Machine Vision Inspection Acceptance: How to Write a Standard Nobody Will Argue About
Guide to acceptance criteria for machine vision inspection projects: it explains how to agree escape rate and over-rejection rate, how to define the sample set, how to describe acceptance conditions, and key practices to avoid disputes at the acceptance stage.
Clarify One Thing First
Turns "inspection capability" into verifiable clauses rather than an adjective
The acceptance standard for machine vision inspection should specify four things: the acceptance sample set, the acceptance conditions, the escape rate and over-rejection rate agreed separately by defect type, and a quantified definition of the acceptance criteria. If any one of these is missing, disputes are likely at the acceptance stage. Acceptance should be carried out on the sample set and under the site conditions confirmed by both parties.
Acceptance of visual inspection projects is prone to dispute because the concept of "inspection capability" is too soft. If the contract only says "inspection must be accurate", the two sides will inevitably understand it differently.
The way to make a soft concept concrete is Break complex things down and specify them separately: which defect type, under what conditions, on which sample set, and what escape and over-rejection rates are allowed. Once broken down this way, the clause becomes testable.
This article gives suggestions on the structure and wording of acceptance criteria; the specific values need to be agreed by both parties according to the business cost — because "how much escape is acceptable" is essentially a business judgement, not a technical one.
selection Factors
Acceptance Sample Set
It must be jointly confirmed and sealed by both parties before acceptance. The sample set should cover all defect types and the batch variation of conforming products. This is the basis of acceptance.
Acceptance Conditions
This includes on-site lighting, material batch, speed, pose and running time. Results change when the conditions change, so they must be stated clearly.
Agreed by Type
Detection difficulty varies greatly between defects, so requirements must be agreed separately by defect type; a single value cannot cover all defects.
escape rate
The opposite side of a good part being judged NG — the proportion that should have been detected but was not. Usually the most critical metric, with a high cost.
over-rejection rate
The proportion of good parts wrongly judged as NG. It directly affects cost, and an upper limit must also be agreed at acceptance.
stability
Performance over a period of continuous operation (for example several hours or several shifts), not just the result of a few trial runs.
Field of View and Pixel Sampling
Do Not Mistake Pixel Scale for Inspection Accuracy
A common acceptance mistake is "trying a few samples and treating a pass as qualification". This cannot reflect real performance, because the number of samples is too small and they are often picked typical defects.
A more reliable approach is to run a blind test with a sample set sealed by both parties, operate continuously for a period, count escapes and over-rejections by defect type, and record anomalies and downtime.
| Acceptance Item | Recommended Acceptance Items | Description |
|---|---|---|
| Defect Detection | Detection performance on the sealed sample set | Counted separately by defect type |
| escape rate | Set separate limits by defect type | Critical defects usually require tighter limits |
| over-rejection rate | Calculate the false call rate from good parts | Affects cost; an upper limit must be agreed |
| throughput | Meets the agreed cycle time | Includes loading and unloading and judgement time |
| stability | Performance over continuous run duration | Record downtime and anomalies |
| Data Functions | Record, query and export functions | Validate against agreed fields |
Light Source and illumination
Sample Sealing
Acceptance samples must be signed by both parties and sealed, to avoid disputes at acceptance over "whether this is the same piece". Where conditions permit, the sample list and images should also be retained.
Realistic Conditions
The acceptance conditions must be written as a reproducible description: which material, which batch, which speed, which illumination, which pose. A vague "normal production conditions" is as good as not writing anything.
Blind Test
At acceptance, it is better not to reveal which samples are NG, to avoid a tendency to "lean toward the expected result".
Exception Logging
All anomalies, stoppages and false calls during acceptance should be recorded, both for evaluation and for later optimization.
algorithm Selection
The most common reason for acceptance failure is not that the equipment falls short, but that the "acceptance criteria do not match the boundaries of the solution"—for example, requiring zero escapes on a defect that was clearly stated to be beyond reach at the solution stage. The way to avoid such problems is to write the boundaries clearly at the solution stage.
Explicit Boundaries
Clarify at the solution stage which defects can and cannot be inspected.
- Write into the technical agreement
- Avoid disputes at acceptance
Layered Acceptance
Set tiered metrics for critical defects and general defects.
- Stricter criteria for critical defects
- General defects with cost in mind
Progressive Acceptance
Validate with a small batch first, then scale up to continuous operation.
- Find problems early
- Reduce risk for both parties
Threshold Negotiation
Threshold adjustments affect escapes and over-rejection and must be confirmed by both parties.
- Re-run validation after adjustment
- Record the final threshold
Communication and interlocking
Acceptance is not just a point in time but a process: preliminary acceptance (function and cycle time) → trial run (stability and statistics) → final acceptance (targets met and documentation delivered). Carrying it out in stages exposes problems early.
- 01 Both parties confirm the acceptance sample set and seal the samples
- 02 Define acceptance conditions and run duration
- 03 Agree on escape and over-rejection limits by defect type
- 04 Pre-acceptance of functions and cycle time
- 05 Continuous trial run
- 06 Compile statistics and review data
- 07 Final acceptance and documentation delivery
- 08 Freeze thresholds and recipes
Common Question
How Many Samples Should Be Used at Acceptance?
Should the metrics for critical defects and general defects be kept separate?
What if acceptance does not pass?
Can the Escape Rate and the Over-Rejection Rate Both Be Kept Very Low?
Who writes the acceptance criteria?
Can it still be adjusted after acceptance?
Submit sample testing
We recommend that the acceptance criteria be drawn up jointly by both parties. Please provide your defect classification and business priorities, and we will give an achievability assessment and recommended metrics on that basis.
Submitted successfully
We have received your sample testing request. A solution engineer will contact you within 1 business day via contact you.
Need to assess imaging conditions, defect criteria and cycle time item by item? Go to the Full Requirement Assessment →
The Most Effective Step in Selection: Test Your Own Sample
For the same workpiece, the inspection result differs greatly with different lenses, light sources, mounting distances and algorithm combinations. Sending us samples for measurement is more reliable than extrapolating from a specification table.