Machine vision dimensional measurement for small parts

Measuring a small part to 0.01 mm requires more than choosing a lens with sufficient magnification. You need stable image scale, measurable edges, controlled part presentation and calibration verified against an independent reference. This guide develops an illustrative setup for a nominally 15 mm square part viewed from approximately 200 mm away. It explains how to size the field of view, compare conventional and telecentric optics, and demonstrate measurement performance before selecting final hardware.

Define the measurement requirement before the lens

For machine vision dimensional measurement, first distinguish display resolution, repeatability and accuracy. Reporting three decimal places does not demonstrate micrometre-level accuracy. A system can repeat a biased measurement very consistently.

Write an acceptance requirement such as: measured length and width must each differ from a reference value by no more than 0.01 mm across the specified operating conditions. Define those conditions, including allowable height, rotation, position, temperature and motion.

Also define which physical edges determine the dimensions. A top face, bottom silhouette, chamfer boundary and maximum external envelope can produce different answers. A part that is 3.2 mm thick does not necessarily move through 3.2 mm of measurement height. Fixture seating variation and the location of the measured surface determine the relevant height range.

Use telecentric optics when height-dependent scale changes would consume the error allowance. Conventional optics remain reasonable when the measured plane is constrained and calibration remains valid. The decision framework in when telecentric optics are needed for gauging explains that distinction.

Calculate coverage and sampling together

Start with the largest part and its placement envelope, not its nominal dimensions. For this example, assume length and width each range from 14.8 mm to 15.2 mm. A 24 mm horizontal by 18 mm vertical field provides useful clearance when orientation is controlled.

Rotation consumes that clearance. For a square rotated in its own plane, its axis-aligned bounding width is:

That envelope fits within the shorter field dimension. Unrestricted rotation would require a larger field: at 45°, the same square spans approximately 21.5 mm. Do not preserve sampling by accidentally cropping valid parts.

Consider an illustrative sensor with an active area of 12.8 mm by 9.6 mm and an image of 4000 by 3000 pixels. These are example design inputs, not specifications of a listed camera.

The sampling is therefore 6 µm per pixel. A 10 µm dimensional change spans approximately 1.67 pixels. That supports further investigation, but it does not prove 10 µm accuracy. Lens blur, calibration, edge bias and mechanical changes still contribute.

Use the camera resolution selection guide to compare sampling requirements. The magnification-to-sampling calculation explains how actual pixel pitch affects this calculation. Select area scan cameras through the camera finder only after confirming active sensor dimensions, image dimensions and the required shutter behavior.

Treat working distance and height variation as separate constraints

A mechanical clearance of approximately 200 mm does not uniquely determine focal length or magnification. Working distance, sensor dimensions and optical design must satisfy the coverage requirement together.

With a conventional lens, changing object distance changes image scale. A first-order perspective estimate is:

This estimate already exceeds the example error limit. It assumes an effective object distance of 200 mm; distance from the lens housing may differ. Use it to flag risk, then measure the actual scale change experimentally.

Object-side telecentric optics reduce perspective-related scale changes within their specified operating range. They do not provide unlimited focus depth or eliminate every dimensional error.

The separate treatments of telecentric distortion and telecentricity and measurement accuracy through telecentric focus depth explain why both scale stability and edge sharpness need verification.

Compare optical approaches against the actual fixture

Choose the architecture that controls the largest error sources without creating unnecessary mechanical constraints.

Bi-telecentric construction alone is not evidence of a smaller total measurement error. Compare documented performance and measurements under your conditions.

Browse telecentric lenses when height sensitivity dominates. Compare machine vision lenses when the fixture provides a stable measurement plane. The lens finder helps narrow either route, while the focal length selection guide supports conventional-lens estimates.

For the illustrative sensor, its diagonal is 16 mm, calculated as √(12.8² + 9.6²). The lens must cover that active diagonal with acceptable image quality. A mount match alone does not establish coverage or measurement performance.

Use catalog examples to expose tradeoffs

These examples illustrate selection constraints rather than a validated parts list. None should be ordered solely because its mount or magnification appears plausible.

The LS-TCL-2194 is an object-side telecentric lens with 0.5x magnification, WD 110 mm, a 1" sensor format and C-mount. Its catalog description also lists built-in coaxial light.

Its magnification is close to the calculated requirement, but its working distance differs substantially from the proposed clearance. At an illustrative magnification of 0.5x, the example sensor would cover 25.6 mm by 19.2 mm, with sampling of 6.4 µm per pixel. That calculation checks coverage only; it does not establish compatibility, edge quality or accuracy.

The LS-TCL-1422 is a bi-telecentric lens with 0.166x magnification, WD 258 mm, a 1" sensor format and C-mount.

This example demonstrates why longer working distance is not enough. Lower magnification expands coverage but reduces pixels across the part. Evaluate the resulting sampling before accepting a lens simply because it provides more mechanical clearance.

The LS-FFL-0679 is a 35 mm, 2/3", 6 MP, C-mount machine vision lens.

