Engineering
Engineering Guides & Machine Vision FAQ
Reference material we use internally, written so an engineer can act on it without a sales call.
Author: VisionAxiom EngineeringReviewed by: VisionAxiom Technical ReviewUpdated: 2026-09-02
Direct answer
Where should an engineer start scoping a vision inspection project?
Start with the defect and the smallest feature to resolve, then work outward: field of view and resolution, optics and working distance, lighting geometry that makes the defect visible, exposure and triggering against line speed, interface bandwidth, compute architecture, and finally the validation plan and PLC integration.
Guides
- Camera Selection — resolution math, shutter, interface and margin.
- Lighting — geometry first: bright field, dark field, dome, backlight, coaxial, telecentric pairing.
- Optics — focal length, working distance, depth of field, distortion, telecentricity.
- SDK & Integration — acquisition APIs, result schemas, PLC handshakes, data retention.
- System Architecture — compute placement, timing budgets and false-reject control.
Machine vision FAQ
The questions below are the ones we are actually asked in feasibility calls. They are answered as engineering positions, including cases where our answer is that you do not need a custom system.
Frequently asked questions
- What resolution do I need for my inspection?
- Compute pixels per millimeter from your field of view and sensor resolution, then require enough pixels across the smallest feature: roughly 2-3 pixels for reliable presence detection, 4-6 for character verification, and 10 or more for dimensional measurement with tight tolerances. Add margin for fixture variation.
- Do I need AI for visual inspection?
- Only when the defect cannot be described as a measurable threshold. Rules-based tooling is deterministic, easier to validate and cheaper to maintain. A learned model is justified for variable cosmetic defects, and it brings obligations: labeled data, retraining, and version control tied to your quality records.
- Is a smart camera enough, or do I need a PC-based system?
- A smart camera is often the right answer for a single stable check on a moderate-speed line. PC-based architectures win when you need multiple synchronized cameras, high bandwidth, heavy image processing, custom algorithms, long image retention or deep data integration.
- How long does a vision project take?
- Feasibility imaging on real samples is typically the fastest and most valuable step. Full build timelines depend on mechanical integration and line access, which are usually the schedule drivers rather than software.
- Can you use our existing cameras and PLC?
- Frequently, yes. We start by evaluating whether existing acquisition hardware can resolve the defect. If it can, reusing it is cheaper; if it cannot, we say so with the imaging evidence to support it.
- What data does an inspection system produce?
- At minimum: timestamp, part or job identifier, per-check results with measured values, overall verdict, and a reference to the retained image. That record is what makes inspection auditable rather than anecdotal.