Machine vision basics is a six-chapter introduction to production line camera inspection, written for production, quality and engineering teams planning their first station. It explains what machine vision inspection is, what a station is made of, how it decides pass or fail, where it sits and how it rejects, what a camera can and cannot see, and how a station is tested and accepted. Each chapter is short and links to the detailed guide on its topic.
Who this series is for and how to use it
The series is for the people who will own a vision station once it is on the line: production and plant managers, quality and food safety staff, and maintenance or automation engineers. It assumes no background in optics or software. Each chapter answers one question in plain words, so that the first conversation with any supplier starts from a shared understanding of what a station can do and what it cannot.
Read the chapters in order for the full picture, or go straight to the question you have. Each chapter ends with links to the deeper guides on its subject, such as the lighting guide, PLC and reject integration and the commissioning procedure. Terms are defined in the glossary.
The content comes from how addaScan builds inspection stations for South African plants on HIKROBOT hardware and VisionMaster software. Vendor statements are cited to HIKROBOT’s own documents, and the series gives no performance figures that are not sourced: what a station achieves is proved on your samples and agreed in writing.
Chapters
- 01What is machine vision inspection?What vision inspection is, what lines use it for, how it differs from guidance and code reading, and when a simpler sensor is enough. · 5 min
- 02What are the components of a vision inspection station?Trigger, camera, lens, light, processing, software, PLC link, reject and enclosure: what each part of a station does. · 5 min
- 03How does machine vision inspection decide pass or fail?Locate, measure, count, read and compare; rule-based versus deep learning; limits and recipes. · 5 min
- 04Where should an inspection station sit, and how does it reject?Placement after the step that causes the fault, tracking to the reject, reject devices, confirmation and fail-safe behaviour. · 5 min
- 05What can machine vision detect, and what can it not see?Visible surfaces only, contrast and lighting, and what to use for leaks, insides, strength and weight. · 5 min
- 06How is a vision inspection station tested and accepted?Sample test, FAT with challenge samples, SAT, false rejects and escapes as agreed figures, and the records to keep. · 5 min
Turn a quality complaint into an inspection question
Start by describing the product condition you want to distinguish, not the equipment you want to buy. “Inspect our packaging” leaves the acceptance decision undefined. “Hold a pack when its printed batch differs from the released production order” gives the team an observable field, a reference and an action. A separate question is whether the printed batch remains legible after storage and handling; that cannot be established from an image taken immediately after printing.
- Define the fault and the reference
- Make the required feature visible
- Acquire, locate and inspect
- Track the item and confirm disposition
The camera converts reflected or transmitted light into an image. Software then searches for a feature, measures a boundary or interprets a symbol. The PLC associates the result with an item and directs the line response. Each step can fail independently. A clear image does not prove correct product tracking, and a moving reject cylinder does not prove that the intended failed item was removed. The series chapters explain these separate responsibilities.
| Question to bring to the first meeting | Useful evidence | Chapter to read |
|---|---|---|
| What exactly is unacceptable? | Quality-approved good, bad and borderline samples | What machine vision inspection is |
| Can the feature be exposed? | Product views, substrate and presentation limits | What a camera can and cannot see |
| How will a result follow the item? | Conveyor layout, trigger, PLC and reject location | Where a station sits and how it rejects |
| How will performance be demonstrated? | A labelled sample list and acceptance criteria | How a station is tested and accepted |
Use the sample list to preserve disagreement rather than hide it. If production calls a mark acceptable and quality calls it a defect, the supplier cannot resolve the commercial specification by choosing a convenient software threshold. Keep that sample in an unresolved group until the product owner decides. This prevents a successful demonstration on obvious defects from being mistaken for agreement about the difficult production cases.
A worked decision example and the records to retain
Hypothetical example: an assembly must contain the specified insert. The trial includes a correct insert, an empty recess, a wrong insert with a similar colour, and a correct assembly rotated within the permitted presentation range. A simple brightness threshold might distinguish the empty recess but fail on the wrong insert. The requirement is therefore identity as well as presence. A shape or code check, or a more controlled presentation, must be evaluated before choosing the final device. No detection percentage is implied by this example.
| Observed fault | Check on the line | Record before changing anything |
|---|---|---|
| Pass with the wrong component | Inspect the actual feature used by the algorithm | Image and location overlay for correct and wrong variants |
| Failures only during fast running | Compare moving and stationary images | Exposure, trigger time and local product motion |
| Correct screen result, wrong item rejected | Follow product identity through the PLC | Trigger, result and reject-confirmation sequence |
| Works before cleaning, fails afterwards | Inspect optical alignment and window condition | Recipe, focus and light settings before and after cleaning |
Use standards and technical sources for the question they actually answer. GS1 General Specifications section 3 defines Application Identifier data where GS1 symbols are used; section 5 covers symbol requirements. Reading a code at the station is different from grading its printed quality. (GS1 General Specifications, Release 26.0) GenICam concerns a camera software interface, not a complete inspection acceptance standard. (EMVA: GenICam)
For dimensions, NIST’s gauge-study guidance helps separate variation from systematic bias. For classification, a reserved set of independently labelled samples tests the locked decision. These are different experiments and should produce different reports. (NIST e-Handbook, 2.4 Gauge R&R studies) A camera can repeatedly produce the same answer while consistently measuring the wrong feature or accepting the wrong product. Ask for evidence linked to the requirement, not only a supplier’s headline accuracy.
South African plants should also identify who owns local maintenance, cleaning, recipe changes and held-product release. Agree the restore procedure before the station is needed for production: the recipe, calibration, camera settings and PLC changes must be recoverable together. A replacement camera that produces an image is not necessarily calibrated or configured to make the same decision. The next step after these basics is a documented sample assessment, followed by an installation scope and an acceptance plan tied to the plant’s actual quality risk.
Questions about this series
Who is the machine vision basics series for?
Production, quality and engineering teams who are considering automated camera inspection for a line and want to understand how a station works before talking to suppliers. No background in optics or programming is assumed.
Do I need to read the chapters in order?
No. Each chapter answers one question and stands on its own. Reading them in order gives the full path from what inspection is to how a finished station is accepted.
Is the series tied to one brand of equipment?
The principles apply to any machine vision platform. Where an example helps, we use HIKROBOT devices and VisionMaster, which we build on as an independent integrator; we do not claim official HIKROBOT partner or distributor status.
What should I do after reading it?
Write down the fault you need to catch, collect a few good and defective samples, and note the line speed and the PLC. With that, a sample test can show whether the fault is visible on your product.
Sources
- HIKROBOT VisionMaster product page (vendor page, checked 2026-10-03)
- HIKROBOT Machine Vision Smart Product Catalog, 2025 Q4 (PDF) (vendor catalogue, checked 2026-10-03)
- BRCGS Food Safety Issue 8 FAQs (clause 6.2.4, online verification equipment) (BRCGS guidance, checked 2026-10-03)
- GS1 General Specifications, Release 26.0 (sections 3 and 5: data and symbol requirements; read 2026-10-27)
- EMVA: GenICam (standard camera programming interface; read 2026-10-27)
- NIST e-Handbook, 2.4 Gauge R&R studies (measurement study design; read 2026-10-27)
Have a line in mind?
Describe the product and the fault you need to catch. We test on your samples and reply with a feasibility note and a budget.