Microstructure evolution and anisotropy of nickel-based superalloy fabricated by LPBF
-
摘要:
通过改变激光粉末床熔融(laser powder bed fusion, LPBF)的扫描速度研究IN738LC合金组织演化及各向异性机制,采用光学显微镜(optical microscopy,OM)及扫描电镜(scanning electron microscopy,SEM)对组织形貌特征进行表征分析,通过X射线衍射(X-ray diffraction,XRD)对其织构性进行测试,使用显微硬度仪对显微硬度及各向异性进行评价. 结果表明,随着扫描速度从800 mm/s提高到
1600 mm/s,晶粒尺寸得到显著细化,且晶粒长轴取向由低扫描速度下沿建造方向择优,转变为高扫描速度下的沿熔池边界法线方向择优. 这是因为低扫描速度下高熔池重熔率导致更多枝晶沿建造方向外延择优生长. 这种沿建造方向的强择优生长同时导致(200)面沿建造方向择优的织构性,且这种织构强度随扫描速度增加而降低. 这种(200)面沿建造方向择优织构还导致水平截面软轴居多,进而导致水平显微硬度低于侧界面显微硬度的各向异性.Abstract:The microstructure evolution and anisotropy mechanism of IN738LC alloy under varying scanning speeds in laser powder bed fusion (LPBF) were investigated. Optical microscopy (OM) and scanning electron microscopy (SEM) were used to characterize the microstructural features, X-ray diffraction (XRD) was employed to examine the texture, and a microhardness tester was used to evaluate microhardness and anisotropy. The results showed that as the scanning speed increased from 800 mm/s to
1600 mm/s, the grain size became significantly smaller. At low scanning speeds, the grain major axis was preferentially aligned along the build direction. However, at high scanning speeds, the grain major axis transitioned to being aligned normal to the melt pool boundary. This was attributed to the higher melt pool remelting rate at low scanning speeds, which promoted more preferential epitaxial growth of dendrites along the build direction. This strong preferential growth along the build direction also resulted in a (200) texture along the build direction. The intensity of this texture decreased as the scanning speed increased. Furthermore, the (200) preferential alignment along the build direction led to an increased presence of soft axes in the horizontal section, resulting in anisotropy where the horizontal microhardness was lower than that of the side section.-
Keywords:
- scanning speed /
- laser powder bed fusion /
- nickel-based superalloy /
- microstructure /
- anisotropy
-
0. 序言
动力电池作为新能源电动汽车的核心组件,其性能和安全性依赖于制造过程中每个环节的质量控制. 激光焊因其高精度、高速度等优点,被广泛应用于方形和软包动力电池模组制造[1],高质量的焊接连接确保电芯与模组之间的电流传输稳定,有效降低接触电阻,减少发热量,提升能量转换效率和延长电池寿命[2]. 然而,传统单模高斯激光焊易导致金属飞溅缺陷,增加电池接头的电阻和热损耗,降低整体性能和安全性[3]. 为应对这些挑战,探索新型激光焊技术,以实现飞溅的抑制成为关键方向.
近年来,可调环模VBP激光作为新型激光热源,通过合理分配激光能量,提升焊接过程的稳定性,但激光焊过程过于复杂,热−力耦合效应对成形质量的影响规律尚不明晰. 前期研究表明[4],在内外环激光功率比为1∶2 ~ 1∶3时,焊接稳定且飞溅率最低.然而,这些研究主要停留在定性分析,缺乏准确的飞溅状态量化研究,需要开发高时空分辨的实时测量方法,为精细调控激光能量提供指导,也为焊接质量闭环控制提供关键反馈.
声发射传感方法和光电同轴传感系统在激光焊缺陷预测和熔透状态分类识别方面取得进展[5–7],但受噪声干扰和间接传感技术的局限.而高分辨率光学相干层析成像(OCT)技术可实现焊接过程中匙孔深度的精确测量[8],但不能直接与实际焊接质量等同,所以需要结合机器学习方法,建立匙孔深度与飞溅状态的数据驱动模型. 为此,搭建基于OCT传感技术的VBP激光焊实时监测系统,通过1DCNN-BiLSTM深度模型,实现对焊接过程飞溅状态的精准预测,最终在动力电池VBP激光焊平台试验中证实方法的有效性和可靠性.
