{"id":{"repo_id":"cambridge","oai_identifier":"oai:www.repository.cam.ac.uk:1810/375809"},"canonical_url":"https://search.dev.ndltd.org/etd/cambridge/oai:www.repository.cam.ac.uk:1810/375809","repository":{"repo_id":"cambridge","name":"Cambridge University","base_url":"https://api.repository.cam.ac.uk/server/oai/request"},"display":{"title":"A Data-Centric Approach to Loss Mechanisms for Compressor Preliminary Design","abstract":"In the preliminary stages of design, low-fidelity models, which describe the underlying design space, are used to guide designers to optimal points in the design space. However, the majority of models currently used in compressor preliminary design are often limited by the human ability to find physical patterns in complex multidimensional data. Detecting patterns in complex multidimensional data is exactly the task for which machine learning and other statistical techniques have been developed. The aim of this thesis, therefore, is to try and use machine learning to augment the human ability to find physical relationships and to consequently develop a more physically accurate and general preliminary design system for compressors. In this work, a data-centric approach is proposed that utilises the strengths of machine learning to highlight patterns in complex data, while also utilising the strengths of human wisdom to ensure the models are physically meaningful. This approach is applied to datasets populated by large numbers of computational simulations for compressor and turbine blades. It is shown using 2D data that a new profile loss model can be developed which halves the error in preliminary design correlations and generalises well to turbines as well as compressors. The reason for these improvements is shown to be the result of a more physically accurate decomposition for mixing loss than current industry standard models. This new mixing loss model utilises flow properties local to the wake and is shown to explain the limiting cases for the mixing of two flows, the mixing out of a trailing edge blockage and the form drag of an aerofoil. It is argued that the reason this model is more accurate and general than existing models is that it retains the physical form for the source of loss creation at the fundamental scale. The data-centric approach is also applied to a 3D RANS compressor linear repeating stage dataset and shown to help develop a model for the total loss in a compressor blade row that is composed of physically accurate models for each of the key loss sources in a blade row. This model is demonstrated to halve the error in loss predictions compared to industry standard preliminary design models, and capture the key sensitivities to trailing edge thickness, lean and sweep, not previously accounted for at the preliminary stages of design. In particular, new, more accurate and general models for tip loss and wake mixing loss are highlighted that have the same physical form as the source of loss creation. The resulting model is finally used to generate Smith Charts to explain how the optimal point on the Smith Chart is very sensitive to the choice of preliminary design parameters and how these parameters can be used to move the optimal point. In particular, it is highlighted that there is a need to consider 3D effects in the preliminary stages of design to move the design point to higher stage loading coefficients.","abstract_html":"In the preliminary stages of design, low-fidelity models, which describe the underlying design space, are used to guide designers to optimal points in the design space. However, the majority of models currently used in compressor preliminary design are often limited by the human ability to find physical patterns in complex multidimensional data. Detecting patterns in complex multidimensional data is exactly the task for which machine learning and other statistical techniques have been developed. The aim of this thesis, therefore, is to try and use machine learning to augment the human ability to find physical relationships and to consequently develop a more physically accurate and general preliminary design system for compressors. In this work, a data-centric approach is proposed that utilises the strengths of machine learning to highlight patterns in complex data, while also utilising the strengths of human wisdom to ensure the models are physically meaningful. This approach is applied to datasets populated by large numbers of computational simulations for compressor and turbine blades. It is shown using 2D data that a new profile loss model can be developed which halves the error in preliminary design correlations and generalises well to turbines as well as compressors. The reason for these improvements is shown to be the result of a more physically accurate decomposition for mixing loss than current industry standard models. This new mixing loss model utilises flow properties local to the wake and is shown to explain the limiting cases for the mixing of two flows, the mixing out of a trailing edge blockage and the form drag of an aerofoil. It is argued that the reason this model is more accurate and general than existing models is that it retains the