基于神经网络的选区激光熔化残余应力预测

景艳龙; 李杰; 石文天; 闫晓玲

doi:10.11884/HPLPB202133.210223

基于神经网络的选区激光熔化残余应力预测

doi: 10.11884/HPLPB202133.210223

北京工商大学人工智能学院, 北京100048

基金项目: 国家自然科学基金项目（51975006）

详细信息

作者简介:
景艳龙，jingyanlong@btbu.edu.cn

通讯作者:
李　杰，lijie0739@btbu.edu.cn

中图分类号: TN249
计量
- 文章访问数: 887
- HTML全文浏览量: 325
- PDF下载量: 69
- 被引次数: 0
出版历程
- 收稿日期: 2021-06-04
- 修回日期: 2021-09-15
- 网络出版日期: 2021-10-19
- 刊出日期: 2021-10-15

Prediction of residual stress in selective laser melting based on neural network

School of Artificial Intelligence, Beijing Technology and Business University, Beijing 100048, China

摘要

摘要: 当前对选区激光熔化产生的残余应力预测方法主要为数值模拟，但由于设备、环境、粉末等因素差异性较大，且具有较大不确定性，很难建立符合实际情况的数值模拟模型。利用神经网络在预测多变量、复杂线性信息处理方面能力强的特点，建立适用于预测316L不锈钢粉末选区激光熔化残余应力的模型。使用选区激光熔化技术打印相当数量的不同工艺参数的试样，采用超声波检测其内部残余应力作为神经网络的训练样本，并使用这些样本对神经网络模型进行训练，获得具有预测功能的神经网络，将验证样本的工艺参数输入神经网络，计算出预测的残余应力值，与实际检测值进行对比。实验结果表明，预测值与实际测量值偏差较小，验证了所提方法的有效性。采用神经网络预测残余应力的方法，可以快速确定不同选区激光熔化工艺参数对应的残余应力，避免设置残余应力较高的工艺参数，有效缩短制备高质量工件试样的周期，降低成本。
- 增材制造 /
- 残余应力 /
- 神经网络 /
- 选区激光熔化 /
- 超声波检测 /
- 自适应
Abstract: At present, numerical simulation is the main method for predicting residual stress produced by selective laser melting. However, due to different factors such as equipment, environment and powder, it is difficult to establish a practical numerical model. The accuracy of numerical simulation needs to be verified. Based on the strong ability of neural network in predicting multivariable and complex linear information processing, a model for predicting residual stress in selective laser melting (SLM) is established. SLM technology is used to print a considerable number of samples with different process parameters, and ultrasonic wave is used to detect the internal residual stress. Training samples of neural network are used to train the neural network model, and the neural network with predictive function is obtained. The process parameters of the validated samples are input into the neural network, and the predicted residual stress value is calculated, which is in accordance with the actual situation. The detection values were compared. The experimental results show that the deviation between the predicted value and the measured value is small and the accuracy is high, which verifies the effectiveness of the proposed method.The method of predicting residual stress by neural network can quickly determine the residual stress corresponding to laser melting process parameters in different selection areas, avoid setting process parameters with high residual stress, effectively shorten the cycle of preparing high-quality workpiece samples and reduce the cost.
- additive manufacturing /
- residual stress /
- neural network /
- selective laser melting /
- ultrasonic detection /
- self-adaptivity

HTML全文

图 1 LCR波传播示意图

Figure 1. Schematic diagram of LCR wave propagation

下载: 全尺寸图片幻灯片

图 2 SLM神经网络模型

Figure 2. SLM neural network model

下载: 全尺寸图片幻灯片

图 3 试样尺寸图

Figure 3. Sample size

下载: 全尺寸图片幻灯片

图 4 试样实物照片

Figure 4. Photographs of samples

下载: 全尺寸图片幻灯片

图 5 扫描电镜下的微观组织图

Figure 5. Microstructure images by scanning electron microscope

下载: 全尺寸图片幻灯片

图 6 声时差-加载应力曲线

Figure 6. Graph of acoustic time difference-loading stress

下载: 全尺寸图片幻灯片

图 7 功率-残余应力曲线

Figure 7. Graph of power-residual stress

下载: 全尺寸图片幻灯片

图 8 扫描速度-残余应力曲线

Figure 8. Graphs of scanning speed-residual stress

下载: 全尺寸图片幻灯片

图 9 训练样本预测值与实测值

Figure 9. Predicted value and measured value of training samples

下载: 全尺寸图片幻灯片

图 10 验证样本预测值与实测值

Figure 10. Predicted value and measured value of verifying samples

下载: 全尺寸图片幻灯片

表 1 316L不锈钢粉末化学成分比重

Table 1. Chemical composition of 316L stainless steel by weight

element	content/%	element	content/%
Fe	balance	N	0.1max
Cr	16～18	O	0.1max
Ni	10～14	P	0.045max
Mo	2～3	C	0.03max
Mn	2max	S	0.03max
Si	1max

