Scholarlink

<p>Optimizing the magnetic properties of Co–Ni alloy thin films requires understanding complex composition–structure–property relationships that conventional analysis methods cannot adequately capture due to nonlinear interdependencies among synthesis parameters, microstructure, and magnetic behavior. This study introduces the first comprehensive application of triangulated interpretability methods—combining SHAP, perturbation, and Sobol sensitivity analyses—to quantitatively decode the magnetic behavior of electrodeposited Co–Ni thin films, providing unprecedented insights for targeted materials design. Through systematic electrodeposition of four Co–Ni compositions (52–75&#xa0;wt% Co) and comprehensive characterization using XRD, SEM, and VSM, we generated a dataset of 1322 field-dependent magnetic moment measurements. Our custom deep neural network achieved exceptional predictive accuracy (R<sup>2</sup> = 0.973) and, through triangulated interpretability analysis, revealed that applied magnetic field dominates magnetic response (SHAP value = 0.695), followed by cobalt content (0.291) and nickel content (0.384). The integrated framework identified optimal compositions for specific applications: ~ 70&#xa0;wt% Co with 350–380&#xa0;nm grain sizes for high-saturation magnetization (Ms ≈ 120&#xa0;emu/g) in EMI shielding and &lt; 60&#xa0;wt% Co for low coercivity sensor applications. This triangulated interpretability approach provides robust, quantitative guidance for accelerating magnetic materials development, demonstrating how advanced machine learning can transform empirical materials optimization into predictive, knowledge-driven design.</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

Deep learning-based prediction of magnetic properties in electrodeposited Co–Ni alloy thin films

  • 期刊论文

  • 发表日期：

    2025年9月18日

查看全文
  • Hasan Güler,
  • Rasim Özdemir,
  • Adem Coşkun

摘要

Optimizing the magnetic properties of Co–Ni alloy thin films requires understanding complex composition–structure–property relationships that conventional analysis methods cannot adequately capture due to nonlinear interdependencies among synthesis parameters, microstructure, and magnetic behavior. This study introduces the first comprehensive application of triangulated interpretability methods—combining SHAP, perturbation, and Sobol sensitivity analyses—to quantitatively decode the magnetic behavior of electrodeposited Co–Ni thin films, providing unprecedented insights for targeted materials design. Through systematic electrodeposition of four Co–Ni compositions (52–75 wt% Co) and comprehensive characterization using XRD, SEM, and VSM, we generated a dataset of 1322 field-dependent magnetic moment measurements. Our custom deep neural network achieved exceptional predictive accuracy (R2 = 0.973) and, through triangulated interpretability analysis, revealed that applied magnetic field dominates magnetic response (SHAP value = 0.695), followed by cobalt content (0.291) and nickel content (0.384). The integrated framework identified optimal compositions for specific applications: ~ 70 wt% Co with 350–380 nm grain sizes for high-saturation magnetization (Ms ≈ 120 emu/g) in EMI shielding and < 60 wt% Co for low coercivity sensor applications. This triangulated interpretability approach provides robust, quantitative guidance for accelerating magnetic materials development, demonstrating how advanced machine learning can transform empirical materials optimization into predictive, knowledge-driven design.