Glass Container Inspector Product

Overview

Glass container inspection by machine vision is a critical quality gate in beverage and food packaging lines. Every bottle, jar, or carton must be free of cracks (which can cause leakage or injury), bubbles (which indicate glass defects), and dimensional faults (which prevent capping or case packing). Manual visual inspection is impractical at speeds >100 containers/minute and is inconsistent. Automated vision inspection using high-speed cameras and LED illumination can reliably detect defects >0.5 mm at 300–1,200 containers per minute, rejecting defective containers in real-time with >99% sensitivity.

The workhorse design is the rotary starwheel: a vertical wheel with 8–16 pockets spaced radially, each pocket holding one container. As the wheel rotates, containers move sequentially past fixed camera stations. A top camera inspects the rim and opening; a side camera (at 45°) inspects sidewall defects; a bottom camera inspects the base and punt. LED arrays (coaxial, backlit, and glancing-light) provide complementary illumination, with transmitted light revealing internal cracks and bubbles, and reflected light showing surface texture. Real-time image processing (on an embedded PC or smart camera) classifies each image as accept or reject in <10 ms. If a defect is detected, a solenoid-triggered air jet or mechanical gripper removes the container from the line before it reaches the downstream filler or seamer.

How it works

Starwheel indexing: Containers arrive on an input conveyor and are singulated one by one onto the rotating starwheel. A spring-loaded or vacuum-operated pocket grips the container opening, holding it securely as the wheel rotates. The starwheel is precision-balanced and rotates at 2–10 RPM (adjusted via VFD motor control), moving containers from the infeed position through three camera inspection stations and a reject station.

Image acquisition: As each container reaches a camera station (indexed by an encoder pulse from the starwheel shaft), a high-speed camera (line-scan or 2D area, 5–12 MP, 30–60 fps) captures one or more frames. LED lights (coaxial, backlit, and side) are triggered simultaneously, illuminating the container from multiple directions. For example, a backlight reveals internal cracks and bubbles; a coaxial light shows surface texture; a glancing light at 45° emphasizes edge defects.

Real-time image processing: Images are streamed from cameras via Gigabit Ethernet to an industrial PC running vision software (typically custom or licensed libraries like OpenCV, Cognex VisionPro, or similar). Algorithms detect:

• Cracks: High-contrast edge lines indicating glass separation. Template matching or edge-following identifies suspicious linear features.
• Bubbles and inclusions: Circular or irregular dark spots in transmitted light indicating voids in the glass.
• Dimensional faults: Deviations in opening diameter, rim thickness, or bottle height measured from edge detection or Hough-transform circle fitting.
• Threads: Defective or missing threads in screw-cap openings.
• Surface defects: Scratches or abrasions detected by texture analysis or glancing light.

Each image is classified as "accept" or "reject" with a confidence score (0–100%). If confidence is >95% (or a user-set threshold), the defect is flagged immediately.

Rejection: When a defect is detected, the vision PC sends a trigger signal to a fast solenoid valve (<10 ms response) that opens a 80 psi air jet. The jet (positioned at the reject station of the starwheel) blows the defective container laterally into a discard chute, deflecting it away from the good-container stream. Alternatively, a mechanical gripper or vacuum gripper releases the pocket, allowing the container to fall into a reject bin. A confirmation sensor verifies the container was removed.

Data logging: All inspected images are logged to a database (SQL or CSV) with timestamp, container ID (if tracked), defect classification, and confidence score. Production reports (defects per hour, defect types, reject rate) are generated for traceability and process improvement.

Lighting strategies and defect visibility

Different defects require different illumination:

• Internal cracks: Backlit (transmitted light, 10,000+ lux) creates high contrast between the crack (dark line) and surrounding glass (bright). Best for detecting stress cracks and internal fractures.
• Bubbles: Also visible in backlit mode as dark spots. Size and quantity are logged.
• Surface scratches: Glancing light (45° angle, 3,000–5,000 lux) causes scratches to cast small shadows, making them visible.
• Threads: Coaxial light illuminates thread geometry; edge detection reveals broken or missing threads.
• Color and haze: Some applications require color cameras and specific light wavelengths to detect discoloration or cloudiness.

Defect classification and thresholds

Different products have different tolerance levels:

• Beverage bottles: Strict limits on cracks (zero tolerance) and large bubbles (>1 mm); minor scratches may be acceptable.
• Food jars: Similar crack/bubble limits; some applications accept minor surface defects if integrity is maintained.
• Pharmaceutical/medical bottles: Extremely strict; even hairline cracks are rejected; bubbles >0.3 mm trigger rejection.

Thresholds are set during line setup and are specific to the bottle design and product. A 10-minute setup period involves placing sample bottles (known good and known defective) on the line, capturing images, and teaching the vision software what "good" and "bad" look like. This training is critical for accurate operation.

Throughput and integration

At 300 containers/minute (typical for smaller bottles), each container spends ~200 ms on the starwheel (12 pockets × 5 rotations/min × 60 sec). This gives ample time for 3 cameras to capture images and process them; vision processing (100+ fps) completes in <10 ms. The bottleneck is typically the mechanical rotation speed, not vision—making this a true "zero-loss" inspection that does not slow the line.

Integration with upstream and downstream systems (depalletizer, filler, seamer, caser) is via Ethernet signals or hardwired relay pulses. A defective container detected at the inspector is automatically flagged in the MES (Manufacturing Execution System), which can trigger alerts or adjust production recipes.

Common challenges and troubleshooting

Lighting inconsistency: If LED brightness drifts (due to temperature or power supply aging), defect detection confidence degrades. Remedy: use intelligent LED drivers with feedback control; monitor LED brightness in real-time; replace aging LED arrays.

Reflection and glare: Shiny glass surfaces reflect lights, creating glare that obscures defects. Remedy: use polarizing filters or oblique lighting angles; adjust camera exposure automatically via histogram equalization.

False rejects: If threshold sensitivity is too high, good containers may be rejected due to minor surface blemishes. Remedy: set thresholds conservatively during setup; accept some false rejects (1–2%) rather than missing true defects; train operators to recover false rejects manually.

Throughput variability: If the starwheel speed is not synchronized to the upstream conveyor, containers may skip or bunch, causing inspection misses. Remedy: use encoder feedback and PLC synchronization; verify chain tension and bearing condition on starwheel drive.

Pocket wear: Spring-loaded or vacuum pockets degrade over time, losing grip. Remedy: inspect pockets monthly; replace worn springs or vacuum seals; ensure pneumatic lines are clean.

Regulatory and compliance

FDA requires that all containers be inspected for defects that affect product safety or integrity. Documentation of inspection images, defect rates, and rejection logs must be maintained for traceability. Vision systems are typically validated for sensitivity (>99% detection of defects ≥0.5 mm) and false-reject rate (<2%) before production use.