No Cover Image

E-Thesis 546 views 53 downloads

Neural Network Based Constitutive Modelling of Inelastic Materials / REEM ALHAYKI

Swansea University Author: REEM ALHAYKI

DOI (Published version): 10.23889/SUThesis.66509

Abstract

New materials that are associated with novel micro-structures are becoming available at a rapid speed. The mechanical characteristics of solid materials are studied using constitutive models. The formulation of constitutive relations is challenging and may be related with the limiting hypothesis. Ma...

Full description

Published: Swansea University, Wales, UK 2024
Institution: Swansea University
Degree level: Doctoral
Degree name: Ph.D
Supervisor: Dettmer, W., and Peric, D.
URI: https://cronfa.swan.ac.uk/Record/cronfa66509
Tags: Add Tag
No Tags, Be the first to tag this record!
first_indexed 2024-05-24T13:51:04Z
last_indexed 2024-05-24T13:51:04Z
id cronfa66509
recordtype RisThesis
fullrecord <?xml version="1.0" encoding="utf-8"?><rfc1807 xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema"><bib-version>v2</bib-version><id>66509</id><entry>2024-05-24</entry><title>Neural Network Based Constitutive Modelling of Inelastic Materials</title><swanseaauthors><author><sid>fb117ddf7b981f2040bb56559ec6e6c5</sid><firstname>REEM</firstname><surname>ALHAYKI</surname><name>REEM ALHAYKI</name><active>true</active><ethesisStudent>false</ethesisStudent></author></swanseaauthors><date>2024-05-24</date><abstract>New materials that are associated with novel micro-structures are becoming available at a rapid speed. The mechanical characteristics of solid materials are studied using constitutive models. The formulation of constitutive relations is challenging and may be related with the limiting hypothesis. Machine learning approaches have been widely used to simulate material behaviour in recent years. This study presents a neural network-based approach to reproduce the complex constitutive relations of solid materials including the modelling of rate-independent, elastic or inelastic, isotropic or anisotropic material behaviours. The approach can follow a history-based or an internal variables-based strategy. The network is developed using a generic internal variable formalism, with the number of internal variables determined by the nature of the problem and the degree of accuracy desired. It is demonstrated that the history-based and internal variable based techniques accurately describe the von Mises elastoplastic material model in uniaxial stress condition.However, a thorough examination indicates that the internal variable-based strategy is most suited.In this work, the network is trained on data sequences of strains and corresponding stresses that can be generated with physical experiments or numerical simulations based on Representative Volume Element (RVE). The used training strategy is based on gradient-free optimisation. The model performance is evaluated on different numerical examples including uniaxial damage mechanics and multiaxial elastoplasticity under plane strain conditions. The trained model is compared against the corresponding constitutive model to evaluate the accuracy. An efficient criterion for the verification of thermodynamic consistency is proposed and applied to the trained stress update models.The proposed strategy is tested with numerical examples that represent relations of porous solid materials. The training data are generated with numerical multi-scale homogenisation based on an RVE composed of the von Mises elastoplastic matrix with an arbitrary void volume fraction. The RVE simulation was performed by enforcing appropriate RVE boundary conditions, and the stress responses are computed for various macro strain data sequences to generate different loading paths. The neural network based stress update algorithm is trained and validated. The obtained results show the ability of the proposed procedure in describing the material stress/strain relationship with high accuracy. In addition, plots of stress paths in the hydrostaticdeviatoric space show accurate results when compared against the corresponding Gurson model for porous elastoplastic material.</abstract><type>E-Thesis</type><journal/><volume/><journalNumber/><paginationStart/><paginationEnd/><publisher/><placeOfPublication>Swansea University, Wales, UK</placeOfPublication><isbnPrint/><isbnElectronic/><issnPrint/><issnElectronic/><keywords>Data driven computational mechanics, Neural network, Constitutive modelling, Elasto-plasticity, Damage mechanics.