Understanding Depth of Field for Thick Lens Inspection in Machine Vision
Depth of field describes the zone in which a scene remains acceptably sharp. In machine vision it sets the boundary between captured detail and blurred rejection. For thick lens inspection, this boundary becomes the central engineering constraint because the optical element under test is rarely flat. Curved surfaces, stepped edges, and moulded bevels all sit at different axial positions, and a single fixed-focus camera cannot keep them all crisp at once. Engineers therefore design the optical stack around the depth of field first, then choose illumination, sensor, and processing around it.
Australian manufacturers have taken this constraint seriously. In Sydney and Melbourne, a growing number of precision component suppliers produce ophthalmic blanks, safety eyewear, and ruggedised optics for defence and mining clients. Their production lines must verify each lens for centre thickness, surface figure, and cosmetic defect before shipment. The depth of field available in those cells is often less than a few millimetres, which forces a careful match between lens, light source, and software. Choosing poorly leads to scrap, false rejects, and lost throughput on lines that already operate at the edge of margin in a high-cost local labour market.
The following sections walk through the optical foundation of focal range, the special difficulties created by thick transmissive parts, the trade-offs between aperture and working distance, the calibration methods used in modern lines, and finally the Australian industrial settings where these systems are deployed. Readers familiar with basic imaging will find familiar concepts restated for a quality-control context, while those with a more specialised background can move directly to the trade-off discussion.
The Optical Foundation of Focal Range
Depth of field, sometimes called focal range, is governed by the f-number of the lens, the pixel size on the sensor, and the working distance between the lens and the part. A lower f-number opens the aperture and gathers more light, which shortens the in-focus zone. A higher f-number tightens the beam and lengthens that zone, but only at the cost of signal-to-noise. In a machine vision cell running at high line rates, the engineer is rarely choosing between sharpness and blur in a vacuum; the choice is constrained by the shutter speed required to freeze a moving conveyor or a vibrating pick-and-place head.
Numerical aperture behaves the same way on the microscopic scale. Higher numerical aperture resolves finer surface detail, yet it also compresses the depth of focus. When a lens is thick, light entering at the edge travels a longer path through the glass than light passing near the centre, and the apparent position of each surface shifts according to the refractive index. The visible focus plane can therefore float as the camera scans from centre to edge of the part. Engineers respond either with telecentric optics, which keep magnification constant across the working range, or with coaxial illumination that suppresses the most distracting reflections.
The choice between active and passive measurement shapes everything that follows. A comparison of triangulation vs structured light is particularly relevant here, because the two techniques share the goal of recovering three-dimensional shape but solve the depth problem in different ways. Triangulation offers a wide dynamic range at moderate cost, while structured light projects a known pattern and decodes geometry from its deformation. Both rely on a camera that holds the relevant portion of the part within its depth of field, so the optical foundation is the same even when the engineering choices look very different.
Why Thick Lenses Bend the Rules
A thin lens approximation treats every surface as a single refracting plane. A thick lens must be modelled as two refracting surfaces separated by a finite medium, and the principal planes no longer coincide with the physical surfaces. Light passing through the edge of the part therefore focuses at a different axial position than light passing through the centre. For inspection this means that even a perfectly telecentric lens will record a curved field of best focus, and the image of the back surface will appear at a different image distance than the front surface.
The second problem is internal reflection. A thick glass element can trap light between its surfaces, producing ghost images, fringes, and stray contrast that looks like a defect. Polarising filters help, but they also reduce the light budget and can shift the working point of the system. Many lines add a dark-field illumination ring to push the contrast away from these internal reflections, yet that adds another optical component whose focal behaviour must align with the imaging lens.
The following table summarises how three common sensor classes behave when tasked with imaging a thick transmissive part. It is meant as a quick reference rather than a procurement recommendation, and the numbers will vary with the specific optics and lighting chosen.
| Sensor type | Typical pixel size | Best f-number range | Depth of field at 200 mm working distance | Suitability for thick transmissive parts |
|---|---|---|---|---|
| Global shutter CMOS (5 MP) | 3.45 µm | f/4 to f/8 | roughly 6 to 14 mm | Good for front surface only without focus stacking |
| High-resolution CCD (20 MP) | 5.5 µm | f/5.6 to f/11 | roughly 12 to 28 mm | Strong for full-volume capture if telecentric |
| Time-of-flight depth sensor | 15 µm equivalent | fixed optics | up to 60 mm scene range | Useful for relative thickness, weak on fine surface detail |
The figures are conservative and assume a 35 mm image circle. Real cells often use a smaller format sensor, which tightens the depth of field further and makes the choice of working distance even more important.
Aperture, Working Distance, and Sensor Trade-offs
The classic trade-off is between light, resolution, and depth. Opening the aperture by one stop doubles the photons on the sensor and halves the focal range. Closing the aperture buys back the focal range but cuts light and forces a longer exposure, which in turn risks motion blur. Engineers break the trade-off in three directions: by changing the illumination to deliver more usable photons, by selecting a sensor with larger pixels that need less light, or by moving to a telecentric lens that holds magnification constant across the working range.
