Digital AIUpdated 24 October 20268 min read

Machine Vision Inspection: Computer Vision Guide

Machine vision inspection uses cameras, lighting and software to check parts automatically. This guide explains how classic and deep-learning inspection differ, what a system contains, how it talks to the PLC, and how to learn the skills in the UK.

Computer VisionMachine VisionVisual InspectionDeep LearningFactory Automation
Man working at a desk with a mouse and keyboard in an engineering lab

Key takeaways

  • A machine vision inspection system is lighting, optics, camera, trigger, processor and a link to the PLC; lighting is usually the biggest success factor.
  • Use rule-based vision for clear measurements and deep learning for hard-to-describe surface defects; many lines use both.
  • The vision system informs the PLC with a pass or fail result and data; define a safe default for timeouts, power loss and late results.
  • Agree what counts as a defect and keep a reference set of good and bad parts, or the model has no consistent target.
  • UK support includes Made Smarter Adoption and the HVM Catapult; a BAE Systems X-ray inspection trial shows the staged approach.

Machine vision inspection uses an industrial camera, controlled lighting and software to check every part on a line for defects, dimensions, labels or assembly errors, and to send a pass or fail result to the machine in milliseconds. Modern systems combine classic image processing for measurement with deep learning for surface defects that are hard to describe with rules.

What is in a machine vision inspection system?

Every system, from a simple smart camera to a multi-camera cell, has the same building blocks:

  • Lighting. Ring lights, backlights, dome lights and structured light are chosen to make the feature you care about stand out. Lighting is usually the biggest factor in success.
  • Optics and camera. The lens and sensor decide resolution, field of view and speed. Area-scan cameras image a part at once; line-scan cameras build an image of a moving web or roll.
  • Triggering. A sensor or encoder tells the camera exactly when the part is in position, so every image is taken in the same place.
  • Processing. A smart camera, an industrial PC or an edge device runs the software that analyses the image.
  • Communication. The result goes to the PLC, robot or reject gate over discrete I/O, PROFINET, EtherNet/IP or OPC UA.
IDS uEye industrial machine vision camera
IDS uEye industrial machine vision cameraPhoto: Lg123456, CC BY-SA 4.0, via Wikimedia Commons
What is industrial machine vision and what can it do? Vision for factory automation explained

Rule-based vision or deep learning: which should you use?

Use rule-based vision, the traditional approach, when the task is clear and measurable. Examples are measuring a hole diameter, reading a barcode or a date code, checking that a cap is present, or guiding a robot to a fiducial. It is fast, deterministic and easy to validate.

Use deep learning when the defect is hard to describe: a scratch on brushed metal, a cosmetic flaw on a moulded part, variation in natural products such as food or timber. A trained model copes with more variation in appearance than a fixed rule set, but it needs labelled images, a repeatable setup and monitoring for drift.

Many production systems use both. A rule-based step locates and measures the part, and a model inspects the surface. The right choice is the simplest method that meets the escape rate and false reject rate you need.

How does a vision system talk to the PLC?

The camera does not control the machine; it informs it. A typical sequence is: the PLC sees a part arrive, sends a trigger, the vision system captures and processes the image, and then returns a result bit and often a data block such as measured values and a recipe ID. The PLC logic decides what to do: divert the part, stop the line, or record the result against the serial number.

Design the handshake carefully. Define what happens if the camera does not answer, if the result arrives late, or if the vision system loses power. The safe default is almost always to reject or hold the part. Our guide to machine vision and PLC integration covers triggers, timing and data exchange in detail.

How do you choose lighting and a camera?

Start from the smallest defect you must detect. A common rule of thumb is that the defect should cover several pixels in the image, which fixes the resolution you need for the field of view. Then pick the lighting technique: a backlight gives a crisp silhouette for dimensions, a low-angle light shows scratches and surface texture, a diffuse dome light reduces glare on shiny curved parts, and polarising filters cut reflections. Test with real parts, including the worst ones, before buying anything. Run the cell at line speed so exposure time, trigger timing and part movement are realistic, and record a few thousand images, because those become your training and validation data.

What does a good deployment look like in the UK?

