Nova estratanãgia coloca a evolução de estruturas microsca³picas no caminho certo
As estruturas e propriedades microsca³picas dos materiais estãointimamente ligadas e personaliza¡-las éum desafio. Os engenheiros da Rice University estãodeterminados a simplificar o processo por meio do aprendizado de ma¡quina.
![](data:image/jpeg;base64,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)
Engenheiros da Rice University e do Lawrence Livermore National Laboratory estãousando redes neurais para acelerar a previsão de como as microestruturas dos materiais evoluem. Este exemplo prevaª o crescimento de cristal dendratico semelhante a um floco de neve. Crédito: Mesoscale Materials Science Group / Rice University
As estruturas e propriedades microsca³picas dos materiais estãointimamente ligadas e personaliza¡-las éum desafio. Os engenheiros da Rice University estãodeterminados a simplificar o processo por meio do aprendizado de ma¡quina.
Para esse fim, o laboratório Rice do cientista de materiais Ming Tang, em colaboração com o fasico Fei Zhou no Laborata³rio Nacional Lawrence Livermore, introduziu uma técnica para prever a evolução das microestruturas - caracteristicas estruturais entre 10 nana´metros e 100 macrons - em materiais.
Seu artigo de acesso aberto no peria³dico Cell Press Patterns mostra como as redes neurais (modelos de computador que imitam os neura´nios do cérebro) podem se treinar para prever como uma estrutura crescera¡ sob um determinado ambiente, como um floco de neve se forma a partir da umidade na natureza.
Na verdade, estruturas cristalinas dendraticas semelhantes a flocos de neve foram um dos exemplos que o laboratório usou em seu estudo de prova de conceito.
"Na ciência de materiais moderna, éamplamente aceito que a microestrutura muitas vezes desempenha um papel crítico no controle das propriedades de um material", disse Tang. "Vocaª não quer apenas controlar como os a¡tomos são organizados nas redes, mas também a aparaªncia da microestrutura, para ter um bom desempenho e atémesmo novas funcionalidades.
"O Santo Graal de projetar materiais éser capaz de prever como uma microestrutura mudara¡ sob determinadas condições, se nosa aquecemos ou aplicamos estresse ou algum outro tipo de estimulação", disse ele.
Tang trabalhou para refinar a previsão da microestrutura durante toda a sua carreira, mas disse que a abordagem tradicional baseada em equações enfrenta desafios significativos para permitir que os cientistas acompanhem a demanda por novos materiais.
"O tremendo progresso no aprendizado de ma¡quina encorajou Fei em Lawrence Livermore e nosa ver se poderaamos aplica¡-lo aos materiais", disse ele.
Felizmente, havia muitos dados do manãtodo tradicional para ajudar a treinar as redes neurais da equipe, que visualizam a evolução inicial das microestruturas para prever a próxima etapa, a próxima e assim por diante.
"a‰ nisso que as ma¡quinas são boas, ver a correlação de uma forma muito complexa que a mente humana não écapaz de fazer", disse Tang. "Tiramos vantagem disso."
Os pesquisadores testaram suas redes neurais em quatro tipos distintos de microestrutura: propagação de ondas planas, crescimento de gra£os, decomposição espinodal e crescimento de cristal dendratico.
Â
Em cada teste, as redes foram alimentadas entre 1.000 e 2.000 conjuntos de 20 imagens sucessivas ilustrando a evolução da microestrutura de um material conforme previsto pelas equações. Depois de aprender as regras de evolução a partir desses dados, a rede recebeu de 1 a 10 imagens para prever os pra³ximos 50 a 200 quadros, e geralmente o fazia em segundos.
As vantagens da nova técnica rapidamente ficaram claras: as redes neurais, alimentadas por processadores gra¡ficos, aceleraram os ca¡lculos em até718 vezes para o crescimento de gra£os, em comparação com o algoritmo anterior. Quando executados em um processador central padra£o, eles ainda eram até87 vezes mais rápidos do que o manãtodo antigo. A previsão de outros tipos de evolução da microestrutura mostrou aumentos de velocidade semelhantes, embora não tão drama¡ticos.
Comparações com imagens do manãtodo tradicional de simulação provaram que as previsaµes estavam corretas, disse Tang. “Com base nisso, vemos como podemos atualizar os parametros para tornar a previsão cada vez mais precisaâ€, disse. "Então, podemos usar essas previsaµes para ajudar a projetar materiais que não vimos antes.
"Outro benefacio éque ele écapaz de fazer previsaµes mesmo quando não sabemos tudo sobre as propriedades do material em um sistema", disse Tang. "Nãopoderaamos fazer isso com o manãtodo baseado em equações, que precisa conhecer todos os valores dos parametros nas equações para realizar simulações."
Tang disse que a eficiência de computação das redes neurais pode acelerar o desenvolvimento de novos materiais. Ele espera que isso seja útil no projeto contanuo de seu laboratório de baterias mais eficientes. "Estamos pensando em novas estruturas tridimensionais que ajudara£o a carregar e descarregar baterias muito mais rápido do que temos agora", disse Tang. "Este éum problema de otimização perfeito para nossa nova abordagem."