{"id":{"repo_id":"trento","oai_identifier":"oai:iris.unitn.it:11572/380529"},"canonical_url":"https://search.dev.ndltd.org/etd/trento/oai:iris.unitn.it:11572/380529","repository":{"repo_id":"trento","name":"Università degli Studi di Trento","base_url":"https://iris.unitn.it/oai/request"},"display":{"title":"Evolutionary Optimization of Neural Architectures for Remaining Useful Life Prediction","abstract":"Remaining useful life (RUL) predictions are a key enabler to achieving efficient maintenance in the context of Industry 4.0. Data-driven approaches, in particular employing deep neural networks (DNNs), have shown success in the RUL prediction task. However, although their architecture considerably affects performance, DNNs are usually handcrafted by human experts via a labor-intensive design process. To overcome this issue, we propose evolutionary neural architecture search (NAS) techniques that explore a search space using a genetic algorithm (GA). NAS automatically discovers the optimal architectures of neural networks for RUL predictions. Our GA allows an efficient search, finding high-quality solutions based on performance predictions which reduce the needed computational efforts for network training. In particular, first, we apply evolutionary computation to find the best architectures of deep and complex neural networks in terms of prediction accuracy. On the other side, we consider multi-objective optimization (MOO) of rather simple and fast neural networks to search for the best network architectures in terms of the trade-off between RUL prediction error and the number of trainable parameters, the latter being correlated with computational effort. In our experiments, we evaluate the performance of the found solutions on widely-used benchmark datasets, CMAPSS and N-CMAPSS. In comparison with the state-of-the-art, the obtained networks by our single objective NAS approach outperform other handcrafted recent DNNs in terms of prediction error, and the automatically designed networks by the multi-objective NAS approach provide comparable results with manually designed traditional DNNs in terms of the test RMSE, but the number of trainable parameters is considerably smaller and the training time is significantly shorter. Our results demonstrate that the neural networks whose architecture is optimized by evolutionary NAS techniques can be a useful tool to solve the RUL prediction task.","abstract_html":"Remaining useful life (RUL) predictions are a key enabler to achieving efficient maintenance in the context of Industry 4.0. Data-driven approaches, in particular employing deep neural networks (DNNs), have shown success in the RUL prediction task. However, although their architecture considerably affects performance, DNNs are usually handcrafted by human experts via a labor-intensive design process. To overcome this issue, we propose evolutionary neural architecture search (NAS) techniques that explore a search space using a genetic algorithm (GA). NAS automatically discovers the optimal architectures of neural networks for RUL predictions. Our GA allows an efficient search, finding high-quality solutions based on performance predictions which reduce the needed computational efforts for network training. In particular, first, we apply evolutionary computation to find the best architectures of deep and complex neural networks in terms of prediction accuracy. On the other side, we consider multi-objective optimization (MOO) of rather simple and fast neural networks to search for the best network architectures in terms of the trade-off between RUL prediction error and the number of trainable parameters, the latter being correlated with computational effort. In our experiments, we evaluate the performance of the found solutions on widely-used benchmark datasets, CMAPSS and N-CMAPSS. In comparison with the state-of-the-art, the obtained networks by our single objective NAS approach outperform other handcrafted recent DNNs in terms of prediction error, and the automatically designed networks by the multi-objective NAS approach provide comparable results with manually designed traditional DNNs in terms of the test RMSE, but the number of trainable parameters is considerably smaller and the training time is significantly shorter. Our results demonstrate that the neural networks whose architecture is optimized by evolutionary NAS techniques can be a useful tool to solve the RUL prediction task.","abstract_has_math":false,"creators":["Mo, Hyunho"],"institution":"Università degli studi di Trento","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Federico Lucca at BlueTensor","Iacca, Giovanni"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2023,"date_issued":"2023-06-19","date_published":"2023-06-19","updated_at":"2026-07-24T05:04:33Z","subjects":[],"languages":["eng"],"rights":["info:eu-repo/semantics/openAccess","license:Tutti i diritti riservati (All rights reserved)","license uri:iris.PRI01"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["http://dx.doi.org/10.15168/11572_380529","10.15168/11572_380529"],"render_values":[{"text":"http://dx.doi.org/10.15168/11572_380529","href":"http://dx.doi.org/10.15168/11572_380529","code":true},{"text":"10.15168/11572_380529","href":"https://doi.org/10.15168/11572_380529","code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/11572/380529","outbound_label":"Handle","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Federico Lucca at BlueTensor","Mo, Hyunho","Iacca, Giovanni"]},{"key":"dc:creator","label":"Author","values":["Mo, Hyunho"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2023-06-19"]},{"key":"dc:publisher","label":"Institution","values":["Università