A Bayesian regularization network approach to thermal distortion control in 3D printing

• Main Authors: Xie, Yuxi, Li, Boyuan, Wang, Chao, Zhou, Kun, Wu, C. T., Li, Shaofan
• Other Authors: School of Mechanical and Aerospace Engineering
• Format: Article
• Language: English
• Published: 2023
• Subjects: Engineering::Mechanical engineering; 3D Printing; Bayesian Learning
• Online Access: https://hdl.handle.net/10356/172254
• Institution: Nanyang Technological University

In this work, a Bayesian Regularization Network based Geometric Deviation Control (BRN-GDC) algorithm is developed to mitigate thermal distortion in 3D printing. Inspired by points registration in computer vision and function approximation theory, the Bayesian regularization network method is used to quantify thermal distortion in 3D printed products. Because of “shallow” regularization network architecture, the BRN-GDC method is training-free and does not require lots of data. Due to the lack of one-to-one correspondence between the design point data and the scan point data in 3D printing, conventional point set registration methods, e.g. Coherent Point Drift method, may fail in finding the global geometric deviation field, while the Bayesian regularization network approach works. In the two experiments presented in this paper, we showed that the BRN-GDC algorithm has the capability to control the thermal distortion in 3D printing that is parameter- and location-dependent.