Online monitoring and penetration recognition in all-position TIG welding of nuclear power pipeline

Research question

How can the back-side pool be observed inside a pipe to recognize penetration during all-position TIG welding?

Method

An internal crawler supports follow-up imaging of the groove, back-side pool and solidified bead. SegNet extracts pool boundaries, and a random-forest classifier combines image-derived features with process parameters to identify penetration states.

Main findings

Pipe-welding experiments demonstrate the chain from internal back-side observation to penetration-state classification, with comparisons against decision-tree, SVM and other models. The system makes otherwise difficult-to-view internal weld formation observable.