{"id":{"repo_id":"mcmaster","oai_identifier":"oai:macsphere.mcmaster.ca:11375/33400"},"canonical_url":"https://search.ndltd.org/etd/mcmaster/oai:macsphere.mcmaster.ca:11375/33400","repository":{"repo_id":"mcmaster","name":"McMaster University","base_url":"https://macsphere.mcmaster.ca/server/oai/request"},"display":{"title":"Fault Detection and Diagnosis for Hairpin Windings in Electric Motors","abstract":"Electric vehicles have been flourishing owing to the increasing demand for decarbonization and low emissions. Electric motors with hairpin windings are replacing conventional vehicle motors with random-wound wire windings due to hairpin windings presenting a high level of power density, energy efficiency, and production automation. The complex manufacturing processes involved in hairpin welding and insulation introduce subtle defects, which can evolve into faults under thermal, electrical, and mechanical stresses. These faults ultimately result in reduced motor lifetime or complete breakdowns, critically affecting production reliability and economic cost. Therefore, it is critically important to detect incipient welding and insulation faults during the manufacturing stage. Various Fault Detection and Diagnosis (FDD) methods have been explored to assess the health condition of hairpin windings. However, they are subject to environmental constraints and additional testing-induced stresses, typically requiring specialized testing equipment, as well as suffering from susceptibility to noise and limited robustness in detecting incipient faults. This thesis proposes an effective and efficient impedance-based methodology utilizing impedance measurements as the primary source of information for assessing the health condition of hairpin windings. This non-destructive and low-stress methodology is highly sensitive to the incipient faults. It establishes mappings between impedance information and incipient faults, along with facilitating the integration of advanced Artificial Intelligence (AI) methods. Furthermore, the approach employs simple, compact, and cost-effective instrumentation, making it fully compatible with the end-of-line mass production testing. Various diagnostic methods are employed based on collected impedance measurements to extract the underlying patterns of fault conditions. A statistical method is employed first to develop a similarity indicator for the detection of incipient welding faults. The developed similarity indicator is used as a metric to establish the threshold, effectively identifying abnormal hairpin stators and locating faulty phase windings. In addition, a data-driven method based on machine learning is used to detect the insulation faults related to epoxy. With a minimal amount of collected impedance measurement, a machine learning approach based on Support Vector Machine (SVM) establishes the decision boundaries to determine the health condition of hairpin windings. Further, a model-based method based on the Equivalent Circuit Model (ECM) is built to characterize the frequency response of hairpin windings. It reflects the health status of hairpin windings through the variation of the extracted parameters in the distinctive range, providing the interpretation and insights into the failure mechanism. A hybrid diagnostic method is subsequently proposed, integrating both model-based and data-driven methods to detect epoxy faults, leveraging the strengths of both methodologies. Specifically, the model-based method is applied to extract features by identifying parameters within the ECM using system identification techniques, and the data-driven method is adopted for fault classification by employing a machine learning system incorporating Principal Component Analysis (PCA) and SVM. The results achieve higher detection accuracy over pure data-driven methodologies. To overcome the limited data availability during the early stage of manufacturing processes, a data augmentation technique is employed to generate large-scale, high-fidelity, and diverse synthetic data using a physics-informed model-based method. With a substantial amount of generated synthetic data, advanced Deep Neural Networks (DNN) are deployed to further improve FDD performance. Specifically, advanced Convolutional Neural Networks (CNN) and state-of-the-art Transformer networks are modified to adapt to hairpin winding applications, outperforming conventional machine learning methodologies. Overall, the research conducted is essential to enable robust quality assurance for hairpin winding stator production and meet the increasing reliability and safety demands of modern EVs. Moreover, this thesis enables the application of FDD for hairpin winding stators anomaly detection, pushing the boundaries of the research areas in model-based, data-driven, and hybrid diagnostic methods, as well as advanced DNN learning system deployment for diagnostic applications.","abstract_html":"Electric vehicles have been flourishing owing to the increasing demand for decarbonization and low emissions. Electric motors with hairpin windings are replacing conventional vehicle motors with random-wound wire windings due to hairpin windings presenting a high level of power density, energy efficiency, and production automation. The complex manufacturing processes involved in hairpin welding