1. 试验设计与数据采集
1.1 试验材料与设备
试验所用的材料为1060铝合金,材料尺寸为100 mm × 80 mm × 2 mm. 如图1所示,搭建了一套基于OCT传感技术的铝合金VBP激光焊实时监测平台,主要由自主设计研发的谱域光学相干层析成像(SD-OCT)模块和数据采集系统、高功率准直聚焦焊接头(KCTII BLFIW-01)和VBP激光器(EVERFOTON 长飞光坊 FFSC-2000SM/4000)等构成,其中内外环激光峰值功率分别为2 kW和4 kW,内外环芯径为14/100 μm,试验工艺参数设计见表1.
表 1 试验工艺参数Table 1. Experiment parameters design内环功率
P1/W外环功率
P2/W离焦量
f/mm焊接速度
v/(mm·s−1)保护气体流量
q/(L·min−1)0 ~ 1 500 0 ~ 1 500 0 150 25 1.2 光学相干断层扫描分析
如图2所示,采用SD-OCT测量模块和数据采集系统,采样频率为250 kHz,轴向分辨率达20 μm,实现激光焊过程中的匙孔深度在线测量. 具体原理如下:利用低相干光源(SLD)通过光纤耦合器分成两个光束,一个光束经过工件平面返回,另一个光束与激光束同轴到达匙孔底部返回. 两个光束经过光纤耦合器干涉,利用光谱仪线阵相机获取光谱干涉条纹图像,对其功率谱密度进行逆傅里叶变换处理,从而获取匙孔深度信息.
1.3 “匙孔深度−飞溅状态”的数据集构建
根据前期高速摄影观察研究[4]以及图3可知,飞溅发生最小周期为12 ms左右,飞溅缺陷长度一般在2 ~ 6 mm,根据焊接速度150 mm/s换算飞溅发生的周期一般为23 ~ 40个匙孔深度数据,则取单个采集数据量为400个. 进行了30余组不同内外环激光功率试验,并对预处理后的匙孔深度数据进行分类,有飞溅标注为S(spatter),无飞溅标注为NS(no spatter),得到飞溅数据581个,无飞溅数据1 065个.
2. 飞溅评估模型设计
2.1 模型单元设计
2.1.1 1DCNN单元
焊接过程采集到的匙孔深度数据是一维时间数据,其局部时序相关性反映出了焊接质量,考虑到其数据特征,使用1DCNN网络单元也就是一维卷积层对数据进行特征提取并捕捉时序关系[9],且卷积层后通常紧跟一层池化层, 其可对特征进行二次提取,同时减少因卷积运算而产生的参数,进而降低网络的训练代价.
2.1.2 BiLSTM单元
由于匙孔深度数据包含了丰富的焊接飞溅特征信息,但其信号时序波动关系复杂,且包含许多冗余信息. 为有效过滤噪声和冗余信息,并提取有用的深度特征,在1DCNN后使用双向长短期记忆网络(BiLSTM),可在提取局部特征的基础上进一步捕捉全局特征和序列依赖关系,增强对飞溅的评估能力[10].
BiLSTM单元由两层相反的长短时记忆网络(long short-term memory,LSTM)组成,匙孔深度特征通过输入门it进入网络,单元状态Ct中储存了之前累积的匙孔深度特征的长期记忆,遗忘门ft对长期记忆的匙孔深度特征选择性丢弃,只保留重要的信息,输出门ot将当前输入的匙孔深度特征和上阶段积累的前期记忆相结合作为此时刻网络的输出,通过这种方式,可以极大程度地保留每个时间步中的匙孔深度特征,更准确地反映焊接过程的状态和质量.
2.2 超参数优化
贝叶斯优化通过揭示神经网络黑箱函数的全局最优解,能最大化资源利用率,处理噪声数据,并有效利用非连续空间达到全局最小[11]. TPE算法在建模阶段利用评估结果构建概率模型,分别建立x属于更好结果和较差结果时的概率密度函数. 在采样阶段,EI的计算基于两个模型的比值,通过选择使EI(x)最大的x选定效果较好的超参数. 各模型的卷积核尺寸为3,Batchsize为256,Epochs为200,损失函数为交叉熵模型. 通过此法得到各种模型的最佳超参数见表2.
