文档名：机器学习在晶体生长中的应用研究进展
摘要：制备大尺寸优质晶体是材料领域的重要课题,当前晶体生长仍十分依赖于技术人员的生长经验,耗时长,成本高,导致研制进度十分缓慢.随着计算机科学的发展,机器学习有望加快大尺寸晶体生长工艺的探索速度.本文概述了机器学习在晶体生长中的应用,重点介绍了进化算法和神经网络算法,并概括了国内外相关领域的研究成果.在晶体结构预测方面,已成功开发出晶体结构预测软件,如CrystalStructurePredictionNetwork(CRYSPNet)、UniversalStructurePredictor:EvolutionaryXtallography(USPEX)和CrystalstructureAnaLYsisbyParticleSwarmOptimization(CALYPSO).在晶体生长条件优化方面,机器学习已成功用于铸锭炉热场优化、SiC、YAG、Si单晶温场以及生长工艺参数优化,取得了显著的效果.在晶体生长控制方法方面,机器学习已成功改进了激光晶体、Si单晶的直径控制,并在Si单晶晶体质量控制以及模型的自适应控制方法方面也取得了一定的进展.机器学习具有高精度、高效率的特点,可避开复杂的物理化学过程,通过从大量数据中挖掘信息和规律,实现晶体结构预测、晶体生长条件优化以及晶体生长精准控制.但机器学习在晶体生长中的应用也面临诸多挑战,包括大尺寸多元复杂体系的晶体结构预测、模型物理解释和数据处理、晶体生长过程的物理机制和控制规律以及多因素的耦合和复杂系统的模拟等.
Abstract：Withrecentdevelopmentonvariousfunctionalcrystalmaterialsandlarge-sizecrystalpreparationtechnologies,therelianceonexperienceandtechnologicalbottlenecksintraditionalcrystalgrowthbecomesincreasinglyprominent.Tomeetthegrowingtheoreticalandtechnicaldemands,somemethodsandtechnologiesforcrystalgrowtharedeveloped.Also,thedevelopmentofcomputerscienceandmachinelearningtechnologiesprovidessomeopportunitiesinthisfield.Machinelearning,throughtheanalysisofvastamountsofdata,canautomaticallyextracttheunderlyingknowledgeandpatterns,therebyenablingapredictionofcrystalstructuresandanoptimization/controlofthecrystalgrowthprocess.Crystalstructurepredictioninvolvesdeterminingthemicrostructureofacrystalundergivenchemicalcompositions.Traditionally,thestructureofcrystalscanbejustdeterminedthroughtherelatedexperiments.However,theexperimentalmethodsaretime-consumingandcostlyprocesses.Incontrast,machinelearningmethodscanlearnfromthecrystalstructuraldataandpredictthestructureofcrystalswithouttheexperiments.AseriesofcrystalstructurepredictionsoftwaresnamedCRYSPNet,USPEX,andCALYPSOaredeveloped.Conventionalmethodsofoptimizinggrowthconditionsusuallyrequireextensiveexperienceandexperimentaltrial-and-error.Machinelearning,viaanalyzingtheexistingdataoncrystalgrowth,canpredicttheoptimalparametercombinations,guidetheselectionofparametersinactualproductionandacceleratetheoptimizationprocess.ComparedtoconventionalexperimentsandCFDsimulations,machinelearningoffersfasterandmoreaccuratepredictions.Forinstance,modelconstructionbasedonneuralnetworksisapproximately107timesfasterthantheCFDsimulation.Theapplicationofsuchnoveltechnologiesincrystalgrowthcanpromotetheresearchanddevelopmentintherelatedfields.Thecrystalgrowthprocessiscomplexandvariable,involvingtheinteractionofmultiplefactors.Conventionalcontrolmethodslargelyrelyonexperienceandactualoperations.However,withtheapplicationanddevelopmentofmachinelearningandautomaticcontrolmethods,crystalgrowthcontrolisnolongerlimitedtosubjectiveandempiricaljudgments,butcanleveragethecapabilitiesofcomputeralgorithmsanddataanalysistoachievemoreaccurateandprecisecontrolofcrystalgrowth.Machinelearningviautilizinglargedatasetstotrainandoptimizemachinelearningmodelscanpredictanddeterminethedynamicchangesincrystalgrowth,andtimelyadjustandcontrol,therebyachievinganeffectivecontroloverthecrystalgrowthprocess.SummaryandprospectsThisreviewprovidedabriefsummaryoftheresearchprogressintheapplicationofmachinelearningtocrystalgrowth,anddiscussedmainlythreeaspects,i.e.,crystalstructureprediction,optimizationofcrystalgrowthconditions,andmethodsofcontrollingcrystalgrowth.Intermsofcrystalstructureprediction,machinelearningachievessignificantresults,buttheaccuracyandcomputationalefficiencystillneedtobeimproved,especiallyforlarge-size,multi-componentcomplexsystems.Inthefuture,itisnecessarytofurtherimproveanddevelopalgorithmstoincreasetheaccuracyandcomputationalefficiencyofpredictionmodels.Regardingtheoptimizationofcrystalgrowthconditions,machinelearningmethodscanpredicttheoptimumparametercombinations,speedinguptheoptimizationprocess.However,therearestillsomeissueswithdataaccuracy,processcomplexity,andmodelinterpretabilitythatneedtobeaddressed.Futureworkshouldinvolvetheintegrationofmoremachinelearningalgorithms,combinedwiththeoreticalandpracticalresearch,todevelopreliableandinterpretablemodels.Intermsofmethodsforcrystalgrowthcontrol,machinelearningalgorithmscanpreciselycontrolthecrystalgrowthprocess,improvingthestabilityandqualityofcrystalgrowth.However,therearestillsomechallengessuchasin-depthstudiesonthephysicalmechanismsofcrystalgrowth,thelawsofcontrol,andthesimulationofcomplexsystemswithmultiplecoupledfactors.Inthefuture,itisnecessarytocombineadvancedmachinelearningalgorithmsandoptimizationmethodstoenhancethesimulationofmulti-factorcouplingandcomplexsystems,furtherimprovingthecontrolcapabilityofcrystalgrowth.
作者：杨明亮 王瑞仙 孙贵花 王小飞 窦仁勤 何異 张庆礼 Author：YANGMingliang WANGRuixian SUNGuihua WANGXiaofei DOURenqin HEYi ZHANGQingli
作者单位：中国科学院合肥物质科学研究院,安徽光学精密机械研究所光子器件与材料安徽省重点实验室,合肥230031;中国科学技术大学,合肥230026;先进激光技术安徽省实验室,合肥230037中国科学院合肥物质科学研究院,安徽光学精密机械研究所光子器件与材料安徽省重点实验室,合肥230031;先进激光技术安徽省实验室,合肥230037
刊名：硅酸盐学报 ISTICEIPKU
Journal：JournaloftheChineseCeramicSociety
年，卷(期)：2024, 52(7)
分类号：TP181O781
关键词：计算机科学 机器学习 晶体生长 结构预测
Keywords：computerscience machinelearning crystalgrowth structureprediction
机标分类号：OO782TP391.41
在线出版日期：2024年7月24日
基金项目：国家重点研发计划，国家自然科学基金，国家自然科学基金，中国科学院青年创新促进会项目，安徽省实验室重点基金机器学习在晶体生长中的应用研究进展[
期刊论文] 硅酸盐学报--2024, 52(7)杨明亮 王瑞仙 孙贵花 王小飞 窦仁勤 何異 张庆礼制备大尺寸优质晶体是材料领域的重要课题,当前晶体生长仍十分依赖于技术人员的生长经验,耗时长,成本高,导致研制进度十分缓慢.随着计算机科学的发展,机器学习有望加快大尺寸晶体生长工艺的探索速度.本文概述了机器学习...参考文献和引证文献
参考文献
引证文献
本文读者也读过
相似文献
相关博文
机器学习在晶体生长中的应用研究进展 Research Progress on the Application of Machine Learning in Crystal Growth
机器学习在晶体生长中的应用研究进展.pdf
- 文件大小:
- 9.45 MB
- 下载次数:
- 60
-
高速下载
|
|