degli studi di Trento","place:TRENTO"]},{"key":"dc:relation","label":"Dc Relation","values":["firstpage:1","lastpage:119","numberofpages:119","alleditors:Federico Lucca at BlueTensor"]},{"key":"dc:type","label":"Dc Type","values":["info:eu-repo/semantics/doctoralThesis"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["info:eu-repo/semantics/openAccess","license:Tutti i diritti riservati (All rights reserved)","license uri:iris.PRI01"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://hdl.handle.net/11572/380529","http://dx.doi.org/10.15168/11572_380529","10.15168/11572_380529"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Remaining useful life (RUL) predictions are a key enabler to achieving efficient maintenance in the context of Industry 4.0. Data-driven approaches, in particular employing deep neural networks (DNNs), have shown success in the RUL prediction task. However, although their architecture considerably affects performance, DNNs are usually handcrafted by human experts via a labor-intensive design process. To overcome this issue, we propose evolutionary neural architecture search (NAS) techniques that explore a search space using a genetic algorithm (GA). NAS automatically discovers the optimal architectures of neural networks for RUL predictions. Our GA allows an efficient search, finding high-quality solutions based on performance predictions which reduce the needed computational efforts for network training. In particular, first, we apply evolutionary computation to find the best architectures of deep and complex neural networks in terms of prediction accuracy. On the other side, we consider multi-objective optimization (MOO) of rather simple and fast neural networks to search for the best network architectures in terms of the trade-off between RUL prediction error and the number of trainable parameters, the latter being correlated with computational effort. In our experiments, we evaluate the performance of the found solutions on widely-used benchmark datasets, CMAPSS and N-CMAPSS. In comparison with the state-of-the-art, the obtained networks by our single objective NAS approach outperform other handcrafted recent DNNs in terms of prediction error, and the automatically designed networks by the multi-objective NAS approach provide comparable results with manually designed traditional DNNs in terms of the test RMSE, but the number of trainable parameters is considerably smaller and the training time is significantly shorter. Our results demonstrate that the neural networks whose architecture is optimized by evolutionary NAS techniques can be a useful tool to solve the RUL prediction task."]},{"key":"dc:title","label":"Title","values":["Evolutionary Optimization of Neural Architectures for Remaining Useful Life Prediction"]}]}],"canonical_facts":{"dc:contributor":["Federico Lucca at BlueTensor","Mo, Hyunho","Iacca, Giovanni"],"dc:creator":["Mo, Hyunho"],"dc:date":["2023-06-19"],"dc:description":["Remaining useful life (RUL) predictions are a key enabler to achieving efficient maintenance in the context of Industry 4.0. Data-driven approaches, in particular employing deep neural networks (DNNs), have shown success in the RUL prediction task. However, although their architecture considerably affects performance, DNNs are usually handcrafted by human experts via a labor-intensive design process. To overcome this issue, we propose evolutionary neural architecture search (NAS) techniques that explore a search space using a genetic algorithm (GA). NAS automatically discovers the optimal architectures of neural networks for RUL predictions. Our GA allows an efficient search, finding high-quality solutions based on performance predictions which reduce the needed computational efforts for network training. In particular, first, we apply evolutionary computation to find the best architectures of deep and complex neural networks in terms of prediction accuracy. On the other side, we consider multi-objective optimization (MOO) of rather simple and fast neural networks to search for the best network architectures in terms of the trade-off between RUL prediction error and the number of trainable parameters, the latter being correlated with computational effort. In our experiments, we evaluate the performance of the found solutions on widely-used benchmark datasets, CMAPSS and N-CMAPSS. In comparison with the state-of-the-art, the obtained networks by our single objective NAS approach outperform other handcrafted recent DNNs in terms of prediction error, and the automatically designed networks by the multi-objective NAS approach provide comparable results with manually designed traditional DNNs in terms of the test RMSE, but the number of trainable parameters is considerably smaller and the training time is significantly shorter. Our results demonstrate that the neural networks whose architecture is optimized by evolutionary NAS techniques can be a useful tool to solve the RUL prediction task."],"dc:identifier":["https://hdl.handle.net/11572/380529","http://dx.doi.org/10.15168/11572_380529","10.15168/11572_380529"],"dc:language":["eng"],"dc:publisher":["Università degli studi di Trento","place:TRENTO"],"dc:relation":["firstpage:1","lastpage:119","numberofpages:119","alleditors:Federico Lucca at BlueTensor"],"dc:rights":["info:eu-repo/semantics/openAccess","license:Tutti i diritti riservati (All rights reserved)","license uri:iris.PRI01"],"dc:title":["Evolutionary Optimization of Neural Architectures for Remaining Useful Life Prediction"],"dc:type":["info:eu-repo/semantics/doctoralThesis"]},"updated_at":"2026-07-24T05:04:33Z"}