Machine vision lighting stability: test before deployment

A stable inspection needs more than a bright image. If illumination changes between captures, a threshold can move even when the part remains unchanged. Automatic camera adjustments can hide that instability during setup, then expose it when production conditions change.

This qualification plan is for engineers accepting an inspection station or specifying its lighting controls. You will establish a fixed reference, measure variation through warm-up and repeated triggers, and test sensitivity to room lighting. The aim is to distinguish imaging-chain drift from genuine part differences before setting acceptance limits.

Decide what stability means for the inspection

Machine vision lighting stability is the repeatability of useful illumination over the conditions your inspection actually encounters. Those conditions include startup, normal production, idle periods, trigger bursts and changes in background illumination.

Brightness repeatability alone is not enough. A light can maintain its average output while its pulse moves relative to exposure. Conversely, a stable pulse can illuminate a target differently if a mount shifts. Qualify the complete image-forming arrangement, then investigate individual components when measurements change.

Keep three requirements separate:

• Temporal repeatability: the same stationary reference produces similar measurements across captures.
• Spatial repeatability: different image regions retain their relative brightness and contrast.
• Inspection repeatability: the measured feature or classification remains within its allowed variation.

Start with a fixed operating point, not feedback corrections. The guide to closed loop lighting control explains what feedback can address after you understand the underlying variation.

Lighting geometry must already reveal the feature. If contrast is weak even under steady conditions, revisit machine vision illumination types before qualifying stability.

Build a reference measurement that cannot follow the drift

Use a clean, matte reference target that fills the relevant inspection area. Fix its position, angle and distance mechanically. A production part can serve as a second reference, but texture, curvature and placement variation make it a poor first diagnostic target.

Select several regions of interest: one near the center and others near the field edges. Add regions corresponding to critical inspection features. Keep all regions away from glare, shadows, target edges and saturated pixels.

Freeze the camera settings:

• Exposure time and gain.
• Black level and gamma or other tone mapping.
• White balance and color processing, when applicable.
• Pixel format, region size and image-processing pipeline.
• Automatic brightness, contrast and exposure functions.

Use raw or minimally processed images where available. A mean pixel value is a relative imaging measurement, not a calibrated measurement of optical power. Its interpretation depends on the camera response remaining unchanged.

The camera exposure, gain and balance guide helps establish a reproducible acquisition configuration.

Define the numbers before running the test

For each image, calculate the mean intensity within each reference region. Also record the inspection's actual measurement, such as edge position, feature contrast or a classification score.

Useful stability metrics are:

Calculate these consistently on the same signal representation. If you subtract a measured dark offset, subtract it from every applicable sample and document the method. Values near the noise floor make relative percentages unreliable.

CV describes typical spread, while peak-to-peak variation exposes extremes. Neither explains the cause. Plot intensity against capture time and image index to reveal startup drift, alternating frames or changes associated with trigger bursts.

Set limits from inspection margin rather than adopting a universal percentage. For an illustrative station, you might initially investigate a 1% CV, 2% warm-up drift or 5% peak-to-peak spread. These are proposed test limits, not general performance requirements or product specifications.

Finally, test known acceptable and unacceptable samples under the permitted brightness range. A stable mean does not prove that the inspection has adequate separation between those populations.

Run a staged qualification sequence

Change one condition at a time. Keep the target, optical arrangement and image processing fixed throughout the initial tests.

1. Establish the baseline. After the proposed warm-up period, acquire 100 reference images at normal production cadence. Record regional means, saturation counts and inspection measurements.
2. Measure startup. From a representative cold start, collect a short image batch every minute for 20 minutes. Repeat the startup sequence on three occasions. Extend the observation if the signal has not settled.
3. Check repeated triggering. Acquire 1,000 images at the intended cadence. Include the fastest expected burst and the longest normal idle interval.
4. Test ambient sensitivity. Repeat matched measurements with normal room lighting, reduced room lighting and the strongest expected background condition. Do not move the target between conditions.
5. Check recovery after idle. Compare the first captures after a pause with later captures in the same burst. A satisfactory steady-state average can conceal an unreliable first image.
6. Repeat with production samples. Once the reference is stable, introduce representative part and placement variation. Compare its magnitude with the stationary-reference variation.

Record timestamps and controller settings alongside image identifiers. A saved image without its operating condition is difficult to interpret later.

If only the first image changes, use the first-frame exposure and light-timing checks rather than treating the problem as slow thermal drift.

These sample counts are starting points for characterization. Production acceptance may require longer runs to capture rare events or the full operating cycle.

Work through a numerical example

Consider a stationary reference measured in a linear 8-bit image. Assume its independently measured dark offset is negligible for this example, and no pixels in the measurement region are clipped.

A batch has a mean of 150 digital counts and a standard deviation of 1.2 counts. Its minimum and maximum frame means are 146 and 154 counts.

