InsightsThe mechanism
How does a color sorter actually work?
Direct answer
Six steps, in the same order every time: meter the material into an even flow, present it to the view, illuminate it so the target defect is visible, capture every particle in position, classify accept against reject, and fire a timed air pulse at the rejects. A correct decision still cuts wrong if the timing is off.
The six steps
- FeedMaterial is metered into an even, stable flow.
- PresentA chute or belt presents particles to the view.
- IlluminateTuned lighting makes target defects visible.
- CaptureCameras image every particle in position.
- RecogniseRecipe and AI logic classify accept against reject.
- EjectTimed air pulses remove the rejected stream.
Each step has to hold for the result to be repeatable. That is the part most descriptions skip: a sorter is not a camera with a reject gate attached, it is a chain, and the chain is only as good as its weakest link.
Lighting is not a detail
A defect that is not lit is a defect that does not exist. Precision lighting reveals target defects and reduces visual ambiguity across different materials, which is why illumination is tuned on the actual material rather than set once and reused.
Reflective material makes the point plainly. Glass cullet behaves differently under each lighting setup, so the illumination and the accepted color window are both set on the real cullet. The same is true of dusty or fine material, where contrast drops and color can be masked.
Timing is part of accuracy, not separate from it
Detection, particle position and actuation are synchronised to the millisecond so the right pulse meets the right particle. A correct classification that fires late removes a good piece and leaves the reject in the product — the recognition was right and the outcome was still wrong.
This is why compressed air, product speed and material spread all affect final rejection behaviour, and why a recipe is documented per material test rather than copied across unrelated streams.
What the machine is deciding
Only appearance. A color sorter classifies color, transparency, shape and visible defects — which is enough to pull off-color flakes, black fragments, and foreign objects such as wood, metal, stone, rubber and paper out of a stream reliably.
It is not deciding what anything is made of. PET and PVC can look identical, so a color decision cannot separate them at any setting. That is a different question, answered by NIR material recognition, and often staged after the color decision in the same line.
Where this stops applying
- A color sorter does not identify polymer type. Appearance and material are independent properties.
- Lighting recipes are documented per material test and must not be copied across unrelated streams.
- Compressed air, product speed and material spread all move the ejection result, so reject-accuracy claims need approved test data.
- Chute-fed machines need material that travels predictably: pellets and soft film fall outside them regardless of how well the camera sees.
- Final configuration is confirmed on a representative sample of the real stream.
Common follow-ups
What are the steps inside a color sorter?
Feed, present, illuminate, capture, recognise, eject. Material is metered into an even flow, presented to the cameras under tuned lighting, imaged and classified, then removed by timed air pulses.
Why does lighting matter so much?
Because a defect that is not lit is not visible to the camera. Precision lighting reveals target defects and reduces visual ambiguity, and it is tuned on the real material — reflective or dusty streams behave differently under each setup.
What does ejector timing do?
It synchronises detection, particle position and air actuation to the millisecond. Without it, a correct decision can still remove the wrong piece, because the pulse arrives when a different particle is in front of it.
Can a color sorter remove wood, metal and stone?
Yes, reliably. Those objects look clearly different from the surrounding plastic in color, brightness and surface appearance, and that visible difference is exactly what the machine detects.
Can a color sorter tell PET from PVC?
No. They can look identical, and appearance is all a color sorter reads. Separating them needs NIR material recognition.