CNC Treatment Center: How the Machine Actually Holds a Tolerance

• 1
  Key takeaways
• 2
  What a CNC treatment center actually does to the part
• 3
  Where the error comes from before the tool touches metal
• 4
  Setup decisions that decide the first-article result
• 5
  What the machine can and cannot hold
• 6
  When a CNC treatment center is the wrong answer
• 7
  Which setup fits which part

A CNC treatment center is a machining center that holds the part and moves a rotating tool through it under numerical control. The word treatment here means the cutting operation applied to the workpiece, not heat treatment. On a vertical machine the spindle points down and the table carries the work; on a horizontal machine the spindle is sideways and the part sits on a rotary table. Both do the same job: remove material to a programmed path.

The machine does not decide the tolerance. It executes a path. What makes the path land inside ±0.005 mm is the combination of a rigid spindle, a tool that is measured before it cuts, a fixture that does not deflect, and a control that compensates for the ball screw and the thermal state of the machine. Take away any one of those and the same program will produce a different part.

For most parts we machine at GreatLight, the process runs on 3-axis, 4-axis or 5-axis equipment depending on how many faces need work and how many setups the part can tolerate. More faces reached in one setup means less re-clamping error, which matters more than raw spindle speed once you are chasing a two-thousandth of an inch.

So the practical question is never which machine is best. It is which setup gives the fewest chances for error between the first cut and the last measurement.

• 1
  Vertical centerBest for plates, housings and parts with a single dominant face.
• 2
  Horizontal centerBest for long parts and multi-face work where chips fall away cleanly.
• 3
  Five-axis centerBest when two or more angled faces must be cut without re-clamping.

Start with the spindle. A spindle that has been idle overnight is not the same spindle it will be after two hours of running. Bearing preload and housing temperature both change the axial position of the tool tip. On a finishing cut that is worth several micrometres. Most shops run a warm-up cycle of 15 to 30 minutes at moderate speed before critical work, and some run a spindle growth test on the first part of the shift.

Then the tool. A cutter has runout, and runout changes the effective diameter. An end mill with 0.01 mm of runout cuts a slot wider than its nominal size. So the operator measures the tool offset on a presetter or in the machine, and re-measures after the first article if the cut is off. Tool wear adds to this over a run: on stainless steel a cutter can lose 0.01 mm of diameter within a few hundred parts.

Then the fixture. If the part is held on three points and one clamp pushes it upward, the part springs when the cut releases the stress. Thin walls are the classic case. A ribbed wall 1.5 mm thick will deflect away from the cutter, so the finished wall is thin in the middle and thick at the ends. The fix is a support, a lower depth of cut, or a different clamping scheme, not a slower feed.

Finally the machine geometry. Squareness between axes, ball screw pitch error and backlash all shift the tool path. A control can compensate for these if the machine is calibrated, and a laser interferometer check plus ball bar test is how that calibration is verified. That is a maintenance task, not something the operator can tune away.

• 1
  Spindle growthWarm up 15–30 minutes; check the first article after warm-up.
• 2
  Tool runoutMeasure the offset; re-check after the first article.
• 3
  Fixture springSupport thin walls; reduce depth of cut before reducing feed.
• 4
  Axis geometryCalibrate with a ball bar and interferometer on a maintenance schedule.

The datum is the first decision. On a machined part the datum should be a surface the machine itself produced, not a raw casting face. That way the second operation is located from something that is already true to the first. If the drawing calls out a datum that exists only on the raw stock, the shop has to decide whether to machine it first or accept the stock variation. That conversation belongs in the DFM review, not on the shop floor.

Clamping force is the second decision. More force is not better. A vise closed too hard on a thin aluminum bracket will bend it, and the bend shows up as a flatness error after unclamping. For soft materials, low-pressure clamps or a vacuum plate hold the part flat without crushing it. For a part with a finished face, clamping on the finished face risks marking it, so a soft jaw or a sacrificial tab is used instead.

The third decision is how many setups. Every re-clamp adds a locating error, typically 0.01 to 0.03 mm depending on the fixture. A part that needs four faces machined can be done in four setups on a three-axis mill, or in one setup on a five-axis center. The five-axis route removes three re-clamp errors, which is often worth more than the machine's higher hourly rate on a tight-tolerance part.