Under the Made Smarter Innovation Technology Accelerator programme, the Somerset start-up Machine Intelligence worked with BAE Systems to analyse X-ray images and detect defects in the joints of metal parts, assisting human testing. BAE applied it to an existing inspection process and is trialling it on another. The pattern is typical: prove value on one inspection station, keep people in the loop, then extend.

Smaller manufacturers can look to Made Smarter Adoption for advice and grant support towards digital tools, and the High Value Manufacturing Catapult network, which includes centres such as the AMRC and the MTC, for testing and demonstration. Terms change by region and year, so check them before planning.

What are the main pitfalls?

  1. Variable lighting. Ambient light from a window or a change in bulbs can break a model that worked in the demo. Enclose the inspection area or use controlled lighting.
  2. Movement and vibration. Blur and shifting position change the image. Use fast exposures and mechanical fixing.
  3. Too few defect examples. Collect real defects over time; synthetic images help but do not replace them.
  4. No ground truth. If two inspectors disagree on what counts as a defect, the model cannot learn a consistent rule. Agree the standard first and keep a reference set of good and bad parts.
  5. Ignoring change management. A new product, material or supplier is a change that needs re-validation.

What do computer vision engineers do and earn?

Computer vision engineers specify the cell, collect and label data, train and evaluate models, and integrate the result with the control system. ITJobsWatch showed a median UK salary of £80,000 for permanent computer vision roles across 151 advertisements in the six months to 23 October 2026. Many of those roles are software-led, but the skills overlap strongly with industrial inspection work, and engineers who understand PLCs, cameras and models together are unusual.

How to learn machine vision inspection for free

Start free

Enrol on the free Machine Vision and AI Quality Inspection course, try the first five free days of the Computer Vision Engineer job simulation, and check the computer vision jobs page for the skills employers list. The self-paced EDWartens AI Module includes computer vision with graded industrial projects if you want a structured route; see its page for the current price.

Start free

Learn machine vision free, then practise the job

Free job simulation

Computer Vision Engineer

Specify, label and evaluate an inspection model on a production line.

The first five days are free, no card needed.

Live UK jobs

Computer vision jobs in the UK

Updated daily, with licensed visa sponsors marked.

Browse all engineering jobs

Frequently asked questions

What is machine vision inspection?

It is automated checking of parts using cameras, controlled lighting and software. The system captures an image, analyses it for defects, dimensions or presence of features, and sends a pass or fail result to a PLC or robot, usually in milliseconds.

What is the difference between machine vision and computer vision?

Machine vision usually means industrial systems with cameras, lighting and controllers on a production line, built for speed and reliability. Computer vision is the wider field of teaching computers to interpret images, including the deep learning models that many modern inspection systems use.

When should I use deep learning instead of traditional machine vision?

Use traditional rules for clear measurements, codes and presence checks. Use deep learning when defects are cosmetic or varied, such as scratches, texture changes or natural product variation, and when you can supply labelled examples. Many systems combine both.

How does a vision system connect to a PLC?

Usually through discrete I/O for trigger and result bits, or industrial Ethernet protocols such as PROFINET, EtherNet/IP or OPC UA for richer data. The PLC owns the decision about rejecting parts or stopping the line.

Can I learn machine vision without a degree?

Yes. You can learn image processing, model training and integration through free courses and practise on a job simulation. Employers will still look for evidence of projects, so keep a portfolio of what you build.

Sources

  1. Machine Intelligence and BAE Systems case story, Made Smarter Innovation
  2. Made Smarter Adoption
  3. High Value Manufacturing Catapult
  4. Computer vision jobs in the UK, ITJobsWatch (checked 24 October 2026)
  5. NVIDIA Jetson modules

Share this article

About the author

EDWartens Team

EDWartens Instructional Team

The EDWartens instructional team brings together UK-based automation engineers, certified Siemens and Rockwell specialists, and industry practitioners with decades of combined experience in PLC programming, SCADA engineering, and industrial AI.

View all articles by EDWartens

Ready to become job-ready?

Start with a free course today, or train live in Siemens PLC, HMI and SCADA on the Professional Module.