Automated Sorting of Recycled Optical Glass by Clarity

Recycled optical glass can have considerable value, yet its quality varies widely after collection, dismantling and crushing. Clear lenses, transparent cover glass and high-grade borosilicate fragments may sit alongside tinted pieces, frosted material, laminated remnants and dirt. When these streams are processed together, a small amount of contamination can lower the value of an entire batch.

This case study examines a representative Australian recycling line that introduced automated optical inspection to separate recovered glass according to clarity. The system used controlled illumination, industrial cameras, image-processing software and pneumatic rejection to classify material continuously rather than relying on manual inspection.

The project was designed for a processor supplying glass feedstock to manufacturers of laboratory products, lighting components and specialty glazing. Its goals were practical: improve product consistency, reduce sorting labour, increase recovery from mixed loads and create a traceable quality record for every production run.

The approach also suits optical-equipment manufacturers and integrators working with machine vision, surface inspection and material classification. A carefully calibrated line can identify visual differences that are difficult to judge reliably at conveyor speed, especially when fragments vary in size, thickness, edge shape and surface condition.

The Feedstock Problem

The processor received optical glass from damaged instruments, discarded display assemblies, lighting equipment and commercial fit-outs. Collection patterns changed with local contracts and demolition schedules. Loads from Sydney and Melbourne often contained a mixture of clean production offcuts and post-consumer fragments, while regional consignments tended to include more dust, labels and mixed packaging residue.

The existing process used a vibrating feeder, a belt conveyor and two manual inspection stations. Operators removed visibly coloured or opaque pieces, but fatigue and changing ambient light affected decisions. A fragment that appeared clear under one lamp could show a grey cast under another. Fine scratches, embedded particles and thin coatings were also easy to overlook.

The business needed a repeatable definition of clarity. In the new specification, premium material had to transmit light with minimal haze and no visible colour contamination above a defined size. A second grade allowed mild haze or a pale tint, while opaque, heavily coated, cracked or contaminated pieces were directed to a lower-value stream.

Building A Machine Vision Classifier

The automated line began with feed preparation. A vibrating conveyor spread the fragments into a shallow, reasonably even layer and reduced overlap. Air knives removed loose dust, while a magnetic stage captured ferrous particles before imaging. This preparation was essential because a camera cannot make a reliable judgement when one piece hides another.

The inspection enclosure used diffuse backlighting to measure transmission and a separate angled light source to reveal scratches, chips, coatings and surface deposits. High-resolution monochrome cameras recorded brightness through each fragment, while a colour camera identified green, brown, blue and amber contamination. The software calculated average transmission, haze, colour shift, edge opacity and defect area.

A transparent test piece did not automatically qualify as premium glass. Thin fragments could appear bright simply because they allowed more light through, whereas thicker pieces naturally produced a lower signal. The algorithm therefore used conveyor position, calibrated reference targets and an estimated size range to normalise readings. This helped distinguish genuine cloudiness from normal variation in thickness.

The line also used a reject confirmation camera. Once a pneumatic nozzle removed a fragment, the confirmation image checked whether the correct item had entered the intended bin. This closed-loop arrangement provided useful evidence during commissioning and reduced disputes between production and quality teams. For applications that require broader inspection capability, the processor reviewed vision verification tools as part of its equipment selection research.

Defining Clarity Grades

The team established three commercial grades rather than treating clarity as a simple pass-or-fail decision. Premium clear glass required high light transmission, low haze and no significant colour cast. Standard clear glass allowed small areas of surface marking or a slight tint. Reject material included opaque fragments, strongly coloured pieces, laminated sections, ceramic contamination and pieces outside the accepted size range.

The following operating values illustrate the classification logic used during the trial. Exact thresholds vary with glass chemistry, thickness, customer requirements and camera configuration, so they must be confirmed with physical samples rather than copied directly into another installation.

Grade Typical visual condition Indicative transmission Common destination
Premium clear Bright, colour-neutral, low haze Above 92% Optical and laboratory products
Standard clear Slight haze or mild tint 78–92% General glazing and lighting
Coloured or hazy Noticeable tint, coating or cloudiness 45–77% Decorative or secondary applications
Reject Opaque, laminated, heavily contaminated or damaged Below 45% Further processing or disposal

Samples were collected across a full week of production rather than from a single clean delivery. Operators labelled pieces by source, approximate thickness and visible condition. The engineering team then used these samples to train and validate the classification rules. This prevented the system from being optimised for an unusually easy batch.

Clarity data was combined with size and contamination checks. A large clear shard with a glued label could not enter the premium stream, while a small clear fragment below the customer’s minimum size was rejected even if its optical reading was excellent. Linking several inspection criteria in one decision made the output more useful to downstream buyers.

Results From The Trial

During an eight-week pilot, the processor compared the manual line with the automated system under similar feed conditions. Manual sorting achieved acceptable results on clean material but became inconsistent when throughput rose or the incoming mix contained fine fragments. The machine maintained a stable decision threshold throughout the shift and created images for samples close to the grade boundary.

Premium-grade recovery increased because clear material was no longer discarded with uncertain pieces. At the same time, the reject stream became more concentrated, reducing the amount of low-quality glass sent to customers who expected a cleaner feedstock. The processor reported fewer customer complaints linked to haze and colour variation, although the greatest improvement came from better separation of mildly tinted fragments.

