人工神经网络模型辅助优化制备Sm-Co纳米颗粒  

Artificial Neural Network Model Assisted Optimization of Preparation of Sm-Co Nanoparticles

作　　者：权滢滢 白帆 董莹 康达庄 张天丽[1] 蒋成保[1] Quan Yingying;Bai Fan;Dong Ying;Kang Dazhuang;Zhang Tianli;Jiang Chengbao(School of Materials Science and Engineering,Beihang University,Beijing 100191,China;Suzhou Innovation Research Institute of Beihang University,Suzhou 215004,China)

机构地区：[1]北京航空航天大学材料科学与工程学院,北京100191 [2]北京航空航天大学苏州创新研究院,江苏苏州215011

出　　处：《稀有金属》2024年第3期305-316,共12页Chinese Journal of Rare Metals

基　　金：国家重点研发计划项目(2018YFB2003901);国家自然科学基金项目(91960101,52031001)资助。

摘　　要：液相合成与钙热还原结合制备Sm-Co纳米颗粒的方法,具有粒度易于调控的优势,但由于前驱体成分和工艺参数等因素都对合成产物的磁性能具有很大影响,依赖筛选式实验进行多影响因素的优化研究费时费力。将人工神经网络模型与Sm-Co纳米颗粒钙热还原制备实验相结合,以实验数据为基础,利用人工神经网络模型建立实验参数和纳米颗粒磁性能之间的关系,不断优化模型,分析实验参数对磁性能的影响,进而根据分析结果优化实验参数,达到准确预测Sm-Co纳米颗粒磁性能的目标。最后,采用优化参数进行实验制备,获得高磁性能Sm-Co纳米颗粒,并进行测试表征,验证模型的预测精度。将人工神经网络模型应用于快速优化Sm-Co纳米颗粒磁性能,解决制备过程中多因素参量优化问题,实现了对Sm-Co纳米颗粒磁性能的高精度预测,提高了性能优化效率,为更广泛的Sm-Co系永磁体性能优化开辟了新途径。Sm-Co permanent magnets have the advantages of high Curie temperature and high anisotropic field strength,Sm-Co nanoparticles have important applications in the fields of ultra-high-density magnetic recording,nanocomposites,and biomedicine.However,it is difficult to synthesis Sm-Co nanoparticles,and the mechanism of the influence of various components and processes on the magnetic properties of Sm-Co nanoparticles is not clear.Machine learning algorithm enable to quickly find the relationship between multi-factor parameters and the material properties of interest,so it is widely used in the field of materials.This paper adopted a method of calcium thermal reduction to synthesize Sm-Co nanoparticles.With machine learning algorithm,Sm-Co nanoparticle magnetic performance could be predicted.The model was used to analyze the influence of various factors on Sm-Co nanoparticle magnetic properties,and the composition of Sm-Co nanoparticles was optimized.Sm-Co nanoparticles were prepared according to the optimization results.Sm-Co nanoparticles were stably prepared by the inorganic method,and the composition parameters were changed through the experiments.The composition and process parameters of each sample were recorded,and the samples remanence(B_(r))and coercivity(H_(cj))were tested,and the sample data table was finally established,and a neural network model was built.The model structure with three layers was built:input layer,hidden layer and output layer,in which,the input layer was consist of the composition parameters,the output layer was composed with B_(r) and H_(cj).The test set was used on the trained model,and the initial model performance was obtained.To optimize the neural network model,according to the training and prediction results of the model,extreme values in the data were filtered out.By adjusting the model structure,the best fit model structure was set.B_(r) and H_(cj)of the data samples were stratified shuffle split separately,and different iterations of the test to fit B_(r) and H_(cj)of the samp

关 键 词：Sm-Co系永磁体 Sm-Co纳米颗粒 机器学习算法 成分优化 磁性能优化 

分 类 号：TM273[一般工业技术—材料科学与工程]