What comes out of the furnace
Sintered NdFeB is pressed from aligned powder and then densified in a vacuum furnace, and the part shrinks substantially in the process — on the order of 15 to 20% linearly, and not equally in every direction. Shrinkage differs along the alignment axis and across it, and it varies batch to batch with powder characteristics and furnace loading.
An as-sintered part is entirely usable where tolerance is loose, and specifying it that way is the cheapest magnet you can buy. Everything tighter is bought back one ground face at a time.
Sintered NdFeB behaves like a ceramic. It does not yield, it fractures, and it has low tensile strength — roughly 75–80 MPa in tension against several hundred in compression. Nothing in the finishing process can change that, and it constrains what features are practical: sharp internal corners, thin webs, small unsupported sections and threaded holes are all either impossible or expensive.
How the tight dimensions are actually achieved
You cannot turn or mill a sintered magnet in the ordinary sense. The material is too hard and too brittle; a cutting edge would chip it out rather than cut. Everything is abrasive.
| Process | Produces | Achievable | Notes |
|---|---|---|---|
| Surface grinding | Flats, thickness, parallelism | ±0.02 mm routinely | Diamond wheel, flood coolant; parts held on a fixture in batches |
| Centreless / OD grinding | Outside diameter, roundness | ±0.02 mm | Efficient on cylinders in volume |
| ID grinding | Bores in rings | ±0.03 mm | Slower; concentricity to OD is a separate and costlier requirement |
| Slicing | Thin parts from a block | Kerf loss per cut | Multi-wire or diamond blade; the economical route to thin sections |
| Lapping / polishing | Flatness, surface finish | ±0.005 mm, sub-micron finish | Specialist; sensor and optical-grade work only |
| Wire EDM | Profiles, arcs, slots | ±0.03 mm | Works because the alloy conducts; leaves a recast layer that must be removed |
| Core drilling | Holes | ±0.05 mm | Diamond core; countersinks are ground, never cut |
Every one of these is done wet, and for a reason beyond tool life. Dry grinding of NdFeB produces a fine metallic dust that is pyrophoric — it can ignite spontaneously in air, and the fire is a metal fire that water will not put out safely. Reputable producers grind under flood coolant and collect the swarf wet. It is a genuine hazard, and it is one of several reasons magnet finishing does not belong in a general machine shop.
Magnetization is the last step before or after coating, never before machining. A magnetized part collects its own grinding swarf instantly, cannot be held reliably on a fixture, and interferes with the machine. If a supplier quotes machining on magnetized stock, something is wrong with the process description.
What each tolerance band costs
The relationship between tolerance and price is not linear. It steps, because each band change either adds an operation, adds a slower operation, or moves the part from batch grinding into individual handling.
| Band | How it is reached | Relative cost | Reasonable for |
|---|---|---|---|
| ±0.1 mm and looser | As-sintered, no grinding | 1.0× | Holding, latching, anything with a compliant mount |
| ±0.05 mm | Ground on the toleranced faces | 1.15–1.3× | The standard commercial band; most applications |
| ±0.025 mm | Ground with tighter process control and gauging | 1.4–1.8× | Press fits, stacked assemblies, defined air gaps |
| ±0.013 mm | Precision grinding, sorted, often 100% inspected | 2–3× | Sensor targets, precision bearings, optical work |
| Tighter than that | Lapping, individual handling, selective assembly | 3× and up | Rare; question whether the assembly can absorb it instead |
Two multipliers sit on top of the band itself. The number of toleranced faces matters, because each distinct ground surface is a setup: a disc with a toleranced thickness only is cheaper than the same disc with thickness, diameter and parallelism all called out. And geometric callouts — perpendicularity, concentricity, true position — are more expensive than the equivalent dimensional tolerance, because they constrain relationships between faces that were ground in separate operations.
The tolerances and acceptance guide covers how these get written and verified at receiving, and the should-cost model puts the grinding operations into the wider cost build-up.
Pre-plate dimensions and the coating stack
A coated magnet is ground undersize so that the finished part, with coating, lands on the drawing dimension. This sounds obvious and is the source of a steady stream of drawing disputes.
applied on every coated dimension, both sides
t = 10–25 µm for Ni-Cu-Ni, 15–30 µm for epoxy, 5–15 µm for parylene
Coating thickness has its own tolerance, and that tolerance stacks on top of the ground tolerance. A part ground to ±0.02 mm and plated with a coating varying by ±0.008 mm per side arrives at roughly ±0.036 mm on the finished dimension. If you have specified ±0.025 mm on a plated part, you have specified something that the plating variation alone can consume.
A sharp-edged sintered magnet chips in handling, in shipping and during assembly, and every chip is a bare-substrate corrosion site under an otherwise sound coating. A 0.2 to 0.5 mm edge break costs almost nothing at the grinding stage and removes a large fraction of field failures. Specify it by default unless a sharp edge is functionally required.
Geometry that costs, and geometry that does not
Some shapes are close to free and some are expensive for reasons that have nothing to do with how complicated they look on screen.
| Feature | Cost | Why |
|---|---|---|
| Disc, block, ring | Lowest | Standard tooling, batch grinding, and often already in stock |
| Arc / segment | Moderate | Profile grinding or EDM; tooling is part-specific |
| Thin sections (< 1 mm) | High | Breakage in handling and grinding; yield falls sharply |
| High aspect ratio | High | Long thin parts distort in sintering and break in fixturing |
| Countersinks | Moderate | Ground with a form wheel; a separate operation |
| Small bores | High | Slow ID grinding; plating throws poorly; breakage risk |
| Sharp internal corners | Very high or impossible | Stress raisers in a brittle material; wheels have a radius |
| Threads in the magnet | Not practical | Use a cup, an insert or a bonded fastener instead |
| Non-standard sizes near a stocked one | Disproportionate | A 0.3 mm difference from a stocked part can mean tooling, MOQ and 10 weeks |
That last row is the most valuable one commercially. A meaningful proportion of custom magnet requests differ from a stocked standard by a fraction of a millimetre or by a coating choice carrying no functional requirement. Where the design can absorb it, moving to a stocked size collapses the tooling charge, the minimum order quantity and the lead time in a single change. It is the first question worth asking in any cost reduction review, and it is worth asking before the drawing is released rather than after.
Removing material does not damage the magnet's intrinsic properties. It does change the geometry, and geometry sets the permeance coefficient. Grinding a magnet thinner in the magnetization direction lowers its working point and can move it toward the knee, so a part that was comfortable at 5 mm may not be at 3 mm. Check the working point against the new geometry rather than assuming the grade still carries it.
What to put on the drawing
A magnet drawing that quotes accurately and consistently across suppliers carries a short, specific list. Most disputes trace back to something on this list being absent.
Two of these save disproportionate trouble. Magnetization direction stated with an arrow removes the single most common cause of an entire lot being wrong — a diametrically magnetized ring and an axially magnetized one look identical and behave nothing alike; the directions guide covers the conventions. And stating the maximum working temperature lets a supplier tell you before production that the grade will not survive it, rather than after.
The RFQ preparation guide covers the commercial side — volumes, release pattern, and what makes a quote comparable across suppliers.
Ask for the part as drawn and as it would be with commercial tolerances, in the same quotation. The comparison tells you exactly what your callouts are costing, and it converts a tolerance argument into an engineering decision with a number attached.