Additive manufacturing (AM) is a crucial component of smart manufacturing systems that disrupts traditional supply chains. However, the parts built using the state-of-the-art powder-bed 3D printers have noticeable unpredictable mechanical properties. In this paper, we propose a closed-loop machine learning algorithm as a promising way of improving the underlying failure phenomena in 3D metal printing. We employ machine learning approach through a Deep Convolutional Neural Network to automatically detect the defects in printing the layers, thereby turning metal 3D printers into essentially their own inspectors. By comparing three deep learning models, we demonstrate that transfer learning approach based on Inception-v3 model in Tensorflow framework can be used to retrain our images data set consisting of only 200 image samples and achieves a classification accuracy rate of 100 % on the test set. This will generate a precise feedback signal for a smart 3D printer to recognize any issues with the build itself and make proper adjustments and corrections without operator intervention. The closed-loop ML algorithm can enhance the quality of the AM process, leading to manufacturing better parts with fewer quality hiccups, limiting waste of time and materials.