Advanced Manufacturing | 论文推荐—用于监测金属增材制造过程的物理信息机器学习综述

A review on physics-informed machine learning for monitoring metal additive manufacturing process

Shoulan Yang, Shitong Peng∗, Jianan Guo, Fengtao Wang∗

Department of Mechanical Engineering, Shantou University, Shantou, 515063, China

Citation:Yang S, Peng S, Guo J, Wang F. A review on physics-informed machine learning for monitoring metal additive manufacturing process. Adv. Manuf. 2024(2):0008, https://doi.org/10.55092/am20240008. 

DOI: 10.55092/am20240008

Abstract: 

The traditional data-driven models and pure physics models have been widely employed in quality prediction for additive manufacturing (AM). However, data-driven models rely on a large amount of labeled data, while pure physics models suffer from lower computational efficiency and accuracy. The Physics-Informed Neural Network (PINN) model has emerged as a hybrid data-driven paradigm that imbues data-driven models with physical domain knowledge. To refrain from the inherent “black box” or inefficiency of AM process prediction or monitoring, this paper discusses the pros and cons of traditional data driven methods and pure physics models and further elaborates on the principles and architecture of the PINN model along with its applications in AM research. We review and analyze current state-of-the-art PINN applications to AM, focusing on temperature field prediction, fluid dynamics issues, fatigue life prediction, accelerated finite element simulation, and process characteristics prediction. The corresponding embedded physical knowledge, either integrated into loss function or data preprocessing, is also summarized for these applications. Based on this review, we identify the challenges of PINN and provide an outlook for further research of its AM applications.

Keywords: Data-driven models; physics models; additive manufacturing; physics-informed neural network

识别二维码，阅读原文

Preview: 

左右滑动查看全文（共28页）