Small round part inspection: planning views and measurements

A small round part can need several inspection methods even when it fits comfortably inside one image. Diameter, face damage, edge burrs, thickness, and warpage do not present the same optical problem. For parts spanning 3–22 mm, the first decision is therefore not which camera to buy, but which view makes each acceptance requirement measurable.

This guide turns eight representative checks into a trial plan for equipment builders and production teams. You will be able to define the required views, estimate image sampling, compare lighting approaches, and specify evidence before purchasing components or commissioning a complete machine.

1. Turn eight requirements into observable checks

Start with a defect matrix rather than a component list. The eight checks below are a planning example, not an assumption about your drawing. Replace them with the actual requirements, including dimensions, allowable defects, and inspection locations.

For each row, record the smallest unacceptable defect, its likely contrast, and the consequence of a missed defect. A scratch described only as “visible” is not a usable purchase specification. Define its width, length, orientation, and appearance on acceptable production material.

Also separate missing features from damaged features. A hole-presence check may need modest sampling, while a small chip on that hole's edge can require substantially more detail.

The lens and lighting test plan provides a useful framework for recording trials. Establishing this matrix first prevents a successful demonstration of one easy feature from being treated as proof of every requirement.

2. Plan views around presentation and hidden surfaces

A top view is usually the first station to investigate. With a backlight, it can reveal the external silhouette and open holes. With reflected illumination, it can inspect the visible face. These are different image conditions, even if they use the same camera.

A side view adds rim information and an apparent thickness profile. However, one side image does not inspect the entire circumference. A burr on the opposite side may be hidden behind the part. Rotation, additional viewing directions, or a different presentation method may be necessary.

Likewise, a top camera cannot inspect the lower face through an opaque part. Plan a turnover operation, an underside viewing opening, or a separate lower-face station. Include any surface hidden by the nest in the coverage assessment.

Part restraint is part of the optical design. A nest that blocks the outside contour prevents complete silhouette gauging. A nest that allows tilt changes the projected shape and can shift the measured face position.

Document these presentation conditions before comparing area scan cameras:

• Is the part stationary during acquisition, or moving continuously?
• Can the part tilt, overlap another part, or arrive upside down?
• Which surfaces touch the nest, and which become hidden?
• Can the smallest part move laterally inside the largest-part fixture?
• Does each inspection image retain the correct part identity through rejection?

The camera finder becomes more useful once these conditions are known. Choose a shutter and trigger arrangement that fits the actual motion, not a stationary bench demonstration.

3. Calculate sampling from the smallest defect

The largest diameter determines coverage, but the smallest defect determines image sampling. Do not select resolution from part diameter alone.

Consider an illustrative field measuring 32 mm horizontally by 24 mm vertically, sampled by 4000 × 3000 pixels. These are planning assumptions, not specifications assigned to a listed camera. A centered 22 mm part has 1 mm clearance on each side of the shorter field dimension.

A 50 µm feature would span about 6 pixels under these assumptions. That is a starting point for visibility trials, not a promise that every scratch or burr of that width will be detected.

If the initial target is at least 5 pixels across that feature, the maximum sampling interval is 0.050 / 5 = 0.010 mm/pixel. Covering the proposed field then requires at least 3200 horizontal pixels and 2400 vertical pixels, before allowing for optical performance or image cropping.

Use the camera resolution selection guide to distinguish feature visibility from dimensional measurement requirements. Pixel count alone does not establish measurement uncertainty.

At the same image scale, a 3 mm part spans 3 / 0.008 = 375 pixels. That might satisfy presence inspection but fail a much smaller defect requirement. Separate optical setups or adjustable recipes may be preferable when the smallest parts need finer detail.

The LS-CMR-0053 is a rolling-shutter camera candidate for stopped-part trials. Confirm acquisition timing and part stability before treating it as a production selection.

4. Choose optics for coverage and measurement stability

Field of view, sensor dimensions, magnification, working distance, and mechanical access must agree. Sensor format labels are not substitutes for the actual active sensor dimensions needed for optical calculations.

Calculate both field dimensions, not just the longer one. A circular part must fit the shorter dimension with allowance for lateral presentation error, calibration coverage, and any region cropped during processing.

Conventional lenses may be sufficient for face appearance checks when part height is controlled. For dimensional checks, investigate telecentric lenses when permitted height variation would otherwise change apparent size. Telecentric imaging reduces perspective-related magnification changes within its designed operating range; it does not remove defocus or every source of measurement error.

The guide to telecentric lenses for gauging explains the decision in more detail. For distortion and angular effects that remain, consult telecentric distortion and telecentricity.

Review the LS-TCL-2194 as an optics candidate, not a confirmed match to the illustrative field. Check the actual camera sensor dimensions, image coverage, mount, working distance, and required field before pairing components.

Use the lens finder to narrow candidates after establishing these requirements. The broader lens selection guide is useful when comparing appearance inspection against gauging needs.

5. Give silhouette and surface checks different lighting

For opaque parts, backlights provide a strong starting point for outside diameter, holes, and missing edge material. They generally do not reveal scratches confined to the visible face. Those require reflected illumination that produces contrast from the relevant surface change.

A diffuse backlight and a collimated backlight can produce different edge profiles, especially around thick, rounded, or reflective edges. Do not assume that the image with the sharpest-looking boundary automatically gives the most accurate diameter.

Evaluate the LS-BKL-1468 as a standard-backlight candidate and the LS-BKL-0830 as a collimated alternative. Their catalog comparison supports component screening; sample images must determine which approach suits the part.

