Discover HeatCore AI

AI Welding Demo — Weld Pool and Arc Detected Frame by Frame

A short clip of the AI at work: in every frame of a live weld it finds the weld pool and the arc, and shows how sure it is.

AI welding demo: weld pool and arc detection

Who This Demo Is For

Quality managers, welding engineers, and production supervisors evaluating AI weld monitoring for their MIG/MAG, TIG, or robotic welding lines. This 9-second clip shows the first thing the AI does on every frame: find the weld pool and the arc.

What You'll See

  • 0:00–0:02 — The arc strikes; the model boxes the arc in red and a small weld pool in green, with a low score at first
  • 0:02–0:09 — As the pool grows, the green mask follows its shape and the pool score settles around 0.9 while the red box stays on the arc

Key Takeaways

  • The model tells the weld pool and the arc apart in every frame
  • Each detection carries a confidence score, so weak frames are visible
  • Pool size and position over time become signals the monitoring can track
  • Train the AI using your own "good weld" references

From Baseline to Live Verdict

Finding the pool and the arc is the first step of the monitoring chain, and the steps after it rest on a single idea: a sound weld heats and cools in a characteristic way, so a defect is a departure from that learned pattern. During training the model watches known-good joints and fixes the normal temperature field — the shape of the trailing thermal wake, the rate the bead sheds heat, the symmetry across the joint. In live monitoring, a pore or a trapped void changes local conduction and appears as a hot or cold anomaly against that baseline, while incomplete fusion alters how heat crosses the interface. Because the reference is your own process rather than a generic threshold, the same rig adapts across MIG/MAG, TIG, and robotic lines without re-tuning fixed limits. Every flagged frame is written to the record as the kind of documented thermographic NDT evidence that ISO 17635 and ISO 3834 workflows expect.

Related Resources

Related Videos

Transcript (short)

Show transcript

A live weld seen through the camera, with the AI detections drawn on every frame: a red box with its score on the arc, and a green box and mask with its score on the weld pool. The pool score starts low as the arc strikes and settles around 0.9 once the pool has formed.

Progetto cofinanziato nell'ambito del PR Piemonte FSE+ 2021-2027,
Priorità I, Obiettivo Specifico a), Azione 4 – "Sostegno alla nascita delle start up"