Computer vision for automated manufacturing defect inspection

Many defect types, each needing a different inspection method

Manufacturing inspection can involve surface scratches, fiber dust, spots, cracks, edge damage, surface irregularities, dimensional deviations, potting residue and missing or misaligned characters.

The challenge was not simply detecting whether a defect existed. Different defects required different approaches: visual detection, dimensional measurement, edge analysis and character recognition.

Some requirements involved very small defects and pixel-level measurements, adding challenges around camera calibration, image quality, lighting, data availability and processing latency.

A computer vision pipeline, not a single model

Saptum combined deep learning with traditional computer vision, choosing the right technique for each defect being inspected.

Core inspection capabilities

From image to inspection decision

1. 01

   Image capture

2. 02

   Image preprocessing

3. 03

   Detection / classification

4. 04

   Measurement

5. 05

   Defect decision

6. 06

   Inspection output

Engineering impact

• Automated image-based inspection
• Reduced dependence on purely manual visual inspection
• Support for multiple defect categories
• Low-latency edge processing architecture
• Modular approach for adding new defect types
• Foundation for continuous model improvement
• Integration-ready inspection workflow

Technology Development Partner

• Computer vision development
• Machine learning approach
• Image-processing pipeline
• Inspection logic
• System engineering
• Edge-ready architecture