Magnetic Density Separation Gets a Map of Where It Actually Works

Magnetic density separation has long promised a cleaner way to sort crushed waste and ores: instead of relying on chemicals or exhaustive grinding, particles are dropped into a magnetic liquid whose effective density changes from place to place under a magnetic field, letting each particle settle where its own density matches the liquid around it. A new study published in Results in Engineering tackles a deceptively simple question that has held the technology back: for a given machine, which combinations of magnet angle and magnetic-liquid magnetization actually produce a usable separation? The answer, the researchers show, is not a single optimal setting but a bounded operating region, and knowing its shape changes how the process should be run.

The technique, often abbreviated MDS, exploits a phenomenon sometimes called magneto-Archimedes separation. A non-uniform magnetic field acting on a liquid containing suspended magnetic nanoparticles creates a spatially varying effective density, so that a particle of a given density experiences an apparent buoyancy that pushes it toward the region where the liquid’s effective density equals its own. In the inclined-magnet configuration studied here, those density-matching regions correspond to different horizontal positions along the bottom of a tank. Particles of different densities therefore migrate to different collection zones, where splitters can divide them into distinct products. Because the method works on millimetre-sized particles, it could reject low-value material before energy-hungry fine grinding, a step that dominates the energy budget of many mineral and recycling flowsheets.

The two variables an operator can control are the magnetization of the magnetic liquid and the inclination angle of the magnet. Magnetization sets the range of effective densities the liquid can span, while the inclination angle changes how that density field intersects the tank and the collection system. The catch, identified by the team led by Hongli Su and Francesco Di Maio of Delft University of Technology, is that these variables act together. Some combinations push a predicted cut point outside the usable tank region entirely; others squeeze the width available for collecting the middle product below what the particle size and splitter geometry physically allow. Cranking the magnetization up or tilting the magnet further does not necessarily improve anything, because feasibility depends on the coupled effect of both settings on where the cut points land and how far apart they sit.

To map this behavior, the researchers built a quasi-static density-matching model calibrated against measured densities of the magnetic liquid. The liquid’s baseline density was measured at five magnetization values ranging from zero to 12,000 amperes per metre, yielding densities from 1,000 to 1,250 kilograms per cubic metre, and piecewise-linear interpolation filled in the gaps. An exponential expression for the magnetic field decay then converted each pair of angle and magnetization values into three predicted positions along the tank bottom: the high/middle cut, a representative middle position, and the middle/low cut. A condition was declared feasible only when all matching positions existed within the tank and the magnetic field’s coverage, retained the correct ordering, lay downstream of the feed zone, and left a middle-product window at least as wide as the largest particle plus the splitter thickness, a minimum of 4.5 millimetres for the 2 to 4 millimetre feed used in the study.

The resulting operating map, computed on a fine grid of more than 44,000 parameter combinations, revealed that roughly 81 percent of the explored angle-magnetization space is geometrically feasible, with the maximum predicted middle-product width reaching 14.2 centimetres. The map also exposed the failure modes at the edges of the domain. At a low inclination of 6 degrees with high magnetization, one cut point landed near the edge of the usable magnetic field while the other fell outside it altogether. At the same angle with weak magnetization, the liquid could not reach the effective density needed for the high-density cut at all. These are precisely the kinds of surprises an operator would otherwise discover only by trial and error, and the map makes them visible before any material is fed into the machine.

Sensitivity analyses gave the map a measure of robustness. Perturbing the three target cut densities by one percent changed the predicted collection width at the tested condition by less than 0.55 percent and the feasible-domain fraction by less than 0.19 percent. Refining the numerical grid changed the key metrics by less than a third of a percent, confirming that the 201-by-221 grid was more than adequate. The authors are careful to note the limits of this analysis: the device parameters held fixed in the model, such as the magnetic-field amplitude and magnet geometry, were not varied because experimentally supported uncertainty ranges for them were unavailable, so the sensitivity results apply only to the inputs actually tested.

The crucial test came at a representative operating condition of a 12-degree magnet angle and a magnetization of 5,300 amperes per metre, where the model predicted cut positions at 21.92 and 27.58 centimetres along the tank, leaving a middle-product window of 5.66 centimetres, more than twelve times the required minimum. Splitters were placed at these model-predicted positions, and three independent separation runs were performed on a 2 to 4 millimetre feed of rare-earth-element-bearing material. The high-grade fraction came out at a mean total rare-earth oxide concentration of 5.52 percent by weight, more than double the 2.31 percent in the feed, while the low-grade fraction was depleted to 1.17 percent. The middle fraction sat between at 1.75 percent, exactly the layered partitioning the density-matching picture predicts.

The authors are equally candid about what the experiments do and do not prove. Because the splitter positions were prescribed from the model rather than measured independently from the particle streams, the runs demonstrate density-based product partitioning at one feasible condition; they do not independently validate the predicted positions or the shape of the feasibility boundary across the wider operating space. The opacity of the magnetic liquid also prevented direct tracking of particle trajectories, so transit times and hydrodynamic effects remain unquantified, and the quasi-static model deliberately ignores drag, inertia, and particle-particle interactions. No conditions near the predicted boundary were tested, and energy consumption was not measured, leaving the much-touted grinding-energy savings for future work.

Those caveats point directly to the next steps the researchers consider most important: separation experiments near the predicted feasibility boundary, spatial mapping of the actual magnetic field to check the assumed exponential decay, additional calibration points for the liquid density, controlled measurements of where particles are actually fed, and tracer experiments in an optically accessible system. Each would tighten the assumptions on which the operating map rests and extend its credibility beyond the single tested condition.

The broader significance of the work lies in its reframing of how magnetic density separation should be operated. Rather than chasing a single mathematical optimum or simply maximizing magnetization, the framework identifies the entire region of practically collectable operating conditions, making the constraints explicit and transparent. For engineers, that means splitters can be positioned and operating points chosen with confidence about what the geometry allows, and the same map-based approach could be adapted to other separator configurations. As demand grows for recovering rare-earth elements and other critical metals from waste streams, tools that turn uncertain process tuning into principled operating windows may prove as valuable as the separators themselves.

Subject of Research: Operating-window identification for magnetic density separation of rare-earth-bearing particles

Article Title: Operating-window identification for magnetic density separation

Article References: Su, H., Tang, K., van Beek, M., Rem, P., Bo, Z., Bheemireddy, R., Ali, N., Wu, Y., & Di Maio, F. (2026). Operating-window identification for magnetic density separation. Results in Engineering, 32, Article 113344. https://doi.org/10.1016/j.rineng.2026.113344

Image Credits: AI Generated

DOI: Not provided

Keywords: magnetic density separation, magneto-Archimedes separation, rare-earth elements, recycling, mineral processing, operating window, magnetic liquids, magnet inclination angle, magnetization, density matching, splitter placement, waste sorting