基于迁移学习的NiCo-FGM机器人砂带磨削工艺

doi:10.12068/j.issn.1005-3026.2025.20240096

摘要/Abstract

摘要：

为提高镍钴功能梯度材料（nickel-cobalt-functional gradient materials, NiCo-FGM）去除深度的一致性，采用自适应磨削力控制系统对5种不同质量分数IN718的NiCo-FGM进行恒力分区磨削实验，探究工艺参数对材料去除深度及表面粗糙度的影响趋势及程度.然后对迁移学习进行可行性分析并对比迁移学习与经验公式的去除深度建模精度.最后对比恒力与变力磨削的去除深度预测结果.结果表明：法向力对材料去除深度与表面粗糙度的影响最显著.迁移学习预测的平均误差降低了4.07%，且效率更高.恒力磨削下其余含量的IN718与50%IN718去除深度最大差值为8.955 μm，100%IN718与0%IN718去除深度最大差值为15.619 μm，而通过变力磨削可以提高去除深度一致性.

关键词: 镍钴功能梯度材料, 迁移学习, 去除深度一致性, 机器人砂带磨削, 自适应磨削力控制系统

Abstract:

In order to improve the removal depth consistency of nickel-cobalt functional gradient materials （NiCo-FGM）， an adaptive grinding force control system was constructed to carry out constant force zonal grinding experiments on five types of NiCo-FGM with different mass fractions of IN718 to investigate the trend and extent of the influence of the process parameters on the removal depth and surface roughness of the materials. The feasibility of transfer learning was then analyzed and the accuracy of the removal depth modelling was compared with that of empirical formulas. Finally， comparing the removal depth prediction results of constant force and variable force grinding. The results showed that the normal force has the most significant effect on the removal depth and surface roughness of the materials. The average error in the prediction of transfer learning is reduced by 4.07%， and the efficiency is higher. The maximum difference in removal depth between the remaining content of IN718 and 50%IN718 under constant force grinding is 8.955 μm， and the maximum removal depth difference between 100%IN718 and 0%IN718 is 15.619 μm， whereas the removal depth consistency can be improved by variable force grinding.

Key words: nickel-cobalt functional gradient material （NiCo-FGM）, transfer learning, removal depth consistency, robotic belt grinding, adaptive grinding force control system

中图分类号:

TH 161

辛博, 李宏亮, 孙文鑫, 刘洺君. 基于迁移学习的NiCo-FGM机器人砂带磨削工艺[J]. 东北大学学报（自然科学版）, 2025, 46(6): 66-75.

Bo XIN, Hong-liang LI, Wen-xin SUN, Ming-jun LIU. Transfer Learning-Based Robotic Belt Grinding Process for NiCo-FGM[J]. Journal of Northeastern University(Natural Science), 2025, 46(6): 66-75.

图/表 21

图1 机械结构示意图

Fig.1 Mechanical structure schematic diagram

图2 机械结构实物图

Fig.2 Mechanical structure physical drawing

图3 控制算法逻辑图

Fig.3 Control algorithm logic diagram

表1 IN718和Stellite 6的热物性参数(20 ℃)

Table 1 Thermophysical parameters of IN718 and Stellite 6 (20 ℃)

参数	IN718	Stellite 6
弹性模量/GPa	205	230
泊松比	0.3	0.34
线膨胀系数×10⁶/℃^-1	13.0	13.6
导热系数/[W·(m·K)^-1]	11.4	13.6
比热容/[J·(kg·K)^-1]	435	500

表2 实验参数

Table 2 Experimental parameters

参数名称	值
水质量分数×10⁸/%	<4
氧质量分数×10⁸/%	<100
激光功率/W	1 800
扫描速率/(mm∙min^-1)	600
送粉速率/(g∙min^-1)	15

表3 粉末材料的化学成分（质量分数） (%)

Table 3 Chemical composition of powder material (mass fraction)

粉末材料	C	Si	Mn	Cr	Mo	Ti	Fe	Al	Co	Ni
IN718	0.05	0.71	0.16	18.17	2.32	0.93	20.89	0.63	—	余量
Stellite 6	1.15	1.58	0.75	31.25	0.89	—	3.54	—	余量	2.54

图4 4种主要元素质量分数随IN718含量变化规律

Fig.4 Mass fractions of four main elements changing with the content of IN718

图5 不同磨削方向NiCo-FGM连续磨削力变化（a）—仿真示意图；（b）—仿真磨削力结果.

Fig.5 Continuous grinding force change of NiCo-FGM in different grinding directions

图6 NiCo-FGM各梯度硬度

Fig.6 Hardness of NiCo-FGM alloy by gradient

图7 磨削方案示意图

Fig.7 Schematic diagram of the grinding program

表4 不同质量分数IN718的正交实验水平表 (with different mass fractions)

Table 4 Orthogonal experimental level table of IN718

水平	因素
水平	v_s/(m·s^-1)	F_n /N	v_w/(mm·s^-1)
1	4.712	4	3
2	7.069	8	4.5
3	9.425	12	6

图8 正交实验结果（a）—去除深度；（b）—表面粗糙度.

Fig.8 Results of the orthogonal experiments

图9 迁移学习流程图

Fig.9 Flowchart for transfer learning

图10 50%IN718扩充实验结果

Fig.10 Expanded experimental results of 50%IN718

表5 75%IN718正交实验水平表 (75%IN718)

Table 5 Orthogonal experimental level table of

水平	因素
水平	v_s/(m·s^-1)	F_n /N	v_w/(mm·s^-1)
1	5.312	5	3.4
2	7.669	9	4.9
3	10.025	13	6.4

表6 75%IN718去除深度经验公式、神经网络与迁移学习误差对比 (of 75%IN718)

Table 6 Comparison of empirical formulas, neural networks and transfer learning errors in the removal depth

编号	v_w/(mm·s^-1)	F_n/N	v_s/(m·s^-1)	去除深度h/μm	预测误差/%
编号	v_w/(mm·s^-1)	F_n/N	v_s/(m·s^-1)	去除深度h/μm	神经网络	经验公式	迁移学习
平均	—	—	—	—	38.071	9.405	5.336
1	5.312	5	3.4	15.545	24.451	19.735	15.840
2	5.312	9	4.9	28.655	43.600	8.909	5.362
3	5.312	13	6.4	35.296	68.530	11.249	2.596
4	7.669	5	4.9	22.247	94.781	10.042	10.366
5	7.669	9	6.4	28.181	35.695	7.294	2.882
6	7.669	13	3.4	67.068	0.298	6.957	1.644
7	10.025	5	6.4	23.442	53.298	10.014	7.087
8	10.025	9	3.4	59.917	6.311	5.153	0.769
9	10.025	13	4.9	74.106	15.670	5.295	1.477

图11 极差分析结果

Fig.11 Extreme variance analysis results

图12 均值分析结果（a）—砂带线速度对表面粗糙度的影响；（b）—磨削法向力对表面粗糙度的影响；（c）—进给速度对表面粗糙度的影响；（d）—砂带线速度对去除深度的影响；（e）—磨削法向力对去除深度的影响；（f）—进给速度对去除深度的影响.

Fig.12 Mean value analysis results

表7 相同去除深度下不同IN718含量的磨削法向力

Table 7 Grinding normal forces with different IN718 contents at the same removal depth

目标去除深度h/μm	磨削法向力/N
目标去除深度h/μm	100%IN718	75%IN718	50%IN718	25%IN718	0%IN718
20	3.929	5.012	5.370	5.784	7.074
22	4.397	5.459	5.877	6.277	7.591
24	4.865	5.905	6.383	6.761	8.095
26	5.333	6.348	6.885	7.237	8.587
28	5.801	6.789	7.386	7.707	9.071
30	6.270	7.227	7.885	8.174	9.549
32	6.739	7.662	8.382	8.640	10.024
34	7.208	8.095	8.878	9.108	10.498
36	7.678	8.527	9.372	9.580	10.974

表8 相同磨削法向力下不同IN718含量的去除深度

Table 8 Removal depths of different IN718 contents under the same grinding normal force

磨削法向力F_n/N	去除深度/μm
磨削法向力F_n/N	100%IN718	75%IN718	50%IN718	25%IN718	0%IN718
5.370	26.419	21.309	20.000	17.882	15.593
5.877	28.760	23.728	22.000	19.869	17.380
6.383	31.075	26.110	24.000	21.838	19.114
6.885	33.378	28.472	26.000	23.803	20.810
7.386	35.676	30.835	28.000	25.778	22.485
7.885	37.979	33.217	30.000	27.779	24.158
8.382	40.292	35.635	32.000	29.817	25.849
8.878	42.617	38.099	34.000	31.896	27.573
9.372	44.955	40.607	36.000	34.013	29.336

