Caggiano, A.A.CaggianoZhang, J.J.ZhangAlfieri, V.V.AlfieriCaiazzo, F.F.CaiazzoGao, R.R.GaoTeti, R.R.Teti2022-03-062022-03-062019https://publica.fraunhofer.de/handle/publica/25948410.1016/j.cirp.2019.03.021A machine learning approach for on-line fault recognition via automatic image processing is developed to timely identify material defects due to process non-conformities in Selective Laser Melting (SLM) of metal powders. In-process images acquired during the layer-by-layer SLM processing are analyzed via a bi-stream Deep Convolutional Neural Network-based model, and the recognition of SLM defective condition-related pattern is achieved by automated image feature learning and feature fusion. Experimental evaluations confirmed the effectiveness of the machine learning method for on-line detection of defects due to process non-conformities, providing the basis for adaptive SLM process control and part quality assurance.en621journal article