Electronics assembly adopted automated inspection early because the features are small and numerous. The same tools, cameras, lighting, 3D sensors and image analysis, now inspect bottles, castings, welds, medical devices, food, textiles and packages. If a person checks it by looking, a vision system can probably check it, and usually more consistently.
The four things vision systems do
| Task | Question answered | Examples |
|---|---|---|
| Inspection | Is it right? | Surface defects, missing features, assembly completeness, label presence |
| Measurement | How big is it? | Dimensions, gaps, bead volume, flatness, fill level |
| Identification | What is it? | Reading text, barcodes and 2D codes; sorting by type |
| Guidance | Where is it? | Locating parts for robots, alignment before assembly |
2D or 3D?
Most tasks are solved in 2D: a camera and well-designed lighting. Add 3D when the feature of interest is one of height or shape: the volume of an adhesive bead, the flatness of a sealing surface, the depth of a dent, a part that has the same color as its background. Structured light, laser triangulation and stereo cameras are the common 3D methods, each suited to a different range of sizes and speeds.
Lighting is the larger part of the problem
The camera only records what the lighting reveals. A scratch on polished metal is invisible under diffuse light and obvious under low-angle light. Clear plastic needs backlighting. Printed text on a curved, shiny surface needs a dome. Choosing the lighting and optics is where experience counts most, and it is why two systems with the same camera can perform very differently.
Good candidates for a first project
- A manual inspection step that is slow, tiring or inconsistent.
- A defect that keeps reaching customers despite inspection.
- A measurement currently made by sampling with hand gauges.
- A check that is needed on every part for traceability.
- A step where the cost of a missed defect is high.
How to scope it
- Define the defects. Collect physical examples of each one, and of acceptable parts near the limit.
- State the requirement in numbers: smallest defect to detect, measurement tolerance, parts per minute, acceptable false reject rate.
- Describe how parts are presented: moving or stationary, oriented or random, clean or oily.
- Decide what happens on a reject: stop, divert, mark or record.
- Decide where the results go: a light, a PLC, a database.
Insist on a feasibility study
Before anything is built, your real samples should be imaged with the proposed optics and lighting, and the results shown to you. This costs little, takes days, and answers the central question: can the defect be seen reliably? Projects that skip this step are the ones that disappoint.
Off-the-shelf or custom?
Smart cameras and standard vision sensors handle a wide range of simple, well-defined checks and are quick to deploy. A custom system is warranted when the part, the defect or the environment is out of the ordinary, when several kinds of sensing must be combined, when micron-level measurement is needed, or when the vision has to be built into your own product.
What success looks like
A good vision system is one the operators stop noticing. It rejects the bad parts, passes the good ones, and produces a record that nobody argues with. Getting there is mostly a matter of defining the task precisely and proving the imaging before building the machine.