基于BP神经网络的激光熔覆参数多目标优化  

Multi-Objective Optimization of Laser Cladding Parameters Based on BP Neural Network

作　　者：邓德伟[1,3] 江浩 李振华 宋学官 孙奇[3] 张勇[3] Deng Dewei;Jiang Hao;Li Zhenhua;Song Xueguan;Sun Qi;Zhang Yong(Research Center of Laser 3D Printing Equipment and Application Engineering Technology(Liaoning Province),School of Materials Science and Engineering,Dalian University of Technology,Dalian 116024,Liaoning,China;School of Mechanical Engineering,Dalian University of Technology,Dalian 116024,Liaoning,China;Shenyang Blower Group Corporation,Shenyang 110869,Liaoning,China)

机构地区：[1]大连理工大学材料科学与工程学院辽宁省激光3D打印装备及应用工程技术研究中心,辽宁大连116024 [2]大连理工大学机械工程学院,辽宁大连116024 [3]沈阳鼓风机集团股份有限公司,辽宁沈阳110869

出　　处：《激光与光电子学进展》2023年第17期222-235,共14页Laser & Optoelectronics Progress

基　　金：辽宁重大装备制造协同创新中心基金(DUT2017031);高端控制阀产业技术协同创新中心基金(2018WZ003)。

摘　　要：为了获得TiC铁基合金粉末在316L不锈钢上的激光熔覆最佳工艺参数,提出了一种基于遗传算法优化的反向传播(BP)神经网络的激光熔覆参数优化方法。设计三因素五水平的全因子试验,测量了熔覆层的宏观形貌和平均硬度,建立输入参数(激光功率、扫描速度、保护气流量)和响应量(熔覆层宽度、熔覆层高度、稀释率、显微硬度)的神经网络模型。以多元非线性回归分析工艺参数对响应量的影响,并以综合灰关联度表征熔覆层的综合性能,寻优得到最佳参数。试验结果表明,激光功率和扫描速度对熔覆层宽度、稀释率和显微硬度的影响明显,而保护气流量对熔覆层高度影响最显著,遗传算法优化的BP神经网络模型各响应量模型的拟合优度均达到0.85~0.91之间,GA-BP模型精度良好,当参数为1090 W,扫描速度为4.4 mm/s,保护气流量为10 L·min^(−1),综合性能最佳,表明BP神经网络算法适用于激光熔覆层质量控制和参数优化。In order to obtain the optimal process parameters for laser melting of TiC iron-based alloy powder on 316L stainless steel,a back propagation(BP)neural network based on genetic algorithm optimization for laser melting parameters optimization is proposed.A three-factor,five-level full factorial experiment was designed to measure the macroscopic morphology and average hardness of the melted layer,and a neural network model was established for the input parameters(laser power,scanning speed,and protective gas flow rate)and response quantities(melted layer width,melted layer height,dilution rate,and microhardness).The effect of the process parameters on the response quantity was analyzed by multiple non-linear regression,and the overall performance of the clad layer was characterized by the integrated gray correlation,and the optimal parameters were obtained.The experimental results show that the laser power and scanning speed have obvious effects on the width of the molten layer,dilution rate and microhardness,while the protective gas flow rate has the most significant effect on the height of the molten layer.The goodness of fit of each response quantity model of the BP neural network model optimized by the genetic algorithm reaches between 0.85 and 0.91,and the GA-BP model has good accuracy.The best overall performance was achieved when the parameter was 1090 W,the scanning speed was 4.4 mm/s,and the protective gas flow rate was 10 L/min,indicating that the BP neural network algorithm was suitable for the quality control and parameter optimization of the laser cladding layer.

关 键 词：激光熔覆 反向传播神经网络 遗传算法 灰关联度 参数优化 

分 类 号：TN249[电子电信—物理电子学]