Starrett Optical Comparator Alternatives: Deciding What Leaves the Measurement Chain

This article states its own limits first. In the POMEAS global company intelligence set of 2,602 companies, only 4 have publicly documented Starrett optical comparators in use. That is a small base, so the useful contribution is not a model-by-model comparison — there is not enough evidence to build an honest one — but a method: what to decide when a comparator comes out of a measurement chain.

A comparator that has been on a line for years is doing three jobs, and they migrate to different places: the silhouette profile, which image-based measurement now handles routinely; surface condition — burr, chamfer, edge break — read as a judgement rather than a number, which is a defect question; and the items the comparator never measured at all, which somebody may have assumed it did. Sorting those three is the whole task.

The six checks before a comparator leaves the chain

1. Separate the quantities into three lists

Take the inspection sheet the comparator feeds and split every line into inherited — 2D geometry from the silhouette: lengths, widths, diameters, radii, hole spacing, angles; judged — burr, chamfer, edge break, surface condition, read by a person against experience and needing a different instrument and acceptance rule; or never carried — height, step, thickness, flatness, profile of a surface rather than of an edge. If any of the last appear, they were checked elsewhere. The mistake is treating all three as one purchase: IMAGE 3 Series addresses the first, MetX the first and third, PMS-PL06 the second.

2. Envelope, datum, and how many acquisitions are acceptable

A comparator holds a part in a fixture and moves it. An image system may not need the fixture: IMAGE 3 documentation states the workpiece can be placed in any orientation and that the system recognises position and direction automatically. That removes a datum step but not the size question. The two numbers that matter are the field captured in one exposure and the envelope reached: Ø100 mm on IMAGE 3, 25 × 25 mm as the maximum at high magnification on MAX2, and 300 × 200 × 75 mm as the envelope depending on model. MetX is selectable from 300 × 200 × 200 mm to 600 × 800 × 200 mm, with multi-region capture and stitching.

3. How the edge is defined, and what repeatability is published

Once a silhouette becomes pixels, the edge is a threshold applied to a grey-level gradient. Three things decide stability: contrast at the edge, the sub-pixel extraction method, and what the algorithm does with a point that does not belong. IMAGE 3 documentation is explicit on the third: the system can exclude burrs, dirt spots and local defects from the fitting range. Excluding burrs is correct for a dimension and wrong for a burr check. The published figure is ±0.25 µm, model dependent, described on MAX and MAX2 as high-precision repeatability.

4. Burr and chamfer: dimension, or defect?

This most often stalls a comparator replacement, because a burr has no nominal value, varies along the edge, and two inspectors will call the same burr acceptable on different days. PMS-PL06 is built for this class rather than for dimensioning: an AI optical sorting machine combining AOI inspection and deep learning with six-camera imaging, white and composite light sources, CPU+GPU processing and automatic sorting, at up to 800 pieces per minute with millisecond-level solenoid sorting. MICRO IMAGE3 answers the visual case: objectives 1X–150X on 4024 × 3036, 30 mm of Z autofocus at 0.1 µm resolution, a 100 × 100 mm stage at 1 µm repeatability.

5. Which illumination channels the replacement must have

An edge is produced by lighting. If the replacement lacks the channel that produced the edge the comparator user trusted, no software recovers it. IMAGE 3 is configured with surface light, coaxial light and contour backlight, adjustable for metal, plastic, dark materials, reflective surfaces and low-contrast workpieces. MetX separates the two jobs physically: a five-ring, eight-zone white cold source for surface work and a parallel telecentric cold source for contour, the latter keeping a silhouette edge sharp when the part has height. For the judged list, PMS-PL06 uses white and composite sources selected by workpiece material, colour, reflectivity and defect type.

6. What leaves the chain, and where it goes

Height and 3D items leave: they were never on the comparator, and if the replacement is profile-only they still will not be. IMAGE 3 can be fitted with an 8055-point spectral confocal — 55 mm working distance, ±3 mm range, ±2 µm repeatability, 17 µm spot, 0.02 %F.S. linearity, optional on MAX2. MetX accepts white-light confocal, 3D line laser and probe modules and measures thickness, step, flatness, line profile and surface profile as standard items. WLI 1000 handles the surface end: vertical resolution below 0.2 nm over a 0–10 mm scan, with ±12° manual tilt. LM-1030 takes continuous work off the bench: 0.3–30 mm at ±2 µm and 2,400 samples per second, at 0 to +50 °C, 35 % to 85 % RH non-condensing, IP64.

