Optimising machine vision lighting for dome-shaped optical components
The task is to optimise machine vision lighting for dome-shaped optical components without allowing curvature, glare or changing ambient conditions to hide defects. A hemispherical lens, cover, reflector or coated window can produce several different image behaviours at once: bright specular highlights, dark bands near the edge, distorted reflections and uneven illumination across the centre.
A reliable inspection system treats the light source, camera, lens, polarisation, exposure and software as one optical arrangement. Increasing brightness alone rarely solves the problem. In many cases, a softer source, a longer exposure, cross-polarisation or a different viewing angle produces a more useful image than a higher-powered LED.
This is particularly important for Australian production environments. Equipment may operate in a bright Sydney factory, a dusty Queensland solar facility or a remote mining-related workshop where cleaning intervals and ambient temperature vary. A design that works on a laboratory bench must be tested against local dust, daylight, operator routines and the production speed required by the market.
Understand how curved surfaces distort images
A flat part usually presents a comparatively predictable relationship between the light, the surface and the camera. A dome-shaped component redirects light across its changing radius. The centre may reflect the LED array directly into the camera, while the shoulder of the dome sends the same light away from the sensor. This creates a hot spot in one region and insufficient contrast in another.
Surface finish has an equally strong effect. Clear optical plastics can transmit, refract and reflect light simultaneously. A coated glass dome may show pinholes or coating variation only at particular angles. Polished metal reflectors often behave like mirrors, making the camera record the shape of the light source rather than the shape of the defect.
Before selecting hardware, classify the inspection target. Scratches, pits and inclusions are usually detected through local contrast. Shape errors require stable silhouettes, edge positions or fringe patterns. Coating variation may be revealed by colour or spectral response. A useful optical testing resource can support background research, but the final setup should be based on samples from the actual production batch.
The field of view must cover the complete dome without placing the rim too close to the image boundary. When the edge is important, a telecentric lens or a carefully calibrated entocentric lens can reduce apparent size changes. The camera should be centred mechanically rather than corrected later with software, since digital correction cannot recover detail lost through oblique viewing or saturation.
Select a lighting geometry that controls reflections
A large diffuse dome light is often the best starting point for transparent or glossy components. It surrounds the part with a broad, uniform luminance field and reduces the sharp reflection of individual LEDs. For a clear cover, this can reveal dust, bubbles and surface marks against a controlled bright background.
Diffuse overhead lighting does not suit every defect. A low-angle dark-field ring can make scratches and raised particles appear bright against a dark image. A coaxial light can help when the relevant feature is near the optical axis, while a backlight is effective for checking the outer profile, wall thickness or missing material. If the dome is deep, several lighting positions may be required to cover the crown and the perimeter separately.
For highly reflective surfaces, use the light source as a visual mask. A curved diffuser, tunnel light or hemispherical enclosure can create a reflection with controlled boundaries. The goal is not simply to eliminate reflections; it is to make the reflection predictable. A narrow, uncontrolled highlight may conceal a defect, whereas a broad uniform reflection allows software to identify small deviations.
Dark-field illumination is especially useful for edge chips and fine abrasions, but it can exaggerate normal mould lines. Set the incident angle just above the threshold at which the target defect becomes visible, then verify the result across multiple samples. Lighting that is too oblique may create false positives whenever the part is rotated slightly or seated at a different height.
Balance wavelength, polarisation and spectral response
Visible white light is convenient, but it may not provide the highest defect contrast. Blue light has a shorter wavelength and can emphasise fine surface scattering, although it may be absorbed differently by coloured plastics. Red or near-infrared illumination can pass through some materials more effectively. The correct choice depends on the component’s transmission, coating and defect type.
Polarising filters can suppress glare from non-metallic surfaces. Place a linear polariser over the light and a second polariser over the camera, rotating one by roughly 90 degrees for cross-polarised imaging. This arrangement reduces direct reflected light while retaining some scattered light from scratches, stress marks and contamination. It is less effective on metallic mirrors and can reduce the available signal, so exposure and LED power must be checked again.
The following selection guide connects common inspection objectives with practical lighting arrangements:
| Inspection objective | Starting light arrangement | Useful camera control | Common risk |
|---|---|---|---|
| Outer profile or missing section | Collimated or diffuse backlight | Short exposure, edge thresholding | Internal transmission creates a false edge |
| Scratches and chips | Low-angle dark-field ring | Cross-polarisation, moderate gain | Mould lines appear as defects |
| Coating uniformity | Large diffuse dome or coaxial light | Fixed exposure, colour or monochrome analysis | Hot spots hide gradual variation |
| Bubbles and inclusions | Transmitted diffuse panel | Longer exposure, low gain | Background texture resembles inclusions |
| Reflector shape | Controlled directional panels | HDR or multiple exposures | Mirror-like glare saturates the sensor |
| Fine contamination | Dome light with cross-polarisation | Flat-field correction | Dust on the light appears on every image |
Spectral selection should be validated with image data rather than visual impressions. A component that looks uniform to an operator under factory lighting may show strong variation under blue or near-infrared LEDs. For teams comparing wavelength options, multi-spectral guidance offers useful terminology for evaluating several bands, though production decisions still require controlled trials with representative parts.
