A new camera system spots microscopic defects in high-tech glass while it moves along the production line at more than 400 millimetres per second. High-tech glass used in semiconductor wafers, solar panels, and display screens must be manufactured with zero defects. Current inspection methods rely on manual checks or slow automated tools that miss internal flaws such as bubbles, inclusions, and scratches. By the time defects are found—after polishing, cutting, and bonding—entire production batches must be discarded, costing manufacturers millions of pounds. The GlassEye system combines a light-field camera with AI-powered image analysis to detect, locate, and classify defects in unpolished toughened glass in under one second, with a false-negative rate below 0.1 percent. If commercialised, the system could reduce material waste and energy use, cut costs by 15 percent through fewer rejected batches, and replace unreliable manual inspections. The consortium projects £39.5 million in revenue and 82 new jobs over five years. For the general public, the impact is invisible but concrete: cheaper, more reliable electronics and solar panels, and less glass waste sent to landfill.
View original technical description
Advanced high-tech glass is utilised in a growing array of high-value applications, including semiconductor wafers, lithography photomasks, photonics, optical flat panels, displays, photovoltaic (PV) panels, etc. These products require high-throughput, zero-defect manufacturing. Common defects during manufacturing include internal imperfections such as inclusions, seeds, bubbles, and blisters, as well as surface damage caused by improper handling, including holes, dirt, scratches, and spots. These defects often go unnoticed before being polished, cut, cleaned and bonded to a display, at which point they are non-recyclable. Consequently, production lines have to be shut down, and entire production batches are often rejected and discarded if defects are found, costing millions of pounds to the manufacturer.Current glass inspection methods are manual, unreliable, costly, and lack real-time feedback. Existing automated 3D inspection tools struggle with advanced glass, while technologies like confocal scanners and Moiré cameras lack the speed and resolution for accurate high-tech glass inspection. **GlassEye** is a light field camera system for online production quality control. Its AI-powered software uses advanced image processing and machine learning, including CNNs for defect detection, segmentation for localisation, and classification for categorisation. The system enhances image quality, identifies defects with high precision, and provides actionable insights. Defect data is fed back to the production line to enable targeted responses, such as discarding defective glass or optimising process parameters. This ensures faster, more accurate defect detection and continuous process improvement. _The proposed solution results in **several advantages as:**_ * Accurate 3 µm image resolution over a large area ( 80 fps/\\\>400 mm/sec, <1 sec decision time) and categorisation of flaws in unpolished toughened glass * Reliable classification (<0.1% false negative, <1% false positive rate) and relation of defects to production process * Minimisation of energy use, material waste, CO2 emissions, and other resources on defective product and its rework * Significantly reduced labour-intensive OPEX by replacing unreliable manual inspections, enabling glass production in high-wage countries. * 15% reduced costs due to reduce e-waste (rejected batches). **Our initial target market** is the global surface vision and inspection market, valued at $2.53Bn(~£1.98Bn) in 2022\. Project outputs (a plenoptic camera system with integrated AI software platform incorporating machine-learning optimisation) will be commercialised, generating £39.5m in revenues, £31.1m in profit, a 22.9:1 return on grant investment over 5 years post-project, and the creation of 82 jobs within the consortium and the supply chain for the same period.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know