Below about four microns, ferrous wear debris stops behaving like a solid and starts behaving like part of the fluid. Gravity does not remove it. Settling tanks do not remove it. Centrifuges do not remove it. Media filters cannot remove it without stripping out the barite you paid for. It circulates, and it keeps cutting. This is the physics — and the one technology that beats it.
Drop steel wear debris into drilling mud and watch. The coarse swarf a ditch magnet is built for hits the bottom in minutes. The fine fraction — the abrasive that actually cuts your pumps, motors and MWD tools — is still hanging in the fluid weeks later. Press play and run the clock forward.
Compressed view raises every vertical rate to the power 0.75, so all five classes move visibly in one window. Real settling spans 40,000:1 — at any single playback speed the coarse fraction would hit the floor before the fines had twitched. Ordering, ratios and Brownian dominance are all preserved; only the spread is squeezed. The true full-drop times in the readout below are always the real Stokes values. Switch to True Stokes for uncompressed physics.
Settling speed is governed by Stokes' law, and it scales with the square of particle diameter. Halve the particle and it falls four times slower.
Cut a 100 µm chip down to 2 µm and you have not made it 50 times slower. You have made it 2,500 times slower. Twenty-eight minutes becomes forty-eight days.
1 · Brownian motion. Below roughly 1 µm, random molecular bombardment jostles the particle further each second than gravity pulls it down. It performs a random walk, not a descent.
2 · Gel strength. A 2 µm steel particle presses down on the mud at about 0.09 Pa. A typical water-based mud gels at 2.4–14 Pa. The particle is physically not heavy enough to break the gel. It is held, not falling slowly.
3 · Circulation. Its Stokes number is around 6 × 10⁻⁸. It has no inertia of its own. Wherever the fluid goes, it goes — and the fluid goes round the well.
Solids-control equipment was designed to remove drilled cuttings, not machined wear debris. The train is a coarse-to-fine cascade — shakers screen first, because hydrocyclones plug and erode on unscreened flowline returns, and everything downstream depends on the shakers having gone before it. Follow one batch of fluid through and watch what each stage takes out — and what walks straight through to the pump suction.
All counts are normalised to 1,000 particles entering the flowline, so the figures mean something regardless of how long the simulation has run. Every particle arriving at stage 8 has already survived all seven conventional stages — that population is what reaches your pump on a rig with no CleanMud fitted, so the same run gives both answers at once. Hit CleanMud: IN SERVICE to bypass the unit and watch the outlet fill.
That is why the first two cards below disagree so violently. Judge the train on mass and it scores 99.8%. Judge it on the sub-4-micron count that destroys pumps, motors and MWD tools, and it scores about 12%. A rig can be removing 99.8% of its ferrous debris and still be circulating essentially all of the damage.
Watch the two magnet stages. Both do real work — the header box takes about three-quarters of the coarse swarf, and the ditch magnets downstream of the shakers pick up much of the coarse ferrous that got through the screens. Neither touches the fines: across everything reaching it, the ditch magnet set removes roughly 1% of the sub-4-micron fraction. Two sets of conventional magnets on the rig, and the material that actually cuts your pumps and tools passes both.
| # | Stage | Cut point | 100 µm | 40 µm | 4 µm | 2 µm | 0.5 µm |
|---|---|---|---|---|---|---|---|
| 1 | Header box / possum belly — ditch magnet | coarse ferrous | Cut | Some | Pass | Pass | Pass |
| 2 | Shale shakers (API 200 mesh) | ~74 µm | Cut | Some | Pass | Pass | Pass |
| 3 | Ditch magnets — in the ditch, downstream of shakers | coarse ferrous | Cut | Some | Pass | Pass | Pass |
| 4 | Sand trap / settling | gravity only | Some | Pass | Pass | Pass | Pass |
| 5 | Desander (hydrocyclones > 6 in) | ~45 µm | Cut | Some | Pass | Pass | Pass |
| 6 | Desilter (hydrocyclones < 6 in) | ~20 µm | Cut | Cut | Pass | Pass | Pass |
| 7 | Decanting centrifuge | ~5–7 µm | Cut | Cut | Some | Pass | Pass |
| 8 | CleanMud high-gradient magnetic filtration | < 0.5 µm ferrous | Cut | Cut | Cut | Cut | Cut |
"So put a finer filter on it."
You cannot. Barite — the weighting agent you are paying for and depending on for well control — is 2 to 74 microns. A mechanical filter fine enough to catch 2 µm wear debris strips the barite out of your mud at the same time. It blinds off in minutes, the differential pressure runs away, and your mud weight walks off spec.
Mechanical filtration cannot tell the difference between the barite you want and the steel you do not. It only sees size — and at this size, they are the same.
A magnetic field does not sort by size. It sorts by magnetic susceptibility.