Set camera exposure around useful signal
A bright image is not necessarily a good inspection image. If the sensor clips at the dome’s central highlight, the software loses all information in that region. Use the camera’s histogram or pixel statistics to keep critical surfaces below saturation. A small specular highlight may be acceptable if it lies outside the inspection zone, but it should never be allowed to move unpredictably between parts.
Keep analogue gain low wherever possible. Gain raises signal and noise together, which can make a clean plastic surface look granular. Increase illumination or exposure before increasing gain, provided motion remains controlled. If the part rotates during inspection, a pulsed strobe can freeze movement and deliver greater brightness without requiring a long exposure.
High dynamic range imaging is useful when the crown and side wall require very different exposure levels. Capture two or more images with controlled exposures, then combine them for analysis. The camera, part and light must remain stationary between captures. In a fast line, separate cameras or sequential strobes may be more practical than HDR processing.
Lens selection also affects the usable signal. A lens with insufficient resolution will blur fine cracks, while excessive magnification may exclude the dome’s edge. Check the modulation transfer function, working distance and distortion at the actual field of view. For measurement tasks, calibrate with a suitable target at the same height as the component rather than relying on a nominal lens specification.
Build calibration into the production cell
Calibration should establish more than a pixel-to-millimetre scale. Record the camera position, lens focus, light intensity, exposure, polariser angle and trigger timing. Save a reference image of an approved component and, where possible, images of known defect standards. This creates a baseline for detecting drift before rejected parts accumulate.
Flat-field correction can compensate for gradual brightness variation across a dome light or camera sensor. It cannot correct a loose mount, a cracked diffuser or a component that is not seated consistently. Use a mechanical nest with a defined datum, especially when inspecting parts supplied from several moulding or polishing lines.
Australian conditions make environmental testing important. Bright daylight entering a loading-bay door in Melbourne can alter a poorly shielded setup, while dust at a Queensland solar site can settle on a diffuser and reduce contrast. In Perth or Adelaide, dry air and fine particles may increase static contamination on plastic domes. Enclose the optical path, use a replaceable protective window where suitable, and include cleaning in the normal shift schedule rather than waiting for image quality to fail.
Safety and compliance must be part of the design review. Laser-based illumination requires attention to applicable laser safety requirements, including AS/NZS IEC 60825.1 where relevant. Guarding, interlocks and electrical installation should align with Australian workplace safety obligations and the machinery risk assessment used by the site. A lower-powered diffuse LED system is often easier to guard and maintain than a concentrated optical source.
Validate with production samples and defect data
Laboratory samples are useful for initial alignment, but they rarely represent the full range of production variation. Test clear, tinted, coated, scratched, dusty and slightly misaligned components. Include parts from different mould cavities, suppliers and shifts. Record false rejects as carefully as missed defects, since a lighting arrangement that detects every mark may be commercially unusable.
A practical validation sequence starts with mechanical alignment, followed by illumination uniformity, focus, exposure and polarisation. Capture images at the fastest intended line speed. Then repeat the trial after the light has warmed up and after the component has been deliberately shifted within its expected seating tolerance. The system should preserve defect contrast under these normal changes.
Use quantitative measures where possible. Contrast-to-noise ratio, saturated-pixel percentage, edge-position error and repeatability across repeated captures provide stronger evidence than a visual preference for one image. For colour or coating inspection, measure channel response with fixed white balance and disable automatic camera settings. Automatic exposure and gain can make a stable process appear inconsistent.
Production sites in Australia may serve long distances and small specialist markets, so maintainability has a direct effect on return on investment. A local operator should be able to replace a diffuser, clean a window and verify a reference image without waiting for an overseas technician. Standardised LED modules, documented cable paths and accessible adjustment points reduce downtime when a line is running extended shifts or servicing remote customers.
Connect lighting decisions with inspection software
Software should receive images that already contain meaningful optical information. Thresholding, blob analysis, edge measurement and defect classification are more reliable when illumination is stable. If the system depends on aggressive background subtraction to remove a moving highlight, the underlying lighting geometry probably needs refinement.
For dome-shaped parts, divide the image into logical zones when brightness naturally changes with curvature. The crown, shoulder and rim can use separate intensity normalisation or defect thresholds, provided the zones are defined by the calibrated part position. A single global threshold is simple, but it may fail when the optical path creates predictable regional differences.
Machine-learning inspection can classify subtle surface variation, yet training images must include lighting drift, acceptable mould marks and realistic contamination. Keep a versioned record of the light settings used to collect each training image. Changing the diffuser or wavelength without updating the dataset can alter the appearance of both good and defective parts.
An inspection reference should also describe why a setting was chosen, not just list numerical values. Teams comparing equipment or reviewing inspection reference material can then distinguish a repeatable optical principle from a setting that worked only for one sample. The strongest systems combine controlled illumination, calibrated mechanics, stable camera parameters and software rules that operators can audit.
When these elements are designed together, curved optical components become measurable rather than visually unpredictable. The result is a repeatable machine-vision process that protects defect sensitivity, limits false rejects and remains practical across Australian factory and field conditions.