A generalized Bayesian regularization network approach on characterization of geometric defects in lattice structures for topology optimization in preliminary design of 3D printing

摘要

In this work, we developed a Generalized Bayesian Regularization Network (GBRN) approach that can quantitatively identify the defect shapes and locations by mapping the distorted lattice structure to its original designed configuration, making registration between manufactured parts with defects and the perfect design models in the preliminary design stage of 3D printing.. On the one hand, it shows the proposed GBRN method has quantitatively comparable accuracy to the Coherent Point Drift (CPD) method in 2D boundary points registration problems. On the other hand, we have shown that the proposed GBRN method can find the possible geometric defects in the 3D printed lattice structure model and identify inherent defect-prone lattice structure parameters with obvious advantages over those two-dimensional point registration methods, i.e., coherent point drift (CPD) method, in registration of interior points of 3D lattice structures.