The typical spread is small, but the extremes deserve investigation. Examine those images and their timestamps before changing the controller. Target movement, a late pulse or a processing change could produce the same summary statistics.

Suppose the mean rises from 150 to 156 counts during warm-up:

That exceeds an illustrative 2% drift limit. Possible responses include a validated warm-up requirement, a different operating point or improved regulation. Increasing the acceptance limit is justified only if inspection performance remains acceptable across that range.

Now compare active-light-on and active-light-off images under two room-lighting conditions. Suppose the paired regional means are 150 and 12 counts, then 160 and 22 counts:

Under these assumptions, the change is consistent with added ambient illumination rather than changed active-light contribution. This subtraction requires a linear, unsaturated response and stable conditions within each pair. It is not a substitute for optical calibration.

Separate output drift, ambient effects and timing errors

Use the shape of the change to choose the next test.

Do not assign every image change to the LED source. Camera temperature, gain changes and target movement remain confounding variables until controlled.

For periodic modulation, follow the tests in camera flicker under artificial lighting. For output that changes within a short pulse, the LED strobe current-droop diagnostic addresses a different timescale from minute-by-minute warm-up.

Specify controller regulation and timing separately

Browse light controllers by electrical compatibility first. Then ask for documented behavior over the load, operating range and trigger pattern you intend to use. The controller finder helps narrow the available configurations, but a filter label is not a complete qualification specification.

A constant-current output regulates electrical current; it does not necessarily measure or correct optical output. LED temperature and aging can still change illumination. Likewise, a digital setting does not by itself specify optical repeatability, trigger latency or pulse jitter.

The LS-CTL-0180 is listed as a constant-current controller. The LS-CTL-0290 is listed as a digital constant-voltage light controller. Compare their catalog information, then confirm the intended light's electrical requirements.

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

For triggered operation, specify input electrical requirements, polarity, trigger mode, latency, jitter, pulse-width range and recovery between pulses. Request any limits that the available product information does not establish.

The LS-CTL-0070 is cataloged with a strobe feature. Confirm its detailed timing and load compatibility before selecting it for a pulse-sensitive inspection.

Do not infer overdrive permission from the presence of a strobe function. The strobe duty-cycle and timing guide explains why both the light and controller need documented limits.

Verify the exposure window, not just the trigger

With pulsed illumination, the relevant question is whether useful light overlaps the intended exposure consistently. Trigger repeatability alone does not answer that question.

For an illustrative global-exposure arrangement, suppose stable light spans 0.2–1.6 ms after a trigger. An exposure starting at 0.4 ms and lasting 1.0 ms finishes at 1.4 ms. That leaves a nominal 0.2 ms margin at each end.

Those margins must accommodate measured jitter, pulse rise and fall behavior, and operating changes. A rolling-shutter sensor requires a different check because image rows may integrate at different times.

Also calculate duty cycle for the actual sequence:

This arithmetic does not establish a safe operating limit. Check the documented limits for the complete light-and-controller combination, including burst behavior.

Where instrumentation is available, compare an optical detector signal with the camera's documented exposure signal. Use electrically compatible probes and interfaces. Do not assume a software timestamp marks the physical start of exposure.

Check the installed arrangement before ordering

The laboratory setup should represent the intended installation closely enough to expose electrical and mechanical differences. Specify cable length, connector type, pin assignment and permitted extension arrangements.

Use lighting accessories and the accessory finder to identify candidate connection parts. Cable listings do not establish compatibility with every controller or light.

The LS-LAC-0060 is listed as a light extension cable. Verify its connector, wiring and electrical suitability for the selected combination before including it.

With power isolated according to the equipment instructions, check connections before making wiring changes. Repeat the stability test with the final cable arrangement and normal nearby equipment operating.

The usual fixed-reference answer is incomplete when the target itself changes. Fluorescent materials can fade, reflective surfaces can move highlights, and contamination can alter a reference patch. Replacing or repositioning the reference may then change the measurement even with stable illumination.

Color-sensitive inspections also need channel-specific measurements. An unchanged overall grayscale mean can conceal a change in relative color-channel response.

Turn the measurements into an acceptance checklist

Before releasing the station, document the following:

• Reference and settings: identify the target, regions, camera configuration and controller operating point.
• Startup requirement: state the validated warm-up period and how readiness is determined.
• Repeatability limits: record regional CV, extremes, drift and allowable inspection-measurement variation.
• Timing evidence: verify exposure overlap at normal cadence, after idle and during bursts.
• Ambient tolerance: retain measurements from the expected background-lighting conditions.
• Production validation: demonstrate acceptable decisions with representative good and defective samples across the permitted range.
• Maintenance baseline: retain reference images and define when the qualification must be repeated.

A passing reference test establishes a repeatable imaging baseline, not complete inspection capability. Keep the reference measurements separate from the evidence that the algorithm detects the required defects.

Use the controller configurator once electrical and timing requirements are defined. For help narrowing parts, send the target images, operating sequence and measured limits through applications support.