Key points
- Send good samples that cover the normal variation on your line, not only perfect ones.
- Send defective samples for every fault you need caught, including borderline cases, each labelled with its defect type and whether it should pass or fail.
- There is no fixed sample quantity. The set must be representative, and quantities are agreed per project.
- The result is a written feasibility conclusion: what was visible, under which conditions, and what still has to be verified on the line.
Why the test starts with your samples
A vision station can only judge what the image shows. Whether a scratch, a crooked label or a missing cap stands out from the good product depends on lighting, lens, angle and material, and cannot be settled from a specification sheet. So before we propose a station, we image your own products under candidate lighting and optics and check whether each fault is clearly visible.
A set of perfect products and one obvious defect proves very little. A set that shows how your product really varies, and how each fault really looks, gives a conclusion you can plan with.
Good samples: the normal variation
Good samples show what "acceptable" looks like. Do not send only the best products: if the test never sees normal variation, the station may later reject good products that simply look a little different.
- Take samples from different batches, shifts and production days.
- Include every SKU, size and label variant that passes the inspection point.
- Include normal differences: print shade, gloss and matte finishes, moulding cavities, material or packaging from different suppliers.
- Mark each sample as good somewhere the camera will not see, or on a tag or bag.
Defective samples and how to label them
Send at least one example of every defect type on your list, and several where the defect varies in size, position or appearance.
Borderline cases matter most. A product that is just acceptable and one that should just be rejected show where the limit sits. Agree these with your quality team before sending, because the station will be tuned to the line you draw.
Label every sample:
- Give each sample an ID and record, in a simple sample list, the defect type, where the defect is, and whether it should pass, fail or is unsure.
- Do not write on or tape over the area to be inspected; use the underside, a tag or a numbered bag.
- If a defect was made on purpose, mark it as simulated. Simulated faults show whether a defect is visible, but they often look cleaner than real ones.
If real defects are rare, collect them over time from quality holds, rework or customer returns.
How many samples? What makes a set representative
There is no fixed number that suits every project, and we do not publish one. A sample set is representative when it covers:
- every product variant the station will inspect;
- the sources of normal variation on your line, such as batches, suppliers, moulds, print runs and shifts;
- every defect type, in the range of sizes and positions it occurs;
- the borderline cases on both sides of the limit.
How many samples that takes depends on how much your product varies, how many defect types there are, how serious a missed defect is, and what the test must prove. A feasibility test asks "can this be seen and separated?" Acceptance testing at FAT and SAT asks "does the station perform on the line?" and needs a larger, agreed set. Deep-learning inspection usually needs more labelled images than rule-based tools. The rule-based or deep learning guide explains why. We agree quantities with you per project, before anything is shipped.
Photos and video that help
Photos and video do not replace physical samples, but they show us the line in a way samples cannot:
- the product at the inspection point, in its real orientation;
- a short video of the line at normal speed, showing spacing and handling;
- the space around the proposed camera position;
- close-ups of each defect, with a ruler or scale in the frame;
- daylight from roof sheets or doors near the station, which affects the enclosure design.
Product and line information
This information tells us how the station must fit your line. The PLC and reject integration guide lists what we ask about your controls.
| Information | What to send | Why it matters |
|---|---|---|
| Product description | Drawings or photos, dimensions, materials, colours and finishes of the product and the area to inspect. | Sets the field of view, camera resolution and lighting approach. |
| SKU list | Every product, size and label variant that runs past the inspection point. | Each variant needs a recipe; variants that look alike need a reliable way to tell them apart. |
| Defect list | Each fault, what it costs if it reaches a customer, and which faults matter most. | Decides what the station must catch and where limits sit. |
| Line speed and spacing | Products per minute or per hour, whether speed varies, and the gap between products. | Affects exposure, lighting, trigger method and whether an encoder is needed. |
| Handling | Conveyor type, orientation, whether products rotate or touch. | The camera needs a consistent view. |
| Controls | PLC make and model, spare inputs and outputs or network ports, existing reject. | Decides how the result and reject are wired to the line. |
| Environment | Washdown, dust, heat, vibration, and daylight or overhead lighting near the station. | Drives the enclosure, shrouding and mounting. |
| Current method | How the fault is found today and the acceptance criteria. | Defines what "defective" means on your line. |
Packing and shipping samples
- Pack good and defective samples separately, in labelled bags or boxes.
- Protect defective samples from new damage. A scratch sample with fresh scratches from transit is no longer a clean example.
- Agree perishable, fragile or hazardous products with us before sending, including any safety data sheet.
- Put a packing list in the parcel that matches your sample list.
- Agree whether samples are returned, kept as a reference set for acceptance testing, or disposed of.
We confirm the delivery details and method with you before you ship.
Confidentiality of samples and drawings
Samples, drawings and line video can reveal designs and process details. If any of this is confidential, say so before sending, and agree in writing:
- whether a non-disclosure agreement is signed first;
- who may see the samples and images;
- whether test images are kept after the project, and for what purpose;
- what happens to the physical samples afterwards.
Send only what the inspection needs; external dimensions and the inspected area are usually enough.
What the feasibility conclusion contains
The test ends in a written conclusion, not a sales promise. It sets out:
- which defects were clearly visible, under which lighting, lens and viewing angle, with example images;
- which defects were not visible or not separable from normal variation, and why;
- how the borderline samples behaved, and where a limit could sit;
- the proposed station concept: number of views, camera, optics, lighting and inspection method;
- the conditions it depends on, the risks, and what must still be verified on the line.
A small test set cannot prove a long-term missed-defect rate. It shows the defect can be seen; performance at line speed is measured at FAT and SAT on an agreed set. See how we deliver a project for the phases that follow.