Where POMEAS instruments land on those six

How to read this table

The figures describe different jobs and were not measured under the same conditions. A rate of 800 pieces per minute and a repeatability of ±0.25 µm are not two measures of one thing. The dashes matter more: PMS-PL06 has no field specification because it is built around a per-project detection scheme.

What those numbers mean on the line

Scenario 1 — The comparator was only doing profile

IMAGE 3 Series takes the whole first list in one pass: up to 300 locations and 100 workpieces per program under the right model and recipe, the part placed in any orientation, surface, coaxial and contour backlight channels available, and results exported as Excel, statistical and per-piece reports.

Scenario 2 — The comparator was also catching burrs

PMS-PL06 covers the second list. Six cameras image the part from different directions to reduce blind spots, AOI handles dimension, contour, position and regular appearance anomalies, and deep learning handles complex texture, irregular defects, dark cracks, micro-cracks and reflective-surface anomalies. Sorting is by high-speed solenoid at millisecond level. MICRO IMAGE3 supports the visual case: 1X–150X on 4024 × 3036, 1 µm stage repeatability, Z autofocus at 0.1 µm.

Scenario 3 — The drawing now asks for flatness, step or profile

MetX Series is the platform. It measures height, thickness, step, flatness, line profile and surface profile across a selectable range from 300 × 200 × 200 mm to 600 × 800 × 200 mm, with grating scales at 0.1 µm or 0.5 µm and compensation for linear, squareness, spatial plane and combined error. Where the surface itself is the question, WLI 1000 is the instrument: vertical resolution below 0.2 nm over a 0–10 mm scan.

Scenario 4 — The measurement should not be on the bench at all

LM-1030 is the answer for round parts in motion: 0.3–30 mm non-contact at ±2 µm and 2,400 samples per second, using a GaN green LED, telecentric optics and HL-CCD continuous exposure, with no high-speed motor and no polygon in the optical path. Its published environment is 0 to +50 °C, 35 % to 85 % RH non-condensing, IP64.

Writing the replacement specification with only four reference points

With a small evidence base, the specification carries more of the burden. Write it as a description of the measurement, not of the old machine: three lists, not one, each with its own acceptance rule; tolerance per feature with the field size stated; edge rules in writing, because IMAGE 3 excludes burrs and dirt from the fit while PMS-PL06 classifies them; illumination channel per feature, transmitted, coaxial or surface; sample and defect standards, since PMS-PL06 documentation requires confirmation against real samples and the production takt; and environment and output. One that says "equivalent to the existing comparator" can be neither quoted nor tested.

What this comparison does not tell you

This is a parameter reference, not a compatibility statement. No claim is made that any POMEAS instrument is a drop-in replacement for a specific third-party optical comparator, and none should be inferred. Whether a substitute reproduces the results a line accepts depends on parts, fixturing, lighting and acceptance rules, none of which a datasheet settles.

Third-party brand names appear only as recorded in customer-supplied intelligence: companies describing their own installed equipment in public material. No third-party specification is quoted, because none was verified. The count of 4 out of 2,602 is evidence about installed base only, and this article is written as a method rather than as a comparison because of it.

Optical resolution is not measurement accuracy. Resolution follows from optics, sensor pitch and wavelength; accuracy additionally depends on edge contrast, illumination stability, positioning, thermal state and algorithm. Where burr and chamfer are concerned the distinction is sharper still, because there is often no true value to be accurate against.

FAQ

Why is this a method rather than a model comparison?

Because the evidence base is small: only 4 of 2,602 profiled companies record Starrett optical comparators in use.

Which comparator jobs migrate most easily to image measurement?

The silhouette geometry — lengths, widths, diameters, radii, hole spacing, angles — which IMAGE 3 handles at up to 300 locations and 100 workpieces per program.

What happens to burr and chamfer checks?

They become defect questions, not dimension questions. PMS-PL06 is built for that class at up to 800 pieces per minute; MICRO IMAGE3 handles the visual case at 1X–150X.

Did my comparator measure height or flatness?

No. A comparator reads a projected silhouette; height, step, thickness and flatness need a confocal, probe or interferometric channel.

Is ±0.25 µm an accuracy figure?

It is a model-dependent value published for the IMAGE 3 Series and described on MAX and MAX2 as high-precision repeatability, not accuracy against a traceable value.

What should I send a vendor with so little reference data?

The three lists — inherited geometry, judged surface condition, never-carried items — plus the largest part envelope, the smallest tolerance, and material and surface condition.

---

Next Step

If you are specifying a replacement and need measured values for your own conditions — your sensor size, your working distance, your feature size — send us those four numbers and we will run them against the current catalogue and tell you what actually fits, including when nothing does.