Volumetric Error Measurement for Multi-Axis Precision Stages: Diagonal Tests, Isolation and Compensation

What volumetric error actually measures

Every axis of a stage has six geometric error components: one linear along the axis of travel and five parasitic - two straightness, three angular. For a three-axis machine that is eighteen components plus three squareness terms between the axes. Each component alone may be small and well within its datasheet line item, but at a point away from the measurement axes the effects add as vectors. The volumetric error at a target point is the distance between the commanded point and the actual point in space, and it is the quantity that determines pattern placement, feature registration and tool-to-tool matching in real work.

This is why a stage can pass every axis-level acceptance test and still disappoint in use. The certificates describe behaviour along the axes; the process operates in the volume. Planning for volumetric error early also changes procurement: it forces the error budget, the Abbe arm lengths and the feedback selection to be discussed together instead of being discovered as separate problems later.

The body diagonal displacement test

In the ISO 230-6 method, a laser interferometer measures displacement along a body diagonal of the workspace while all axes move simultaneously along it. Four diagonals are measured - the two on each pair of opposing faces - in both directions, typically with several cycles per direction. The paired diagonals are what give the test its diagnostic power: comparing them separates squareness from straightness effects, because a squareness error changes the lengths of opposing diagonals with opposite signs, while many other components appear with the same sign.

The test has limits that a buyer should understand before writing it into a specification. A single diagonal measurement cannot decompose the error into its components, and some error combinations remain ambiguous even with all four diagonals. The diagonal test is therefore a strong verification instrument - it proves how large the combined error is, before and after compensation - but not a complete diagnostic instrument on its own.

Isolating the sources behind the diagonals

Attribution starts with single-axis data. Positioning error along each axis is measured with the laser in the conventional axis-aligned setup; straightness and the angular components are measured on the same axes with appropriate optics. When these measurements are available, the diagonal results can be reproduced by the error model within its uncertainty, and the residuals become a quality check on the measurements instead of a source of argument. If the model cannot reproduce the diagonals, something in either dataset is wrong - a mounting issue, an Abbe offset that was ignored, or an environment that drifted during the run.

Two practical confounders deserve explicit attention. Random error - repeatability, vibration, servo noise - appears in the diagonals exactly as it appears everywhere else, and no compensation model can remove it; the repeated-run scatter of the diagonal test sets the floor that systematic compensation can reach. Thermal drift mimics a slow systematic error and is best excluded by repeating one diagonal at intervals and logging temperatures, so that a trend attributable to warm-up is not baked into the compensation table.

• Measure single-axis positioning, straightness and angular errors before interpreting the diagonals
• Use the paired-diagonal sign differences to separate squareness effects
• Quantify repeatability scatter first; it defines the achievable residual

Building a compensation model that holds

The output of isolation is a parametric model: a small set of geometric parameters - scale factors, straightness polynomials, angular error curves, squareness values - that predict the volumetric error anywhere in the workspace. Loading these parameters into the controller, or applying them through the position feedback loop, corrects the systematic part of the error. A parametric model is strongly preferable to a lookup table that interpolates only the measured test points, because it behaves sensibly between and beyond the points and can be re-fitted when the machine is re-measured.

The quality of the feedback underneath the compensation matters directly. Compensation corrects what the error model describes, based on the encoder readings the controller sees; if the scale itself has uncorrected cyclic or installation errors, the compensation inherits them. This is one of the reasons encoder accuracy, scale installation quality and the error model should be reviewed as one package rather than by separate teams in separate documents.

Verifying and re-verifying

The acceptance logic is straightforward: run the diagonal test before compensation, record it, apply the model, and run the same test again under the same conditions. The residual volumetric error, together with the repeatability scatter, becomes the acceptance number. What matters for the future is that the measurement is repeatable - same target points, same instrument, same warm-up state - so that the next verification, months later, can be compared against the original without the comparison being dominated by setup differences.

Stages change with use and environment. A sensible maintenance plan re-runs the diagonals at defined intervals or after events that plausibly move the geometry - a collision, a relocation, a major service - and re-fits the model when the residual grows. Treated this way, volumetric error stops being a one-time acceptance hurdle and becomes a monitored property of the machine, which is exactly what processes with tight placement budgets need from their motion supplier.