Toward online layer-wise surface morphology measurement in additive manufacturing using a deep learning-based approach

Layer-wise surface morphology information plays a critical role in the quality monitoring and control of additive manufacturing (AM) processes. 3D scan technologies can provide effective means to obtain accurate surface morphological data. However, most of the existing 3D scan technologies are time consuming due to either contact mode or algorithm complexity, which are not capable of obtaining the surface morphology data in an online manner during the printing process. To implement online layer-wise surface morphological data acquisition in AM processes, one practical solution is to model the correlation between 2D images and 3D point cloud data. In practice, since this correlation is usually highly complex due to the high dimensionality and non-linearity, it is usually impractical to find an explicit mathematical transfer function to quantify this correlation effectively. To address this issue, a deep learning-based model is developed in this study, in which a powerful deep learning algorithm, namely, convolutional neural network (CNN), is incorporated. With the trained CNN model, the 3D surface data can be predicted directly without the relatively time consuming triangulation computation by the 3D scanner. Thus, the speed of surface data acquisition and morphology measurement can be improved. To validate the effectiveness and efficiency of the proposed methodology, both simulation and real-world AM case studies were performed. The results show that the prediction accuracy using the proposed method is promising. In terms of averaged relative prediction error, it can be mostly lower than 10% in the experiments. Therefore, the proposed method has a great potential for online layer-wise surface morphology measurement in AM.