The fourth decision is in-process probing. A probe that measures the part between operations catches drift before the whole batch is wrong. On a run of 500 parts we probe a sample at fixed intervals rather than every part, because probe time adds up and the machine is not cutting while it measures.

• 1
  Datum from machined facesLocate the second operation from a surface the machine already produced.
• 2
  Minimum clamping forceEnough to hold the part against cutting load, no more.
• 3
  Fewer setupsEach re-clamp adds 0.01–0.03 mm of locating error.
• 4
  Sampled probingProbe at fixed intervals on long runs, not every part.

A CNC treatment center holds a tolerance when the tolerance is wider than the sum of the errors in the chain. If the machine contributes 0.005 mm, the fixture 0.008 mm and thermal drift 0.004 mm, the practical floor is around 0.017 mm unless something is changed. That is why a shop that quotes ±0.005 mm on a thin-wall part in a soft material is either planning to change the setup or planning to rework.

Feature size matters too. A Ø2 mm end mill cutting a 20 mm deep pocket has a length-to-diameter ratio of 10 to 1, so it deflects. The wall of that pocket will taper. The same pocket cut with a Ø6 mm cutter has a ratio of 3.3 to 1 and holds size far better. If the drawing allows a corner radius of 3 mm instead of 1 mm, the shop can use the stiffer tool and the tolerance becomes realistic.

Surface finish follows the same logic. A fine finish of Ra 0.2–0.8 μm is achievable on a rigid setup with a sharp tool and a small stepover, but it costs time. A high finish of Ra 0.8–1.6 μm is the normal target for functional sealing faces and bearing bores. An as-machined finish of Ra 1.6–3.2 μm is fine for non-critical surfaces and is what you get when the priority is cycle time.

The material sets the boundary. Aluminum 6061 and 7075 cut cleanly and hold a good finish. Stainless 316L work-hardens, so the cutter must keep moving and the feed must not be too light. Titanium TC4 (Ti-6Al-4V) generates heat at the cutting edge and needs lower speed plus more coolant. Inconel is slower still. A tolerance that is easy in aluminum can be a problem in Inconel on the same machine.

• 1
  Stack the errorsMachine plus fixture plus thermal drift sets the real floor.
• 2
  Watch L/D ratioA 10:1 cutter deflects; a 3:1 cutter holds size.
• 3
  Match finish to functionSealing faces need Ra 0.8–1.6 μm; cosmetic faces do not.
• 4
  Material changes the rulesInconel and 316L are harder to hold than 6061.

The machine is the wrong answer when the feature is not a machined feature. A deep internal channel with a curved path, a part with an internal lattice, or a hollow shape with no access for a cutter is a job for additive manufacturing. A machined version would need the part split into pieces and joined, which adds cost and a joint that may not be allowed.

It is also the wrong answer when the quantity is high and the geometry is stable. At several thousand parts per year, die casting or injection molding spreads the tooling cost across enough units to beat machining. Machining wins on low volume, on tight tolerance, and on parts that will change before the next run. It loses when the design is frozen and the volume is large.

There is a middle case that catches people out: a part that is machined from a casting. The casting saves material and roughing time, but the as-cast surface is not a datum, and the casting may have porosity. If a pore opens into a sealing face, the part is scrap. For pressure-tight parts, the DFM review should flag where porosity is likely and whether a machined-from-solid version is safer.

Finally, the machine is the wrong answer when the tolerance is finer than the process can hold and the drawing does not need it. A ±0.005 mm callout on a mounting hole that only needs location, not fit, adds cost with no function. Relaxing it to ±0.05 mm can cut cycle time and inspection time without changing how the part works.

• 1
  Internal channelsUse additive; a cutter cannot reach a curved internal path.
• 2
  High stable volumeDie casting or molding beats machining once tooling is amortized.
• 3
  Machined-from-castingPorosity can scrap a pressure-tight part; review before committing.
• 4
  Over-tight calloutsA tolerance with no function still costs money to inspect.