Use RF-DETR for Glass Defect Detection
RF-DETR is a real-time transformer architecture for object detection and instance segmentation that handles the subtle morphology of inclusions, bubbles, and coating defects on transparent product.
Add a real-time inspection layer to every glass product on the line with Vision AI for glass defect detection. Built for the operations where one missed inclusion in float glass destined for premium architectural glazing, undetected stone in a container destined for premium beverage, coating defect on a tempered automotive windshield, or edge crack on a specialty display glass can mean rework downstream, a warranty claim, an automotive recall on a deployed safety glass program, a customer chargeback that erodes margin and supplier reputation, or a field-failure that ends in litigation on a regulated safety glazing program. Whether you're inspecting float glass coming off the tin bath, container glass on the forming line, automotive glass at tempering and laminating, specialty display glass coating, edge inspection on finished panels, or final QC before shipping, Roboflow extends your QC coverage to every glass product on the line, on the cameras and inspection stations your facility already runs.
Float, Container, and Forming Defects:
Surface, Edge, and Coating Defects:
Tempering, Lamination, and Final Safety QC:
Bring intelligence to every glass product today. Stop glass defects from becoming warranty claims, recalls, customer chargebacks, or safety glazing field failures.
What is glass defect detection with Vision AI?
Glass defect detection with Vision AI uses computer vision models to inspect glass products at every stage of manufacturing, from float glass coming off the tin bath through container forming, automotive tempering and laminating, specialty display glass coating, edge inspection on finished panels, and final QC before shipping. The system extends QC coverage to every glass product on the line, catching inclusions and stones (refractory particles from the furnace), bubbles and seeds (gas pockets formed during melting), knots and gobs of un-melted material, cords and birefringence patterns (optical distortion from un-mixed glass), scratches and digs and abrasions, tin-side defects from the float process, edge chips and cracks, coating defects on low-E coatings, anti-reflective coatings, and mirror silvering, lamination defects on PVB-laminated assemblies, and stress and warp defects from tempering. Float glass producers (Pilkington/NSG, Saint-Gobain, Guardian, AGC, Vitro), container glass manufacturers (Owens-Illinois, Ardagh, Verallia, Vidrala), automotive glass suppliers (Saint-Gobain Sekurit, AGC Automotive, Pilkington Automotive, Fuyao), and specialty display and optical glass producers (Corning, Schott, NEG) use it to cut rework, prevent customer chargebacks, reduce recall risk on automotive safety glazing, defend against field-failure investigations, and document compliance under ASTM C1036, ASTM C1048, ASTM C1503, ANSI Z97.1, EN 12150, EN 1279, FMVSS 205, ECE R43, IATF 16949, USP <660>, and customer-specific PPAP and APQP requirements.
Can Vision AI catch the inclusions and bubbles that rule-based vision struggles with on transparent product?
Yes. Inclusions, bubbles, and subtle defects in transparent product are exactly where rule-based and template-based machine vision systems feel the most pressure. Rule-based vision excels at deterministic measurement tasks with high-contrast features, fixed lighting, and consistent product presentation (precise 2D dimensional measurement of glass sheet edges, presence-or-absence of high-contrast features), but struggles with glass defects because glass is transparent and refractive (rule-based detection has limited contrast to work with), defect morphology varies lot to lot (an inclusion in one batch looks different from another batch from the same furnace), product thickness and tint varies (thicker or tinted glass refracts differently), and SKU complexity spans hundreds of product specifications across float, container, automotive, and specialty glass programs. Roboflow models add a deep-learning inspection layer trained on your actual product appearance, lot variation, and grade-specific characteristics, catching the defect categories rule-based vision struggles with and co-piloting existing glass surface inspection systems from ISRA Vision (now Atlas Copco), Sparklike, Glassrobots, Light Works, and specialty container inspection systems from Krones, Heuft, Tiama, and Iris Inspection Machines by adding visual verification on borderline rejects (reducing false-positive scrap from over-sensitive thresholds, increasing true-positive confidence on safety-critical automotive and architectural safety glazing programs).
Does glass defect detection support ASTM C1036, ANSI Z97.1, FMVSS 205, and IATF 16949?
Yes. Roboflow models can be trained against your specific ASTM C162 (Standard Terminology of Glass and Glass Products), ASTM C1036 (Standard Specification for Flat Glass), ASTM C1048 (Standard Specification for Heat-Strengthened and Fully Tempered Flat Glass), ASTM C1503 (Standard Specification for Silvered Flat Glass Mirror), ANSI Z97.1 (Safety Glazing Materials Used in Buildings), EN 12150 (European tempered glass), EN 1279 (European insulating glass), FMVSS 205 (US automotive safety glazing), ECE R43 (European automotive safety glazing), IATF 16949 for automotive glass quality management, AS9100 for aerospace glazing, USP <660> for pharmaceutical glass containers (specifically relevant for Type I borosilicate pharma vials), ISO 9001, and customer-specific PPAP (Production Part Approval Process) and APQP (Advanced Product Quality Planning) acceptance criteria. The system applies the same pass/fail logic your trained quality inspectors, float bath operators, and tempering process leads use, against your written specifications, ASTM and EN acceptance criteria, and customer release documentation, and produces validated inspection records that support customer audits, automotive Tier 1 PPAP submissions, FMVSS recall investigation defense on automotive safety glazing, regulatory enforcement on architectural safety glazing, and traceability to the heat, lot, and forming run for downstream quality investigation. Your glass quality teams own the acceptance criteria; Roboflow provides the inspection engine that enforces them at line speed across every glass product.
Can it integrate with our float line PLCs, container forming line, tempering oven control, MES, eQMS, and ERP?
Yes. Roboflow Inference exposes a standard API and supports common glass manufacturing automation protocols, so Vision AI glass defect detection events flow into your existing float line PLCs, container forming line, tempering oven control, MES, eQMS, ERP, and field-failure traceability platforms. Customers integrate with float line PLCs from Allen-Bradley, Siemens, and ABB, container forming line PLCs from Bottero, GPS, and Heye, tempering oven control from Glaston, LandGlass, and HHH Tempering, lamination line equipment from Glaston, Lisec, and Bystronic Glass, specialty glass inspection systems from ISRA Vision (now Atlas Copco), Sparklike, Glassrobots, and Light Works, container glass inspection from Krones, Heuft, Tiama, and Iris Inspection Machines, glass MES platforms (Lisec FENZI, Glaston platforms, custom plant systems), eQMS platforms (MasterControl, Veeva Vault QMS, Sparta TrackWise, ETQ Reliance), and ERP systems (SAP, Oracle) through REST, MQTT, OPC UA, and direct database writes, with PLC-level integration to float line cooling rates, container forming gob distribution, tempering oven cycle decisions, and downstream cutting and sorting where pass/fail decisions need to drive line behavior. Models support full IQ/OQ/PQ documentation, audit trails for training data, model versions, and inspection results that pass customer audits, automotive Tier 1 PPAP submissions, FMVSS recall investigation requirements, architectural safety glazing regulatory enforcement defense, and USP <660> pharmaceutical glass container compliance documentation.