This conventional option belongs in a different evaluation path: match it to a suitable sensor, calculate coverage and test height sensitivity. It is not a substitute for the illustrative larger sensor combination without checking image-circle coverage.

If the proposed working distance is mandatory, request a candidate that satisfies it together with magnification, sensor coverage and dimensional performance. Do not change a fixed-magnification lens's installation distance arbitrarily to recover the desired field.

Make edge formation part of the measurement design

For external length and width, a transmitted silhouette is often a useful starting point. Reflective top lighting may instead emphasize texture, chamfers or reflections rather than the required outer boundary.

Lighting geometry can shift the apparent edge, particularly on thick, rounded or transparent parts. The comparison of collimated and diffuse illumination for edge gauging explains when illumination direction affects the measured silhouette.

Fit straight edges over suitable regions, excluding corner radii, burrs and chamfers unless the specification includes them. Measure perpendicular separation between fitted lines rather than using an axis-aligned bounding box. In-plane rotation otherwise changes the reported bounding dimensions.

Keep gain, exposure and image processing fixed during validation. Saturation, sharpening, compression and unstable automatic adjustments can alter edge location. Subpixel fitting can improve localization, but it cannot recover a stable boundary from a changing optical signal.

If the part moves, estimate exposure from the allowed blur:

This is a design example, not a camera capability claim. A global shutter avoids rolling-shutter geometric skew, but it does not remove blur during exposure. For stopped inspection, verify that vibration has settled before acquisition.

Use an image sharpness metric to support focus adjustment, then choose the setting that delivers stable dimensions rather than merely the highest contrast score.

Calibrate the measurement plane, then check an independent reference

Calibrate with a dimensionally characterized target suitable for the required uncertainty. Place its reference surface at the actual measurement plane, using the final focus, aperture, illumination and camera settings.

A single scale factor may suffice over a small, well-controlled region, but it must be demonstrated. Map residual error across the usable field. Distortion correction or a planar calibration can help when the geometry warrants it.

The guide to length calibration from image markers describes the coordinate relationship. Use the calibration target coverage guidance to avoid checking only the center while measuring parts elsewhere.

Tilt introduces another error that telecentricity does not remove. An out-of-plane tilted feature is foreshortened:

That nearly consumes the example allowance. A calibration for a fixed plane does not automatically correct subsequent changes in part tilt.

Finally, check a reference that was not used to fit the calibration. Otherwise, a close match may demonstrate only that the calibration reproduces its own inputs. Record reference uncertainty and measurement conditions alongside the error.

Verify the complete error allowance

Validation must cover the conditions under which you will accept or reject parts. Start with repeated captures of a stationary reference, then repeat after removing and reseating it. These tests separate image repeatability from handling variation.

A practical initial study could include 100 stationary captures, five positions across the usable field, and the minimum, nominal and maximum permitted measurement heights. Add rotation, reseating and operating-temperature conditions relevant to production. This is a screening plan, not a universal qualification standard.

For each condition, record reference error, repeatability, focus quality and edge contrast. Poor performance at one corner or height is not repaired by averaging it with good measurements elsewhere.

An illustrative uncertainty budget shows why sampling cannot be the only criterion:

If these contributions are adequately characterized and uncorrelated, their root-sum-square is approximately 2.75 µm. Multiplication by a coverage factor of 2 gives an illustrative expanded uncertainty of 5.5 µm. The coverage factor requires justification; it does not automatically establish a particular confidence level for every dataset.

Do not combine uncorrected biases as though they were independent random noise. Correct known biases where appropriate, account for correlation and avoid counting the same contribution twice.

One conservative acceptance rule is:

This is an example decision rule, not a measured capability claim. Agree the actual rule and uncertainty treatment before qualification.

Check before ordering and know when to change the approach

Use this short checklist to close the design:

• Define the measured edges, dimensional range and error limit. This prevents selecting optics for the wrong boundary.
• Include rotation and placement in the field envelope. Nominal part size alone understates required coverage.
• Confirm active sensor dimensions, pixel count and image-circle coverage. Format labels are only a starting point.
• Confirm working distance, lens dimensions and mounting clearance. Optical suitability does not establish mechanical fit.
• Specify allowed height and tilt variation. Both can dominate a small error allowance.
• Establish illumination, exposure and edge-processing settings. Calibration depends on stable image formation.
• Reserve independent references and validation time. Product specifications cannot replace system qualification.

Change the usual approach when the physical boundary is ambiguous, the part deforms during handling, or tilt cannot be constrained. More pixels will not resolve those problems. A different fixture, another view or a different measurement method may be necessary.

For a stable planar part, conventional optics may pass the verification plan. For meaningful height variation, telecentric optics deserve priority. Neither route should be accepted solely from nominal part dimensions.

Take the field envelope, sampling calculation, clearance drawing and validation requirement to the lens configurator. If those constraints conflict, contact the applications team before ordering a lens and camera combination.