Motionwell has delivered vision inspection systems for electronics manufacturing in Singapore since 2014, working with electronics and semiconductor manufacturers on projects including a vision-guided sensor panel assembly line holding ±0.01 mm repeat accuracy with ABB SCARA robots, cleanroom automated test equipment, and a copper-plate thickness measurement machine built around Keyence displacement sensors. This guide covers how to specify vision inspection for an electronics line: what is worth inspecting, how to size the camera, how to choose the lighting, and how the system talks to the rest of the machine.
The general architecture question, smart camera versus PC-based, false-reject tuning, and how results are stored for traceability, is covered on our machine vision inspection systems page. This article is the electronics-specific version.
What does a vision system actually have to do on an electronics line?
A vision inspection system captures an image under controlled lighting, evaluates it against defined criteria, and produces a pass/fail decision or a measurement that the line controller can act on.
| Component | Function | What determines the choice |
|---|---|---|
| Camera | Captures the image, area scan or line scan | Resolution needed, field of view, frame rate |
| Lens | Sets magnification, focus, and depth of field | Working distance, feature size, depth of field |
| Lighting | Makes the feature or defect visible and repeatable | Defect type, surface finish, ambient light |
| Processor and software | Runs the algorithms, stores results, talks to the PLC | Inspection complexity, cycle time, data retention |
The failure point is usually not the camera or the software. It is the lighting, closely followed by part presentation. A defect that is invisible under one lighting configuration is obvious under another, and a part that sits differently in the nest each cycle sets a detection floor no algorithm can beat. This is why we test production parts under several lighting setups before specifying anything.
Which inspection tasks come up in electronics manufacturing?
Component presence, position, and polarity
The highest-value and simplest check. The camera confirms that each component is present, in the right place, and the right way round. Polarity in particular is cheap to catch inline and expensive to catch at functional test.
Lighting is usually diffuse or dome to suppress reflections off component bodies and solder. Resolution is set by the smallest marking you need to resolve, not by the smallest component.
Connector pin alignment and coplanarity
Bent, missing, or non-coplanar pins are a classic pre-mate failure. Pin position in the image plane is a 2D problem and solves with a backlit or dark-field setup. Coplanarity is a height measurement and needs 3D: laser profiling or structured light.
If the connector mates blind during assembly, this check pays for itself the first time it prevents a damaged mating half.
Dimensional measurement and panel datums
High-resolution cameras with telecentric lenses measure hole positions, gap widths, edge offsets, and datum locations at production speed, feeding statistical process control.
The engineering problem in inline measurement is repeatability across thermal drift, vibration, and part-to-part variation, not raw accuracy in a lab. Measurement stations need thermal compensation, isolated mounting, and a calibration routine an operator can actually run on shift.
Code reading and grading
Reading a Data Matrix code is easy. Grading its print quality is the part that matters for traceability. A grading system evaluates symbol contrast, modulation, decodability, defects, and quiet zone, then assigns a grade under ISO/IEC 15415 for 2D codes or ISO/IEC 15416 for 1D barcodes, and stores that grade with the production record.
Grade the code immediately after it is applied and before the unit is aggregated. A bad code found after aggregation forces you to break the case and reconcile the record. The line-level version of this problem is covered under serialization and traceability lines.
Vision-guided assembly
The camera locates the part or the panel datum and sends a position offset to a robot or servo stage. This removes the need for precise fixturing, because the robot absorbs the placement variation instead of the fixture.
On the sensor panel line, the vision system locates panel datums, passes offsets to the SCARA controller, and verifies assembly completion after each operation, holding ±0.01 mm repeat accuracy. The critical procedure is the camera-to-robot calibration: we use multi-point calibration with verification runs so that accuracy holds across the whole working envelope, not just at the calibration point. The mechanical side of that cell is described in the SCARA panel assembly case study and in our guide to SCARA robots in electronics assembly.
Keyence or Cognex: which platform fits your inspection?
These are the two platforms Motionwell deploys for production inspection. The choice is driven by how variable the defect is, not by preference.
| Platform | Strengths | Best fit |
|---|---|---|
| Keyence IV3 | Built-in lighting, fast setup, self-contained | Presence and absence, orientation, simple pass/fail at a single point |
| Keyence CV-X | Multi-camera on one controller, high-speed processing, large tool library | Multi-point inspection, high-speed lines, pattern matching |
| Cognex In-Sight | Spreadsheet programming, AI-based defect classification | Variable or visually complex defects where rule-based logic produces false rejects |
Rule-based tools work when you can describe the defect. When you can only show examples of it, a scratch that matters versus a scratch that does not, a trained classifier will usually beat a threshold. That is the practical dividing line between the two approaches.
Where an application needs non-standard optics or an existing in-house image-processing stack, a generic GigE Vision camera with third-party software is an alternative architecture. It buys flexibility and costs you the integrated toolchain, the vendor support path, and usually several weeks of development.
