Accuracy and Reliability of Machine Tools: What Engineers Should Verify

• 1
  Geometric accuracy: where the machine actually is
• 2
  Thermal behavior: the accuracy and reliability of machine tools drift with heat
• 3
  Servo and control response: following error under load
• 4
  Tool, fixture and measurement: the loop outside the machine
• 5
  Capability, Cp and Cpk: what the numbers actually say
• 6
  Error source vs. typical size and the feature it ruins

Geometric accuracy is the static truth of the machine. It covers squareness between axes, straightness of each travel, parallelism of the spindle to the Z axis, and the real position of the rotary table center. A machine can hold ±0.005 mm on a test cut and still be squareness-limited on a long part, because squareness error grows with distance.

Think in microns per 300 mm. A 10 μm squareness error between X and Z shows up as 10 μm of taper over 300 mm of travel, and 33 μm over 1,000 mm. On a 4,000 mm part the same error becomes 133 μm. That is the number that decides whether a long weldment can be bored in one setup or has to be re-fixtured.

Linear scale feedback changes the picture. A machine with glass scales on X, Y and Z compensates ballscrew pitch error and thermal growth in the screw, so positioning accuracy stays near the scale rather than the screw. Machines with rotary encoders only are more sensitive to screw heating and to backlash after a crash.

Check geometry with a ballbar or a laser interferometer, not with a dial indicator on a vise. Ballbar circularity data also shows servo mismatch and backlash, which are dynamic problems that a static check will miss.

• 1
  SquarenessMeasured between axes, quoted in μm per 300 mm.
• 2
  StraightnessDeviation of one axis travel from a true line.
• 3
  Scale vs encoderScales reduce sensitivity to screw growth.
• 4
  Rotary centerOff-center tables add a sine error on the part.

A machine tool is a structure that changes size as it warms. Spindle bearings, ballscrews, ways and the bed all heat at different rates, and the whole loop settles over 30 to 90 minutes. Until it settles, part dimensions drift. This is why the first part off a cold machine and the tenth part often differ.

The dominant term is usually spindle growth in Z, often 10 to 30 μm from cold to warm on an air-cooled spindle. Ballscrew growth adds position error along the travel, worst at the far end from the thrust bearing. A 1,000 mm screw at 5 °C rise moves roughly 60 μm if it is not pre-tensioned or scale-compensated.

You cannot remove heat, so you manage it. Warm-up cycles run the spindle and axes through the working envelope before the first cut. Coolant temperature is held close to ambient. Oil-air lubrication replaces grease in high-speed spindles. Some shops run a spindle chiller on aluminum jobs where Ra 0.2–0.8 μm and ±0.005 mm both matter.

For tight work, we check the first article, let the machine run, and check again after 30 minutes. If the drift exceeds a third of the tolerance, the process is not stable yet and the offset is adjusted before the run continues.

• 1
  Warm-upRun axes and spindle through the envelope before cutting.
• 2
  Coolant temperatureKeep it near ambient to limit gradient.
• 3
  Spindle growthTypically 10–30 μm in Z from cold to warm.
• 4
  Screw growthWorst at the end farthest from the thrust bearing.

Static geometry does not cut a part. The servo loop has to follow the commanded path while the cutter pushes back. Following error is the gap between commanded and actual position during motion, and it is proportional to feed rate and inversely proportional to loop gain. Push the feed and the error grows.

This shows up as corner rounding, overshoot at reversals, and chatter marks on walls. On a circle, X and Y servos with mismatched gain produce an ellipse. A ballbar test reports this as circularity deviation and often as a figure-eight trace. The fix is gain matching, feedforward tuning, or simply a lower feed on the finishing pass.

Mechanical backlash is a separate term. Any lost motion at reversal appears directly on the part as a step. On a 12 mm end mill taking a 0.3 mm finishing pass, 5 μm of backlash is visible on a wall under light. Ballscrew preload, thrust bearing condition and way clearance all set this number.

Chatter is the reliability problem hiding inside accuracy. A tool that chatters cuts oversize and destroys inserts. Stability lobes depend on spindle speed, tool overhang and holder stiffness. Shortening overhang from 60 mm to 40 mm often moves the stable window enough to finish a deep pocket.

• 1
  Following errorGrows with feed rate, shrinks with loop gain.
• 2
  Gain mismatchTurns a circle into an ellipse.
• 3
  BacklashShows as a step at every axis reversal.
• 4
  ChatterCut oversize and poor finish at the same time.

A machine holding ±0.005 mm can still ship a bad part because of the tool, the fixture or the gauge. Tool runout of 10 μm on a 6 mm end mill cuts 10 μm off the wall on one side. Thermal growth of a solid carbide tool at 12,000 rpm adds axial length, which shows up on a Z-depth feature.

Fixtures decide repeatability. A three-point locating scheme with a clamp directly over the support is stiff. Clamping over air bends the part, and the spring-back after unclamping shows as flatness error. For thin walls, we cut with light radial depth, 0.1–0.2 mm, and support the back with a low-melt fixture or a soft jaw.

Measurement closes the loop. A CMM at 20 °C and a caliper on the shop floor at 28 °C disagree on an aluminum part of 200 mm by roughly 20 μm, because aluminum moves about 23 μm per meter per degree. That is four times the tolerance. Gauge and part have to be at the same temperature before the number means anything.

This is why we inspect 100% before shipment and issue reports on request. Raw material check, in-process monitoring and final inspection catch different failure modes, and no single one of them covers the others.

• 1
  Tool runoutBecomes a direct wall offset on the part.
• 2
  ClampingClamp over the support, not over air.
• 3
  Thermal mismatchAluminum moves ~23 μm/m per °C.
• 4
  InspectionMaterial, in-process and final checks catch different faults.

Accuracy is closeness to nominal. Precision is spread. A process can be accurate and imprecise, or precise and offset. Cp measures spread against the tolerance band; Cpk measures spread plus how far the mean sits from the center. A Cpk of 1.33 means the process has room; a Cpk of 1.0 means the tails are already touching the limits.

Tolerance is not capability. If a print says ±0.05 mm and the process spread is 0.08 mm, the machine is not capable of the print even though it is accurate on average. Running 30 parts and computing the spread tells you more than checking one part three times.

For one-off and prototype work, capability studies are not practical. There we control the first article, hold the process, and confirm at the end. For the 10,000+ part runs where we hold ±0.005 mm, we monitor the spread and adjust offsets before the trend reaches the limit.

A single good part proves nothing about reliability. Reliability is the probability that the next part is good, and it only comes from a stable process that has been measured over time.

• 1
  CpSpread against the tolerance band only.
• 2
  CpkSpread plus mean offset from center.
• 3
  Cpk 1.33A common minimum for controlled production.
• 4
  One partNot a capability study.