Buyers of optical inspection and sorting equipment rarely see their own product running before they sign. Catalog sheets describe resolution, throughput and accuracy in general terms, while the decision that matters is narrower: can this system see the defect your customer keeps returning parts for? A sample test answers that with your surface finish and your defect mix, not a demonstration part chosen because it images beautifully.
A sample test before buying means sending a defined set of your own parts, good, defective and borderline, to a supplier who images them on a real machine with the proposed optics and software, then reports what was detected, what was missed and what was falsely rejected. It is a feasibility check on your parts, not a performance guarantee for your line, and it is the cheapest moment to discover that a defect is not optically visible at all.
The sections below follow that process in order, from choosing which parts to send to writing the test conditions into the contract.
What a Sample Test Can and Cannot Prove
A sample test proves whether your defects are detectable under a specific optical setup, and roughly how fast one part can be judged. It does not prove how the system behaves across shifts, batches and operators, and no honest supplier will turn a few dozen parts into a promise of zero escapes on millions.
What the test can confirm
The test first settles whether the defect has enough contrast to be imaged at all. A dark stain on matte black moulding, a hairline crack on a polished chamfer, a missing pin in a deep bore: each is a lighting problem before it is a software problem. Once the image is stable and unambiguous, the rest becomes engineering.
It also answers questions that specifications rarely address, such as whether one view is enough or the part needs top, bottom and side imaging, and the smallest defect size that survives imaging at working distance.
For parts with tight dimensional requirements, it shows whether one optical sorting machine configuration can handle appearance and gauging together, or whether those need separate stations.
What the test cannot confirm
Stability over time is invisible in a two hour session on a clean bench. Lamps age, lenses collect dust, and incoming material drifts between batches. Nor can the test predict a changeover to a second variant. Thirty parts can show that a defect type is detectable; they cannot estimate an escape rate of one in a thousand.
What to Prepare Before You Ship Your Parts
The quality of the test is bounded by the quality of the sample set and the brief that travels with it. Parts without a defect description produce a polite report that answers nothing, while a small documented set of representative and borderline parts produces a report you can decide with.
The three part groups that matter most
Send the typical production part from a normal batch, because it sets the baseline for geometry, colour, coating and surface texture. Send real defects including borderline ones, because a scratch your customer argued about is worth more than fifty obvious cracks, and if two inspectors would disagree about a part, it belongs in the shipment. Send high risk variants too: a darker finish, a transparent grade, a deep pocket, a part that runs twice a year.
Documentation to send with the parts
| Document | Why the supplier needs it | What goes wrong without it |
| Drawing with critical tolerances | Defines a dimensional failure | Results judged against wrong limits |
| Material, finish and coating | Drives the lighting choice | Setup tuned for the wrong surface |
| Defect list with accept and reject boundaries | Converts complaints into criteria | Report describes “anomalies” instead of defects |
| Line rate and part presentation | Sets the required cycle time | Passes at bench speed, fails in production |
| Variant list and photos on the line | Shows recipes and physical access | Proposed station cannot fit your machine |
State which surfaces are customer visible, since a mark accepted on an internal rib is a reject on a front panel. If the defect list is hard to write, start from the checks your inspectors already perform: describing the visual inspection functions in your current process turns tribal knowledge into a specification.
How Many Parts Should You Send
There is no universal number, but there are sensible ranges. Ten to thirty mixed parts settle a feasibility question, fifty to a hundred cover several defect types properly, and several hundred identical parts are needed before anyone can discuss repeatability.
| Test objective | Typical quantity | What it answers |
| Feasibility screen | 10 to 30 mixed parts | Is the defect visible, does handling work |
| Defect coverage | 50 to 100 parts across all types | Which defects are found, where the boundary sits |
| Repeatability | 200 to 500 identical parts | Does the same part get the same verdict |
This follows standard acceptance sampling, where confidence depends on how many units were examined and how many failures appeared, not on the percentage printed at the end, a logic the NIST/SEMATECH handbook sets out under lot acceptance sampling.
Keep tuning and validation samples apart too: if thresholds are adjusted on the same parts used to demonstrate success, the demonstration is circular.
Write the Test Brief and Agree the Acceptance Criteria First
Put the pass and fail rules on one page and get them agreed in writing before the machine is set up. Criteria invented after the results are known are not criteria, they are negotiation.
Turn goals into measurable criteria
“High accuracy” has to become “detects a scratch 0.3 mm long and 0.05 mm wide on the visible face at 200 parts per minute”. “Checks quality” becomes “identifies missing insert, reversed orientation, flash above 0.2 mm and contamination above 0.5 mm”. The list should cover the defect set with size and location, the surfaces to be imaged, the production rate and presentation, and the rule for borderline parts.
Separate escapes from false rejects
| Metric | Definition | Commercial consequence |
| Escape | Defective part judged good | Reaches the customer, drives returns |
| False reject | Good part judged defective | Consumes reinspection labour |
| Detection rate by type | Share found per defect type | Averages hide the type that fails completely |
| Cycle time | Time per part including handling | Decides whether one machine or two are needed |
Tightening a threshold reduces escapes and raises false rejects, and no configuration minimises both, so the buyer must state which side of that trade the plant prefers. A safety critical fastener and a cosmetic panel should not run to the same rule, even on similar hardware.