How do you size camera resolution for a PCB or a panel?
Start from the smallest feature you must resolve and work backwards. The arithmetic is simple and it is where most specifications go wrong.
Resolution at the part = field of view width ÷ sensor width in pixels. You then need enough pixels across the feature: roughly 3 for reliable detection, and 5 or more if you intend to measure it.
| Field of view width | Sensor width | Resolution at the part | Pixels across a 0.1 mm feature |
|---|---|---|---|
| 100 mm | 2448 px (5 MP) | 0.041 mm/px | ~2.4 |
| 100 mm | 4096 px (12 MP) | 0.024 mm/px | ~4.1 |
| 200 mm | 4096 px (12 MP) | 0.049 mm/px | ~2.0 |
| 200 mm | 5472 px (20 MP) | 0.037 mm/px | ~2.7 |
| 400 mm | 5472 px (20 MP) | 0.073 mm/px | ~1.4 |
Read the bottom row carefully. If you need to detect a 0.1 mm defect across a 400 mm panel, a single camera will not do it, whatever the marketing says. The answer is to split the field between multiple cameras, index the panel under one camera in several positions, or use a line-scan camera and move the part. Deciding that at concept stage costs nothing. Discovering it at commissioning costs a station.
Two more constraints that come from the optics rather than the sensor: depth of field shrinks as you magnify, so a warped panel can drop out of focus at the edges, and a telecentric lens removes perspective error but restricts you to a field of view no larger than the lens itself.
Which lighting technique will reveal your defect?
Lighting is the most critical and most underestimated decision. The same defect can be invisible or obvious depending on angle, type, and colour.
| Technique | Best for | How it works |
|---|---|---|
| Diffuse (dome) | Reflective component bodies, suppressing glare | Light arrives from all angles, minimising specular highlights |
| Directional (bar or ring at an angle) | Surface defects, scratches, texture | Controlled angle casts shadows at defect edges |
| Dark field (low angle) | Fine scratches, raised or engraved features | Near-horizontal light; only defects scatter light back to the camera |
| Backlight | Silhouettes, pin positions, edge measurement | Light behind the part; the camera sees the outline only |
| Structured light | Coplanarity, height, solder paste volume | A projected pattern deforms on the surface; triangulation gives height |
| Coaxial | Flat specular surfaces, wafers, lead frames | Light along the camera axis; specular surfaces return bright |
One practical point specific to electronics: mixed surface finishes in a single field of view, matte substrate, glossy solder, bright metal shielding, often cannot be lit well by one source. Two exposures under different lighting, combined in software, will beat a compromise setup nearly every time.
How does the vision system talk to the PLC and the robot?
A vision system is a subsystem of a machine, not a standalone instrument. It has to trigger at the right moment, finish inside the cycle time budget, and hand its result to whatever acts on it.
| Integration point | Protocol | What is communicated |
|---|---|---|
| PLC trigger and result | EtherNet/IP or discrete I/O | Trigger, pass/fail, measurement values |
| Robot offset data | TCP/IP socket or a vendor protocol | X, Y, theta offsets for guided alignment |
| MES or SCADA logging | OPC UA or a database write | Images, measurements, statistical summaries |
| Reject mechanism | Discrete I/O from the PLC | Diverter, pusher, or robot sort command |
Allen-Bradley CompactLogix and Siemens S7-1500 are the platforms we most often integrate with. The PLC owns sequencing, safety interlocks, and reject control; the vision system owns image processing. Keeping that split clean is what makes the machine debuggable two years later.
Every inspection result should be tied to a part identifier, serial number, panel ID, or a sequence counter, at the moment of inspection. Retrofitting that link is painful, and without it the images you archived are evidence of nothing in particular.
Where does the same method apply outside electronics?
The sizing, lighting, and integration method above transfers directly to other regulated production. The differences are in what counts as a defect and how much evidence you have to keep.
- Medical devices. Assembly verification and dimensional conformance at each station of a rotary machine, with gauge repeatability studies to prove the measurement system itself. See medical device automation.
- Pharmaceutical packaging. Code grading and label verification tied to a serialization record, on pharma and packaging lines.
What should you send us for a feasibility study?
The most productive first step is always a feasibility study with real parts. Nothing else settles the lighting question.
- Sample parts: 5 to 10 good ones, and 5 to 10 with each defect type you care about
- Your current inspection criteria and the tolerances that matter
- Line speed, cycle time, and how the part is presented today
- Existing PLC platform and how the reject mechanism should behave
We test the parts, recommend a configuration, and give you a budget range before you commit. If the study shows the defect cannot be lit reliably, we will say so, that answer is worth more than a system that works in a demo and fails on the line. The way we run that study inside a larger build is described in our guide to special purpose machine design, and if you are still comparing suppliers, the system integrator evaluation framework covers what to ask.
Contact the engineering team to arrange a feasibility study, or read more about our work in electronics and semiconductor manufacturing.