What is machine vision inspection?

What does machine vision inspection mean?

Machine vision is the industrial use of cameras, optics, lighting and software to make decisions about products under controlled conditions. Computer vision is the wider field of software that interprets images; machine vision adds what a line needs: fixed lighting and positions, a trigger, a link to the PLC and a way to remove failed products. The glossary entry gives the short definition.

Inspection is the job most plants buy it for. Every product that passes the station is imaged and judged against the same limits, at line speed, on every shift. The result is a pass or a fail, often with data attached: the code that was read, a measured value, or the image of a rejected product.

The camera is only one link. A station that works also needs a trigger, lighting designed around the defect, a link to the line PLC and a reject that removes failed products. Chapter 2 covers each part.

What is it used for on production lines?

Plants usually start from a fault that has already cost them: a wrong label after a changeover, a date code that rubbed off, a missing cap, a short-filled bottle, a part missing from a kit. The checks a station makes fall into a few plain questions:

• Is it there? Caps, labels, inserts, components and the items in a pack.
• Is it the right one? The label for the product running, the right variant, the right date and batch code.
• Is it in the right place? Label position and skew, cap height and tilt, part orientation.
• Is it the right size? Lengths, diameters, hole positions, and fill height in clear containers.
• Is the surface acceptable? Scratches, dents, flash, short shots and contamination that lighting can make visible.
• What does it say? Barcodes, 2D codes and printed text, read and matched against the expected data.

HIKROBOT describes the uses of its VisionMaster software in similar terms: visual positioning, size measurement, defect detection and information recognition (VisionMaster product page). The applications overview sorts these checks by the fault you need to catch.

How does inspection differ from guidance and code reading?

Three jobs share cameras and software but answer different questions, and the difference decides what you buy.

In practice they overlap. An inspection station often reads a code to confirm that the right product is running, and the locating tools that guide a robot are the same tools that let an inspection follow a product that has shifted; HIKROBOT notes that VisionMaster’s positioning output can be used for robot guidance (VisionMaster product page). Reading a code is also not the same as grading its print quality; the barcode verification guide explains the difference. This series is about inspection; addaScan takes on guidance projects case by case.

When is a simpler sensor enough?

Not every check needs a camera system. Start from the question you need answered, and choose the simplest device that answers it:

• Photoelectric sensor (photo eye): enough when the question is whether something is present at one point, or to count spaced products. It cannot tell a right item from a wrong one.
• Vision sensor: a compact camera with a built-in light and a small set of tools, for one or a few simple checks from one view, such as a cap present or a part the right way round. HIKROBOT lists its SC1000, SC2000 and SC3000 series as vision sensors (HIKROBOT catalogue 2025 Q4).
• Fixed code reader: enough when the only question is whether a code reads and carries the right content.
• Inspection station: needed when there are several checks or views, many products with their own recipes, image records, or rejects that must be tracked to the right product.

The simplest device that answers the question keeps cost and maintenance down. The sample test shows whether it really does.

What can machine vision inspection not do?

A camera judges the surface it is pointed at. It does not see inside opaque material, cannot prove that a pack is leak-tight and does not weigh anything. Chapter 5 sets out what a camera can and cannot see, and what to use instead.

Nor does it fix the process. It finds faults after they happen, which is why a station sits straight after the step that causes them; chapter 4 explains placement and rejection. Saved reject images and counts do help find the cause upstream.

Write a decision that can be demonstrated

An inspection specification needs a visible feature, an authorised reference and a disposition. For a cap check, “cap present” means something different from “correct cap fitted” or “closure is leak-tight”. The first may be a silhouette decision, the second may require colour, geometry or printed identity, and the last may need another test. State those boundaries before choosing a camera. A device cannot infer an unobserved property from a familiar-looking package.

Worked example — hypothetical: In a hypothetical carton check, a readable product code is compared with the order being packed. An old label can decode perfectly but still fail the order comparison. A blank label gives a no-read result and must not reuse the previous carton’s value. This example distinguishes reading, validation and item association; it is not a claim about a particular reader’s speed.

GS1 General Specifications section 3 defines the meaning of GS1 Application Identifiers. It does not decide whether the identified product is the one your line should currently be packing. (GS1 General Specifications, Release 26.0) (EMVA: GenICam)

Keep the original image together with the active recipe and the reference judgement when investigating a disagreement. Change one controlled setting at a time and re-run the relevant challenge samples. Once a correction works, verify the neighbouring requirements too: improving contrast for one feature can hide another. Record the revised configuration and the evidence for release so maintenance does not have to reconstruct the decision from memory.