Advanced clustering technique for automatic labelling of welding signals

In the field of additive manufacturing, effective monitoring is indispensable for ensuring the quality of produced components. This task is usually solved using supervised learning approaches, whom need to deal with manually labelled datasets. In this research we leverage advanced artificial intelligence techniques, such as data-driven features extraction, dimensionality reduction and clustering algorithms, welding current and welding voltage during a Wire Arc Additive Manufacturing deposition have been labelled automatically. Features from both time and frequency domain have been automatically extracted, while Uniform Manifold Approximation and Projection (UMAP) algorithm is employed to reduce the number of features from 889 to 10 using a non-linear projection. Finally, Hierarchical clustering has been employed to generate labels.