, offers a dry, passive alternative: when a granular bed is shaken, larger particles tend to rise while smaller ones sink.

To test whether this effect works on real lunar material, the team used Apollo 15 sample 15601, a well-characterised soil collected near Hadley Rille at the base of the Apennine Front. The sample is chemically rich in iron, with bulk FeO exceeding 19 weight percent, and its median particle diameter sits near 89 micrometres, spanning an exceptionally broad range from about 8 micrometres to a full millimetre. Alongside this precious material, the researchers tested three simulants: JSC-1, a basaltic ash developed at NASA Johnson Space Center; LMS-1, a mare-derived simulant from Exolith Lab with the finest and narrowest size distribution of the three; and TUBS-M, a basalt-based simulant from TU Braunschweig designed to reproduce the density, flowability, and particle morphology of mare regolith. Each sample, roughly one gram, was sealed in a small glass vial and mounted on a shaker driven at 15, 30, and 70 Hertz for three minutes at a time.

The analytical workhorse of the study was X-ray micro-computed tomography, performed at the European Space Research and Technology Centre in the Netherlands. Scanning each vial before and after shaking at a resolution of 3.7 micrometres allowed the team to reconstruct the internal architecture of the granular bed in three dimensions. Individual particles were labelled with an 18-neighbour connectivity algorithm, each vial was divided into five axial layers, and particle size distributions were computed for every layer. Two complementary statistical tools then quantified the results: Rosin–Rammler functions characterised each distribution with a mean diameter and a spread parameter, while Jensen–Shannon Divergence provided a symmetric, bounded measure of how closely each simulant’s layered structure matched that of the Apollo soil, validated with permutation testing against a null model of random similarity.

The findings were strikingly material-dependent. The Apollo sample developed a reproducible three-layer structure: coarse particles accumulated at the top, a compacted intermediate zone formed beneath, and fine particles settled at the base. This stratigraphy sharpened with increasing frequency, with the coarse cap reaching a thickness of 1.13 millimetres after shaking at 70 Hertz. LMS-1 produced a similar three-layer architecture, though with weaker compaction, while JSC-1 displayed classical two-layer segregation with a sharp interface and particles larger than 1.8 millimetres surfacing at high frequency. TUBS-M, by contrast, largely resisted separation, with over 90 percent of the sample remaining static throughout the test sequence and only transient layering at 30 Hertz that dissipated by 70 Hertz.

To explain these differences, the researchers built two mechanistic models. The first, a particle-scale analysis, examined how aerodynamic drag inside the sealed vial affects grains of different sizes. The conclusion was unambiguous: particles below roughly 20 micrometres couple so strongly to the oscillating air column that they lose momentum almost immediately and never achieve ballistic flight, even when inertial launch thresholds are exceeded. Coarser grains, experiencing minimal damping, are consistently lofted. This drag-mediated trapping of fines, which would be absent in the lunar vacuum, explains the persistent fines-rich basal layers observed in both the Apollo sample and LMS-1, both of which contain substantial sub-20 micrometre fractions. Comparative trajectory calculations confirmed that in vacuum the fines would behave entirely differently.

The second model addressed bulk behaviour using a Janssen-style force balance, in which vertical loads in a granular column are partially transferred to the container walls through friction. Incorporating Mohr–Coulomb cohesion yielded critical mobilisation thresholds for each material, expressed as the ratio of vibrational to gravitational acceleration. LMS-1 was predicted to mobilise most readily, followed by TUBS-M, the Apollo sample, and JSC-1. Yet the experiments showed that crossing this threshold does not guarantee segregation. LMS-1, easiest to mobilise, segregated only weakly, while JSC-1, hardest to mobilise, separated efficiently at every frequency. The discrepancy underscores that bulk mobilisation is necessary but not sufficient: the formation of stable layers depends on particle size distribution breadth, fines content, and drag sensitivity interacting dynamically with the imposed vibration.

Perhaps the most consequential finding concerns compaction. All four samples densified under vibration, but the degree of compaction correlated cleanly with the breadth of each particle size distribution rather than with cohesion, friction angle, or bulk density. The Apollo sample, with the widest distribution, compacted most dramatically, its bed height falling by more than three millimetres, while LMS-1 and JSC-1, with narrower distributions, showed only modest densification. This has direct implications for laboratory practice: vibration is routinely used to prepare simulant beds at target densities before geotechnical testing, but the study shows that such preparation simultaneously drives size segregation and depth-dependent density gradients. Beds prepared this way may not be homogeneous, potentially skewing results in cone penetration, shear box, wheel mobility, and excavation tests that assume uniformity.

On the question of which simulant best mimics real lunar soil, the answer depends on the metric. TUBS-M delivered the closest statistical match to the Apollo sample in Jensen–Shannon Divergence terms, particularly in the middle and lower sub-volumes, despite being omitted from recent geotechnical benchmarking studies. LMS-1 more closely reproduced the visual and structural layering, including the jammed intermediate zone, while JSC-1 diverged significantly across all metrics, behaving more like a synthetic bimodal mixture than a true analogue. The authors argue that simulant selection for ISRU process development should therefore consider dynamic stratification behaviour alongside static geotechnical benchmarks, rather than relying on averaged properties alone.

The team is candid about the limitations of the work. Sealed vials under Earth gravity introduce air drag and boundary effects that would not exist on the Moon, and one-gram samples cannot fully capture bulk geotechnical behaviour, which recent research shows depends on both density and sample volume. Still, the framework they have built, combining micro-CT imaging, Rosin–Rammler modelling, and rigorous statistical divergence analysis, offers a practical and transferable method for evaluating segregation potential in granular materials. Future work could extend the approach to reduced-pressure and reduced-gravity environments, explore how cohesion and electrostatic charging evolve under lunar conditions, and develop three-dimensional metrics based on particle packing and contact networks. For now, the study provides the first direct comparison of vibrational segregation in Apollo regolith and its terrestrial stand-ins, and a sobering reminder that even something as simple as shaking dust is governed by a rich interplay of physics that must be understood before humanity can live off the lunar land.

Subject of Research: Vibration-induced size segregation of lunar regolith and simulants for in situ resource utilisation

Article Title: Vibration-induced size segregation of lunar regolith and simulants for in situ resource utilisation (ISRU)

Article References: Rasera, J. N., Salinas-Farran, L. E., Starr, S. O., Schein, V., Lomax, B., McDonald, F., Cilliers, J. J., & Hadler, K. (2025). Vibration-induced size segregation of lunar regolith and simulants for in situ resource utilisation (ISRU). Space and Planetary Resources, 1(1), Article 2. https://doi.org/10.1007/s44461-025-00004-5

Image Credits: AI Generated

DOI: 10.1007/s44461-025-00004-5

Keywords: lunar regolith, ISRU, vibrational segregation, Brazil Nut Effect, Apollo 15, lunar simulants, micro-CT, beneficiation, size classification, granular materials, TUBS-M, LMS-1