Tolerances and finish: what can be ordered in raw form and what needs to be machined

Of all the conversations that go awry with a new client, half are the same: a drawing arrives with machining tolerances in all dimensions, and you have to explain that an LPBF part does not come off the machine like this, that it can end up like this, and that this has a price.

This article is to help you avoid that conversation. Or to have it once, using numbers.

What does "raw" mean?

part fresh off the machine, with supports removed and stress relief applied, but nothing else touched. This is the state in which everything comes out unless otherwise specified, and it's a perfectly usable state for many parts: tooling, housings, manifolds, functional prototypes, and assembly parts where nothing fits.

What it is not: a machined surface. And what it almost never is: a dimension of two hundredths.

Raw dimensional accuracy

The typical orders of magnitude of the process, not of a specific machine:

Dimension range	Standard deviation (as measured)
Up to 20 mm	±0.1 to ±0.2 mm
20-80 mm	±0.2 mm or ±0.2% of dimension
Over 80 mm	±0.2% of dimension, increasing with size
Flatness	Depends heavily on the surface and how it was held
Perpendicularity between faces	Modest: do not rely on it without machining

Translated into drawing terms: a rough LPBF part falls, more or less, within the range of a medium to coarse general tolerance class. It serves the purpose of a well-made casting, not the purpose of a grinding machine.

And there is an asymmetry that should be known: the dimensions in the layer plane are more reliable than those that grow vertically, because in the plane the optics rule —which is calibrated and checked— and vertically the accumulated layer thickness and the contraction rule.

Roughness: depends on which way you look at the face

This is what surprises those who come from machining the most: the same part has very different roughness depending on the face, and not because of a defect, but because of how it is manufactured.

Face	Ra roughness (indicative)	Why
Top, horizontal	Best of all	It's the last layer cast, with no powder on top
Vertical wall	Intermediate	Layer steps and partially adhered powder are visible
Upward sloping face	Intermediate, worsens when laid flat	Staircase effect
Downward sloping face	By far the worst	Casts against loose powder, not solid
Face with support removed	Marked, with residue	Trace remains where the support was

Hence the most cost-effective practical consequence: if a surface needs to be smooth, orient it upwards or vertically, never downwards. Changing the orientation of a part is free; polishing an underside is not.

And the one that prevents the most problems: roughness isn't just aesthetic. It's the point where a fatigue crack starts. In a part subjected to fatigue, the rough underside isn't a matter of finish, it's a structural issue — as explained in anisotropy and orientation.

What changes in each post-process

Ordered from cheapest to most expensive, with what they offer and what they really cost:

Process	What it achieves	What it costs	When to use it
Shot blasting	Matte and uniform surface, removes adhered powder	Very cheap, almost always included	Default in almost everything
Vibration or tribo-finishing	Rounds edges and significantly reduces roughness	Cheap, but charged by the hour	Batches of small parts
Machining	The dimension and finish you request	Tooling, clamping, and machine time	Functional faces only
Grinding/lapping	High precision	Expensive	Seats, guides, seals
Hand polishing	Mirror finish	Expensive and not repeatable	Aesthetics, molds, medical
Electropolishing	Very low roughness, even internally	Medium, requires a polishing tank and chemicals	Fluids, sanitary, medical
Hot isostatic pressing	Seals internal porosity, does not improve the exterior	Expensive	Fatigue-critical parts

Two money-saving tips.

Shot blasting doesn't fix a bad underside. It evens out the appearance; it doesn't change the geometry. If the underside had steps, it still has them.

None of these methods reach the inside of a narrow channel, except for electropolishing and, to some extent, abrasive flow methods. If your part requires a fine interior finish, that decision must be made in CAD, not in the shop floor.

The rule that determines the cost: mechanize only what needs it

Here's the money. A fully printed and machined part costs the same as the printing plus the machining, and it usually turns out worse than machining from a block: it's the fifth case of parts you shouldn't print.

The correct way to approach this is the opposite of how it's usually done: don't ask what tolerance you want, ask which faces make contact. There are usually three or four: two seats, a joint surface, and a bearing housing. The rest of the part doesn't touch anything and doesn't require any tolerance.

Written like this, a typical part becomes:

Area	What's required	How to achieve it
Seats and housings	The actual tolerance	Machining, with planned allowance
Joint or sealing surfaces	Flatness and finish	Milling or grinding
Functional threads	Actual thread	Drilled and tapped afterward
Everything else	General rough tolerance	Nothing. It comes out like this

How to design for cheap machining

Three things, and none of them cost money at the CAD:

Add allowance where you will be machining. On the order of half a millimeter to one millimeter per face, more for large parts or those prone to warping. Without allowance, if the part arrives even slightly short, it's no good.

Clamping references from the start. An LPBF part with organic geometry is a nightmare to clamp on a milling machine. Two flat faces, a couple of positioning holes, or appendages that are cut off later turn an impossible clamping into a five-minute setup. They are designed at the same time as the part, not afterward.

Machining occurs after heat treatment, not before. The workpiece moves when stress is relieved. Machining beforehand means machining a workpiece that will change shape — see deformation and stress relief.

What to write on the plan

A well-made additive part drawing says five things, and that's it for the surprises:

The general rough tolerance, explicitly accepted for everything not marked.
The functional dimensions, marked and with their actual tolerance, which are the dimensions to be machined.
The required finish per face, not for the entire part.
The manufacturing orientation, indicated on the model. It takes precedence over almost everything else.
The complete heat treatment, with its cycle.

A plan that asks for two hundredths across the entire part isn't asking for quality: it's asking for a budget that nobody is going to accept, or an argument.

How we approach it

When a geometry arrives, the first thing we look at isn't whether it's printable—it almost always is—but which faces do something. That determines the orientation, the supports are determined by the orientation, and the price is determined by the supports. In that order.

If you want a rough estimate before talking to anyone, the quote calculator will give you one. And if you'd prefer to see the part in person, that's what the open days are for.

What we're not going to do is accept a plan with impossible tolerances and think, 'We'll see what happens.' It's costly for everyone and, above all, it's discovered too late.

—

The complete dimensional rules are found in design rules for LPBF. Why the part moves and when it needs to be treated is explained in deformation and stress relief. And what is frozen when the part goes to a regulated sector is explained in how to qualify an LPBF part.