physical form for the source of loss creation at the fundamental scale. The data-centric approach is also applied to a 3D RANS compressor linear repeating stage dataset and shown to help develop a model for the total loss in a compressor blade row that is composed of physically accurate models for each of the key loss sources in a blade row. This model is demonstrated to halve the error in loss predictions compared to industry standard preliminary design models, and capture the key sensitivities to trailing edge thickness, lean and sweep, not previously accounted for at the preliminary stages of design. In particular, new, more accurate and general models for tip loss and wake mixing loss are highlighted that have the same physical form as the source of loss creation. The resulting model is finally used to generate Smith Charts to explain how the optimal point on the Smith Chart is very sensitive to the choice of preliminary design parameters and how these parameters can be used to move the optimal point. In particular, it is highlighted that there is a need to consider 3D effects in the preliminary stages of design to move the design point to higher stage loading coefficients.","abstract_has_math":false,"creators":["Senior, Alistair"],"institution":"University of Cambridge","degree_name":"Doctor of Philosophy (PhD)","degree_level":"Doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Miller, Robert"],"committee_chairs":[],"committee_members":[],"year":2024,"date_issued":"2024-01-12","date_published":"2024-01-12","updated_at":"2026-07-22T22:24:17Z","subjects":["Machine Learning","Compressor","Loss","Data-Centric","Loss models","Loss modelling","Compressor Design"],"languages":["eng"],"rights":[],"rights_urls":["https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/a86f3055-bc86-4ded-ae00-8946e2bff32c/download","http://purl.org/NET/rdflicense/allrightsreserved"],"identifier_entries":[]},"links":{"outbound_url":"https://doi.org/10.17863/CAM.113330","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Miller, Robert"]},{"key":"dc:creator","label":"Author","values":["Senior, Alistair"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2024-01-12"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Cambridge"]},{"key":"dc:relation.isreferencedby.uri","label":"Dc Relation Isreferencedby URI","values":["https://www.repository.cam.ac.uk/handle/1810/375809"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["Doctoral"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["Doctor of Philosophy (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Machine Learning","Compressor","Loss","Data-Centric","Loss models","Loss modelling","Compressor Design"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/a86f3055-bc86-4ded-ae00-8946e2bff32c/download","http://purl.org/NET/rdflicense/allrightsreserved"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.17863/CAM.113330"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/024d32bc-6ce9-4ef1-b3c0-bc2c3259357b/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["In the preliminary stages of design, low-fidelity models, which describe the underlying design space, are used to guide designers to optimal points in the design space. However, the majority of models currently used in compressor preliminary design are often limited by the human ability to find physical patterns in complex multidimensional data. Detecting patterns in complex multidimensional data is exactly the task for which machine learning and other statistical techniques have been developed. The aim of this thesis, therefore, is to try and use machine learning to augment the human ability to find physical relationships and to consequently develop a more physically accurate and general preliminary design system for compressors. In this work, a data-centric approach is proposed that utilises the strengths of machine learning to highlight patterns in complex data, while also utilising the strengths of human wisdom to ensure the models are physically meaningful. This approach is applied to datasets populated by large numbers of computational simulations for compressor and turbine blades. It is shown using 2D data that a new profile loss model can be developed which halves the error in preliminary design correlations and generalises well to turbines as well as compressors. The reason for these improvements is shown to be the result of a more physically accurate decomposition for mixing loss than current industry standard models. This new mixing loss model utilises flow properties local to the wake and is shown to explain the limiting cases for the mixing of two flows, the mixing out of a trailing edge blockage and the form drag of an aerofoil. It is argued that the reason this model is more accurate and general than existing models is that it retains the physical form for the source of loss creation at the fundamental scale. The data-centric approach is also applied to a 3D RANS compressor linear repeating stage dataset and shown to help develop a model for the total loss in a compressor blade row that is composed of physically accurate models for each of the key loss sources in a blade row. This model is demonstrated