Automated visual inspection: catching defects on the production line before they ship
How computer vision counts units, finds defects, reads dates and codes, and signals the PLC to reject faulty items, at full line speed.
Why manual inspection misses defects
Human inspectors are good at judging unusual problems, but checking thousands of identical bottles or boxes per hour is tiring work. Sampling helps, yet a defect that appears between samples still reaches the customer.
A camera over the line checks every single unit with the same attention at the end of the shift as at the start.
Four jobs one camera can do
| Job | Examples |
|---|---|
| Count | Units per minute, output against the shift plan, items per carton, cases per pallet. |
| Detect defects | Missing caps, underfill, cracks, dents, skewed labels, wrong color, damaged seals. |
| Read | Expiry dates, batch codes, barcodes and serial numbers, checked against the product being run. |
| Act | Signal a reject arm or air jet, stop the line, or light a tower lamp through the PLC. |
Connecting to the line
Detection only creates value when it triggers an action fast enough. The inspection result is sent to the PLC using standard industrial protocols such as Modbus or OPC UA, or through digital I/O, so the faulty unit is removed at the reject station without slowing the line.
- Use stable lighting at the inspection point; it matters more than camera resolution.
- Trigger the camera from a sensor so each unit is captured in the same position.
- Log every rejection with its image so quality teams can see patterns by shift and machine.
Training models on your products
- 01Collect images
Capture good and defective units on the real line, under real lighting.
- 02Define defects with QA
Agree what counts as a reject, with examples, so the model and the inspectors use the same standard.
- 03Run in shadow mode
Compare the model with your current inspection for a period before letting it reject units.
Questions we often hear
Can one system handle several product types?
Yes. Each product or SKU gets its own inspection profile, selected manually or automatically from the production order.
How fast does it react?
Fast enough to reject a unit at the reject station on typical lines, measured in milliseconds. The exact budget depends on line speed and the distance to the reject point.