US What automated visual inspection is
Automated visual inspection (AVI) is the use of cameras, controlled lighting and image-processing software to inspect products automatically, usually on a moving production line. It replaces or supports human inspectors at the points where defects are visible: surfaces, shapes, dimensions, labels, seals and fill levels.
AVI is one part of automated quality control, the wider practice of measuring and checking products with machines rather than by hand. Automated quality control also includes weighing, leak testing, X-ray and electrical tests, but visual inspection is usually the largest piece, because so many defects can be seen.
The technology behind it is machine vision. What makes it AVI is the job: every part, or a defined sample, is checked against a pass or fail standard, and the result drives a reject mechanism and a quality record.
How an AVI system works
An inspection station is a chain of parts, and the weakest link decides the result. In practice, lighting and part handling cause more problems than software.
-
//
Part handling and trigger. Conveyors, star wheels or robots present each item in the same position, and a sensor or encoder triggers the camera at the right moment.
-
//
Lighting. Backlights, ring lights, dome lights, low-angle and structured light make the defect stand out. Most inspection problems are solved here.
-
//
Camera and optics. Area-scan cameras for discrete parts and line-scan cameras for webs, sheets and cylinders. Global shutter sensors avoid distortion on moving parts.
-
//
Inspection computer and software. Rule-based tools measure edges, blobs, colors and codes; deep learning models classify defects that are hard to describe with rules.
-
//
Reject and record. The result goes to the PLC, which diverts the faulty part, and to the quality system, which stores images and counts for traceability.
Manual vs automated inspection
Manual inspection is flexible and cheap to start, but people tire and disagree, especially on fast lines. Rule-based AVI is fast and exact for defects that can be measured. Deep learning AVI handles variable, natural-looking defects that rules struggle with.
| Aspect | Manual inspection | Rule-based AVI | Deep learning AVI |
|---|---|---|---|
| Speed | Limited by the inspector | Hundreds to thousands of parts a minute | High, depending on model size and computer |
| Consistency | Varies with fatigue and person | Repeatable to the pixel | Repeatable once validated |
| Best at | Rare, unusual or new defects | Dimensions, presence, alignment, codes | Scratches, dents, stains, texture and cosmetic flaws |
| Set-up | Training and work instructions | Engineer programs each tool and tolerance | Labeled images of good and bad parts |
| Product changeover | Quick | New recipe per product | Retraining or fine-tuning |
| Records | Tally sheets | Images, measurements and counts | Images, classes and confidence scores |
Many lines combine the two automated methods: rules for measurements and codes, a deep learning model for cosmetic defects, and a person reviewing the borderline rejects.
What AVI can detect
AVI finds anything that changes how a part looks under the right light: scratches, cracks, chips, burrs, dents, contamination and discolouration; missing or misplaced components; wrong dimensions or hole positions; damaged or misprinted labels, unreadable barcodes and date codes; poor seals and caps; and under- or over-filled containers. It cannot see inside an opaque part, so internal flaws need X-ray, ultrasound or other tests.
In regulated sectors AVI is often a requirement as much as an efficiency. Injectable medicines must be inspected for visible particles, with USP <790> and <1790> setting the expectations for the US market, and EU GMP Annex 1, which the USA MHRA also applies, calling for every filled container to be inspected. Food, automotive and electronics plants use AVI to support traceability and customer audits.
Setting up an AVI system
Successful projects are defined by the defects, not the camera. Start by agreeing what counts as a defect, with physical samples and limits that quality and production both sign off. Collect images of good parts and every known defect type before choosing hardware. Then specify lighting and optics so the smallest defect covers several pixels, and choose the camera to suit: global shutter vs rolling shutter matters as soon as parts move.
Multi-camera stations often use GigE cameras on PoE frame grabbers, with each camera triggered and time-stamped together; our guide to synchronizing machine vision with IEEE 1588 PTP explains how. Deep learning inspection runs on edge AI computers beside the line, or on industrial PCs with GPU slots where several high-resolution cameras feed one model. Finally, validate the system against a known set of parts, measure false rejects and missed defects, and connect results to the PLC and quality records.
Quick answers on automated visual inspection
Automated visual inspection: common questions
Is AVI the same as AOI?
Closely related. Automated optical inspection (AOI) usually means inspecting printed circuit boards for solder and component faults. AVI is the broader term used across packaging, pharmaceuticals, automotive and food.
What is automated quality control?
Checking products with machines rather than by hand. It covers visual inspection, weighing, leak, X-ray and functional tests, with the results fed back to production in real time.
Can AI replace human visual inspectors?
For most routine checks, yes. Deep learning handles variable cosmetic defects well, but people still define the standards, review borderline rejects and investigate new defect types.
How accurate is automated visual inspection?
It depends on the set-up. A well-lit, well-validated system typically detects defects more consistently than people, but accuracy is measured per defect type as missed defects and false rejects, not as a single figure.
What camera should I use for visual inspection?
Start from the smallest defect and the line speed. These set the resolution, frame rate and shutter type. Area-scan global shutter cameras suit discrete parts; line-scan cameras suit continuous material.
We supply and integrate the inspection computers and frame grabbers as standard, with the drivers and vision software installed and every camera port checked before the system ships.