What Actually Happens During the Test
A competent supplier follows a predictable sequence: log the parts, define the imaging task, build the optical setup, configure the logic, run good parts, run defects, repeat, then measure throughput and write the report. Knowing the sequence lets you ask the right question at each stage.
- Intake. Parts are photographed as received and the claimed defect recorded per part. Ask for this log, it becomes the reference for every later claim.
- Optical setup. Camera, lens, working distance and lighting are selected, and the supplier restates which features are checked and which are measured.
- Logic configuration. Thresholds, regions of interest and measurement tools are set, usually on a subset of parts.
- Good part run. Conforming parts establish the false reject baseline.
- Defect run. Each defective and borderline part is presented, with the verdict recorded and imaged.
- Repeat and throughput runs. Selected parts are presented again in a different order, then a continuous batch measures realistic cycle time.
- Report. Results, setup description, images and stated limitations are compiled.
Vision performance is the product of the whole imaging chain rather than one component, a point the A3 Association for Advancing Automation makes repeatedly in its technical material. A high resolution sensor with poor lighting on a reflective surface loses to a modest sensor with a well designed light path.
Questions to ask while the test is running
- Is this the same machine model, lens and lighting that will be delivered?
- Who set the thresholds, and can I see the raw images?
- What happened to the borderline parts specifically, not to the average?
- What would change if my defect standard were twice as strict?
How to Read the Sample Test Report
Read the structure and the sample base before the headline percentage. A report stating its sample counts, listing each defect type, showing the missed parts and naming its limitations is worth more than one reporting a single impressive number.
| Report element | What it tells you | What to check |
| Sample base | How many good and defective parts ran | Counts per defect type, not just a total |
| Detection by type | Which defects are found reliably | Whether a type is missing from the table |
| Escape list | Parts the system let through | Images and cause for each one |
| Setup description | Optics, lighting, thresholds | Whether it matches the quotation |
When the test includes dimensional checks, ask which optical measuring machine configuration produced the numbers and how it was calibrated.
Red flags and fair comparison
Slow the purchase down when a report shows percentages without counts, results on good parts only, processed screenshots with no raw images, or a setup description that differs from the quotation. If two suppliers test your parts, hold the inputs constant: the same parts, defect list, criteria and order of presentation, then compare how each handles the borderline group rather than one aggregate score.
Ship Your Parts or Test On Site
Start by shipping parts, then visit only the shortlisted suppliers. A sample test before buying filters out unsuitable proposals cheaply, while an on site visit adds handling behaviour, changeover reality and the people you will work with.
| Criterion | Shipping parts | Testing on site |
| Cost and turnaround | Postage and a brief, days | Travel and scheduling, one supplier per trip |
| What you observe | Results, images, setup photos | Handling, changeover, workflow |
| Best for | Feasibility screening | Final validation |
A part that images perfectly on a bench may tumble, overlap or bounce when fed at production rate, and the difference between a glass turntable sorting machine and a belt fed unit only becomes obvious with parts actually moving.
From Sample Test to Purchase Order
Carry the tested configuration into the contract. The most common disappointment after a good sample test is a delivered machine whose optics, thresholds or handling differ from the setup that produced the result.
What to write into the specification
- The optical setup used: camera, lens, working distance, lighting type
- The stored recipes and thresholds behind the reported results
- Acceptance metrics stated per defect type, not as one average
- A retest clause running the same part set at acceptance, with the same criteria
- Handling, integration and training scope, including who wires and commissions
- Future variants listed as untested risks rather than assumed capability
A variant nobody tested should appear as a known gap with a validation step, not a silent promise, the same discipline of documented scope and traceability that the ISO 9000 family of standards describes.
Plan the second test on your own line
Factory acceptance confirms the built machine matches the tested configuration; site acceptance confirms it on your line, with your material and operators. Between the two, insist on a full shift at production rate with false rejects and escapes logged by defect type, then review those logs after the first week: a system that performs on day one and degrades by day ten usually reveals a lighting or cleaning issue no bench test could show. If the same line must also tally parts into batches, settle early whether an optical counting machine sits inline or downstream, because that choice changes the handling layout you tested.
FAQ
Is a sample test usually free, and who pays the shipping?
Feasibility tests are commonly offered at no charge, since they cost a few hours of engineering time and act as a qualification step. Incoming freight is normally paid by the buyer. Ask about customs handling too, because a parcel held at a border for a week can delay a project more than the test itself.
Can I run a sample test if I have no defective parts yet?
Yes, and it is often worth doing for the imaging and handling questions alone. Constructed samples can validate the optical path but cannot confirm coverage of your real defect population. Ask the supplier to state which defect types stay unverified, then plan a second test once production rejects exist.
How long does a sample test take, and how long do the results stay valid?
A straightforward test on a documented sample set usually takes a few working days from receipt of the parts, longer when several variants or measurement tasks are involved. Results stay valid while the part, material, surface finish and defect standard remain unchanged. A new coating, a new mould or a stricter customer requirement is a reason to retest, because all three alter the contrast the system depends on.