搜索

x
中国物理学会期刊

基于分子动力学模拟与机器学习的Janus纳米流体导热特性分析

Predicting the thermal conductivity of Janus nanofluids based on molecular dynamics simulation and machine learning

PDF
导出引用
  • 在悬浮液中添加纳米尺寸的金属、金属氧化物或非金属颗粒,可制备具有较高导热性能的纳米流体。Janus纳米流体是将表面性质经不对称修饰的Janus纳米颗粒添加到基础流体中所形成的纳米流体。相较于普通纳米流体,Janus纳米流体具有更高的热导率。本文将平衡分子动力学模拟方法与机器学习方法相结合,计算并分析了银、铜或铁的Janus纳米流体热导率。基于分子动力学模拟计算所得Janus纳米流体热导率数据,分别采用反向传播神经网络、支持向量回归、数据分组处理方法以及随机森林四种机器学习模型,用于预测Janus纳米流体的热导率。机器学习模型的输入参数包括纳米颗粒粒径、体积分数、Janus纳米颗粒差异性参数、纳米流体密度、纳米颗粒质量以及颗粒材料的热导率。通过对比各机器学习模型的预测结果与误差,验证了机器学习预测Janus纳米流体热导率的可行性。综合分析表明,数据分组处理方法在预测Janus纳米流体的热导率中表现最优,均方根误差为0.0058 W/(m·K),决定系数为0.9844,具有较高的预测精度。分析表明,各参数对Janus纳米流体热导率的正向影响规律与已有研究结论相符合,并得到了各影响参数的重要性排序。

    Janus nanofluids are formed by dispersing Janus nanoparticles (particles with asymmetrically modified surface properties) into a base fluid, which is reported to have a high thermal conductivity owing to excellent diffusion properties of the Janus particles. The thermal conductivity of nanofluids can be influenced by multiple parameters, including nanoparticle size, density, volume fraction, and others. It is a significant challenge to analyze the influence of these parameters on the thermal conductivity of nanofluids using traditional experimental and simulation approaches. Machine learning offers a powerful approach to handling nonlinear relationships and high-dimensional data. In this work, equilibrium molecular dynamics simulations are employed to calculate the thermal conductivity of Janus nanofluids containing silver, copper, or iron Janus nanoparticles. Based on the molecular dynamics simulation results, four machine learning models—Backpropagation Neural Network, Support Vector Regression, Group Method of Data Handling, and Random Forest—are applied to predict and analyze the thermal conductivity of Janus nanofluids. The input parameters for the machine learning models include nanoparticle size, nanofluid density, volume fraction, Janus nanoparticle asymmetry parameter, nanoparticle mass, and the thermal conductivity of the bulk nanoparticle material. It is found that the machine learning methods can effectively predict the thermal conductivity of Janus nanofluids. Among these machine learning models, Group Method of Data Handing has the most accurate predictions, with a root mean square error of 0.0058 W/(m·K) and a coefficient of determination of 0.9844. Feature importance analysis using SHAP for different input parameters indicates that the positive effects of each parameter are consistent with existing findings in open literature. This work validates the feasibility and robustness of the machine learning approach for predicting the thermal conductivity of Janus nanofluids and provides a valuable methodology for investigating and rapidly assessing Janus nanofluid thermal conductivity.

    目录

    返回文章
    返回
    Baidu
    map