</keywords><publishedDay>27</publishedDay><publishedMonth>4</publishedMonth><publishedYear>2024</publishedYear><publishedDate>2024-04-27</publishedDate><doi>10.23889/SUThesis.66509</doi><url/><notes/><college>COLLEGE NANME</college><CollegeCode>COLLEGE CODE</CollegeCode><institution>Swansea University</institution><supervisor>Dettmer, W., and Peric, D.</supervisor><degreelevel>Doctoral</degreelevel><degreename>Ph.D</degreename><apcterm/><funders/><projectreference/><lastEdited>2024-06-24T16:24:16.1372066</lastEdited><Created>2024-05-24T13:54:38.2256451</Created><path><level id="1">Faculty of Science and Engineering</level><level id="2">School of Engineering and Applied Sciences - Materials Science and Engineering</level></path><authors><author><firstname>REEM</firstname><surname>ALHAYKI</surname><order>1</order></author></authors><documents><document><filename>66509__30457__490d1a7224dc4a56bac8f09010ef7bb3.pdf</filename><originalFilename>2024_Alhayki_R.final.66509.pdf</originalFilename><uploaded>2024-05-24T14:46:47.0374454</uploaded><type>Output</type><contentLength>49585472</contentLength><contentType>application/pdf</contentType><version>E-Thesis – open access</version><cronfaStatus>true</cronfaStatus><documentNotes>Copyright: The Author, Reem Allhayki, 2024.</documentNotes><copyrightCorrect>true</copyrightCorrect><language>eng</language></document></documents><OutputDurs/></rfc1807>
spelling v2 66509 2024-05-24 Neural Network Based Constitutive Modelling of Inelastic Materials fb117ddf7b981f2040bb56559ec6e6c5 REEM ALHAYKI REEM ALHAYKI true false 2024-05-24 New materials that are associated with novel micro-structures are becoming available at a rapid speed. The mechanical characteristics of solid materials are studied using constitutive models. The formulation of constitutive relations is challenging and may be related with the limiting hypothesis. Machine learning approaches have been widely used to simulate material behaviour in recent years. This study presents a neural network-based approach to reproduce the complex constitutive relations of solid materials including the modelling of rate-independent, elastic or inelastic, isotropic or anisotropic material behaviours. The approach can follow a history-based or an internal variables-based strategy. The network is developed using a generic internal variable formalism, with the number of internal variables determined by the nature of the problem and the degree of accuracy desired. It is demonstrated that the history-based and internal variable based techniques accurately describe the von Mises elastoplastic material model in uniaxial stress condition.However, a thorough examination indicates that the internal variable-based strategy is most suited.In this work, the network is trained on data sequences of strains and corresponding stresses that can be generated with physical experiments or numerical simulations based on Representative Volume Element (RVE). The used training strategy is based on gradient-free optimisation. The model performance is evaluated on different numerical examples including uniaxial damage mechanics and multiaxial elastoplasticity under plane strain conditions. The trained model is compared against the corresponding constitutive model to evaluate the accuracy. An efficient criterion for the verification of thermodynamic consistency is proposed and applied to the trained stress update models.The proposed strategy is tested with numerical examples that represent relations of porous solid materials. The training data are generated with numerical multi-scale homogenisation based on an RVE composed of the von Mises elastoplastic matrix with an arbitrary void volume fraction. The RVE simulation was performed by enforcing appropriate RVE boundary conditions, and the stress responses are computed for various macro strain data sequences to generate different loading paths. The neural network based stress update algorithm is trained and validated. The obtained results show the ability of the proposed procedure in describing the material stress/strain relationship with high accuracy. In addition, plots of stress paths in the hydrostaticdeviatoric space show accurate results when compared against the corresponding Gurson model for porous elastoplastic material. E-Thesis Swansea University, Wales, UK Data driven computational mechanics, Neural network, Constitutive modelling, Elasto-plasticity, Damage mechanics. 27 4 2024 2024-04-27 10.23889/SUThesis.66509 COLLEGE NANME COLLEGE CODE Swansea University Dettmer, W., and Peric, D. Doctoral Ph.D 2024-06-24T16:24:16.1372066 2024-05-24T13:54:38.2256451 Faculty of Science and Engineering School of Engineering and Applied Sciences - Materials Science and Engineering REEM ALHAYKI 1 66509__30457__490d1a7224dc4a56bac8f09010ef7bb3.pdf 2024_Alhayki_R.final.66509.pdf 2024-05-24T14:46:47.0374454 Output 49585472 application/pdf E-Thesis – open access true Copyright: The Author, Reem Allhayki, 2024. true eng
