Part identification by dimensions to prevent mixed parts

Similar parts can look almost identical while differing enough in length, width or height to identify reliably. A camera station can measure those differences and compare them with stored recipes, but only when the dimensional separation exceeds the measurement variation. This guide helps you decide whether that approach will prevent mixed parts, choose useful camera views, and define what happens when an image supports several identities—or none.

Decide whether dimensions can distinguish the parts

Start with a dimensional separation study, not a software selection. List every part that could enter the station, including incorrect variants, and record its permitted dimensional ranges. Use drawing limits and representative production samples rather than nominal dimensions alone.

Two different part numbers may share exactly the same external dimensions. Conversely, one part number may vary substantially within its permitted manufacturing tolerances. Neither a higher-resolution camera nor a more complicated classifier resolves an identity difference that the image cannot observe.

For each candidate dimension, calculate the gap between neighboring permitted ranges:

This is a screening condition, not a complete acceptance rule. It assumes the error allowance applies across the operating conditions you validated. Guarded acceptance can still hold conforming parts near their dimensional limits.

If the ranges overlap, add another observable feature: width, hole spacing, shoulder position or height. If every observable feature overlaps, use a different identification method. The guide to barcode-based verification of similar assemblies explains an alternative when dimensional identity is insufficient.

Choose views that expose the differences

A top view measures projected length and width. It does not directly measure height. A side view can measure height above a controlled datum, provided the relevant edges remain visible.

Use backlights when the required information is a silhouette. They separate an opaque part from the background without depending heavily on surface colour or texture. Compare illumination arrangements in the lighting finder.

Keep the part's measurement datum repeatable. A tilted component can change its projected length even when its actual length is unchanged. Positioning guides should control pose without hiding the edges being measured.

For parts that move toward or away from the camera, telecentric lenses reduce magnification changes caused by depth variation within their specified operating range. They do not remove defocus, tilt or occlusion.

The guide to when telecentric optics are needed for gauging helps determine whether ordinary perspective is acceptable. Explore compatible options in the lens finder.

Turn separation into an imaging requirement

Pixel count sets sampling, not measurement accuracy. Start with the field of view required to contain the largest part, including placement variation and edge margins.

That gap looks promising, but the actual error allowance must also cover calibration residuals, edge contrast, fixture variation, focus, motion and temperature effects. Subpixel edge fitting can improve repeatability without removing systematic errors.

Use the camera resolution selection guide to separate sampling requirements from dimensional performance.

Choose area scan cameras only after defining the required sampling, exposure and part motion. The camera finder helps narrow those requirements to suitable interfaces and shutter types.

For moving parts, estimate the exposure limit from allowable motion blur:

This limits blur during exposure; it does not eliminate rolling-shutter distortion. A stopped-part station is often simpler when cycle time permits. For continuous motion, verify shutter behaviour, trigger timing and available illumination together.

Work through a guarded recipe example

Consider these illustrative recipes. The dimensions describe an application example, not any catalog product.

The length gap between A and B is 50.80 − 50.20 = 0.60 mm. If the validated length error allowance is ±0.10 mm, the gap exceeds twice that allowance: 0.60 mm > 0.20 mm.

A measured length of 50.05 mm produces a guarded interval of 49.95–50.15 mm. That interval lies inside A's length range, but it also lies inside C's. Length and width alone therefore cannot distinguish A from C.

Suppose a side view measures height as 8.02 mm with an allowance of ±0.05 mm. Its guarded interval is 7.97–8.07 mm, which lies inside A's height range and excludes C. Provided the width and image-quality checks also pass, the combined measurements support A alone.

Now consider a measured length of 50.15 mm. Its guarded interval becomes 50.05–50.25 mm, extending beyond A's permitted range. A conservative rule holds the part rather than rounding the result into acceptance.

Calibration establishes the relationship between pixels and physical dimensions. The procedure in camera calibration using image markers provides the foundation, but production validation must also include handling and lighting variation.

Select candidate hardware without assuming compatibility

The following parts illustrate relevant component classes. They are not a validated package; confirm field coverage, mounting, exposure performance, lighting geometry and electrical compatibility before ordering.

The LS-CMR-0166 is a rolling-shutter USB3 camera candidate for a stopped-part arrangement. Its suitability depends on the complete acquisition sequence, including the settling interval before exposure.

The LS-TCL-1421 is a bi-telecentric lens option. Check the required magnification against the camera's actual active sensor dimensions, then verify working distance and mechanical clearance.

The LS-BKL-1468 is a standard backlight candidate. The LS-BKL-0830 provides a collimated backlight alternative for evaluation where illumination direction affects measured edges.

Do not select illumination from the external part footprint alone. Check the luminous area, placement variation, fixture openings and the full ray paths reaching the lens. The comparison of collimated and diffuse backlighting for edge gauging explains why edge position can change with illumination geometry.