表 2 各模型最佳超参数Table 2. Optimal hyperparameters for each model模型名称 各卷积核个数N/个 LSTM层数n1/层 隐藏层数n2/层 Dropout值 学习率α(10−5) 总参数数量M 1DCNN 32/256/512 — — 0.40 17 444 386 LSTM — 4 128 0.20 5.7 465 970 BiLSTM — 3 256 0.30 8.1 369 464 1DCNN-LSTM 16/256 2 128 0.37 1.0 344 482 1DCNN-BiLSTM 32/128 2 128 0.14 1.4 676 242 2.3 1DCNN-BiLSTM复合模型
1DCNN-BiLSTM模型架构如图4所示. 模型首先通过一维卷积层提取匙孔深度数据的局部特征,然后通过最大池化层下采样,减少特征长度和计算复杂度. 经过两次这样的操作后,使用双向长短期记忆网络(BiLSTM)提取数据的全局特征和序列依赖关系. 引入Adam优化器,并通过Dropout层随机丢弃部分神经元,防止过拟合,提高泛化能力,最后通过全连接层将匙孔深度的飞溅特征转换为飞溅(S)状态和无飞溅(NS)状态的概率分布.
3. 试验结果与分析
应对识别飞溅使用二分类中常用的准确率A(accuracy)精确率P(precision)、召回率R(recall)和 F1值(F)4个指标对模型性能进行评估. 各指标对比结果见表3,1DCNN-BiLSTM模型的测试集准确率和损失曲线如图5所示,可以看出1DCNN模型在精准度上优于LSTM和BiLSTM模型,表明匙孔深度数据在飞溅时会导致局部数据波动,这使得对局部敏感的卷积模型表现更好. 1DCNN-BiLSTM模型随着迭代次数增加,损失值逐渐下降且稳定,这说明1DCNN-BiLSTM模型结合了1DCNN和BiLSTM的优点,在挖掘复杂时序波动关系以及飞溅特征和应对冗余信息方面表现优异,有效实现了激光焊过程的飞溅评估.
表 3 各模型指标Table 3. Metrics for each model模型名称 准确率
A(%)精确率
P(%)召回率
R(%)F1值
F(%)1DCNN 86.93 86.95 86.93 86.94 LSTM 80.24 83.21 80.24 78.87 BiLSTM 82.06 85.77 82.06 80.67 1DCNN-LSTM 97.87 97.82 97.87 97.85 1DCNN-BiLSTM 99.69 99.68 99.69 99.65 4. 结论
(1) 针对动力电池铝合金VBP激光焊过程,设计并搭建了基于SD-OCT传感技术的激光焊过程监测平台,实时获取了动态焊接过程的匙孔深度信息,并构建了“匙孔深度−飞溅状态”的样本数据集.
(2) 构建了1DCNN-BiLSTM算法的复合模型,深度挖掘了反映飞溅动力学的局部—全部时序特征.
(3) 相比1DCNN,LSTM和BiLSTM模型,提出的1DCNN-BiLSTM复合模型能够准确地判断飞溅状态,其识别准确率达到99.69%,能够满足VBP焊接飞溅定量评价要求,并在实际焊接平台得到验证.
