Vision Defect Inspection Platform
Multi-model inspection with configurable defect rules
Overview
Visual inspection fails in two directions. A rule-based system misses defects it was never taught, and an over-trained model rejects good parts until the yield loss becomes unacceptable. The practical answer is usually a hybrid: geometric rules for what can be measured, and a learned model for what can only be recognised.
This platform combines machine vision, deep learning, and data analysis, supporting one-machine multi-inspection and model switching. Inspection items are configured rather than hard-coded, so adding a defect class does not require a rebuild.
The honest constraint: camera and lens selection, lighting, defect sample coverage, escape and false-call targets, and cycle time all have to be proven on project samples. A published capability statement is a starting point for that validation, not a substitute for it.
Specifications
Features
- Configurable inspection items, so adding a defect class does not require a rebuild
- Multi-model switching on one machine
- Combines geometric measurement with learned recognition
- Inspection data stored for trend analysis
- Escape and false-call targets proven on project samples
Typical applications
- Component appearance inspection
- PCB defect detection
- Assembly verification
Discuss this configuration
Send the device details and the measurement you need to make. We will confirm which specifications are fixed and which need verification on your sample.
Contact engineering
