Skip to content
PixelDevSolutions home

Edge Anomaly Detection for Manufacturing

On the edge defect detection running on-device at the station, no cloud.

Client
Confidential
Year
2024
Category
Computer Vision
Role
Design + Build
Timeline
4–6 weeks
Edge Anomaly Detection for Manufacturing: State-of-the-art anomaly detection (ANOMALIB) for manufactured parts, deployed to run on edge hardware at the line.

The problem

Sending every part’s image to the cloud for inspection is slow and adds a dependency the line can’t afford. Inspection has to happen at the station.

What we built

An ANOMALIB-based model trained on good parts, then optimised and deployed to edge hardware so it inspects each part on-device in the time it takes to place the next one. It flags anything that departs from normal, so nobody has to enumerate defect types up front.

The result

Inspection at the station, at line speed, with no network in the loop.

runs at the station, no cloud round-trip
on-deviceruns at the station, no cloud round-trip
trained on good parts only
unsupervisedtrained on good parts only
a pass/flag decision before the next one lands
per parta pass/flag decision before the next one lands
Edge Anomaly Detection for Manufacturing, screen 1
Edge Anomaly Detection for Manufacturing, screen 2

Built with

  • Python
  • ANOMALIB
  • PyTorch
  • ONNX
  • Edge AI
  • OpenCV
Next projectDocument OCR & Field ExtractionAI

We deliver what we commit.

Tell us what you're trying to build.

We'll come back within 24 hours with honest feedback on scope, timeline and cost, whether or not we turn out to be the right fit.

Start a project

Or email pixeldevsolutions@gmail.com