下载: 导出CSV

表 2 训练样本SLM工艺参数

Table 2. Training process parameters of experiments in SLM

process parameter	laser power/W	scanning speed/(m·s⁻¹)	powder thickness/μm	preheating temperature/℃
level 1	50	0.25	50	50
level 2	100	0.375	80	100
level 3	150	0.5	−	−
level 4	200	0.625	−	−
level 5	250	0.75	−	−
level 6	300	0.875	−	−
level 7	350	1	−	−
level 8	400	1.125	−	−
level 9	−	1.25	−	−
level 10	−	1.5	−	−

下载: 导出CSV

参考文献(21)

[1]	Sing S L, An Jia, Yeong W Y, et al. Laser and electron-beam powder-bed additive manufacturing of metallic implants: a review on processes, materials and designs[J]. Journal of Orthopaedic Research, 2016, 34(3): 369-385. doi: 10.1002/jor.23075
[2]	王晨光, 沈显峰, 王国伟, 等. 金属面曝光选区激光熔化原理装置及试验研究[J]. 强激光与粒子束, 2021, 33:029001. (Wang Chenguang, Shen Xianfeng, Wang Guowei, et al. Principle device and experimental research of surface exposure selective laser melting for metal powder[J]. High Power Laser and Particle Beams, 2021, 33: 029001 doi: 10.11884/HPLPB202133.200221
[3]	龚丞, 王丽芳, 朱刚贤, 等. 激光增材制造工艺参量对熔覆层残余应力的影响[J]. 激光技术, 2019, 43(2):263-268. (Gong Chen, Wang Lifang, Zhu Gangxian, et al. Influence of process parameters on the residual stress of cladding layers by laser additive manufacturing[J]. Laser Technology, 2019, 43(2): 263-268 doi: 10.7510/jgjs.issn.1001-3806.2019.02.021
[4]	杨健, 陈静, 杨海欧, 等. 激光快速成形过程中残余应力分布的实验研究[J]. 稀有金属材料与工程, 2004, 33(12):1304-1307. (Yang Jian, Chen Jing, Yang Haiou, et al. Experimental study on residual stress distribution of laser rapid forming process[J]. Rare Metal Materials and Engineering, 2004, 33(12): 1304-1307 doi: 10.3321/j.issn:1002-185X.2004.12.017
[5]	Yan Xiaoling, Pang Jincheng, Jing Yanlong. Ultrasonic measurement of stress in SLM 316L stainless steel forming parts manufactured using different scanning strategies[J]. Materials, 2019, 12: 2719. doi: 10.3390/ma12172719
[6]	李九霄, 李鸣佩, 杨东野, 等. 选择性激光熔化制造金属构件残余应力的研究进展[J]. 机械工程材料, 2018, 42(8):1-6. (Li Jiuxiao, Li Mingpei, Yang Dongye, et al. Research progress on residual stress in metal component manufactured by selective laser melting[J]. Materials for Mechanical Engineering, 2018, 42(8): 1-6 doi: 10.11973/jxgccl201808001
[7]	杜畅, 张津, 连勇, 等. 激光增材制造残余应力研究现状[J]. 表面技术, 2019, 48(1):200-207. (Du Chang, Zhang Jin, Lian Yong, et al. Research progress on residual stress in laser additive manufacturing[J]. Surface Technology, 2019, 48(1): 200-207
[8]	梁祖磊, 孙中刚, 张少驰, 等. 数值模拟在激光选区熔化中的应用及研究现状[J]. 航空制造技术, 2018, 61(22):87-91,97. (Liang Zulei, Sun Zhonggang, Zhang Shaochi, et al. Application and research status of numerical simulation in laser selective melting[J]. Aeronautical Manufacturing Technology, 2018, 61(22): 87-91,97
[9]	倪立斌, 刘继常, 伍耀庭, 等. 基于神经网络和粒子群算法的激光熔覆工艺优化[J]. 中国激光, 2011, 38:0203003. (Ni Libin, Liu Jichang, Wu Yaoting, et al. Optimization of laser cladding process variables based on neural network and particle swarm optimization algorithms[J]. Chinese Journal of Lasers, 2011, 38: 0203003 doi: 10.3788/CJL201138.0203003
[10]	Akbari M, Saedodin S, Panjehpour A, et al. Numerical simulation and designing artificial neural network for estimating melt pool geometry and temperature distribution in laser welding of Ti6Al4V alloy[J]. Optik, 2016, 127(23): 11161-11172. doi: 10.1016/j.ijleo.2016.09.042
