文献类型：期刊文献

英文题名：Material-process-performance integrated computational prediction approach for laser powder bed fusion additive manufacturing

作者：Shen, Tao[1,2];Li, Bo[1,2,3]

机构：[1]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Addit Mfg & Intelligent Equipment Res Inst, Shanghai 200237, Peoples R China;[3]Shanghai Collaborat Innovat Ctr High end Equipment, Shanghai 200237, Peoples R China

年份：2025

卷号：155

起止页码：55

外文期刊名：JOURNAL OF MANUFACTURING PROCESSES

收录：;EI(收录号：20254119314760);WOS:【SCI-EXPANDED(收录号:WOS:001596473300001)】；

基金：This research work is sponsored by Shanghai Explorer Program (Grant No. 24TS1411800) , National Natural Science Foundation of China (No. 52175140) , and AECC Industry-University-Research Coop-eration Project (No. HFZL2023CXY024) .

语种：英文

外文关键词：Laser powder bed fusion; Material-process-performance relationship; Physical informed hybrid neural network

摘要：Metal additive manufacturing (AM), particularly laser powder bed fusion (L-PBF), plays a vital role in producing complex metal components with precise microstructure control. However, optimizing AM processes remains challenging due to the complex and not fully understood relationships between material, process, and performance. This study proposes a novel machine learning (ML)-based framework to predict the interactions between AM process parameters, microstructure, and mechanical properties. Using CoCrFeMnNi alloy as a case study, the framework integrates cellular automaton (CA) modeling, finite volume method (FVM) temperature field calculations, and crystal plasticity finite element (CPFE) simulations to examine stress-strain behavior under different microstructural conditions. The framework utilizes a back-propagation neural network (BPNN) for modeling the process-structure link and a hybrid neural network (HNN), enhanced by physical insights (PIHNN), to predict structure-property relationships. This approach accounts for defects affecting mechanical properties, improving the generalizability of the predictive model. The PIHNN model outperforms traditional convolutional neural networks (CNNs) in capturing structure-property interactions. Feature importance analysis highlights the significant role of grain and grain boundary characteristics on mechanical properties, enhancing the model's interpretability. The results demonstrate the framework's effectiveness in addressing the complexity of materialprocess-performance relationships in AM and offer a powerful tool for optimizing AM processes and predicting component properties.