Skip to content
OXODES

OXO-DETECT / AI camera inspection

Let no defect slip past.

Cameras and AI models, set up for your part, check every part on the line.

Not one product: a solution built around your problem.

For every problem we bring the right tools together. The light, lens and fixture are chosen for the part as well.

  • Smart cameras
  • AI models trained for your parts
  • Interfaces made for your process
  • Camera control made for your station
  • Integration with your line and systems

What the cameras check.

  • Measure

    Dimensions, positions, gaps and angles, in line.

  • Detect

    Presence, absence and position of a feature or part.

  • Recognise

    Colour, shape and which variant is in front of the camera.

  • Count

    Parts, fasteners, components on a board or in a pack.

  • Compare

    The part against a reference, feature by feature.

  • Read

    Barcodes, printed text and labels.

Typical targets: cracks, surface anomalies, sealing beads, stitching, labels, colour, missing or misplaced components.

Some defects have no fixed shape. That is what the models are for.

A crack, a scratch or a broken sealing bead looks different every time. Fixed rules find that hard; anomaly models are built for it. They learn what a good part looks like and flag what differs.

The models run at the line, so the verdict is made while the part is still at the station. Where a defect has a known shape, we train models on your own parts.

Training takes minutes, hours or days, depending on the part and the defect.

  • Anomaly detection

    Learns good parts; flags what differs.

  • A verdict at the line

    The decision is made while the part is still at the station.

  • Models trained on your parts

    For defects with a known shape.

VisandTrace / The station software

VisandTrace: made to improve the inspection process.

VisandTrace is our own software for the station. It takes the result and the raw image from the cameras, draws the result on the picture and keeps it on the PC. What is kept is set for each camera, so months later you can still see why a part was rejected. Not every station needs it; it is used where the project calls for it.

Two tiles showing the same frame under two different overlay rule sets.
  1. 1

    The camera’s verdict

    Each tile shows the verdict as a colour and a word: OK or NOK.

  2. 2

    The result drawn on the picture

    The boxes measured, the values and the verdict are drawn on the raw image from the camera.

  3. 3

    The record

    The station says on screen how many results arrived, how many were recorded and how many were written to disk.

    1204 seen1204 recorded1180 written3 waiting21 skipped to keep up

  • Several cameras, one verdict.

    As in OXO, a program triggers the cameras, waits for their results and gives the part one verdict.

  • You decide what is kept.

    Per camera: every result, NOK only, or every NOK plus a sample; the raw image, the picture with the result drawn on it, or both. Searched by date, camera and verdict, exported to CSV and Excel.

  • Connected to your systems.

    Every part goes to a folder, TCP, HTTP or Modbus; PLC I/O is addressed by name.

VisandTrace specifications

Station

Cameras on one station
Tested to nine at 30 fps
Result rate
270 results/s, every one recorded
Platform
Windows 10 / 11, 64-bit
Languages
Turkish, English
Sign-in
RFID badge or password; three roles

Cameras and images

Cameras
Hikrobot, Basler, Daheng and other smart cameras
Wire format
JPEG, BMP or raw, chosen per camera
Images kept
Frame, overlay or both, per camera, by rule
Deferred overlays
Re-derivable from a 4 KB file, months later

Record and integration

Record
One file per station; CSV and XLSX export
Field I/O
Modbus TCP, addressed by name
Sends parts to
Folder, TCP, HTTP, Modbus

Questions buyers ask

Is OXO-DETECT a product?

Not a single product: a solution. The camera, lens, light, fixture, AI model and software are chosen and set up for each project, for the part and the defect that has to be caught.

Where do you use AI?

In the inspection: anomaly models that learn good parts and flag what differs, and models trained on your own parts for defects with a known shape. The models run at the line. VisandTrace does not judge parts; it records and shows the cameras' results.

How long does it take to train a model?

It depends on the project: minutes for some parts, hours or days for others. Once we have seen your part and the defect, we give you a realistic figure.

Does every station need VisandTrace?

No. It is used where it makes the process better: to run several cameras on one screen, to keep each result with its picture, or to give one part verdict from several cameras.

Can it send results to our MES?

Yes. Every part goes out in the same structured format to a watched folder, a TCP socket, an HTTP endpoint or Modbus registers, in any combination, behind a queue so a slow MES never delays a part.

Are the pictures kept?

Yes, by the rule you set. For each camera: every result, NOK only, or every NOK plus a sample; the raw image, the picture with the result drawn on it, or both. Searchable by date, camera and verdict.

Tell us the defect you need to catch.

Tell us about the part and the defect. We will show you how the cameras would check it.

Book a demonstration
Email
info@oxodes.com
Phone
+90 542 489 84 31