to halve the error in loss predictions compared to industry standard preliminary design models, and capture the key sensitivities to trailing edge thickness, lean and sweep, not previously accounted for at the preliminary stages of design. In particular, new, more accurate and general models for tip loss and wake mixing loss are highlighted that have the same physical form as the source of loss creation. The resulting model is finally used to generate Smith Charts to explain how the optimal point on the Smith Chart is very sensitive to the choice of preliminary design parameters and how these parameters can be used to move the optimal point. In particular, it is highlighted that there is a need to consider 3D effects in the preliminary stages of design to move the design point to higher stage loading coefficients."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["69f299d3f009c35920629a51aa978205","87eda9de84448d1f82354d60eee3eb5f"]},{"key":"dc:title","label":"Title","values":["A Data-Centric Approach to Loss Mechanisms for Compressor Preliminary Design"]}]}],"canonical_facts":{"dc:contributor.advisor":["Miller, Robert"],"dc:creator":["Senior, Alistair"],"dc:date.issued":["2024-01-12"],"dc:description.abstract":["In the preliminary stages of design, low-fidelity models, which describe the underlying design space, are used to guide designers to optimal points in the design space. However, the majority of models currently used in compressor preliminary design are often limited by the human ability to find physical patterns in complex multidimensional data. Detecting patterns in complex multidimensional data is exactly the task for which machine learning and other statistical techniques have been developed. The aim of this thesis, therefore, is to try and use machine learning to augment the human ability to find physical relationships and to consequently develop a more physically accurate and general preliminary design system for compressors. In this work, a data-centric approach is proposed that utilises the strengths of machine learning to highlight patterns in complex data, while also utilising the strengths of human wisdom to ensure the models are physically meaningful. This approach is applied to datasets populated by large numbers of computational simulations for compressor and turbine blades. It is shown using 2D data that a new profile loss model can be developed which halves the error in preliminary design correlations and generalises well to turbines as well as compressors. The reason for these improvements is shown to be the result of a more physically accurate decomposition for mixing loss than current industry standard models. This new mixing loss model utilises flow properties local to the wake and is shown to explain the limiting cases for the mixing of two flows, the mixing out of a trailing edge blockage and the form drag of an aerofoil. It is argued that the reason this model is more accurate and general than existing models is that it retains the physical form for the source of loss creation at the fundamental scale. The data-centric approach is also applied to a 3D RANS compressor linear repeating stage dataset and shown to help develop a model for the total loss in a compressor blade row that is composed of physically accurate models for each of the key loss sources in a blade row. This model is demonstrated to halve the error in loss predictions compared to industry standard preliminary design models, and capture the key sensitivities to trailing edge thickness, lean and sweep, not previously accounted for at the preliminary stages of design. In particular, new, more accurate and general models for tip loss and wake mixing loss are highlighted that have the same physical form as the source of loss creation. The resulting model is finally used to generate Smith Charts to explain how the optimal point on the Smith Chart is very sensitive to the choice of preliminary design parameters and how these parameters can be used to move the optimal point. In particular, it is highlighted that there is a need to consider 3D effects in the preliminary stages of design to move the design point to higher stage loading coefficients."],"dc:format.checksum.md5":["69f299d3f009c35920629a51aa978205","87eda9de84448d1f82354d60eee3eb5f"],"dc:identifier.doi":["https://doi.org/10.17863/CAM.113330"],"dc:identifier.uri":["https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/024d32bc-6ce9-4ef1-b3c0-bc2c3259357b/download"],"dc:language":["eng"],"dc:publisher.institution":["University of Cambridge"],"dc:relation.isreferencedby.uri":["https://www.repository.cam.ac.uk/handle/1810/375809"],"dc:rights":["https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/a86f3055-bc86-4ded-ae00-8946e2bff32c/download","http://purl.org/NET/rdflicense/allrightsreserved"],"dc:subject":["Machine Learning","Compressor","Loss","Data-Centric","Loss models","Loss modelling","Compressor Design"],"dc:title":["A Data-Centric Approach to Loss Mechanisms for Compressor Preliminary Design"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["Doctoral"],"dc:type.qualificationname":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-07-22T22:24:17Z"}