基于深度学习的激光选区熔化残余应力预测与调控

褚昭洁; 王瀚绪; 王俊杰; 惠记庄; 阎志强

doi:10.3969/j.issn.1674-6457.2026.06.013

PDF(18608 KB)

精密成形工程 ›› 2026, Vol. 18 ›› Issue (6) : 135-150. DOI: 10.3969/j.issn.1674-6457.2026.06.013

增材制造

基于深度学习的激光选区熔化残余应力预测与调控

褚昭洁¹, 王瀚绪¹, 王俊杰^1,2, 惠记庄¹, 阎志强^1,*

作者信息 +

Prediction and Control of Residual Stress in Laser Selective Melting Based on Deep Learning

CHU Zhaojie¹, WANG Hanxu¹, WANG Junjie^1,2, HUI Jizhuang¹, YAN Zhiqiang^1,*

Author information +

文章历史 +

摘要

目的将深度学习与数值模拟相结合建立选区激光熔化（SLM）过程中的残余应力预测与调控模型。通过精确预测不同工艺参数和零件几何结构下的残余应力分布,进而实现对残余应力的主动调控。方法提出深度学习与数值模拟相结合的方法,采用Adam算法优化的卷积神经网络（Convolutional Neural Networks,CNN）模型与AlSi10Mg粉末床的数值有限元分析相结合的方法,实现残余应力的预测与逆向调控。设计正交实验,获取不同工艺参数下残余应力的分布数据,用于训练深度网络模型并进行模型验证,同时开发了预测残余应力的专用软件。结果在预测x、y两个方向的残余应力时,各项回归评价指标均显示出优异的预测精度,决定系数（R²）均大于0.99;尤为值得注意的是,x方向残余应力预测的平均绝对百分比误差（MAPE）低于0.5%。在工艺参数优化方面,针对不同目标函数得到了相应的最优工艺组合：当以最小化最大残余应力为目标时,最优参数为激光功率360 W、扫描速度695 mm/s;当以提升残余应力分布均匀性为目标时,最佳参数为激光功率361.21 W、扫描速度495 mm/s。结论基于CNN模型建立的平板打印残余应力预测系统有优异的预测精度与泛化能力,可指导实践保证加工质量。

Abstract

The work aims to combine deep learning and numerical simulation to establish a residual stress prediction and control model in the selective laser melting (SLM) process, to achieve active regulation of residual stress by precisely predicting the distribution of residual stress under different process parameters and part geometric structures. A method combining deep learning with numerical simulation was proposed. A convolutional neural network (CNN) model optimized with the Adam algorithm was integrated with finite element numerical analysis of the AlSi10Mg powder bed to achieve prediction and inverse control of residual stress. An orthogonal experimental design was used to obtain residual stress distribution data under different process parameters, which served to train and validate the deep learning model. Additionally, specialized software for predicting residual stress was developed. When predicting residual stress in the x and y directions, all regression evaluation metrics demonstrated excellent prediction accuracy, with coefficients of determination (R²) exceeding 0.99. Notably, the mean absolute percentage error (MAPE) for residual stress prediction in the x direction was below 0.5%. Regarding process parameter optimization, optimal parameter sets were identified for different objective functions: to minimize the maximum residual stress, the optimal parameters were a laser power of 360 W and a scanning speed of 695 mm/s; to improve the uniformity of residual stress distribution, the best parameters were a laser power of 361.21 W and a scanning speed of 495 mm/s. The flatbed printing residual stress prediction system based on the CNN model demonstrates excellent prediction accuracy and generalization ability, which can guide practical applications to ensure processing quality.

导出引用

褚昭洁, 王瀚绪, 王俊杰, 惠记庄, 阎志强. 基于深度学习的激光选区熔化残余应力预测与调控[J]. 精密成形工程. 2026, 18(6): 135-150 https://doi.org/10.3969/j.issn.1674-6457.2026.06.013

CHU Zhaojie, WANG Hanxu, WANG Junjie, HUI Jizhuang, YAN Zhiqiang. Prediction and Control of Residual Stress in Laser Selective Melting Based on Deep Learning[J]. Journal of Netshape Forming Engineering. 2026, 18(6): 135-150 https://doi.org/10.3969/j.issn.1674-6457.2026.06.013

