{"id":{"repo_id":"nus","oai_identifier":"oai:scholarbank.nus.edu.sg:10635/233976"},"canonical_url":"https://search.dev.ndltd.org/etd/nus/oai:scholarbank.nus.edu.sg:10635/233976","repository":{"repo_id":"nus","name":"National University of Singapore","base_url":"https://scholarbank.nus.edu.sg/oai/request"},"display":{"title":"MACHINE LEARNING FOR CONSTITUTIVE MODELLING","abstract":"Machine learning (ML) techniques have been applied as surrogate models in the micro scale of multi-scale analysis to reduce the computational cost. These surrogate models provide the constitutive law of the micro scale simulations after a training process, called off-line learning. For history dependent materials, the recurrent neural network (RNN) is adopted for the training of path dependent mechanical responses, due to its advanced learning ability for history dependent behavior. However, the huge data generation time of RNN surrogate model limits the off-line learning efficiency. This thesis aims to address this limitation. Two sequential training strategies are considered. (i) An Error Correction (EC) strategy, where a reference surrogate model is first trained using limited data, and later refined using another trained model on the error correction. (ii) A Knowledge Transfer (KT) strategy, where a source surrogate model is first trained based on a simplified RVE, followed by a refinement of surrogate model based on target data generated with detailed RVEs. This thesis shows that the proposed two strategies, EC and KT, can improve the off-line learning efficiency.","abstract_html":"Machine learning (ML) techniques have been applied as surrogate models in the micro scale of multi-scale analysis to reduce the computational cost. These surrogate models provide the constitutive law of the micro scale simulations after a training process, called off-line learning. For history dependent materials, the recurrent neural network (RNN) is adopted for the training of path dependent mechanical responses, due to its advanced learning ability for history dependent behavior. However, the huge data generation time of RNN surrogate model limits the off-line learning efficiency. This thesis aims to address this limitation. Two sequential training strategies are considered. (i) An Error Correction (EC) strategy, where a reference surrogate model is first trained using limited data, and later refined using another trained model on the error correction. (ii) A Knowledge Transfer (KT) strategy, where a source surrogate model is first trained based on a simplified RVE, followed by a refinement of surrogate model based on target data generated with detailed RVEs. This thesis shows that the proposed two strategies, EC and KT, can improve the off-line learning efficiency.","abstract_has_math":false,"creators":["DENG HAOXIANG"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2022,"date_issued":"2022-07-12","date_published":"2022-07-12","updated_at":"2026-07-24T03:30:47Z","subjects":["Recurrent neural network, error correction method, knowledge transfer, computational homogenization, multi-scale modelling, history-dependent behavior"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":null,"outbound_label":null,"outbound_source":null},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["DENG HAOXIANG"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2022-07-12"]},{"key":"dc:relation.isreferencedby","label":"Dc Relation Isreferencedby","values":["https://scholarbank.nus.edu.sg/handle/10635/233976"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Recurrent neural network, error correction method, knowledge transfer, computational homogenization, multi-scale modelling, history-dependent behavior"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://scholarbank.nus.edu.sg/bitstreams/ff916851-801b-461f-908e-31e8cdddcb32/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Machine learning (ML) techniques have been applied as surrogate models in the micro scale of multi-scale analysis to reduce the computational cost. These surrogate models provide the constitutive law of the micro scale simulations after a training process, called off-line learning. For history dependent materials, the recurrent neural network (RNN) is adopted for the training of path dependent mechanical responses, due to its advanced learning ability for history dependent behavior. However, the huge data generation time of RNN surrogate model limits the off-line learning efficiency. This thesis aims to address this limitation. Two sequential training strategies are considered. (i) An Error Correction (EC) strategy, where a reference surrogate model is first trained using limited data, and later refined using another trained model on the error correction. (ii) A Knowledge Transfer (KT) strategy, where a source surrogate model is first trained based on a simplified RVE, followed by a refinement of surrogate model based on target data generated with detailed RVEs. This thesis shows that the proposed two strategies, EC and KT, can improve the off-line learning efficiency."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["43d5716c6aa317b1a2d5177ea81a0ad6","1c4e80f2881658bb0064c2f49005a2ea"]},{"key":"dc:title","label":"Title","values":["MACHINE LEARNING FOR CONSTITUTIVE MODELLING"]}]}],"canonical_facts":{"dc:creator":["DENG HAOXIANG"],"dc:date.issued":["2022-07-12"],"dc:description.abstract":["Machine learning (ML) techniques have been applied as surrogate models in the micro scale of multi-scale analysis to reduce the computational cost. These surrogate models provide the constitutive law of the micro scale simulations after a training process, called off-line learning. For history dependent materials, the recurrent neural network (RNN) is adopted for the training of path dependent mechanical responses, due to its advanced learning ability for history dependent behavior. However, the huge data generation time of RNN surrogate model limits the off-line learning efficiency. This thesis aims to address this limitation. Two sequential training strategies are considered. (i) An Error Correction (EC) strategy, where a reference surrogate model is first trained using limited data, and later refined using another trained model on the error correction. (ii) A Knowledge Transfer (KT) strategy, where a source surrogate model is first trained based on a simplified RVE, followed by a refinement of surrogate model based on target data generated with detailed RVEs. This thesis shows that the proposed two strategies, EC and KT, can improve the off-line learning efficiency."],"dc:format.checksum.md5":["43d5716c6aa317b1a2d5177ea81a0ad6","1c4e80f2881658bb0064c2f49005a2ea"],"dc:identifier.uri":["https://scholarbank.nus.edu.sg/bitstreams/ff916851-801b-461f-908e-31e8cdddcb32/download"],"dc:relation.isreferencedby":["https://scholarbank.nus.edu.sg/handle/10635/233976"],"dc:subject":["Recurrent neural network, error correction method, knowledge transfer, computational homogenization, multi-scale modelling, history-dependent behavior"],"dc:title":["MACHINE LEARNING FOR CONSTITUTIVE MODELLING"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T03:30:47Z"}