基于机器学习的磁流变弹性体磁致储能模量的快速准确表征  

Fast and accurate characterization of magnetorheological elastomers based on machine learning

作　　者：任航 赵丹[1] 董立强 刘少刚[1] 杨金水 Ren Hang;Zhao Dan;Dong Li-Qiang;Liu Shao-Gang;Yang Jin-Shui(School of Mechanical and Electrical Engineering,Harbin Engineering University,Harbin 150000,China;Qingdao Innovation and Development Base,Harbin Engineering University,Qingdao 266000,China)

机构地区：[1]哈尔滨工程大学机电工程学院,哈尔滨150000 [2]哈尔滨工程大学,青岛创新发展基地,青岛266000

出　　处：《物理学报》2024年第16期152-161,共10页Acta Physica Sinica

基　　金：国家自然科学基金(批准号:52275098,52075111,51675111)资助的课题.

摘　　要：磁流变弹性体在振动控制领域展现出巨大的潜力,但其磁致力学性能的测量过程往往需投入较高的人工与时间成本.本研究旨在利用机器学习方法在小样本试验数据驱动下实现磁流变弹性体磁致力学性能的快速准确预测.基于加装可控磁场的剪切流变仪测试了磁流变弹性体(9种配比,4种加载频率)的磁致储能模量.每种样品取5个测试点作为训练集并搭建支持向量回归机器学习模型,从而表征磁流变弹性体的磁致储能模量.结果表明,相较于典型的理论模型,SVR模型仅使用5个样本点即可更准确表征磁流变弹性体磁致储能模量,相关系数高达0.998.另外,SVR模型训练时间仅为0.02 s,可显著加速磁流变弹性体表征的进程.更重要的是,SVR模型具有良好的泛化性,对于不同硅油配比和不同加载频率的磁流变弹性体预测结果的相关系数仍可达0.998以上.因此,机器学习模型可实现磁流变弹性体磁致储能模量的快速准确表征,为新型磁流变材料的研发提供参考.Magnetorheological elastomers(MREs)are smart materials with a wide range of applications,particularly in reducing vibrations and noise.Traditional methods of testing their magnetically-induced properties,although thorough,are labor-intensive and time-consuming.In this work,we introduce an innovative method that harnesses machine learning to rapidly characterize MREs by using a smallest dataset,thus simplifying the characterization process.Initially,12 types of MREs are prepared and tested on a shear rheometer with a controllable magnetic field.From these data,we strategically select five representative data points from each sample to form a training dataset.Using this dataset,we develop a support vector regression(SVR)model to characterize the magnetically-induced storage modulus of the MRE.The SVR model exhibits remarkable accuracy,with a correlation coefficient(R2)of 0.998 or higher,exceeding the precision of traditional models.The training time of this model is very brief,only 0.02 seconds,thus greatly accelerating the characterization speed of MRE.Moreover,the SVR model demonstrates strong generalization ability,maintaining a high correlation coefficient of 0.998 or greater even when silicone oil is added to the MREs or tested under various loading frequencies.In a word,the machine learning model not only accelerates the evaluation process but also provides a valuable reference for developing innovative MREs,marking a significant advancement in the field of smart materials research.

关 键 词：磁流变弹性体 支持向量回归 磁致储能模量 

分 类 号：O361.3[理学—流体力学] TP181[理学—力学]