Appearance
Machine Vision
Last updated
License feature: Machine VisionMachine Vision turns a camera image into tags. A vision device is a camera (or up to four) plus one or more image tools — count parts, check something is present, match a color, find a taught feature, measure a width or a bore, read a printed lot code, read barcodes. Each tool gives its own result and its own pass or fail, and the device publishes one pass tag for the whole part.
It runs as an ordinary device driver. That is the important part: what it finds arrives as tags like any others, so it can go on a dashboard, raise an alarm, be trended in the historian, feed an SPC chart, or be read by a script or ladder logic with nothing further to wire up.
Use it for go/no-go checks at a station: a tray short a part, a cap missing or the wrong color, a label with the wrong lot code, a bore out of tolerance. It is not a replacement for a vendor's smart camera on high-speed or sub-millisecond applications — see Limits below.
Quick start
- Go to Devices → Add Device and choose Machine Vision (camera) (under Vision). Give it a name, pick the Camera type and fill in what that type needs — an address, a URL, a camera number, or nothing for an uploaded test image. See Cameras and connection.
- Open Vision in the sidebar and pick the device from Choose a vision device….
- Press Grab one. The picture appears with the default tool (Tool 1, Circles) drawn on it.
- In the tool card, pick the Tool type you need, press Draw region and drag a box round the area to inspect.
- Adjust the tool's sliders while pressing Grab one (or press Start live to refresh continuously) until the result is right on a known-good part. Then check it fails on a bad one.
- Under Pass / fail, tick Include this tool in the verdict and set the target or limits.
- Press Save. The running device reloads with the new settings and its tags are created or renamed to match.
- Bind a dashboard or an alarm to the device's
passtag.
Nothing you change on the page reaches the running device until you press Save. See Tuning on the Vision page.
How the page is laid out
| Area | What it does |
|---|---|
| Toolbar | Device picker, Start live / Stop, Grab one, Reset tool (sets the selected tool's parameters back to defaults), Save |
| Verdict and picture | Pass, Fail or Nothing checked in large type, "N of M tools judged", processing time, one row per tool, and the annotated image |
| Statistics | Live counts, yield over 1, 8 or 24 hours, SPC capability. See Statistics and SPC |
| Camera / Cameras | Camera type and address, When to inspect, Scale, Keep failed images |
| Trigger & lighting | A trigger from any tag, plus steps before and after the picture. See Triggers and lighting |
| Programs | One inspection per product. See Programs and changeover |
| Tools | The tabs for each tool, with Add tool; the selected tool's card below |
| Failed parts | The archive of kept images (hidden when Keep nothing is chosen). See Failed-part images |
One frame, every tool
All the tools on a camera run against the same frame. Grabbing separately per tool would let two tools see the part in different positions, and a combined verdict on two different moments means nothing. Each tool has its own region, parameters and pass rule; the tabs show each tool's last result as a colored dot (green pass, red fail, amber not judged).
Several cameras
Some parts cannot be judged from one side — a label on top and a date code on the side. A vision device can have up to four cameras, and one verdict covers all of them. Every camera is grabbed at the same time for each inspection, each tool looks through one camera, and pass is true only when the judged tools on every camera pass. A camera that gives no picture fails its own tools and the combined verdict. See Several cameras for the full rules.
Who can do what
| Action | Roles |
|---|---|
| Open the page and see a device's settings, statistics and failed-part images | Any signed-in user with access to the page |
| Grab one, Start live, Teach, upload a test image, Save, Run it once, Track in SPC, Reset counts | Admin, Editor |
| Clear all failed-part images | Admin |
| Write a vision device's tags (trigger, changeover, settings) from the tag browser, a dashboard or the API | Admin, Editor (scripts, workflows and ladder logic write on the appliance's behalf) |
Machine Vision is a licensed feature of its own. Without it the Vision page shows a lock, the Machine Vision (camera) card in the Add Device wizard is locked, and existing vision devices stop until the feature is licensed; they start again by themselves once it is. With it, users without the Admin or Editor role can open the page and see settings, statistics and failed-part images, but Start live, Grab one, Save, Teach, Teach characters and the test image upload are hidden from them.
Requirements
- Vision needs the appliance's image-processing and barcode components, which a standard installation includes. If they are missing, the device fails to connect with "Vision cannot run: …", and if only the barcode component is missing, only QR codes are read. Reinstall or update the appliance software to restore them.
- The camera must be reachable from the appliance: an HTTP still, an RTSP/MJPEG stream, or a USB camera plugged into the appliance.
Limits
- It inspects stills, one inspection at a time. How long one takes depends on the camera, the resolution and the tools — watch
last_run_ms. It is not suited to lines that need a result within a few milliseconds of a trigger. - Pattern matching is not scale-invariant, text reading is for printed characters in a taught font, and there is no deep-learning model. See the individual tool sections.
- It does not drive the camera's own outputs and has no built-in reject logic. Once the result is a tag, alarms, scripts, workflows and ladder logic decide what to do about it — and they can be seen and changed, which logic buried inside a driver cannot.
- Dashboards can show the most recent kept image with the Camera widget, not a live video feed. See Failed-part images.