White LED color temperature: specify repeatable inspection

A replacement white light can look identical to the original yet move a color threshold enough to misclassify parts. Matching the label on the light is therefore not a complete replacement specification. This guide is for inspection teams selecting white illumination or qualifying a second unit. You will learn which spectral information to request, how to separate illumination changes from camera adjustments, and how to approve replacements against measured inspection limits rather than appearance alone.

Make camera response the acceptance criterion

Specify white LED color temperature as one part of an illumination specification, not as the final acceptance test. The decisive question is whether the replacement preserves separation between acceptable parts, unacceptable parts and neighboring product classes.

Start by deciding whether white illumination is necessary. Color identification usually needs broadband illumination and a color camera. Some presence, edge and print inspections work better with a selected narrow wavelength. The guide to choosing an LED colour for contrast explains that earlier decision.

When white light is appropriate, qualify the light, camera, lens, acquisition settings and inspection algorithm together. A nominal CCT match cannot establish that this combination will behave identically.

Your acceptance specification should have two layers:

• Source requirements: geometry, emission option, documented CCT range, available spectral information and operating conditions.
• Application requirements: permitted camera-response drift, minimum class separation, saturation limits and inspection outcomes on representative samples.

The first layer makes procurement repeatable. The second checks whether the purchased light actually serves the inspection.

Understand what CCT and CRI describe

Correlated color temperature, or CCT, describes the chromaticity of a near-white source by relating it to the appearance of a thermal radiator. It is expressed in kelvins. Lower values generally appear warmer; higher values generally appear cooler.

A hypothetical source labeled 3000 K appears warmer than one labeled 6500 K. These numbers describe appearance, not the physical temperature of the LED assembly, and they do not specify the complete emitted spectrum.

CCT also leaves out displacement away from the reference locus. Two sources can have the same reported CCT while differing in a greenish or pinkish direction. Chromaticity coordinates and a stated tolerance provide more information than a nominal kelvin value alone.

The color-rendering index, or CRI, describes object-color rendering for a standardized human observer relative to a reference illuminant. The commonly reported general index, Ra, averages performance over eight specified test colors. It is not a camera calibration score.

A higher CRI can be useful when visual color rendering matters. It is not automatically the best choice for separating two industrial materials. For measurements intended to represent color, follow a calibration approach such as color inspection beyond raw camera RGB.

Why matching CCT can still change the image

A camera channel responds to the combination of illumination, material reflectance, optical transmission and sensor sensitivity. CCT compresses information about the source into a single number; it cannot preserve all those interactions.

For a reflective, nonfluorescent surface, a simplified linear-response model is:

Here, C_j is the dark-subtracted signal in camera channel j, t is exposure time, and k collects fixed scale factors. E is the spectral illumination at the part, rho is material reflectance, T is optical transmission, and S_j is channel sensitivity.

This model assumes unsaturated, approximately linear acquisition with fixed geometry and gain. It omits effects such as fluorescence and complicated directional reflection, so treat it as an explanation rather than a complete prediction.

Two lights with matching CCT can have different spectral peaks and valleys. Those differences may fall where a pigment reflects strongly or where one camera channel is especially sensitive. A neutral reference may remain balanced while a colored component changes significantly.

White balance cannot generally remove this problem. It rescales camera channels, but it cannot recreate wavelengths missing from the illumination. The multi-camera white-balance consistency test addresses camera-to-camera variation; replacement-light qualification must additionally test material-dependent changes.

Monochrome cameras are not exempt. Their single channel still weights wavelength according to sensor sensitivity. The discussion of monochrome sorting under changing illumination explains why material contrast can shift without a color image.

Build a controlled replacement-light test

Use a recorded acquisition recipe and representative parts before comparing candidate lights. Otherwise, automatic adjustments can conceal the very difference you need to measure.

1. Freeze the optical arrangement. Record camera position, lens aperture, focus, light position and part orientation. Use locating features so a replacement can return to the same arrangement.
2. Freeze acquisition and processing. Record exposure, gain, black-level settings, white balance, gamma, pixel format and color processing. Disable automatic adjustments for the initial comparison. The guide to exposure, gain and balance settings provides the camera-side starting point.
3. Record the electrical recipe. Document the driver mode, commanded intensity and any pulse timing. Equal brightness commands do not establish equal optical output across different lights.
4. Establish a stable operating condition. Measure reference-patch response during startup and choose a stabilization criterion appropriate to the station. Do not assume a universal warm-up time.
5. Capture a baseline sample set. Include normal production variation, nearby reject conditions and samples close to classification boundaries. Add neutral and colored reference patches, but do not substitute them for real parts.
6. Install the candidate without retuning. Repeat the captures with the stored recipe. Measure both signal-level changes and changes in the inspection's decision features.
7. Evaluate any permitted adjustment separately. If production allows recalibration, document its limits and validate the adjusted recipe across the full sample set.