中图分类号： TG665

参考文献

[1] 周金宇, 陈逸飞. 基于多尺度模拟的选区激光熔化金属件疲劳性能预测[J]. 中国机械工程, 2025, 36(9): 2087-2096.
ZHOU J Y, CHEN Y F.Prediction of Fatigue Property of SLM Metal Parts Based on Multi-Scale Simulations[J]. China Mechanical Engineering, 2025, 36(9): 2087-2096.
[2] YAN Z Q, CHU Z J, HUI J Z, et al.Prediction, Assessment and Optimization of Energy Consumption in Selective Laser Melting Using Hybrid Mechanism-Based and Data-Driven Approaches[J]. Measurement, 2026, 257: 118895.
[3] 王文泉, 赵新甜, 于文慧, 等. 激光冲击强化对增材制造镁合金表面完整性的影响[J]. 表面技术, 2025, 54(11): 173-183.
WANG W Q, ZHAO X T, YU W H, et al.Effect of Laser Shock Peening on Surface Integrity of Additively Manufactured Mg Alloy[J]. Surface Technology, 2025, 54(11): 173-183.
[4] 张长春, 张军利, 倪允强, 等. 多光束激光选区熔化AlSi₁₀Mg合金工艺、组织及性能[J]. 应用激光, 2025, 45(8): 30-47.
ZHANG C C, ZHANG J L, NI Y Q, et al.Process, Microstructure and Mechanical Properties of AlSi₁₀Mg Alloy Fabricated by Multi-Beam Selective Laser Melting[J]. Applied Laser, 2025, 45(8): 30-47.
[5] CHEN H Q, LIU P, REN X C, et al.Fatigue and Corrosion Fatigue Performance of Selective Laser Melted AlSi10Mg and Die Cast A360 Aluminum Alloys[J]. Corrosion Science, 2025, 245: 112711.
[6] KANG C, KIM G H, KIM W R, et al.Effect of Hot Isostatic Pressing Temperature on Microstructures and Characteristics of AlSi10Mg Alloy Fabricated by Selective Laser Melting[J]. Journal of Materials Research and Technology, 2025, 37: 1443-1449.
[7] ZHANG X, CHEN S S, WANG Z Y, et al.Effect of Cooling Rates and Heat Treatment Time on Si Phase of AlSi10Mg Alloy Manufactured by Selective Laser Melting: Microstructure Evolution and Strengthening Mechanism[J]. Journal of Alloys and Compounds, 2024, 1007: 176494.
[8] 惠记庄, 骆伟, 阎志强, 等. AlSi10Mg选区激光熔化表面粗糙度预测、优化及表面形貌分析[J]. 表面技术, 2024, 53(15): 129-140.
HUI J Z, LUO W, YAN Z Q, et al.Surface Roughness Prediction, Optimization and Surface Morphology Analysis of AlSi10Mg by Selective Laser Melting[J]. Surface Technology, 2024, 53(15): 129-140.
[9] GUPTA A, BENNETT C J, SUN W.An Experimental Investigation on the Progressive Failure of an Additively Manufactured Laser Powder Bed Fusion Ti-6Al-4V Aero-Engine Bracket under Low Cycle Fatigue[J]. Engineering Failure Analysis, 2022, 139: 106455.
[10] GUPTA A, BENNETT C J, SUN W.Low Cycle Fatigue Performance of SLM Ti-6Al-4V Aero-Engine Bracket at 200 ℃: An Experimental Study[J]. Procedia Structural Integrity, 2023, 46: 35-41.
[11] WU T, LI C, SUN F, et al.Reduction in Residual Stress and Distortion of Thin-Walled Inconel 718 Specimens Fabricated by Selective Laser Melting: Experiment and Numerical Simulation[J]. International Journal of Pressure Vessels and Piping, 2024, 212: 105292.
[12] REN K, CHEW Y, FUH J Y H, et al. Thermo-Mechanical Analyses for Optimized Path Planning in Laser Aided Additive Manufacturing Processes[J]. Materials & Design, 2019, 162: 80-93.
[13] SALEM M, LE ROUX S, HOR A, et al.A New Insight on the Analysis of Residual Stresses Related Distortions in Selective Laser Melting of Ti-6Al-4V Using the Improved Bridge Curvature Method[J]. Additive Manufacturing, 2020, 36: 101586.
[14] YAN Z Q, CHU Z J, XU Z G, et al.The Influence of Size-Dependence Effect on Residual Stress of Annular Component in Selective Laser Melting via Numerical Modelling and Experiments[J]. Journal of Materials Research and Technology, 2025, 38: 242-261.