Steel wear debris is ferromagnetic. Barite is not — barium sulphate is effectively non-magnetic. Bentonite and drilled solids are not. The field reaches into the fluid, takes hold of the ferrous abrasive, and lets everything you actually want flow straight past.
Zero differential pressure. Zero media to blind off. Zero barite loss. That is not a better filter. It is a different physical principle.
Left: a ditch magnet. The fluid goes past once, the coarse swarf sticks, the fines carry on. Right: a high-gradient magnetic filtration unit running as a continuous recirculating slip-stream. Modest capture per pass, compounded over hundreds of passes a shift. Watch the two cleanliness curves separate.
No single pass through any device is 100% efficient. What matters is how many passes you get.
At a modest 20% capture per pass: five passes gets you to 67%, ten passes to 89%, twenty passes to 99%. A kidney loop on a live active system delivers those passes every shift, automatically, while the well is drilling.
A ditch magnet gets one pass. Ever. That is not a tuning problem — it is the architecture.
Wear debris is autocatalytic. Fine abrasive circulating through a pump generates more fine abrasive, which circulates and generates more. Left in the fluid, contamination compounds.
Remove it continuously and the loop inverts: less abrasive means less wear means less abrasive. That is where 30%+ NPT reduction and ~52% longer component life come from. Not from a stronger magnet — from finally taking the fines out.
Steel ρ = 7,800 kg/m³ · mud ρ = 1,200 kg/m³ (10 ppg) · plastic viscosity 30 cP · 60 °C. Brownian displacement from the Stokes–Einstein relation, RMS over one second. The final column is the ratio of random Brownian wander to gravitational fall — above 1.0, the particle is being shaken around faster than it is sinking.
| Diameter | Settling velocity | Time to fall 2 m | Brownian (RMS/s) | Brownian ÷ gravity | Stokes number |
|---|---|---|---|---|---|
| 100 µm | 1,199 µm/s | 28 minutes | 0.018 µm | 0.00 | 1.4 × 10⁻⁴ |
| 40 µm | 192 µm/s | 2.9 hours | 0.029 µm | 0.00 | 2.3 × 10⁻⁵ |
| 25 µm | 74.9 µm/s | 7.4 hours | 0.036 µm | 0.00 | 9.0 × 10⁻⁶ |
| 10 µm | 12.0 µm/s | 1.9 days | 0.057 µm | 0.00 | 1.4 × 10⁻⁶ |
| 4 µm | 1.92 µm/s | 12.1 days | 0.090 µm | 0.05 | 2.3 × 10⁻⁷ |
| 2 µm | 0.48 µm/s | 48.3 days | 0.128 µm | 0.27 | 5.8 × 10⁻⁸ |
| 1 µm | 0.12 µm/s | 193 days | 0.180 µm | 1.50 | 1.4 × 10⁻⁸ |
| 0.5 µm | 0.030 µm/s | 2.1 years | 0.255 µm | 8.51 | 3.6 × 10⁻⁹ |
Buoyant weight spread over the particle's cross-section, versus the shear stress a mud gel resists.
| Diameter | Buoyant stress | vs 2.4–14 Pa gel |
|---|---|---|
| 100 µm | 4.32 Pa | Marginal |
| 40 µm | 1.73 Pa | Held |
| 10 µm | 0.43 Pa | Held |
| 2 µm | 0.086 Pa | Held |
| 0.5 µm | 0.022 Pa | Held |
A 2 µm particle exerts roughly 1/100th of the stress needed to move through a gelled mud. When circulation stops, it does not slowly sink. It is locked in place.
Saturation magnetisation 1.7 × 10⁶ A/m. Migration velocity from force balance against Stokes drag.
| Field gradient | Force vs gravity | Migration velocity |
|---|---|---|
| 500 T/m | 13,100× | 6.3 mm/s |
| 1,000 T/m | 26,300× | 12.6 mm/s |
| 5,000 T/m | 131,000× | 63.0 mm/s |
| — gravity — | 1× | 0.00048 mm/s |
The magnetic force scales with particle volume, so it does fall away for the very finest debris — which is exactly why high gradient and many passes both matter. Field strength alone is not the answer; field gradient plus a recirculating loop is.
Cumulative removal = 1 − (1 − p)ⁿ, for per-pass efficiency p over n passes.
| Per-pass efficiency | 1 pass | 5 | 10 | 20 | 50 | 100 |
|---|---|---|---|---|---|---|
| 10% | 10.0% | 41.0% | 65.1% | 87.8% | 99.5% | 99.997% |
| 20% | 20.0% | 67.2% | 89.3% | 98.8% | 99.999% | ~100% |
| 30% | 30.0% | 83.2% | 97.2% | 99.9% | ~100% | ~100% |
A ditch magnet lives permanently in the "1 pass" column. That is the entire difference between the two technologies, expressed as a single number.