For a preliminary repeatability screen, capture 20 frames per sample under each light. This can expose short-term variation, but it does not establish a production failure rate or replace testing across multiple units and production lots.

Worked example: a neutral patch passes but a part fails

Consider a station distinguishing two colored plastic classes using dark-subtracted, approximately linear camera values. Its feature is the red-channel fraction:

This ratio reduces sensitivity to a common brightness scale change. It does not remove spectral differences, channel offsets or clipping, and it is not a standardized color coordinate.

Under the baseline light, a representative class A sample gives:

Class B has a measured mean of 0.360. A provisional classifier places its boundary at 0.330. Suppose repeated measurements give a standard deviation of 0.003 for each class.

A simple screening envelope uses the mean plus or minus three standard deviations:

These envelopes summarize the measured spread; they do not guarantee a particular error rate without additional distribution and sampling evidence.

Now install a replacement with the same nominal CCT. The neutral reference remains within its brightness limit, but the class A sample changes:

The total signal is unchanged, yet the feature has reached the classification boundary. Brightness matching alone would miss this failure.

An illustrative acceptance rule might limit feature drift to 0.010 while requiring every tested class envelope to remain at least 0.010 from the boundary. This candidate fails both the drift check and the separation check. Actual limits must come from your required error rates, sample variation and measurement uncertainty.

For broader image comparisons, use the distinctions in meaningful metrics for comparing image color. Whole-image averages can hide changes in the small region that determines acceptance.

Choose geometry before comparing white-light labels

A spectral comparison is difficult to interpret if the illumination angle also changes. Surface finish, curvature and texture can alter the camera signal even when the source spectrum remains constant.

Use ring lights when illumination around the optical axis fits the task. The diffused white ring-light candidate provides a geometry to evaluate; its catalog information does not establish a CCT or CRI guarantee.

Use bar lights when you need control over illumination direction. The diffused white bar-light candidate is another arrangement to evaluate against representative surfaces.

Consider dome lights when broad-angle illumination helps reduce localized reflections on curved or glossy parts. A dome surrounds the viewing region with diffuse illumination, but its camera opening can still produce a visible reflection.

Review the dome-light candidate for the mechanical arrangement, and confirm the required emission option before selection. No CCT or CRI value is established by the supplied catalog entry.

The guide to dome illumination for reflective parts covers coverage and reflection limits. Compare physical formats in the lighting finder, then keep the chosen geometry fixed during spectral qualification.

Write purchasing requirements that can be verified

Request supporting information rather than assuming that a white-light designation includes spectral control. The supplied catalog entries identify useful candidate geometries, but they do not provide CCT ranges, CRI values or spectral tolerances.

A practical purchasing specification should ask for:

• The required emission option and any available CCT range or chromaticity tolerance.
• A spectral power distribution, including the operating conditions used to measure it.
• The CRI reporting method and value, if human-oriented color rendering is relevant.
• Available information about unit-to-unit variation and changes during operation.
• Identification sufficient to trace the supplied configuration and approved replacement.
• Advance notification requirements for changes that could affect optical performance.

Separate requested documentation from confirmed capabilities. If a parameter cannot be guaranteed, record that limitation and decide whether incoming application testing provides adequate control.

Keep an approved reference light and a protected sample set. Store raw captures, acquisition settings, measurement regions, feature calculations and acceptance limits together. A saved image without its acquisition recipe is a weak replacement reference.

When the usual answer is wrong

Choosing the same nominal CCT is reasonable for narrowing candidates, but inadequate when materials sit close to a decision boundary. Choosing the highest available CRI is also insufficient if your camera or materials respond differently from the standardized human observer.

A uniform intensity change may be correctable through an approved exposure or intensity adjustment. A material-dependent change is more serious: correcting one sample may move another in the wrong direction. Test that distinction before allowing automatic normalization.

Fluorescent materials require additional care because they emit light at wavelengths different from those absorbed. The simple reflectance formula no longer describes the complete response. Include every relevant material family in qualification rather than extrapolating from a neutral target.

For color conformity to a customer-defined standard, a camera classifier may not be the appropriate measurement instrument. Establish the required colorimetric method and viewing conditions first. Inspection repeatability and agreement with human appearance are related goals, but they are not interchangeable.

Approval checklist and next step

Before approving a replacement white light, check the following:

• Geometry is repeatable: position and angle changes have been separated from spectral changes.
• Settings are controlled: automatic exposure and balance cannot conceal a mismatch.
• Signals remain usable: no relevant region clips or falls too close to the noise floor.
• Samples represent production: boundary cases and material variations are included.
• Limits are explicit: permitted drift and required decision margins are written down.
• Multiple units are considered: one passing candidate does not establish population variation.
• Recalibration is governed: any permitted adjustment has a documented verification step.
• Records support replacement: approved configuration, captures and acceptance calculations are retained.

Start with the lighting configuration step to select a suitable geometry. For spectral requirements or replacement qualification, send your sample descriptions, acquisition recipe and acceptance limits through application support before ordering.