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

The diffuse versus collimated backlight comparison explains why illumination direction can influence measured edges. Check light coverage at the intended part position, including the full range of allowed lateral movement.

For upper-face appearance, test more than one illumination direction. Shallow-angle lighting can emphasize raised or recessed features, but directional defects may disappear when aligned unfavorably. Broad, diffuse illumination may reduce glare while also reducing contrast from a shallow scratch.

The lighting selection guide and surface-versus-silhouette lighting explanation help separate these tasks. Use the lighting finder after identifying which image conditions work.

6. Decide whether thickness belongs to vision

A dedicated thickness instrument may be preferable when thickness tolerance is tighter than the demonstrated side-view measurement uncertainty. The decision should follow a comparison of methods, not an assumption that adding another camera is always simpler.

A side-view silhouette can measure separation between visible boundaries. It may be useful for accessible, stable parts, particularly with suitable telecentric optics. However, tilt, rounded rims, burrs, and fixture obstruction can change which boundaries the algorithm measures.

A displacement sensor above a conveyor measures top-surface position relative to its reference. That is not automatically part thickness. A contact thickness gauge depends on its anvil geometry and force. Opposed measurements require a defined relationship between their reference surfaces.

For an illustrative thickness acceptance band from 1.95 mm to 2.05 mm, the total tolerance width is 0.10 mm. If the agreed measurement uncertainty target is 10% of that width, the target is 0.010 mm. This allocation is an example, not a universal acceptance rule.

Specify whether the reported quantity is center thickness, maximum thickness, minimum thickness, or thickness at defined locations. A single center reading cannot establish flatness. Flatness requires suitable surface sampling and evaluation relative to the specified geometric definition.

Before purchasing, have each proposed method measure the same labeled samples. Compare readings against an agreed reference method, including tilted parts, burrs, and the production range of surface finishes.

7. Validate with labeled samples and fixed criteria

A useful trial set includes acceptable parts, clear rejects, and borderline examples near each acceptance limit. Include normal changes in material, finish, tooling condition, orientation, and presentation. Keep reference measurements and defect labels traceable to individual samples.

Begin with feasibility images, then separate recipe-development samples from acceptance samples. If every difficult part is used to tune the algorithm, the final demonstration does not independently test the recipe.

A practical validation sequence is:

1. Establish the required views and save unprocessed images for every sample.
2. Verify field coverage at the extremes of permitted part position and height.
3. Fix exposure, illumination, focus, and measurement regions before evaluating repeatability.
4. Inspect repeated images without moving the part to isolate image and processing variation.
5. Remove and reload the same part to expose presentation and handling variation.
6. Run a separate labeled set and record false accepts and false rejects by defect class.
7. Repeat at the intended production timing and save the inspection decision with its images.

For an initial repeatability screen, capture 30 repeated measurements without reloading, then 30 measurements with reloading. These sample counts help reveal obvious variation; they are not a complete measurement-system study.

Report minimum, maximum, spread, and bias against the reference. A stable average does not compensate for individual measurements crossing an acceptance boundary.

For similar-looking part variants, the dimension-based part identification guide helps distinguish identification logic from defect acceptance.

8. Compare components and complete machines on the same scope

At 120 parts per minute, the average cycle allowance is 60 / 120 = 0.5 seconds per part. That allowance must cover the relevant handling, image acquisition, processing, communication, and rejection operations.

For a non-overlapping example, allocate 100 ms to settling, 40 ms to acquisition, 60 ms to processing, and 50 ms to the decision handshake. Their sum is 250 ms, leaving 250 ms for the remaining transfer operations and margin.

These are hypothetical timing allocations, not capabilities assigned to any listed component. Acquisition time must include the necessary trigger response, exposure, readout, and transfer behavior. Where operations overlap, document the pipeline rather than simply adding every duration.

Compare purchasing approaches using the same defect matrix and acceptance protocol:

Request a coverage map, representative raw images, measured repeatability, defect-class results, and a clear list of exclusions. “Two cameras included” says little about whether all required surfaces are visible.

Treat duplicated machines as separate acceptance items. Each needs verified calibration, controlled recipes, and checks that its fixtures reproduce the validated presentation.

9. Check the specification before placing an order

Use this checklist as the release gate for component selection or a machine specification:

• Defect definitions: Every check has a measurable boundary between acceptable and unacceptable parts.
• View coverage: Both faces, the required rim regions, and any hidden features have an inspection path.
• Sampling: The smallest relevant feature has been evaluated at the proposed production image scale.
• Optical compatibility: Sensor dimensions, mount, field, image coverage, focus range, and working distance agree.
• Lighting: Silhouette and appearance checks have independently demonstrated useful contrast.
• Thickness method: The measured quantity, datum, uncertainty target, and reference method are documented.
• Presentation: Tests include permitted translation, tilt, rotation, and changeover variation.
• Timing: Acquisition and decision timing have been measured with production handling, not only on a bench.
• Acceptance: Independent samples, false-accept limits, false-reject limits, and evidence requirements are agreed.
• Maintenance: Calibration checks, cleaning access, recipe control, and fault-response responsibilities are assigned.

The usual top-view camera solution is wrong when the critical defect is hidden, the thickness datum is undefined, or the needed contrast cannot be produced. Resolve those gaps before adding resolution or more processing.

Bring your defect matrix, drawing tolerances, cycle target, and labeled sample images to inspection planning support. Once the views and sampling are supported by evidence, use the camera configuration starting point to organize a component proposal around the validated inspection plan.