Scholarlink

<p>Developing high-performance coatings is highly desirable; however, the efficiency of the traditional trial-and-error method is often insufficient, particularly when designing multi-component coatings with a broad compositional range. Here, a high-throughput preparation method combined with machine learning is proposed to accelerate the design of refractory boride coating with complex compositions. A database was firstly constructed based on 48 groups of compositions and hardness detected in high-throughput deposited coating via NbB, WB, and MoB co-sputtering. Five features were filtered as key features related to hardness, and corresponding regression models were applied to train and test the set which also have been cross validated by Leave-One-Out Cross-Validation. Then, the relationship between the filtered factors and hardness was described by developing a Partial Dependence Plot model to reveal the composition-dependent hardness of boride coatings. As an illustrative example, we successfully predicted an Nb-Mo-W-B boride coating with a high hardness of 20.39&#xa0;GPa, characterized by the composition Nb<sub>15</sub>Mo<sub>33</sub>W<sub>52</sub>B<sub>100</sub>. Subsequently, the predicted optimal composition was used to sinter a target to sputter a coating which exhibits a hardness of 20.7&#xa0;GPa, agreeing well with the predicted value. Our results show the high-efficiency and powerful of Machine Learning model combined with High-Throughput Experiment to accurately predict hardness of multi-component coatings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Composition design and process optimization of refractory boride coatings based on high-throughput preparation and machine learning

  • 期刊论文

  • 发表日期：

    2025年8月23日

查看全文
  • Binyuan Jia,
  • Wenhao Ma,
  • Qiang Wan,
  • Cai Lu,
  • Qiwen Wei,
  • Bing Yang,
  • Yangyi Xiao

摘要

Developing high-performance coatings is highly desirable; however, the efficiency of the traditional trial-and-error method is often insufficient, particularly when designing multi-component coatings with a broad compositional range. Here, a high-throughput preparation method combined with machine learning is proposed to accelerate the design of refractory boride coating with complex compositions. A database was firstly constructed based on 48 groups of compositions and hardness detected in high-throughput deposited coating via NbB, WB, and MoB co-sputtering. Five features were filtered as key features related to hardness, and corresponding regression models were applied to train and test the set which also have been cross validated by Leave-One-Out Cross-Validation. Then, the relationship between the filtered factors and hardness was described by developing a Partial Dependence Plot model to reveal the composition-dependent hardness of boride coatings. As an illustrative example, we successfully predicted an Nb-Mo-W-B boride coating with a high hardness of 20.39 GPa, characterized by the composition Nb15Mo33W52B100. Subsequently, the predicted optimal composition was used to sinter a target to sputter a coating which exhibits a hardness of 20.7 GPa, agreeing well with the predicted value. Our results show the high-efficiency and powerful of Machine Learning model combined with High-Throughput Experiment to accurately predict hardness of multi-component coatings.