基于多层感知器模型的激光粉末床熔融高熵合金孔隙率与工艺区间预测  

Prediction of porosity and process window for high-entropy alloy fabricated by laser powder bed fusion based on multilayer perceptron model

作　　者：乐嘉顺 雷瑛 于博 白凌龙 石家铭 韦辉亮 Yue Jiashun;Lei Ying;Yu Bo;Bai Linglong;Shi Jiaming;Wei Huiliang(School of Mechanical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学机械工程学院,江苏南京210094

出　　处：《南京理工大学学报》2025年第1期92-103,共12页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(52175330);中央高校基本科研业务费专项资金(30921011202)。

摘　　要：为解决激光粉末床熔融(LPBF)增材制造成形孔隙缺陷的定量预测难题,该文融合了LPBF实验、机理模型和深度学习模型数据,实现了LPBF过程中未熔合孔隙率和工艺区间的预测。LPBF实验结果提供基础数据,机理模型用于增强数据集,深度学习多层感知器(MLP)模型则根据实验和机理模型构建的数据集预测未熔合孔隙率和工艺区间。研究发现,深度学习MLP模型的预测准确率与构成数据集的数据量密切相关,通过机理模型增强数据集后MLP模型的预测精度显著提高,最优MLP模型预测独立测试集孔隙率的平均准确率为93.55%。此外,对比分析发现数据融合后预测的工艺区间最小误差为2.62%,最大误差为6.90%,仅使用实验数据则无法准确预测工艺区间。该文结果表明,MLP可用于LPBF高熵合金孔隙率与工艺参数复杂关系预测。To address the challenging issue of quantitatively predicting porosity defects in laser powder bed fusion(LPBF)additive manufacturing,this study integrates LPBF experiments,a mechanistic model and deep learning model data,achieving the prediction of lack-of-fusion porosity and process window during LPBF.The LPBF experimental results provide foundational data,the mechanistic model augments the dataset,and a multilayer perceptron(MLP)deep learning model predicts lack-of-fusion porosity and process window based on the dataset constructed by using the experiments and the mechanistic model.The study finds that the predictive accuracy of the MLP model is closely related to the quantity of data comprising the dataset.A significant improvement in prediction accuracy on the independent test dataset is observed after enhancing the dataset by a mechanistic model.This suggests that the average accuracy of predicting porosity on the independent test dataset using optimal MLP model is 93.55%.Additionally,comparative analysis reveals that the maximum error of the predicted process window diagram formed by data fusion is 6.90%,while the minimum error is 2.62%,indicating that using experimental data alone cannot accurately predict the process window.The research results demonstrate that prediction of complex correlations between technological parameters and porosity using MLP are feasible in high-entropy alloy fabricated by LPBF.

关 键 词：增材制造 深度学习 高熵合金 孔隙率 工艺区间 

分 类 号：TN249[电子电信—物理电子学] TF124[冶金工程—粉末冶金] TG132.3[冶金工程—冶金物理化学] TP391[一般工业技术—材料科学与工程]