Skip to content
RoboFab

Engineering study

Edge AIInspection Camera

Defect detection that runs on the device without a PC on the line

Engineering studyArtificial IntelligenceEmbedded SystemsElectronics Engineering
Camera lens reflecting points of light in the dark
Reference visual

This is a concept study that shows how RoboFab would engineer this kind of systemIt is not a delivered client project and its images are reference visuals

01The problem

Visual inspection is repetitive workand defects slip through when attention drops late in a shift

02The approach

  1. 01A compact camera unit runs a trained defect model directly on the device
  2. 02A failed part triggers a reject signal on the line
  3. 03Every image is stored for review and for retraining the model

03Architecture

Systemarchitecture

System architecture

Camera moduleOn device modelReject signalImage archiveRetraining

Technologies

  • Raspberry Pi
  • Camera module
  • PyTorch
  • OpenCV
  • Model optimization
  • GPIO triggers

A delivery would include

  • Camera and lighting setup
  • Dataset labeling
  • Model training and optimization
  • Enclosure and mounting

04Value

The valueit creates

  • Fewer escaped defects

    Every part inspected with the same attention

  • Simple installation

    One compact unit at the inspection point

  • A model that improves

    Reviewed images make the next version better

Raspberry Pi single board computer on a dark tableReference visual

Need somethinglike this

Start a project

More studies

Otherengineering studies

Other engineering studies