[11]	Li Kun, Yan Shilin, Pan Wenfeng, et al. Warpage optimization of fiber-reinforced composite injection molding by combining back propagation neural network and genetic algorithm[J]. The International Journal of Advanced Manufacturing Technology, 2017, 90(1/4): 963-970.
[12]	雷凯云, 秦训鹏, 刘华明, 等. 基于神经网络的宽带激光熔覆熔池特征参数预测[J]. 光电子·激光, 2018, 29(11):1212-1220. (Lei Kaiyun, Qin Xunpeng, Liu Huaming, et al. Prediction on characteristics of molten pool in wide-band laser cladding based on neural network[J]. Journal of Optoelectronics·Laser, 2018, 29(11): 1212-1220
[13]	姜淑娟, 刘伟军, 南亮亮. 基于神经网络的激光熔覆高度预测[J]. 机械工程学报, 2009, 45(3):269-274,281. (Jiang Shujuan, Liu Weijun, Nan Liangliang. Laser cladding height prediction based on neural network[J]. Journal of Mechanical Engineering, 2009, 45(3): 269-274,281 doi: 10.3901/JME.2009.03.269
[14]	徐浪, 潘勤学, 宿亮, 等. 焊接残余应力超声无损检测技术[J]. 计测技术, 2012, 32(6):29-32,53. (Xu Lang, Pan Qinxue, Su Liang, et al. Compensation technology for nondestructive testing of welded residual stress by ultrasonic[J]. Metrology & Measurement Technology, 2012, 32(6): 29-32,53 doi: 10.3969/j.issn.1674-5795.2012.06.008
[15]	徐春广, 宋文涛, 潘勤学, 等. 残余应力的超声检测方法[J]. 无损检测, 2014, 36(7):25-31. (Xu Chunguang, Song Wentao, Pan Qinxue, et al. Residual stress nondestructive testing method using ultrasonic[J]. Nondestructive Testing, 2014, 36(7): 25-31
[16]	Liu Huaming, Qin Xunpeng, Huang Song, et al. Geometry characteristics prediction of single track cladding deposited by high power diode laser based on genetic algorithm and neural network[J]. International Journal of Precision Engineering and Manufacturing, 2018, 19(7): 1061-1070. doi: 10.1007/s12541-018-0126-8
[17]	朱剑英. 智能系统非经典数学方法[M]. 武汉: 华中科技大学出版社, 2001 Zhu Jianying. Non-classical mathematics for intelligent systems[M]. Wuhan: Huazhong University of Science and Technology Press, 2001
[18]	王东生, 杨友文, 田宗军, 等. 基于神经网络和遗传算法的激光多层熔覆厚纳米陶瓷涂层工艺优化[J]. 中国激光, 2013, 40:0903001. (Wang Dongsheng, Yang Youwen, Tian Zongjun, et al. Process optimization of thick nanostructured ceramic coating by laser multi-layer cladding based on neural network and genetic algorithm[J]. Chinese Journal of Lasers, 2013, 40: 0903001 doi: 10.3788/CJL201340.0903001
[19]	徐春广, 李焕新, 王俊峰, 等. 残余应力的超声横纵波检测方法[J]. 声学学报, 2017, 42(2):195-204. (Xu Chunguang, Li Huanxin, Wang Junfeng, et al. Ultrasonic shear and longitudinal wave testing method of residual stress[J]. Acta Acustica, 2017, 42(2): 195-204
[20]	黄本生, 陈权, 杨江, 等. Q345/316L异种钢焊接残余应力与变形数值模拟[J]. 焊接学报, 2019, 40(2):138-144. (Huang Bensheng, Chen Quan, Yang Jiang, et al. Numerical simulation of welding residual stress and distortion in Q345/316L dissimilar steel[J]. Transactions of the China Welding Institution, 2019, 40(2): 138-144
[21]	于宝义, 何亮, 郑黎, 等. SLM成形ALSi10Mg合金残余应力数值模拟及组织性能分析[J]. 特种铸造及有色合金, 2020, 40(4):349-355. (Yu Baoyi, He Liang, Zheng Li, et al. Numerical simulation and microstructure analysis of residual stress in SLM formed AlSi10Mg[J]. Special Casting & Nonferrous Alloys, 2020, 40(4): 349-355