Telecentric optics are the most common answer for thick glass inspection. A double-telecentric lens makes chief rays parallel on both sides, so magnification is locked to the lens, not to the working distance. The depth of field shrinks slightly compared to a conventional lens of the same f-number, but the field curvature disappears. The operator can position the part anywhere within a defined axial window and still read the same scale. For an Australian cell running ophthalmic lenses at high mix and low volume, that flexibility is worth the price premium, because changeover time disappears almost entirely.
Sensor choice is downstream of the optics but upstream of the software. For high-volume lines that read a single feature per part, a small-pixel global shutter CMOS is hard to beat. For lines that must verify both surfaces in one shot, a larger-pixel sensor behind a telecentric lens gives a more comfortable depth budget. Beyond the sensor itself, the choice of frame grabber, lens mount, and vibration isolation will shift the usable depth of field by perhaps ten to twenty percent. A solid mechanical design often matters as much as the lens specification itself.
For procurement teams looking to compare vendor offerings across these axes, working with optical inspection specialists helps frame the conversation in measurable terms. The relevant questions are not abstract optical theory but specific data: the working distance envelope, the in-focus zone for the chosen f-number, the maximum line rate, and the reject rate observed in a paid trial. Vendors that quote a single depth-of-field figure without naming the f-number, working distance, and pixel size are usually hiding the trade-off rather than resolving it.
Calibration Strategies for Variable Thickness
Calibration starts with a known reference artefact. For thick lens inspection that artefact is usually a calibrated glass gauge with traceable centre thickness and surface figure, supplied by a national metrology institute. The imaging system is then taught to recognise the gauge, lock focus at a known position, and report the deviation of subsequent parts. Modern lines store several calibration profiles, one per part family, and switch between them automatically when the part identifier changes. Engineers who want to keep pace with new algorithms and reference designs can follow industry updates and benchmark their own cell against published case studies.
Focus stacking extends the available depth of field without paying the resolution penalty of stopping down. A small voice-coil stage or a liquid lens moves the focal plane through the part, and the software merges the in-focus regions into a single sharp image. The technique is well known in microscopy and is now migrating into production cells for ophthalmic and safety eyewear. The cost is more images per part, more processing time, and a stricter requirement on vibration control. Australian lines that already use a atomicbands.com workflow for image archiving will find that stacking fits naturally into the same review pipeline.
For lines that cannot afford the cycle time of stacking, software-only defocus deconvolution is sometimes used. The point spread function is measured for the specific optical stack, and a deconvolution kernel restores contrast outside the strict depth of field. The result is not the same as a true in-focus image, but for cosmetic inspection the recovered contrast is usually good enough. The technique is sensitive to changes in the optical stack, so any swap of lens, extension tube, or filter requires a fresh characterisation.
Calibration should be revisited whenever a new part is introduced, whenever the ambient temperature drifts by more than a few degrees, and whenever a consumable such as a light source is replaced. Australian facilities in regions with strong diurnal swings, such as inland Queensland or the Western Australian Pilbara, often find that their depth of field shifts measurably between a cool morning shift and a hot afternoon shift. Scheduled recalibration absorbs that drift before it becomes a quality escape.
Australian Industrial and Research Context
The Australian optics and photonics industry is anchored by research at CSIRO, the Australian National University, and the University of Sydney, each of which runs programmes in computational imaging, biomedical optics, and precision manufacturing. Adelaide hosts a long-standing cluster of optical fabricators serving defence and astronomy, while Melbourne has built a strong base around biomedical devices and ophthalmic instruments. The presence of those research groups matters for production lines because they offer traceable calibration, failure analysis, and access to specialist metrology that smaller companies could not sustain alone.
Industrial demand comes from sectors that Australia has in unusual concentration. Mining operations in Western Australia and the Northern Territory deploy machine vision to inspect wear on crusher liners, slurry pipelines, and conveyor idlers, and a growing number of those cells now also grade mineral samples by optical signature. Food processing in the Murray-Darling basin, particularly the wine and citrus industries around the Barossa Valley and the Riverina, uses vision to sort fruit and cull damaged stock. Defence procurement, increasingly guided by the national Defence Industry Strategy, specifies locally built optical assemblies and asks for verification data that only a depth-of-field aware inspection cell can supply.
Local regulation shapes equipment choice in ways that overseas suppliers sometimes miss. Australian electrical safety standards, electromagnetic compatibility rules, and the privacy provisions of the federal Privacy Act all touch vision systems in different ways. A vision cell in Brisbane that captures identifiable facial data, for example, must be reviewed against the Privacy Act even if the actual product under inspection is a non-optical item on the same line. Working with integrators familiar with those rules saves commissioning time and reduces the risk of a late redesign.
The market for thick lens inspection in Australia remains small by global standards but is steadily rising as more lens and eyewear production returns from offshore. Quotes are typically issued in Australian dollars, lead times run between eight and sixteen weeks for a fully integrated cell, and local service support is part of the procurement criteria. A line specified with a generous depth of field, a sensible aperture budget, and a clear calibration routine will continue to deliver value long after the original integrator has handed over the keys.