Industrial AI Podcast AI and spot welding inspection: Reinventing Quality Control
Oct 15, 2025
In this engaging discussion, Christian Wagner, Managing Director at Vogt Ultrasonics, and Jan Kutnik, Founder of Evoptima, delve into how AI is transforming spot weld inspection in the automotive industry. They tackle the challenges of massive ultrasonic datasets and share how AI enhances prediction accuracy by 20-30%. The duo explores the integration of AI with next-gen inspection devices and the role of collaborative robots. They also discuss the value of robust partnerships and courageous innovation for mid-sized companies navigating AI adoption.
AI Snips
Chapters
Transcript
Episode notes
Data Volume Enables AI Transformation
- Vogt saw vast phased-array ultrasonic data (729 signals per spot) as an opportunity to apply AI instead of incremental software tweaks.
- Christian launched a full software rewrite to embed AI as a core capability and improve UX and measurement precision.
Predict Nugget Size From Raw Signals
- Evoptima trained models to predict nugget diameter directly from raw ultrasonic signals, skipping hand-engineered signal processing.
- They used destructive tear-down measurements as ground truth to learn the mapping from signals to nugget size.
Robustness Means Reasonable Outputs Everywhere
- Model robustness means producing reasonable nugget-size estimates across manufacturing variations and poor probe coupling.
- The team focuses on covering measurement distributions so outputs stay stable even on noisy or bad data.