Context
This was a new production line where defects occurred only once or twice per year.
Problem
Most inspection systems required large defect datasets to function. With so few defects:
- Data collection was unrealistic
- Competing solutions could not achieve reliable detection
- Training on "normal-only" data produced poor results in practice
Additionally, the customer could not export data and required all setup and training to be done on-site.
What the Customer Needed
A system that could start with very limited data and improve over time under strict on-site conditions.
Hypernology's Solution
Hypernology:
- Generated initial defect data using controlled demo samples
- Provided on-site labeling and training tools
- Enabled the customer to continuously improve the model as new data appeared
All configuration, training, and support were performed on-site.
Outcome
- Inspection enabled despite near-zero defect data
- Customer gained long-term control over model improvement
- System performance improved gradually without vendor dependency
Lesson
Limited data should not prevent automation. Practical systems are designed to work under real-world constraints and improve over time.