[15] CHEN Y, LIU Y, CHEN H, et al.Multi-Scale Residual Stress Prediction for Selective Laser Melting of High Strength Steel Considering Solid-State Phase Transformation[J]. Optics & Laser Technology, 2022, 146: 107578.
[16] FANG Z C, WU Z L, ZHAO L, et al.Effects of Thermal Cycling on Residual Stress in Alloy Parts via Selective Laser Melting[J]. Optics and Lasers in Engineering, 2024, 180: 108277.
[17] CHEN S G, ZHANG Y D, WU Q, et al.Effect of Solid-State Phase Transformation on Residual Stress of Selective Laser Melting Ti6Al4V[J]. Materials Science and Engineering: A, 2021, 819: 141299.
[18] QIN Y L, MA C W, MEI L, et al.The Prediction of Residual Stress of Welding Process Based on Deep Neural Network[J]. Materials Today Communications, 2024, 39: 108595.
[19] WOO M, KI H.Deep Learning-Based Prediction of Thermal Residual Stress and Melt Pool Characteristics in Laser-Irradiated Carbon Steel[J]. International Communications in Heat and Mass Transfer, 2024, 155: 107536.
[20] NGUYEN D S, PARK H S, LEE C M.Optimization of Selective Laser Melting Process Parameters for Ti-6Al-4V Alloy Manufacturing Using Deep Learning[J]. Journal of Manufacturing Processes, 2020, 55: 230-235.
[21] XING W, LYU T Y, CHU X, et al.Recognition and Classification of Single Melt Tracks Using Deep Neural Network: A Fast and Effective Method to Determine Process Windows in Selective Laser Melting[J]. Journal of Manufacturing Processes, 2021, 68: 1746-1757.
[22] 桑卓越, 王帅, 尹万兵, 等. 不同热处理工艺对新型Ti421合金组织及性能的影响[J]. 精密成形工程, 2025, 17(9): 70-78.
SANG Z Y, WANG S, YIN W B, et al.Effect of Different Heat Treatments on the Microstructure and Mechanical Properties of a Novel Ti421 Alloy[J]. Journal of Netshape Forming Engineering, 2025, 17(9): 70-78.
[23] 岳峰丽, 郭威, 刘明华, 等. 基于机器学习算法的超声波预测TP2铜材晶粒度模型优化研究[J]. 精密成形工程, 2025, 17(9): 185-194.
YUE F L, GUO W, LIU M H, et al.Optimization of Ultrasonic Prediction of TP2 Copper Grain Size Model Based on Machine Learning Algorithm[J]. Journal of Netshape Forming Engineering, 2025, 17(9): 185-194.
[24] EREN O, YÜKSEL N, BÖRKLÜ H R, et al. Deep Learning-Enabled Design for Tailored Mechanical Properties of SLM-Manufactured Metallic Lattice Structures[J]. Engineering Applications of Artificial Intelligence, 2024, 130: 107685.
[25] WANG R X, CHEUNG C F, WANG C J, et al.Deep Learning Characterization of Surface Defects in the Selective Laser Melting Process[J]. Computers in Industry, 2022, 140: 103662.
[26] ZHENG F L, PENG K, ZHU Y Y, et al.Intelligent Monitoring of Porosity in Laser Melting Deposition Based on Deep Transfer Learning[J]. Additive Manufacturing, 2025, 109: 104905.
[27] OGOKE F, FARIMANI A B.Thermal Control of Laser Powder Bed Fusion Using Deep Reinforcement Learning[J]. Additive Manufacturing, 2021, 46: 102033.
[28] CARTER F M, PORTER C, KOZJEK D, et al.Machine Learning Guided Adaptive Laser Power Control in Selective Laser Melting for Pore Reduction[J]. CIRP Annals, 2024, 73(1): 149-152.
[29] LIU H S, GE J G, ZHANG Y H, et al.Study of Microstructural Evolution and Mechanical Performance of Selective Laser Melting Pure Tungsten Deposit for Process Parameter Optimization[J]. Journal of Materials Research and Technology, 2025, 38: 3909-3925.
[30] YE D S, HSI FUH J Y, ZHANG Y J, et al. In Situ Monitoring of Selective Laser Melting Using Plume and Spatter Signatures by Deep Belief Networks[J]. ISA Transactions, 2018, 81: 96-104.