图13 去除深度差值

Fig.13 Removal depth difference

参考文献 18

[1]	Loh G H， Pei E， Harrison D， et al. An overview of functionally graded additive manufacturing［J］. Additive Manufacturing， 2018， 23： 34-44.
[2]	Wang Y F， Chen X Z， Su C C. Microstructure and mechanical properties of Inconel 625 fabricated by wire-arc additive manufacturing［J］. Surface and Coatings Technology， 2019， 374： 116-123.
[3]	Bobbio L D， Otis R A， Borgonia J P， et al. Additive manufacturing of a functionally graded material from Ti-6Al-4V to Invar： experimental characterization and thermodynamic calculations［J］. Acta Materialia， 2017， 127： 133-142.
[4]	于富明. 基于工件曲率的变压力砂带磨削技术研究［D］. 沈阳：东北大学，2017.
	Yu Fu-ming. Research on variable pressure abrasive belt grinding technology based on workpiece curvature ［D］. Shenyang： Northeastern University，2017.
[5]	王恭硕. 面向航空发动机整体叶盘材料去除精度要求的机器人磨抛技术研究［D］. 武汉：华中科技大学，2022.
	Wang Gong-shuo. Research on robotic grinding and polishing technology for aero-engine integral blisk material removal accuracy requirements ［D］. Wuhan： Huazhong University of Science and Technology， 2022.
[6]	巩亚东，赵显力，张伟健，等.机器人砂带磨削单磨粒材料去除影响因素［J］.东北大学学报（自然科学版），2023，44（9）：1285-1291.
	Gong Ya-dong， Zhao Xian-li， Zhang Wei-jian，et al. Factors influencing single abrasive material removal for robotic abrasive belt grinding［J］. Journal of Northeastern University（Natural Science）， 2023， 44（9）： 1285-1291.
[7]	Preston F W. The theory and design of plate glass polishing machines ［J］. Journal of the Society of Glass Technology， 1927（11）：277-281.
[8]	计时鸣，李琛，谭大鹏，等. 基于Preston方程的软性磨粒流加工特性［J］.机械工程学报， 2011，47（17）：156-163.
	Ji Shi-ming， Li Chen， Tan Da-peng， et al. Study on machinability of softness abrasive flow based on Preston equation［J］. Journal of Mechanical Engineering， 2011，47（17）：156-163.
[9]	张雷，袁楚明，周祖德，等. 模具曲面抛光时表面去除的建模与试验研究［J］. 机械工程学报， 2002， 38（12）：98-102.
	Zhang Lei， Yuan Chu-ming， Zhou Zu-de， et al. Modeling and experiment of material removal in polishing on mold curved surfaces［J］. Chinese Journal of Mechanical Engineering， 2002， 38（12）：98-102.
[10]	Cabaravdic M， Kuhlenköetter B. Optimising belt grinding processes［J］. Metal Surface， 2005（4）：44-47.
[11]	Himeur Y， Elnour M， Fadli F， et al. Next-generation energy systems for sustainable smart cities： roles of transfer learning［J］. Sustainable Cities and Society， 2022， 85： 104059.
[12]	Hazarika D， Poria S， Zimmermann R， et al. Conversational transfer learning for emotion recognition［J］. Information Fusion， 2021， 65： 1-12.
[13]	Lu J， Bebbood V， Hao P， et al. Transfer learning using computational intelligence： a survey［J］. Knowledge-Based Systems， 2015， 80： 14-23.
[14]	Zhuang F Z， Qi Z Y， Duan K Y， et al. A comprehensive survey on transfer learning［J］. Proceedings of the IEEE， 2021， 109（1）： 43-76.
[15]	Wang J D， Chen Y Q， Feng W J， et al. Transfer learning with dynamic distribution adaptation［J］. ACM Transactions on Intelligent Systems and Technology， 2020， 11（1）： 1-25.
[16]	Tzeng E， Hoffman J， Darrell T， et al. Simultaneous deep transfer across domains and tasks［C］// 2015 IEEE International Conference on Computer Vision（ICCV）. Santiago， 2015： 4068-4076.
[17]	Pardoe D， Stone P. Boosting for regression transfer［C］// 27th International Conference on Machine Learning. Haifa， 2010： 863-870.
[18]	杨吉祥，王恭硕，叶葱葱，等. 一种基于迁移学习的材料去除率模型的建立方法和装置： CN202110230749.5［P］. 2021-07-06.
	Yang Ji-xiang， Wang Gong-shuo， Ye Cong-cong， et al. A method and device for establishing a material removal rate model based on transfer learning： CN202110230749.5 ［P］. 2021-07-06.