Moritz Flaschel , Paul Steinmann , Laura De Lorenzis , Ellen Kuhl
{"title":"Convex neural networks learn generalized standard material models","authors":"Moritz Flaschel , Paul Steinmann , Laura De Lorenzis , Ellen Kuhl","doi":"10.1016/j.jmps.2025.106103","DOIUrl":null,"url":null,"abstract":"<div><div>We propose Generalized Standard Material Networks, a machine learning framework based on convex neural networks for learning the mechanical behavior of generalized standard materials. The theory of these materials postulates the existence of two thermodynamic potentials, the Helmholtz free energy density and the dissipation rate density potential, which alone determine the constitutive material response with guaranteed thermodynamic consistency. We parameterize the two potentials with two artificial neural networks and, due to a specifically designed network architecture, we satisfy by construction all the needed properties of the two potentials. Using automatic differentiation, an implicit time integration scheme and the Newton-Raphson method, we can thus describe a multitude of different material behaviors within a single unified overarching framework, including elastic, viscoelastic, plastic, and viscoplastic material responses with hardening. By probing our framework on the synthetic data generated by five benchmark material models, we demonstrate satisfactory prediction accuracy to unseen data and a high robustness to noise. In this context, we observe a non-uniqueness of thermodynamic potentials and discuss how this affects the results of the training process. Finally, we show that a carefully chosen number of internal variables strikes a balance between fitting accuracy and model complexity.</div></div>","PeriodicalId":17331,"journal":{"name":"Journal of The Mechanics and Physics of Solids","volume":"200 ","pages":"Article 106103"},"PeriodicalIF":5.0000,"publicationDate":"2025-03-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of The Mechanics and Physics of Solids","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0022509625000791","RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"MATERIALS SCIENCE, MULTIDISCIPLINARY","Score":null,"Total":0}
引用次数: 0
Abstract
We propose Generalized Standard Material Networks, a machine learning framework based on convex neural networks for learning the mechanical behavior of generalized standard materials. The theory of these materials postulates the existence of two thermodynamic potentials, the Helmholtz free energy density and the dissipation rate density potential, which alone determine the constitutive material response with guaranteed thermodynamic consistency. We parameterize the two potentials with two artificial neural networks and, due to a specifically designed network architecture, we satisfy by construction all the needed properties of the two potentials. Using automatic differentiation, an implicit time integration scheme and the Newton-Raphson method, we can thus describe a multitude of different material behaviors within a single unified overarching framework, including elastic, viscoelastic, plastic, and viscoplastic material responses with hardening. By probing our framework on the synthetic data generated by five benchmark material models, we demonstrate satisfactory prediction accuracy to unseen data and a high robustness to noise. In this context, we observe a non-uniqueness of thermodynamic potentials and discuss how this affects the results of the training process. Finally, we show that a carefully chosen number of internal variables strikes a balance between fitting accuracy and model complexity.
期刊介绍:
The aim of Journal of The Mechanics and Physics of Solids is to publish research of the highest quality and of lasting significance on the mechanics of solids. The scope is broad, from fundamental concepts in mechanics to the analysis of novel phenomena and applications. Solids are interpreted broadly to include both hard and soft materials as well as natural and synthetic structures. The approach can be theoretical, experimental or computational.This research activity sits within engineering science and the allied areas of applied mathematics, materials science, bio-mechanics, applied physics, and geophysics.
The Journal was founded in 1952 by Rodney Hill, who was its Editor-in-Chief until 1968. The topics of interest to the Journal evolve with developments in the subject but its basic ethos remains the same: to publish research of the highest quality relating to the mechanics of solids. Thus, emphasis is placed on the development of fundamental concepts of mechanics and novel applications of these concepts based on theoretical, experimental or computational approaches, drawing upon the various branches of engineering science and the allied areas within applied mathematics, materials science, structural engineering, applied physics, and geophysics.
The main purpose of the Journal is to foster scientific understanding of the processes of deformation and mechanical failure of all solid materials, both technological and natural, and the connections between these processes and their underlying physical mechanisms. In this sense, the content of the Journal should reflect the current state of the discipline in analysis, experimental observation, and numerical simulation. In the interest of achieving this goal, authors are encouraged to consider the significance of their contributions for the field of mechanics and the implications of their results, in addition to describing the details of their work.