-
表 1 IN738LC合金粉末的名义成分 (质量分数,%)
Table 1 Nominal compositions of IN738LC alloy powder
含量范围 Ni Cr Co Mo W Ta Al Ti Nb C B Zr 最小值 余量 15.7 8.0 1.5 2.4 1.5 3.2 3.2 0.6 0.09 0.07 0.02 最大值 余量 16.3 9.0 2.0 2.8 2.0 3.7 3.7 1.1 0.13 0.012 0.08 -
[1] Tan C L, Weng F, Sui S, et al. Progress and perspectives in laser additive manufacturing of key aeroengine materials[J]. International Journal of Machine Tools and Manufacture, 2021, 170: 103804. doi: 10.1016/j.ijmachtools.2021.103804
[2] Yang C Y, Hu R, Liu X X, et al. Microstructural evolution and high-temperature strengthening mechanisms of the IN 738LC superalloy prepared by selective laser melting[J]. Journal of Materials Research and Technology, 2024, 29: 5304 − 5316. doi: 10.1016/j.jmrt.2024.02.219
[3] Dang X F, Li Y, Chen K, et al. Avoiding cracks in additively manufactured non-weldable directionally solidified Ni-based superalloys[J]. Additive Manufacturing, 2022, 59: 103095. doi: 10.1016/j.addma.2022.103095
[4] Lim B, Nomoto K, Clarke A J, et al. On the interplay of internal voids, mechanical properties, and residual stresses in additively manufactured Haynes 282[J]. Additive Manufacturing, 2023, 75: 103749. doi: 10.1016/j.addma.2023.103749
[5] Lu X F, Chen C, Zhang G H, et al. Thermo-mechanical simulation of annealing heat treatment of Ni-based GH4099 superalloy made by laser powder bed fusion[J]. Additive Manufacturing, 2023, 73: 103703. doi: 10.1016/j.addma.2023.103703
[6] 刘鹏宇, 李辉, 张瑞华, 等. 不同振镜精度下选区激光熔化Ti6Al4V零件性能的对比分析[J]. 焊接学报, 2024, 45(8): 52 − 61. doi: 10.12073/j.hjxb.20230616003 Liu Pengyu, Li Hui, Zhang Ruihua, et al. Research on the properties of Ti6Al4V parts prepared by selective laser melting under different galvanometer accuracy[J]. Transactions of the China Welding Institution, 2024, 45(8): 52 − 61. doi: 10.12073/j.hjxb.20230616003
[7] Song X P, Huang J K, Fan D. Review of functionally graded materials processed by additive manufacturing[J]. China Welding, 2023, 32(3): 41 − 50.
[8] Diegel O, Schutte J, Ferreira A, et al. Design for additive manufacturing process for a lightweight hydraulic manifold[J]. Additive Manufacturing, 2020, 36: 101446. doi: 10.1016/j.addma.2020.101446
[9] Wang H, Zhang X, Wang G B, et al. Selective laser melting of the hard-to-weld IN738LC superalloy: efforts to mitigate defects and the resultant microstructural and mechanical properties[J]. Journal of Alloys and Compounds, 2019, 807: 151662. doi: 10.1016/j.jallcom.2019.151662
[10] Zhou W Z, Tian Y S, Tan Q B, et al. Effect of carbon content on the microstructure, tensile properties and cracking susceptibility of IN738 superalloy processed by laser powder bed fusion[J]. Additive Manufacturing, 2022, 58: 103016. doi: 10.1016/j.addma.2022.103016
[11] Xu J Y, Ding Y T, Gao Y B, et al. Grain refinement and crack inhibition of hard-to-weld Inconel 738 alloy by altering the scanning strategy during selective laser melting[J]. Materials & Design, 2021, 209: 109940.
[12] Sun Z J, Ma Y, Ponge D, et al. Thermodynamics-guided alloy and process design for additive manufacturing[J]. Nature communications, 2022, 13(1): 4361. doi: 10.1038/s41467-022-31969-y
[13] Hu Y, Yang X K, Kang W J, et al. Effect of Zr content on crack formation and mechanical properties of IN738LC processed by selective laser melting[J]. Transactions of Nonferrous Metals Society of China, 2021, 31(5): 1350 − 1362. doi: 10.1016/S1003-6326(21)65582-6
[14] Li Y, Kan W B, Zhang Y M, et al. Microstructure, mechanical properties and strengthening mechanisms of IN738LC alloy produced by electron beam selective melting[J]. Additive Manufacturing, 2021, 47: 102371. doi: 10.1016/j.addma.2021.102371
[15] Yu Z R, Guo C, Han S, et al. The effect of Hf on solidification cracking inhibition of IN738LC processed by selective laser melting[J]. Materials Science & Engineering A, 2021, 804: 140733.
[16] Megahed S, Kramer K M, Heinze C, et al. Influence of build orientation on the creep behavior of IN738LC manufactured with laser powder bed fusion[J]. Materials Science & Engineering A, 2023, 878: 145197.
[17] Marquez J A, Rodrigue C M, Herrera C M, et al. Effect of surface morphology of ZnO electrodeposited on photocatalytic oxidation of methylene blue dye part I: analytical study[J]. International Journal of Electrochemical Science, 2011, 6(9): 4059 − 4069. doi: 10.1016/S1452-3981(23)18310-6
[18] Sui S, Tan H, Chen J, et al. The influence of Laves phases on the room temperature tensile properties of Inconel 718 fabricated by powder feeding laser additive manufacturing[J]. Acta Materialia, 2019, 164: 413 − 427. doi: 10.1016/j.actamat.2018.10.032