Machine vision systems are used in manufacturing for five jobs: finding defects, measuring parts, verifying that an assembly is complete, guiding robots and equipment, and counting or sorting. Each one replaces a repetitive visual check with a camera, controlled lighting and software that makes the same decision on every part, at line speed, and records the result.
This article maps those applications by task and by industry, and shows which imaging method fits each one. It covers what a machine vision system is, the jobs it does on a production line, applications by industry, and how to match the imaging method to the application.
What is a machine vision system?
A machine vision system is an integrated set of industrial cameras, lenses, lighting, image-processing software and controls that inspects, measures or locates parts automatically and sends a decision to the production equipment. Unlike a general-purpose camera, it works under controlled lighting and presentation, so it can make the same reliable call on every part at line speed. The core components are:
- Lighting that makes the feature or defect visible.
- Optics and cameras: 2D area scan, line scan or 3D sensors, chosen for the field of view and resolution.
- Software, rule-based or AI, that turns the image into a pass, fail or measurement.
- Integration with PLCs, robots, reject gates and quality databases.
Our complete machine vision guide covers each component in depth. The rest of this article focuses on how manufacturers use machine vision systems, task by task and industry by industry.
The five jobs machine vision systems do on a production line
Almost every application is one of these five, or a combination. Naming the job first is the quickest way to scope a project, because it decides the camera, the lighting and how the pass or fail decision is made.
| Job | Question it answers | Typical method | Example output |
|---|---|---|---|
| Defect inspection | Is anything wrong with this part? | 2D imaging with tuned lighting; rule-based or AI classification | Pass or fail, defect type, size and location |
| Measurement and gauging | Is it the right size and shape? | Calibrated 2D edges, or 3D laser profiling for height | Dimensions in mm or µm against tolerance |
| Presence and assembly verification | Is everything there and in the right place? | 2D pattern and feature matching | Missing, extra or misplaced components |
| Guidance and alignment | Where is the part, and how is it oriented? | 2D or 3D location in robot coordinates | X, Y, Z and rotation sent to a robot or stage |
| Counting and sorting | How many, and which kind? | Image analysis of single or touching parts | A verified count, or a sort decision per part |
Applications by task
Defect detection and surface inspection
Defect inspection is the most common application. Cameras find scratches, dents, cracks, contamination, voids, coating gaps and cosmetic flaws that a person would miss after a few hours of the same task. Reflective and textured surfaces are the hard cases: the lighting has to make the defect visible before any software can find it. Our articles on scratch detection on reflective parts and AI defect detection go deeper, and the automated inspection examples and thin film defect detection system are examples Sciotex has built.
Continuous web and extrusion inspection
Film, foil, paper, nonwovens and extruded profiles are inspected as they move, usually with line scan cameras that build an unbroken image of the material. The system has to keep up with line speed, map each defect to a position on the roll, and avoid flagging normal texture as a flaw. See machine vision for web inspection, the web inspection system and the extrusion inspection system.
Measurement, gauging and dimensional inspection
Calibrated vision replaces calipers, pin gauges and optical comparators for 100% inspection. 2D edge measurement handles outlines, hole positions and diameters; 3D laser profiling adds heights, flatness, warp and volume. The limit is set by pixels across the tolerance, the lens and the calibration, not by the software alone. Read more in automated dimensional inspection and how a laser profilometer works.
Presence, completeness and assembly verification
Is the O-ring seated, the clip installed, the label present, the right part in the right pocket? Presence and assembly checks are usually the fastest payback in machine vision, because a single missed component can mean a field failure or a recall. The gasket inspection system and the O-ring, washer and disk inspection system are typical examples.
Guidance, alignment and robotic inspection
Vision tells a robot or motion stage where a part is so it can pick, place, align or inspect it. In robotic inspection, the robot moves a camera, or the part, through several views so one station can check every face. Sciotex’s robotic vision inspection system for small, high-tolerance parts combines handling, inspection and counting in one cell.
Counting and sorting
Image-based counting is faster and more accurate than weighing for small parts that vary in weight, and it can sort by size, shape or color at the same time. The automated counting guide covers the methods, and PerfectCount is Sciotex’s standard counting machine.
Safety and handling checks
Vision also protects people and equipment: checking pallet loads before they move, confirming guards and fixtures are in place, and catching tip or spill risks early. See pallet safety inspection and the pallet inspection system.