title Neural Network Based Constitutive Modelling of Inelastic Materials
spellingShingle Neural Network Based Constitutive Modelling of Inelastic Materials
REEM ALHAYKI
title_short Neural Network Based Constitutive Modelling of Inelastic Materials
title_full Neural Network Based Constitutive Modelling of Inelastic Materials
title_fullStr Neural Network Based Constitutive Modelling of Inelastic Materials
title_full_unstemmed Neural Network Based Constitutive Modelling of Inelastic Materials
title_sort Neural Network Based Constitutive Modelling of Inelastic Materials
author_id_str_mv fb117ddf7b981f2040bb56559ec6e6c5
author_id_fullname_str_mv fb117ddf7b981f2040bb56559ec6e6c5_***_REEM ALHAYKI
author REEM ALHAYKI
author2 REEM ALHAYKI
format E-Thesis
publishDate 2024
institution Swansea University
doi_str_mv 10.23889/SUThesis.66509
college_str Faculty of Science and Engineering
hierarchytype
hierarchy_top_id facultyofscienceandengineering
hierarchy_top_title Faculty of Science and Engineering
hierarchy_parent_id facultyofscienceandengineering
hierarchy_parent_title Faculty of Science and Engineering
department_str School of Engineering and Applied Sciences - Materials Science and Engineering{{{_:::_}}}Faculty of Science and Engineering{{{_:::_}}}School of Engineering and Applied Sciences - Materials Science and Engineering
document_store_str 1
active_str 0
description New materials that are associated with novel micro-structures are becoming available at a rapid speed. The mechanical characteristics of solid materials are studied using constitutive models. The formulation of constitutive relations is challenging and may be related with the limiting hypothesis. Machine learning approaches have been widely used to simulate material behaviour in recent years. This study presents a neural network-based approach to reproduce the complex constitutive relations of solid materials including the modelling of rate-independent, elastic or inelastic, isotropic or anisotropic material behaviours. The approach can follow a history-based or an internal variables-based strategy. The network is developed using a generic internal variable formalism, with the number of internal variables determined by the nature of the problem and the degree of accuracy desired. It is demonstrated that the history-based and internal variable based techniques accurately describe the von Mises elastoplastic material model in uniaxial stress condition.However, a thorough examination indicates that the internal variable-based strategy is most suited.In this work, the network is trained on data sequences of strains and corresponding stresses that can be generated with physical experiments or numerical simulations based on Representative Volume Element (RVE). The used training strategy is based on gradient-free optimisation. The model performance is evaluated on different numerical examples including uniaxial damage mechanics and multiaxial elastoplasticity under plane strain conditions. The trained model is compared against the corresponding constitutive model to evaluate the accuracy. An efficient criterion for the verification of thermodynamic consistency is proposed and applied to the trained stress update models.The proposed strategy is tested with numerical examples that represent relations of porous solid materials. The training data are generated with numerical multi-scale homogenisation based on an RVE composed of the von Mises elastoplastic matrix with an arbitrary void volume fraction. The RVE simulation was performed by enforcing appropriate RVE boundary conditions, and the stress responses are computed for various macro strain data sequences to generate different loading paths. The neural network based stress update algorithm is trained and validated. The obtained results show the ability of the proposed procedure in describing the material stress/strain relationship with high accuracy. In addition, plots of stress paths in the hydrostaticdeviatoric space show accurate results when compared against the corresponding Gurson model for porous elastoplastic material.
published_date 2024-04-27T16:24:14Z
_version_ 1802756585658777600
score 11.035349