The trial also reduced direct visual inspection time. Staff moved from continuously judging every piece to supervising the feeder, checking exception samples and managing routine quality tests. This changed the role of the sorting team rather than eliminating the need for skilled workers. Human review remained important when a new source material entered the plant or when a customer changed its specification.

The financial case depended on several factors: recovered value per tonne, reject rates, labour costs, compressed-air consumption and equipment uptime. A line processing modest volumes could justify automation through improved product pricing, while a high-volume plant might gain more from reduced rework and reliable supply contracts. The result was measured across the whole operation rather than through labour savings alone.

Adapting The System To Australia

Australian recycling facilities operate across very different conditions. A processor near Brisbane may deal with humid air and condensation on early-morning shifts, while a plant in Adelaide or Perth may face dry dust and high summer temperatures. Enclosures, air handling, lens protection and cleaning intervals must suit the site instead of relying on laboratory conditions.

Feedstock is also shaped by local collection systems. Container Deposit Scheme material in New South Wales, Queensland and Western Australia is generally more controlled than mixed kerbside material, while commercial demolition and manufacturing offcuts can contain larger, cleaner pieces. A line serving Melbourne’s industrial market may therefore need different feeder settings from one handling mixed recovery near Newcastle or regional Queensland.

Regulatory obligations influence the plant layout and records. Operators must account for state environmental protection requirements, workplace safety duties, waste transport documentation and the National Waste Policy Action Plan. The system should preserve batch information showing source, grade, reject reason and destination, helping the recycler demonstrate responsible handling when customers or auditors request evidence.

Local buying habits matter as well. Australian manufacturers often prefer dependable specifications and regular deliveries over occasional loads with a higher headline purity. Automated inspection supports that expectation by giving a supplier repeatable measurements and trend data. It can also help a processor identify whether quality problems originate with a particular collection contractor, product stream or shift.

Integration With Existing Equipment

The inspection module was installed between size separation and final packing. This position prevented very small fines from confusing the optical classifier and allowed the machine to inspect fragments after the main dust-removal stage. Existing conveyors were retained where belt speed and tracking could be controlled accurately, reducing the cost and disruption of the retrofit.

Lighting geometry was the most important integration issue. Ambient sunlight entering through a roller door caused false readings during the first site test, and reflections from polished fragments produced bright spots in the image. The team enclosed the inspection zone, added matte internal surfaces and used reference tiles at scheduled intervals to check illumination stability.

The software interface displayed production rate, grade proportions, reject causes and alarm status. Operators could select a validated recipe for different thickness ranges or customer specifications, while recipe changes required supervisor access. This limited accidental adjustments and created a record of which settings were active for each batch.

Remote diagnostics were useful when the plant operated outside normal office hours, but access was controlled through the customer’s network policy. Camera images and inspection logs were retained according to the processor’s quality system. For equipment teams studying process efficiency in related reflective materials, research into solar-thermal yield offers a useful parallel: stable illumination and consistent surface data are central to both applications.

Quality Assurance And Maintenance

Automated sorting works best when calibration is treated as a production task rather than a one-off installation activity. At the start of each shift, the operator passed certified clear, tinted and opaque reference pieces through the inspection zone. The system compared measured values with permitted ranges and issued an alert when the camera, light source or conveyor position needed attention.

The plant scheduled lens cleaning, air-filter replacement and nozzle checks according to operating hours and contamination levels. Dust on a camera window gradually reduced contrast, while a partially blocked nozzle created misplaced rejects. Maintenance staff reviewed reject confirmation images as part of their daily checks, allowing small mechanical problems to be corrected before they affected a full shipment.

Statistical process control provided another layer of confidence. The software tracked the percentage of premium, standard and rejected material by source batch. A sudden change in premium recovery could indicate a lighting fault, a feeder problem or a genuine change in incoming glass. These trends helped the quality manager separate equipment performance from variations in the waste stream.

The processor retained manual sampling for verification. Every defined quantity of sorted material was checked against the machine’s grade decision, with particular attention to pieces near the clarity threshold. This combination of automated consistency and human audit produced a defensible quality record for customers purchasing recycled optical glass.

Commercial Value And Wider Applications

The clearest commercial benefit was the creation of a reliable premium stream. Buyers could specify transmission and haze limits with greater confidence, while the processor could price material according to measurable quality rather than visual claims. Better separation also reduced the risk that a few coloured or cloudy pieces would downgrade an entire shipment.

The same platform can be expanded beyond clarity. Additional cameras and lighting angles can identify scratches, edge chips, printed markings, coatings, bubbles and foreign particles. With suitable training samples, the software can classify glass by colour, defect type or intended end use. This is relevant to optical lenses, protective covers, solar components and precision glass products.

The equipment manufacturer’s broader optical engineering capability is relevant when designing these systems. Experience with lenses, surface-profile testing and machine-vision monitoring helps teams select suitable cameras, illumination and algorithms for difficult transparent materials. Background information about company credentials can help procurement teams assess a supplier’s manufacturing and integration experience before specifying a custom line.

For Australian recyclers, the case demonstrates that optical sorting is a quality-control investment as much as a labour-saving measure. By combining controlled imaging, clear grading rules, traceable software and routine calibration, a plant can turn inconsistent recovered glass into several dependable product streams. The result is a process that responds more effectively to local collection patterns, customer specifications and the increasing demand for usable recycled materials.