How Jidoka AI Transformed Dowel Pin Quality Control in High-Speed Wood Processing

In a PoC with a European wood processor, Jidoka’s AI-powered inspection achieved 99.8% accuracy at 1,200 parts/min, detecting cracks, knots, burns, and grain flaws — marking a breakthrough in wood and furniture component quality control.

How Jidoka AI Transformed Dowel Pin Quality Control in High-Speed Wood Processing

Industry

Wood Processing / Furniture Components

Result 1

99.8% defect detection accuracy across 10,000+ annotated images

Deployments

KOMPASS AI integrated with existing cutting machine (POC)

Result 2

1,200 parts inspected per minute with <50ms decision speed

Overview

Furniture manufacturing relies on wooden dowel pins that demand both dimensional and visual quality checks. While dimension sorting is often automated, visual defects such as discolouration, cracks, or burn marks typically require manual inspection.

For a high-volume European manufacturer, this dependency created challenges: inspection was slow, labour-intensive, and inconsistent. Given the unpredictable defect types, irregular spacing, and high-speed movement, traditional inspection methods were not viable.

‍

The Opportunity

The proof-of-concept addressed four core challenges in dowel pin quality control:‍

Unpredictable defect types (cracks, knots, grain imperfections, burns)
‍High SKU variation across dowel designs‍
High throughput requirement (>1,000 parts/minute)‍
Over-reliance on manual inspectors for visual quality

"Dowel AI Inspection was one of the most exciting and challenging projects we’ve had the chance to work on. Achieving millisecond-level detection and ejection without altering the existing machine shows what’s possible when deep-tech meets real-world manufacturing."- Jidoka's European technology collaborator

JIDOKA's Approach

Jidoka integrated its Kompass AI visual inspection engine directly with the dowel pin cutting machine:

High-Speed AI Detection: Trained on >10,000 images, the system achieved 99.8% accuracy — outperforming both manual inspection and traditional vision systems.
Real-Time Integration: Image data was streamed from the machine camera to Jidoka’s AI, which returned defect decisions in under 50 milliseconds, enabling ejection at 1,200 parts/minute.
Non-Intrusive Deployment: The solution was layered onto the existing machine setup without requiring process changes or extra equipment.