{"id":{"repo_id":"cambridge","oai_identifier":"oai:www.repository.cam.ac.uk:1810/397938"},"canonical_url":"https://search.dev.ndltd.org/etd/cambridge/oai:www.repository.cam.ac.uk:1810/397938","repository":{"repo_id":"cambridge","name":"Cambridge University","base_url":"https://api.repository.cam.ac.uk/server/oai/request"},"display":{"title":"Probabilistic machine learning for the elimination of thermoacoustic instabilities","abstract":"Thermoacoustic instabilities have hindered the development of high energy density combustors, such as those in jet engines, gas turbines or rockets, for decades. They are notoriously difficult to model and despite efforts to eliminate them, can show up unexpectedly. The current thesis demonstrates how Bayesian machine learning techniques may be of benefit when modeling, designing against and trying to avoid thermoacoustic instabilities. We show that Bayesian Neural Network can be used to assimilate model parameters from flame data and make our qualitative instability models quantitatively accurate. Next, we use gradient-augmented Bayesian optimization to globally optimize the geometric parameters of a thermoacoustically unstable combustor design. The Bayesian algorithm automatically manages the trade-off between exploration and exploitation and therefore, requires fewer evaluations of the underlying adjoint model to arrive at the global optimum. We also use Bayesian neural networks to learn functional relationships between sensor data and measures of thermoacoustic stability. First, we demonstrate on a laboratory-scale Rijke tube driven by a turbulent flame that it is possible to predict decay rates of acoustic pulses from the spectrum of 100 millisecond combustion noise samples, thus enabling us to monitor the combustor's thermoacoustic stability in real time. We then apply these ideas to predict instabilities in an experimental rocket chamber, where we use the history of sensor data, including measurements of dynamic pressure, temperature, static pressure, etc. to find precursors of instabilities upto 500 milliseconds before they occur. The Bayesian nature of our algorithms allows principled estimates of uncertainty to accompany each prediction while the technique of Integrated Gradients lets us interpret our models. It is hoped that this thesis will serve as a first step towards establishing Bayesian machine learning techniques as tools to help us model combustion instabilities better, design against them and discover trustworthy, robust and reliable instability prognostics.","abstract_html":"Thermoacoustic instabilities have hindered the development of high energy density combustors, such as those in jet engines, gas turbines or rockets, for decades. They are notoriously difficult to model and despite efforts to eliminate them, can show up unexpectedly. The current thesis demonstrates how Bayesian machine learning techniques may be of benefit when modeling, designing against and trying to avoid thermoacoustic instabilities. We show that Bayesian Neural Network can be used to assimilate model parameters from flame data and make our qualitative instability models quantitatively accurate. Next, we use gradient-augmented Bayesian optimization to globally optimize the geometric parameters of a thermoacoustically unstable combustor design. The Bayesian algorithm automatically manages the trade-off between exploration and exploitation and therefore, requires fewer evaluations of the underlying adjoint model to arrive at the global optimum. We also use Bayesian neural networks to learn functional relationships between sensor data and measures of thermoacoustic stability. First, we demonstrate on a laboratory-scale Rijke tube driven by a turbulent flame that it is possible to predict decay rates of acoustic pulses from the spectrum of 100 millisecond combustion noise samples, thus enabling us to monitor the combustor's thermoacoustic stability in real time. We then apply these ideas to predict instabilities in an experimental rocket chamber, where we use the history of sensor data, including measurements of dynamic pressure, temperature, static pressure, etc. to find precursors of instabilities upto 500 milliseconds before they occur. The Bayesian nature of our algorithms allows principled estimates of uncertainty to accompany each prediction while the technique of Integrated Gradients lets us interpret our models. It is hoped that this thesis will serve as a first step towards establishing Bayesian machine learning techniques as tools to help us model combustion instabilities better, design against them and discover trustworthy, robust and reliable instability prognostics.","abstract_has_math":false,"creators":["Sengupta, Ushnish"],"institution":"University of Cambridge","degree_name":"Doctor of Philosophy (PhD)","degree_level":"Doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Rasmussen, Carl Edward"],"committee_chairs":[],"committee_members":[],"year":2023,"date_issued":"2023-06-02","date_published":"2023-06-02","updated_at":"2026-07-22T22:24:20Z","subjects":["bayesian machine learning","thermoacoustic instabilities","rocket engines","bayesian optimization","jet engines"],"languages":["eng"],"rights":[],"rights_urls":["https://www.repository.cam.ac.uk/bitstreams/89d4de25-72cd-42ca-8ef8-621816cc7cac/download","http://purl.org/NET/rdflicense/allrightsreserved"],"identifier_entries":[{"key":"dc:creator.authoridentifier","label":"Author Identifier","values":["000000032633946X"],"render_values":[{"text":"0000-0003-2633-946X","href":"https://orcid.org/0000-0003-2633-946X","code":true}]}]},"links":{"outbound_url":"https://doi.org/10.17863/CAM.126919","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Rasmussen, Carl Edward"]},{"key":"dc:contributor.sponsor","label":"Sponsor","values":["I am thankful for the funding I received from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement number 766264."]