How machine vision systems are used by industry
The same five jobs appear in every industry; what changes is the part, the tolerance and the regulations. This table lists the inspections we see most often.
| Industry | Typical applications | Common imaging method | Sciotex example |
|---|---|---|---|
| Electronics and semiconductor | Wafer and chip defects, solder and component presence, connector pins | High-magnification 2D, 3D for coplanarity | Chip and wafer inspection |
| Pharmaceutical and medical | Tablet hole and coating checks, vial and well-plate inspection, device assembly | 2D with diffuse lighting, controlled color | Tablet hole inspection |
| Optics and photonics | Fiber end-face and component defects, mirror and coating flaws | Microscope optics, coaxial or dark-field lighting | Fiber optic inspection |
| Plastics, film and packaging | Web defects, extrusion profiles, printed graphics | Line scan cameras, backlighting | Screen printing vision system |
| Automotive and industrial parts | Machined dimensions, gaskets and seals, surface finish | 2D gauging, 3D laser profiling | Gasket inspection |
| Life science and lab automation | Well-plate imaging, sample handling, cell imaging | Microscopy and custom optics | Well-plate inspection |
Matching the imaging method to the application
Most applications start with 2D imaging, but several need something else. A defect that is a change in height, not in color, needs 3D. A coating thickness or material mix-up may only be visible outside the visible spectrum. Choosing the method before choosing the camera avoids the most expensive mistake in vision projects: a system that physically cannot see the defect.
| If the feature is… | Consider | Read more |
|---|---|---|
| A change in color, contrast or outline | 2D area scan imaging | Camera types |
| A change in height, depth or shape | 3D laser profiling or structured light | 2D vs 3D vision |
| On a continuous moving material | Line scan imaging | Web inspection |
| Hidden by glare on shiny parts | Polarized or angled lighting | Polarization in machine vision |
| Subtle and variable, like natural texture | AI or deep learning classification | AI in vision systems |
What makes an application a good fit
Not every visual task should be automated. The applications that pay back reliably share four traits:
- A defined decision. The defect or dimension can be written down with a size, a tolerance or a reference sample, not just “looks wrong.”
- Repeatable presentation. Parts arrive at a known position, or the system can be designed to handle the variation.
- Enough volume or risk. The labor saved, scrap avoided or escapes prevented outweigh the cost of the system.
- Somewhere to act on the result. A reject gate, a robot, an operator alert or a data record that someone will use.
A quick labor check shows why volume matters. One inspector per shift on two shifts, at a loaded rate of $25 an hour for 8 hours a day and 250 days a year, costs $25 × 8 × 2 × 250 = $100,000 a year, before counting escapes and rework. Sciotex vision systems typically pay back in 7 to 12 months. Our article on building a machine vision ROI case walks through the full calculation, and machine vision vs manual inspection compares accuracy, speed and cost.
Common mistakes when choosing an application
- Starting with a camera instead of the defect: the lighting and optics decide what is visible.
- Testing on a handful of good parts only. A feasibility test needs real defects, including the borderline ones.
- Asking one system to do too many jobs at once. Splitting inspection across stations is often cheaper and more reliable.
- Ignoring how parts are presented. Fixturing and handling are frequently half the engineering effort.
- Leaving out the reject path. Finding a defect is only useful if the bad part is removed and logged.
Have an application in mind?
Sciotex designs and builds custom machine vision systems around your parts, with free testing of your sample parts to confirm what the system can see before you commit. We work with any camera and software, AI or traditional.
Frequently asked questions
The main applications are defect inspection, measurement and gauging, presence and assembly verification, robot guidance and alignment, and counting or sorting. Most production systems combine two or more of these.
Electronics and semiconductor, pharmaceutical and medical device, automotive, packaging, plastics and film, and food production are the heaviest users, but any process with a repetitive visual check can benefit.
For well-defined defects and dimensions, yes, and it inspects every part instead of a sample. Highly subjective cosmetic judgments may still need a person to review borderline cases flagged by the system.
Run a feasibility test on real parts, including known defects. Sciotex offers free testing of sample parts to confirm the right lighting, optics and method before a system is designed.
Machine vision applications are industrial: inspecting, measuring and guiding on a production line with controlled lighting and optics. Computer vision is the broader field of image understanding, used in phones, cars and medical imaging as well as factories.