A machine vision system analyses images from a camera with software and lets a machine or robot make decisions based on what it sees. It is used to locate parts, measure them, reject faulty products, read codes and guide robots. In this article we explain the parts of a machine vision system, where it is used, the difference between classic image processing and AI, and what to watch for on the shop floor.
Parts of a machine vision system
- Camera: An area-scan camera captures a full frame at once; a line-scan camera scans a moving surface line by line (rolls, belts, sheets).
- Lens: Sets the field of view and working distance.
- Lighting: The most critical and most neglected part. Backlights, ring lights and dome lights make the feature you are looking for stand out.
- Processor and software: Software running inside a smart camera or on an industrial PC that analyses the image.
- Communication: The link that sends the result to the PLC, the robot or the operator screen.
Where is machine vision used in industry?
| Application | Example |
|---|---|
| Quality inspection | Scratches, missing parts, wrong assembly, colour defects. |
| Measurement | Hole diameter, edge distance, gap checks. |
| Reading | Barcodes, QR codes and printed text on labels. |
| Robot guidance | Locating a part on a conveyor so the robot can pick it. |
| Seam tracking | Keeping the welding torch on the real seam. |
Classic image processing or AI?
Classic image processing works with rules such as edge detection, pattern matching and measurement. When the target can be clearly defined it is fast, predictable and needs little data. Finding the centre of a weld seam is such a task.
AI (deep learning) learns from example images. It is strong at recognising things whose appearance changes every time, such as varied surface defects or hand-drawn marks. In return it needs example images to be collected for training.
A good system uses each where it fits. DKT welding robots do exactly that: on the welding carriage seam tracking is done with classic image processing, while on the robot arm AI recognises the chalk mark drawn by the welder.
What makes machine vision hard on the shop floor
- Changing light: Daylight from windows or ceiling lamps changes the result. Use your own lighting and shield outside light.
- Shiny surfaces: Reflections on metal create false edges; the right lighting angle and filters are needed.
- Dust, oil and spatter: Camera windows get dirty; a protective window and regular cleaning are a must.
- Welding arc: Arc light overwhelms the image; camera position and filtering are chosen accordingly.
- Calibration: Camera and robot coordinates must be matched, and recalibrated if the camera moves.
The DKT vision system
We develop the vision system on our welding robots ourselves. A camera continuously scans the weld zone, finds the seam centre and corrects the torch if it drifts. 2 laser distance sensors keep the distance between torch and part constant. See how does weld seam tracking work? for details.
Summary
- A machine vision system consists of camera, lens, lighting, software and communication.
- It is used for inspection, measurement, reading, robot guidance and seam tracking.
- Classic image processing suits clearly defined tasks; AI suits things whose appearance varies.
- Lighting, cleanliness and calibration decide success on the shop floor.
Should your robot or machine see?
We develop our vision system for our welding robots. If your machine has a similar need, tell us and we will assess together whether it fits.
For our robot solutions see the industrial robot arm and welding robots pages.