For round components, the diameter and eccentricity inspection guide provides a useful example of choosing measurements that correspond to the actual inspection requirement.

Make recipe matching explicit and conservative

Store more than a nominal length and a photograph. Each recipe should contain permitted feature ranges, required views, datum definitions, calibration identity and a controlled revision. Keep the error allowance tied to the validated measurement method.

A practical classification sequence is:

1. Confirm that exactly one part is present and all required views are available.
2. Check exposure, focus, edge visibility and whether the part lies inside the calibrated measurement area.
3. Determine orientation from defined features rather than assuming the longest image direction is always length.
4. Measure the selected features using fitted edges or geometric models appropriate to the drawing.
5. Compare the guarded measurement intervals with every permitted recipe.
6. Accept only when the selected identity satisfies the acceptance policy and competing identities are excluded.

For a conservative dimensional-conformity policy, require each guarded interval to lie wholly inside the selected recipe's limits:

Also exclude competing recipes across the combined feature set. A competing recipe remains plausible when its permitted range intersects the guarded measurement interval for every required feature.

Keep outcomes separate: accepted identity, ambiguous identity, no matching identity, and invalid measurement. A blurred image is not evidence that the part has the wrong dimensions. Reacquire only under a controlled policy; repeated captures must not become repeated opportunities to obtain a passing result.

Connect the result to the correct physical part

A warning lamp can notify an operator, but it does not by itself prevent mixed parts from continuing downstream. Define the required containment action before choosing output hardware.

The result should carry a part or trigger identifier, recipe revision, measurement status and timestamp or encoder position. The controller must associate that result with the same physical item at the reject mechanism or destination.

For example, a reject point 450 mm downstream at a constant speed of 150 mm/s has a nominal travel time of 450 / 150 = 3 s. That calculation is insufficient when speed changes or parts slip. Use validated position tracking and account for actuator response, part spacing and the available reject window.

The guide to routing one inspection result to different actuators addresses this result-to-actuator association.

Define responses to missing images, late results, communication loss, blocked reject paths and failed reject confirmation. Depending on process risk, the station may hold flow, divert to a controlled location or request an operator action. Do not silently treat missing data as acceptance.

Validate variation, not just a demonstration sample

A demonstration with one example of each part proves very little about production discrimination. Include parts near dimensional limits, different surface conditions, expected pose variation and all relevant machine states.

Repeat captures of the same stationary part to assess short-term measurement noise. Then remove and reload it to include fixturing variation. Repeat across the operating field and after expected warm-up changes. Compare results with an independent measurement method whose uncertainty is suitable for the task.

As an initial diagnostic, 30 reloads per selected challenge sample can reveal obvious positioning problems. That count does not establish a production error rate or prove a low false-accept probability. Sample planning must reflect the consequence of accepting the wrong part and the confidence required.

Track false acceptance and false rejection separately. For identification, also record ambiguous results and invalid images. An apparently low reject rate can hide an unsafe policy that assigns every uncertain image to the nearest nominal recipe.

Challenge the complete sequence with deliberate wrong-part insertions, missing views, interrupted communication and unsuccessful rejection. Confirm physical containment, not merely the appearance of a red indicator on a screen.

Know when dimensional identification is the wrong answer

Dimensions are unsuitable as the sole identity source when variants share the same observable geometry. Material grade, heat treatment, internal construction and revision may differ without changing the silhouette.

Flexible parts can deform under their own weight or fixture force. Transparent parts may not produce a stable silhouette with a basic backlight arrangement. Burrs, flash or loose attachments may move apparent extrema without changing the intended reference dimensions.

In these cases, change the feature or identification method rather than widening tolerance bands until every sample passes. A readable code, a controlled carrier identity, another optical feature or a different sensing method may be more appropriate.

A second camera helps only when it observes a genuinely discriminating feature. Likewise, more pixels help only when sampling—not overlapping part definitions or unstable presentation—is the limiting factor.

Complete the pre-order and acceptance checklist

• Document all plausible variants. Include wrong parts that can realistically reach the station, not just the scheduled production recipe.
• Calculate separation against measurement variation. Identify overlapping ranges before investing in hardware.
• Confirm every required view. Check that fixtures, neighbouring parts and handling equipment cannot hide the discriminating edges.
• Verify the optical package. Confirm sensor coverage, magnification, working distance, luminous area and mechanical access.
• Define ambiguous and invalid outcomes. Neither should become an automatic nearest-match acceptance.
• Control recipes and calibration. Record revisions and restrict changes through the plant's normal approval process.
• Prove containment. Test result tracking, reject actuation, confirmation and recovery after an interrupted cycle.
• Agree on acceptance evidence. Document the challenge samples, operating conditions and error measures before commissioning.

Bring the part drawings, dimensional ranges, presentation method and cycle-time target to the camera configuration starting point. If the separation study leaves competing identities unresolved, use application support to review the measurement approach before selecting parts.