CTE Matching for Optical Encoder Scales: When the Scale and the Machine Expand Differently

Why CTE mismatch is a first-order error, not a footnote

A linear encoder does not measure the position of the tool; it measures the position of the readhead relative to the scale, so whatever happens to the scale is indistinguishable from real motion. When the scale substrate and the surface beneath it have different expansion coefficients, a temperature change creates a strain the scale cannot avoid absorbing: bonded scales stretch or compress, clamped scales bow between their fixing points. Both effects convert directly into position error, scaling linearly with temperature change and differential CTE—no servo, filter or interpolation setting can remove them, because the encoder itself is the source.

The magnitude is easy to underestimate because the numbers look small. A differential of 10 ppm/K over 1 K on a 300 mm scale is 3 micrometers, comfortably inside most accuracy grades. But workshops see 5 to 15 K between a cold morning start and an afternoon fully soaked, and the error accumulates along the whole measuring length: 600 mm of scale at 12 ppm/K over 10 K is 72 micrometers of disagreement between scale and substrate. Against a typical glass scale grade of ±5 micrometers, the mounting decision dominates installed accuracy by an order of magnitude.

The choice only works if it is made against the real mounting surface. Glass grades, steel tapes, Invar and glass ceramics exist precisely because the substrate can be matched to the application; the scale material guide covers each substrate family in detail, and the thermal-drift guide explains how scale behavior combines with the machine's own warm-up into the axis's total drift budget.

How the error shows up: strain, bow and gradients

Mismatched expansion attacks the scale in three ways, each with a different fingerprint. The first is uniform strain: if the scale is continuously bonded to a faster-expanding substrate, it is put into tension proportionally to the differential and its graduations move apart. The encoder reads this as a true length change—the graduations really did move—and the error is proportional to position along the axis, so it masquerades as a linearity error that worsens toward stroke ends.

The second is bowing. Held at discrete points—clips at the ends, or fasteners at intervals—the unconstrained portions arch as the differential grows. A bowed scale no longer lies in the readhead's measuring plane: the readhead sees graduations at varying angle and height, which shortens the effective pitch and adds a sinusoidal error component tied to the bow geometry. Severe bow also perturbs air-gap and signal balance, so bad cases surface first as degraded signal quality rather than as an obvious position error.

The third is the gradient case, which needs no mismatch at all: a heat source on one side of the machine—motor, spindle, sunload—creates a temperature difference between scale and substrate even when their CTEs are identical. Gradients are transient and asymmetric, which makes them hardest to compensate and the most common reason a machine reproduces perfectly at steady state but scatters after startup. Telling the three mechanisms apart requires correlating position error with temperature over time, not a static laser measurement at soak.

• Strain: error proportional to axis position, looks like linearity error
• Bow: sinusoidal component at clip pitch, may degrade signal quality first
• Gradient: transient, asymmetric drift after startup even with matched CTEs

Choosing the substrate and mounting surface as a matched pair

The matching decision is made between two materials, not one. Write down what the scale actually bonds to—not the material the machine datasheet names, but the specific casting, granite rail or aluminum bridge at that location—then compute the differential over the working range. A practical target for precision axes is keeping differential expansion below 2 to 3 micrometers over the largest expected temperature step at full measuring length; if the target cannot be met with the substrate at hand, the scale material changes before the mounting method does.

The common combinations each have a known character, summarized in the table above. Glass on steel is workable because glass grades spanning roughly 3 to 13 ppm/K exist, and a grade near 11 to 12 ppm/K closely tracks cast steel—this is why most machine-tool scales are glass on steel, and why the glass grade should be stated at ordering rather than left to the supplier's default. Glass on aluminum is the worst common pair: aluminum expands about twice as fast as most glasses and responds quickly, so steel-tape scales, which follow the substrate both structurally and thermally, are usually the better answer. Granite changes slowly, convenient for stability but quick to hide mismatch until the room has fully drifted. Invar and glass-ceramic scales shift the error budget onto the machine structure instead—at a cost easy to justify on a 200 mm nanometer-class axis and hard to justify across a 4 m travel.

Whatever pair is chosen, record the calculation in the axis file: substrate material and CTE, scale substrate and CTE, working range, and the resulting differential in micrometers. This one number, written down before ordering, is what separates a defensible design from an accident that happened to pass acceptance.

Mounting methods that let the scale follow its substrate

The mounting method's job is to transfer position, not force. Elastic adhesive tape—the default for glass scales up to roughly 2 to 4 meters depending on manufacturer—couples the scale to the substrate through a thin compliant layer that absorbs the differential shear continuously along the length, so the scale follows its substrate without building up tension. Flexible two-part epoxy serves the same purpose where cleanliness or acceleration demands it, but it must be a low-modulus, gap-controlled bond: a rigid bond turns the mismatch back into strain. Both routes require honest surface preparation and the specified bond-line thickness, because the compliance lives in a layer only tenths of a millimeter thick.

Mechanical clip mounting is the alternative for long scales and adhesive-hostile environments. Done correctly, clips hold the scale at defined points and leave the segments between them free; done incorrectly—every hole used, fasteners hard-torqued, no slip allowance—the scale becomes a constrained beam and every kelvin of mismatch turns into bow. Clip spacing follows the manufacturer's tables precisely because those spacings bound the bow amplitude; deviation is a design change, not an assembly liberty. Steel-tape scales invert the logic: they are tensioned so the tape carries the differential structurally, and the correct tension specification replaces the compliance discussion entirely.

Installation temperature completes the method. Substrate and scale should be thermally soaked together at the documented reference temperature before bonding or tensioning, and the machine's mechanical zero established at that same temperature. A pair installed cold and zeroed hot carries a permanent offset frozen into the bond; a pair installed and zeroed at the same documented point gives every later excursion symmetric, calculable behavior around the reference.

Verifying the pair before qualification

Verification is a temperature test, not a temperature assumption. Run the axis through a representative thermal cycle—cold start, warm-up, soak—with an independent length reference (laser interferometer for shorter axes, a step gauge or a thermally decoupled second encoder for longer ones) and log substrate temperature simultaneously. The signature tells the story: linear error tracking temperature and reversing on cooling points at uniform mismatch strain; a sinusoidal component at the clip pitch points at bow; a fast error right after motion starts with slow recovery points at a gradient, usually from the drive motor on the substrate.

Acceptance is a number against the axis error budget: the temperature-correlated contribution of the scale mounting must fit its allocated share, with the design-file calculation confirming the measured slope. A machine that passes owns its thermal behavior explicitly; a machine that never runs the test owns the mismatch invisibly, and it will be rediscovered—slower and more expensively—the first time ambient conditions move outside the commissioning window.