},{"key":"dc:creator","label":"Author","values":["Sengupta, Ushnish"]},{"key":"dc:creator.authoridentifier","label":"Author Identifier","values":["000000032633946X"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2023-06-02"]},{"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/1810/397938"]},{"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":["bayesian machine learning","thermoacoustic instabilities","rocket engines","bayesian optimization","jet engines"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["https://www.repository.cam.ac.uk/bitstreams/89d4de25-72cd-42ca-8ef8-621816cc7cac/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.126919"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://www.repository.cam.ac.uk/bitstreams/698085b7-5859-476a-a7a9-91e547d2214c/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Thermoacoustic instabilities have hindered the development of high energy density combustors, such as those in jet engines, gas turbines or rockets, for decades. They are notoriously difficult to model and despite efforts to eliminate them, can show up unexpectedly. The current thesis demonstrates how Bayesian machine learning techniques may be of benefit when modeling, designing against and trying to avoid thermoacoustic instabilities. We show that Bayesian Neural Network can be used to assimilate model parameters from flame data and make our qualitative instability models quantitatively accurate. Next, we use gradient-augmented Bayesian optimization to globally optimize the geometric parameters of a thermoacoustically unstable combustor design. The Bayesian algorithm automatically manages the trade-off between exploration and exploitation and therefore, requires fewer evaluations of the underlying adjoint model to arrive at the global optimum. We also use Bayesian neural networks to learn functional relationships between sensor data and measures of thermoacoustic stability. First, we demonstrate on a laboratory-scale Rijke tube driven by a turbulent flame that it is possible to predict decay rates of acoustic pulses from the spectrum of 100 millisecond combustion noise samples, thus enabling us to monitor the combustor's thermoacoustic stability in real time. We then apply these ideas to predict instabilities in an experimental rocket chamber, where we use the history of sensor data, including measurements of dynamic pressure, temperature, static pressure, etc. to find precursors of instabilities upto 500 milliseconds before they occur. The Bayesian nature of our algorithms allows principled estimates of uncertainty to accompany each prediction while the technique of Integrated Gradients lets us interpret our models. It is hoped that this thesis will serve as a first step towards establishing Bayesian machine learning techniques as tools to help us model combustion instabilities better, design against them and discover trustworthy, robust and reliable instability prognostics."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["9f2f200290e13d1f22024e799ee47a19","87eda9de84448d1f82354d60eee3eb5f"]},{"key":"dc:title","label":"Title","values":["Probabilistic machine learning for the elimination of thermoacoustic instabilities"]}]}],"canonical_facts":{"dc:contributor.advisor":["Rasmussen, Carl Edward"],"dc:contributor.sponsor":["I am thankful for the funding I received from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement number 766264."],"dc:creator":["Sengupta, Ushnish"],"dc:creator.authoridentifier":["000000032633946X"],"dc:date.issued":["2023-06-02"],"dc:description.abstract":["Thermoacoustic instabilities have hindered the development of high energy density combustors, such as those in jet engines, gas turbines or rockets, for decades. They are notoriously difficult to model and despite efforts to eliminate them, can show up unexpectedly. The current thesis demonstrates how Bayesian machine learning techniques may be of benefit when modeling, designing against and trying to avoid thermoacoustic instabilities. We show that Bayesian Neural Network can be used to assimilate model parameters from flame data and make our qualitative instability models quantitatively accurate. Next, we use gradient-augmented Bayesian optimization to globally optimize the geometric parameters of a thermoacoustically unstable combustor design. The Bayesian algorithm automatically manages the trade-off between exploration and exploitation and therefore, requires fewer evaluations of the underlying adjoint model to arrive at the global optimum. We also use Bayesian neural networks to learn functional relationships between sensor data and measures of thermoacoustic stability. First, we demonstrate on a laboratory-scale Rijke tube driven by a turbulent flame that it is possible to predict decay rates of acoustic pulses from the spectrum of 100 millisecond combustion noise samples, thus enabling us to monitor the combustor's thermoacoustic stability in real time. We then apply these ideas to predict instabilities in an experimental rocket chamber, where we use the history of sensor data, including measurements of dynamic pressure, temperature, static pressure, etc. to find precursors of instabilities upto 500 milliseconds before they occur. The Bayesian nature of our algorithms allows principled estimates of uncertainty to accompany each prediction while the technique of Integrated Gradients lets us interpret our models. It is hoped that this thesis will serve as a first step towards establishing Bayesian machine learning techniques as tools to help us model combustion instabilities better, design against them and discover trustworthy, robust and reliable instability prognostics."],"dc:format.checksum.md5":["9f2f200290e13d1f22024e799ee47a19","87eda9de84448d1f82354d60eee3eb5f"],"dc:identifier.doi":["https://doi.org/10.17863/CAM.126919"],"dc:identifier.uri":["https://www.repository.cam.ac.uk/bitstreams/698085b7-5859-476a-a7a9-91e547d2214c/download"],"dc:language":["eng"],"dc:publisher.institution":["University of Cambridge"],"dc:relation.isreferencedby.uri":["https://www.repository.cam.ac.uk/handle/1810/397938"],"dc:rights":["https://www.repository.cam.ac.uk/bitstreams/89d4de25-72cd-42ca-8ef8-621816cc7cac/download","http://purl.org/NET/rdflicense/allrightsreserved"],"dc:subject":["bayesian machine learning","thermoacoustic instabilities","rocket engines","bayesian optimization","jet engines"],"dc:title":["Probabilistic machine learning for the elimination of thermoacoustic instabilities"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["Doctoral"],"dc:type.qualificationname":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-07-22T22:24:20Z"}