黄中庆,彭爽,孙德恩,邱恒鑫,崔芹,张健,LIU Shiyu.机器学习辅助设计高强韧(TiZrNbCrSi)N高熵氮化物涂层[J].表面技术,2025,54(1):74-83, 160.
HUANG Zhongqing,PENG Shuang,SUN Deen,QIU Hengxin,CUI Qin,ZHANG Jian,LIU Shiyu.Machine Learning Assisted Design of High Hard Yet Tough (TiZrNbCrSi)N High-entropy Nitride Coatings[J].Surface Technology,2025,54(1):74-83, 160 |
| 机器学习辅助设计高强韧(TiZrNbCrSi)N高熵氮化物涂层 |
| Machine Learning Assisted Design of High Hard Yet Tough (TiZrNbCrSi)N High-entropy Nitride Coatings |
| 投稿时间：2024-05-17 修订日期：2024-09-02 |
| DOI：10.16490/j.cnki.issn.1001-3660.2025.01.007 |
| 中文关键词: 机器学习 高通量制备 高熵氮化物 高强韧涂层 成分设计 |
| 英文关键词:machine learning high-throughput preparation high-entropy nitrides hard yet tough coating composition design |
| 基金项目:中央高校基本科研业务费引进人才项目(SWU-KR22011) |
|
|
| Author | Institution |
| HUANG Zhongqing | Center for New Thin Film Materials and Devices, School of Materials and Energy, Southwest University, Chongqing 400715, China |
| PENG Shuang | Center for New Thin Film Materials and Devices, School of Materials and Energy, Southwest University, Chongqing 400715, China |
| SUN Deen | Center for New Thin Film Materials and Devices, School of Materials and Energy, Southwest University, Chongqing 400715, China |
| QIU Hengxin | Center for New Thin Film Materials and Devices, School of Materials and Energy, Southwest University, Chongqing 400715, China |
| CUI Qin | Center for New Thin Film Materials and Devices, School of Materials and Energy, Southwest University, Chongqing 400715, China |
| ZHANG Jian | Chongqing Chuanyi Control Valve Co., Ltd., Chongqing 400707, China |
| LIU Shiyu | Innovation and Enterprise I&E Agency of Science, Technology and Research AStar, Singapore 138632 |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 通过机器学习构建(TiZrNbCrSi)N体系的硬度算法模型和韧性算法模型，结合高通量制备完成对兼具高硬度和高韧性的高熵氮化物涂层的成分设计和高效筛选。方法 采用磁控溅射多靶共沉积技术制备强韧一体化的(TiZrNbCrSi)N高熵氮化物涂层。采用热场发射扫描电镜和能谱仪对涂层表、截面形貌及成分进行分析，利用纳米压痕仪测量涂层的硬度和弹性模量，采用压痕法定量表征涂层的韧性。同时，引入高通量制备技术大幅缩短样品制备周期，采用随机森林算法构建机器学习模型，对涂层的成分和性能进行高效分析和预测。结果 (TiZrNbCrSi)N涂层的厚度约为720 nm，呈现柱状晶结构，其晶体结构为FCC。涂层的硬度为12~28 GPa，韧性值为1~10 MPa.m1/2。机器学习对涂层硬度预测的均方根误差为1.118 GPa，对韧性预测的均方根误差为1.292 MPa.m1/2。结论 通过机器学习构建的硬度算法模型及韧性算法模型，对(TiZrNbCrSi)N涂层体系的性能预测结果具有较高的准确性。筛选获得的(Ti0.079Zr0.081Nb0.089Cr0.119Si0.068)N0.564高熵氮化物涂层兼具高硬度和高韧性，其硬度为25.6 GPa、韧性值为8 MPa.m1/2。 |
| 英文摘要: |
| In recent years, high-entropy nitride coatings have attracted significant attention due to their unique design concept and excellent comprehensive mechanical properties. However, the preparation and optimization of high-entropy nitride coatings is a time consuming process due to the large number of elements and simulations required to obtain accurate estimates of parameters. To solve that, machine learning and high-throughput preparation were conducted in this study to prepare and optimize a (TiZrNbCrSi)N high-entropy nitride coating by constructing a hardness algorithm model and a toughness algorithm model. High-throughput preparation was conducted using a magnetron sputtering multi-targets co-deposition system. Silicon wafer with a radius of 50 mm were used as the substrate. After ultrasonic cleaning, it was sent into a deposition chamber. During the deposition process, substrate bias voltage of −270 V, Si target power of 50 W, Cr target power of 80 W, TiZrNb alloy target power of 200 W and N2 flow rate of 12 sccm were set to prepare a (TiZrNbCrSi)N high-entropy nitride coating in three stages for a total of 8 000 seconds. The prepared coating was divided into grids with a spacing distance of 3 mm, and the intersection area of the grids was treated as one component point, with a total of 78 component points. Then, the surface and cross-sectional morphology of the sample were observed using JSM-7800F field emission scanning electron microscope (FESEM) and Oxford XMax-80 energy spectrometer, and the elemental composition and content were analyzed. XRD detection was performed using a D8 ADVANCE X-ray diffractometer with a step size of 0.08° and a scanning range of 2θ from 20° to 80°. The mechanical properties of the coating were measured using the Hysitron nanoindentation instrument according to the continuous stiffness method. In order to reduce experimental errors and improve data reliability, five experimental measurements were conducted at each component point to obtain the average value, and the spacing between each indentation was not less than 50 μm. A Shimadzu HMV-G-FA micro Vickers hardness tester with a load of 1-2 N was applied to the coating using the indentation method to characterize the coating toughness. The toughness KIC was calculated by measuring the crack length and the half diagonal length of the indentation. Based on the data obtained from the above experiments, the training and testing sets were divided at a 4∶1 ratio. A total of 9 sets of hyperparameters were obtained by combining the different numbers of estimator (50, 100 and 150) and the different maximum depth (5, 10 and 15). Grid search was performed using 10 fold cross validation. Python was applied to establish a random forest hardness algorithm model and a toughness algorithm model based on the grid search method for component design and efficient screening. The (TiZrNbCrSi)N high-entropy nitride coating was successfully prepared on the surface of silicon wafer. The coating had a thickness of approximately 720 nm and exhibited a columnar crystal with FCC structure. The hardness distribution of the coating ranged within 12-28 GPa, and the toughness was 1-10 MPa.m1/2. The root mean square error for the hardness prediction results was 1.118 GPa, and that for toughness 1.292 MPa.m1/2. The hardness algorithm model and toughness algorithm model built by machine learning in this paper show a high accuracy in predicting the mechanical properties of the (TiZrNbCrSi)N high-entropy nitride coating system. The filtered result (Ti0.079Zr0.081Nb0.089Cr0.119Si0.068)N0.564 coating shows both high hardness and toughness, with a hardness of 25.6 GPa and a toughness of 8 MPa.m1/2. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
渝公网安备 50010702501715号