About this Abstract |
| Meeting |
11th Conference on Trends in Welding Research + Additive Manufacturing (TWR+AM)
|
| Symposium
|
11th Conference on Trends in Welding Research + Additive Manufacturing (TWR+AM)
|
| Presentation Title |
A Deep Learning Framework for Quantitative Radiographic Inspection of Foil-Tab Laser Welds |
| Author(s) |
Peijun Hou, Wei Zhang, Whitney Poling, Hassan Ghassemi-Armaki, Dmitriy Bruder, Blair E. Carlson, Zhili Feng |
| On-Site Speaker (Planned) |
Peijun Hou |
| Abstract Scope |
Digital radiography (DR) inspection is widely used to assess the quality of welded joints, while traditional operator-dependent visual checks often lack the speed and repeatability required for high-precision manufacturing. To address this, we developed a unified deep learning framework that transforms 2D DR into a high-precision, quantitative diagnostic system. By utilizing cross-modal supervision, the framework leverages 3D X-ray computed tomography reconstructions to provide high-fidelity ground-truth annotations for 2D features. A multi-decoder head architecture was designed to simultaneously detect, classify, and quantify complex laser welding defects, such as pores, cracks, and material loss. The model was trained on a heterogeneous dataset of aluminum and copper battery foil-tab welds to ensure robust generalization across varying thicknesses and morphologies. Validated against numerical benchmarks and engineering assessments, the framework provides a highly repeatable and accurate solution for defect characterization, significantly enhancing quality assurance for industrial laser-welded components. |
| Proceedings Inclusion? |
Undecided |