A neural network for identification and classification of systematic internal flaws in laser powder bed fusion

Quality control of mechanical components is crucial to ensure their expected performance and prevent failure. For manufactured additively, quality performed in-process particularly interesting, as the sequential deposition remelting layers represent a possibility mitigate existing flaws. The first step towards closed-loop that monitoring setup data analytics approach can flag discriminate This study aims assess potential layerwise system associated with supervised machine learning identify classify internal flaws in laser powder bed fusion Hastelloy X. that, systematically generated were mapped ex-situ 72 distinct process conditions. outputs near-infrared long-exposure acquisition labeled according characterization used train fully convolutional neural network. network was then previously unseen images into three classes, predominant flaw type expected, lack fusion, keyhole porosity, or residual porosity. Accuracy, precision recall over 96% are obtained, indicating combined this successfully identifies classifies