A cognitive approach for quality assessment in laser welding

Quality assessment in laser welding is of outmost importance. A plethora of in-line inspection techniques have been developed identifying melt pool geometry and weld defects for quality evaluation. This paper aims to introduce a cognitive assessment method for the prediction of weld quality and classification into different quality categories. The study corresponds to camera-based monitoring approaches utilizing thermal images obtained from process simulation models where artificial defects were inserted. A dimensionality reduction technique is deployed, and an image processing technique is afterwards implemented to identify weld defects based on specific melt pool features. A classification algorithm has also been developed and validated.