InsightsBefore a quotation
What samples should you send for a sorting test?
Direct answer
Material that reflects real production rather than a clean selection of it: clear accepted product, the reject classes that actually appear in your stream, and borderline pieces from the edge between them. Size distribution, moisture and contaminant load matter as much as the material name, because they move the decision boundary.
Why a sample and not a specification
A material name tells the machine very little. Two streams both described as "washed PET flake" can behave completely differently once particle size, thickness, presentation and contaminant rate are taken into account — and those are the things that move the decision boundary.
This is why accepted and reject classes are built from examples rather than from a list, and why final model selection is confirmed on samples rather than promised in advance.
Three groups, not one
| What to include | Why it is needed |
|---|---|
| Accepted product | Defines what the machine must keep — the output your buyer actually accepts |
| Reject classes | Named categories that must leave: for PET flakes, non-PET such as PVC, PE, PP, PS, PMMA and PC; plus off-color, black and foreign objects where they occur |
| Borderline pieces | The edge between accept and reject, where the recipe is actually decided |
The third group is the one most often left out, and it is the most valuable. A configuration built only from obvious accepts and obvious rejects has no defined behaviour for the pieces that decide whether a batch passes.
Send it in the condition it arrives in
Cleanliness, moisture and label residue affect the practical sorting boundary before any setting does. A sample washed and dried more carefully than production is a different material, and a recipe tuned on it will not hold.
The same applies to size distribution. Send the real production size range, including the fines, rather than a screened middle fraction.
What the test settles
A test confirms the separation direction, the accepted and reject class definitions, and the platform that fits — bottle stage or flake stage, color decision or material decision or both.
It is also the point at which figures become real. Purity, yield and throughput depend on the material, defect load, target boundary and machine configuration, so they come from a tested result for your specific stream rather than from a general claim.
Where this stops applying
- A test result belongs to the stream it was run on. Recipes are documented per material test and must not be copied across unrelated streams.
- The online AI visual preview is a visual simulation for early stream review, not a real machine test.
- Purity, yield, throughput and reject accuracy are not promised in advance; they come from the tested result.
- Some material forms cannot be tested into suitability — pellets and soft film remain outside the plastic flake platforms regardless of sample quality.
- Reject classes are mapped from examples that appear in the real stream, not assumed from a material list.
Common follow-ups
How much material should be sent?
Enough to represent the real production range rather than a single handful. What matters more than volume is that the sample carries the full size distribution, the actual contaminant load, and examples of every class you need decided.
Should the sample be cleaned before sending?
No. Send it in the condition it arrives in. Cleanliness, moisture and label residue affect the practical sorting boundary, so a cleaner-than-production sample produces a configuration that will not survive production.
What is a borderline sample and why does it matter?
A piece sitting at the edge between accepted and reject. It is where the recipe is actually decided, so a configuration built without borderline material has no defined behaviour for the pieces that matter most.
Do I need a test before a quotation?
Yes. The configuration is built from accepted product, target rejects, borderline pieces and the real size distribution, so those have to exist before a machine can be specified.
Is the online AI visual preview the same as a real test?
No. It is an online visual simulation designed to speed up early stream review and application discussion. Use the real test path when machine selection, reject class stability and production configuration need confirmation.