CNC Assembly Guide: How Machine Build Sets Part Tolerance

• 1
  What a CNC assembly guide actually covers
• 2
  Bed leveling and frame geometry set the baseline
• 3
  Rails, ball screws and preload: where repeatability comes from
• 4
  Spindle seating, runout and thermal growth
• 5
  Squareness, calibration and the metrology loop
• 6
  What changes on a five-axis machine
• 7
  Assembly checks and what they mean for your part

A CNC assembly guide is not a manual for bolting a machine together. It is the chain of decisions that decide how much of your drawing tolerance survives contact with the machine: how the bed is leveled and preloaded, how the linear rails are referenced to the same datum, how the ball screw nut is aligned to the guideway, and how the spindle is seated in its housing. Errors at any link show up at the tool tip, multiplied by axis travel.

The order matters more than the tools. Geometry is set first because nothing downstream can correct a twisted bed. Rails and screws come next, then the spindle, then the metrology loop that measures the machine and writes compensation tables. Skip a step and the next one hides the fault instead of fixing it.

This page is written for two readers. One is the engineer who has to explain why a part drifted 0.02 mm between the first article and the production run. The other is the buyer comparing suppliers and wondering whether the difference in price reflects real machine condition or just marketing. Both need the same answer: which build details are measurable, and which are claims.

Nothing here is a substitute for an inspection report on the specific machine that will run your part. Treat it as a checklist of what to ask about, and what to look at when you visit a shop floor.

Every machine starts with a frame and a bed. Cast iron or welded steel, the casting is stress-relieved before it is machined, because residual stress released after final cutting will move the guideway surfaces over months. A frame that was never stress-relieved can measure flat on delivery and twisted a year later.

Leveling is done with precision levels and, on larger frames, with a granite square and dial indicators read across the full travel. A typical target for a machining center bed is 0.02 mm per meter in both directions, checked before any rail is mounted. On our larger gantry-style platforms, travel reaches 4,000 × 400 × 150 mm, so a small angular error at one end becomes a visible error at the other.

Bolting the frame to the floor is part of geometry, not a finishing touch. Uneven floor contact or over-torqued anchor bolts bend the bed. We torque anchors in a cross pattern in stages, then re-check level after 24 hours, because concrete and cast iron settle at different rates.

The practical meaning for your part: a bed that is out of flat by 0.03 mm over the work envelope shows up as a taper or a step in long parts, and no amount of tool offset will remove it. Short parts may pass. Long parts will not.

Linear rails are mounted against a reference edge machined into the bed. The rail is pushed against that edge, not centered by eye, then clamped progressively from the middle outward. Clamping from one end traps a wave of stress that shows up as a tight spot mid-travel.

Preload class decides stiffness and friction. A light preload rail is easy to move but deflects under cutting load. A medium or heavy preload rail is stiffer but generates more heat and wears faster if lubrication is marginal. For a machine cutting aluminum at moderate load, medium preload is the common choice. For heavy steel roughing, heavier preload with a larger rail size.

Ball screw alignment is the second half. The nut must be concentric with the guideway within a few hundredths of a millimeter over the screw length, and the bearing support at each end must be angular-contact, preloaded, and aligned to the same datum. A screw that is parallel but offset by 0.05 mm will bind at the ends and run free in the middle.

Backlash is measured, not assumed. We sweep each axis with a dial indicator against a known stop and record reversal error. On a well-assembled axis, backlash after compensation sits inside the positioning tolerance, and the compensation value is written into the control rather than left as a mechanical gap.

The spindle is the last mechanical link before the tool, so its seating error passes straight into the cut. The housing bore is checked for roundness and taper before the spindle cartridge goes in. A burr or a chip left in the bore tilts the cartridge, and the resulting runout grows with distance from the nose.

Runout is measured at the taper with a test bar, both near the nose and 300 mm out, because angular error shows up as a difference between the two readings. We also check spindle taper contact with bluing: a good fit shows contact over most of the taper, and a poor fit shows a ring near one end, which lets the tool shift under load.

Heat is the quiet variable. A spindle at 12,000 rpm warms and grows along its axis, pushing the tool deeper into the part. Machines that hold tight tolerance over a long run either warm up for a fixed period before the first cut or use spindle growth compensation. Both are assembly decisions, not operator habits.

On a five-axis head, the same logic applies to the rotary axes. The C-axis table and the trunnion must share a center point, and the tool center point offset is measured and written into the kinematic model. If that offset is wrong, every rotary move drags the tool off the intended path.

Squareness between axes is checked with a granite square and indicators, or with a laser interferometer on larger machines. A squareness error of 0.01 mm over 300 mm sounds small until you cut a 500 mm part with two setups, where the error doubles and the faces no longer meet cleanly.

Volumetric calibration goes further than single-axis checks. It maps positioning error across the whole work envelope, including straightness and angular deviations, then writes compensation tables into the control. This is what makes a machine accurate at the corners of its travel, not just near home.

The environment during this work has to be stable. We assemble and calibrate in a temperature-controlled area held at 20 °C ± 1 °C, because a 5 °C swing moves a 1,000 mm steel scale by roughly 0.06 mm. Measure in a hot afternoon and the compensation table is wrong by morning.

After calibration, we run a test cut and measure it. The test part covers circular interpolation, a long straight pass and a rotary move, so squareness, backlash and spindle error all appear in one result. A machine that passes the test cut is ready to quote against tight drawings.

A three-axis machine has three linear errors to manage. A five-axis machine adds two rotary axes, and each rotary error is amplified by the distance from the rotary center to the tool tip. That distance can be 300 mm or more on a trunnion machine, so a 0.005° angular error becomes a 0.026 mm position error at the tool.

The kinematic model is the heart of it. The control needs the actual position of the rotary centers, the tool center point offset, and the geometry linking them. These values are measured during assembly, not copied from a drawing. If they are wrong, simultaneous five-axis motion produces a smooth-looking path that is off by a constant amount.

Rotary axis preload, encoder mounting and brake function all get checked at this stage. A rotary table with a loose encoder coupling will pass a static test and fail under reversing load. We run each rotary axis through repeated reversals and log the following error before signing off.

For parts with tight true-position callouts, this is where the tolerance is won or lost. A well-calibrated five-axis machine can hold ±0.005 mm on a contoured surface in one setup, which removes the stack-up error of multiple fixtures. A poorly calibrated one cannot, no matter how good the CAM toolpath is.