Deep Learning-Based Super-Resolution for the Finite Element Analysis of Additive Manufacturing Process

Abstract

Finite element analysis (FEA) of fused deposition modeling (FDM) has recently been recognized in additive manufacturing (AM) for predictions in temperature gradient of three-dimensions (3D) printed components. These predictions can be invaluable for making corrections to the printing process to improve quality of printed components. However, FEA has its limitations. For example, models with fine mesh (small element size) yield more accurate results than ones with coarse mesh (large element size). Comparing with the coarse mesh model, a fine mesh model can take considerably longer computational times and discourages most manufacturers from using FEA. In this work, an innovative deep-learning (DL) based super-resolution approach is used to improve the result accuracy of a coarse mesh model to the higher accuracy level of a fine mesh model and reduce the computational time. The element in the FEA was treated as the physical pixel in an image, so the fine temperature grid and coarse temperature grid in the FEA were analogous to high resolution (HR) images and low resolution (LR) images, respectively. The result shows that the difference value HS between HR image and super resolution (SR) image is much smaller than the one HL between HR image and LR image, which demonstrated that our proposed DL-based super-resolution approach was effective to enhance the result accuracy of the coarse mesh model. Besides, both the increased Peak Signal-to-Nosie Ratio (PSNR) value and Structural Similarity Index (SSIM) value indicated that the quality of the images was also improved through the super-resolution approach.