and insulation introduce subtle defects, which can evolve into faults under thermal, electrical, and mechanical stresses. These faults ultimately result in reduced motor lifetime or complete breakdowns, critically affecting production reliability and economic cost. Therefore, it is critically important to detect incipient welding and insulation faults during the manufacturing stage. Various Fault Detection and Diagnosis (FDD) methods have been explored to assess the health condition of hairpin windings. However, they are subject to environmental constraints and additional testing-induced stresses, typically requiring specialized testing equipment, as well as suffering from susceptibility to noise and limited robustness in detecting incipient faults. This thesis proposes an effective and efficient impedance-based methodology utilizing impedance measurements as the primary source of information for assessing the health condition of hairpin windings. This non-destructive and low-stress methodology is highly sensitive to the incipient faults. It establishes mappings between impedance information and incipient faults, along with facilitating the integration of advanced Artificial Intelligence (AI) methods. Furthermore, the approach employs simple, compact, and cost-effective instrumentation, making it fully compatible with the end-of-line mass production testing. Various diagnostic methods are employed based on collected impedance measurements to extract the underlying patterns of fault conditions. A statistical method is employed first to develop a similarity indicator for the detection of incipient welding faults. The developed similarity indicator is used as a metric to establish the threshold, effectively identifying abnormal hairpin stators and locating faulty phase windings. In addition, a data-driven method based on machine learning is used to detect the insulation faults related to epoxy. With a minimal amount of collected impedance measurement, a machine learning approach based on Support Vector Machine (SVM) establishes the decision boundaries to determine the health condition of hairpin windings. Further, a model-based method based on the Equivalent Circuit Model (ECM) is built to characterize the frequency response of hairpin windings. It reflects the health status of hairpin windings through the variation of the extracted parameters in the distinctive range, providing the interpretation and insights into the failure mechanism. A hybrid diagnostic method is subsequently proposed, integrating both model-based and data-driven methods to detect epoxy faults, leveraging the strengths of both methodologies. Specifically, the model-based method is applied to extract features by identifying parameters within the ECM using system identification techniques, and the data-driven method is adopted for fault classification by employing a machine learning system incorporating Principal Component Analysis (PCA) and SVM. The results achieve higher detection accuracy over pure data-driven methodologies. To overcome the limited data availability during the early stage of manufacturing processes, a data augmentation technique is employed to generate large-scale, high-fidelity, and diverse synthetic data using a physics-informed model-based method. With a substantial amount of generated synthetic data, advanced Deep Neural Networks (DNN) are deployed to further improve FDD performance. Specifically, advanced Convolutional Neural Networks (CNN) and state-of-the-art Transformer networks are modified to adapt to hairpin winding applications, outperforming conventional machine learning methodologies. Overall, the research conducted is essential to enable robust quality assurance for hairpin winding stator production and meet the increasing reliability and safety demands of modern EVs. Moreover, this thesis enables the application of FDD for hairpin winding stators anomaly detection, pushing the boundaries of the research areas in model-based, data-driven, and hybrid diagnostic methods, as well as advanced DNN learning system deployment for diagnostic applications.","abstract_has_math":false,"creators":["Zhang, Yu"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":"Mechanical Engineering","school":null,"contributors":[],"advisors":["Habibi, Saeid","Ziada, Youssef"],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T19:30:39Z","subjects":["Hairpin winding","fault detection","machine learning","model-based"],"languages":["en"],"rights":["Attribution-NonCommercial-NoDerivatives 4.0 International"],"rights_urls":["http://creativecommons.org/licenses/by-nc-nd/4.0/"],"identifier_entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://doi.org/10.71548/1065"],"render_values":[{"text":"https://doi.org/10.71548/1065","href":"https://doi.org/10.71548/1065","code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/11375/33400","outbound_label":"Handle","outbound_source":"dc:identifier.uri","record_url":"https://hdl.handle.net/11375/33400","record_label":"Repository record","record_kind":"handle","record_source":"dc:identifier.uri","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://macsphere.mcmaster.ca/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Amacsphere.mcmaster.ca%3A11375%2F33400","prefix":"dim"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Habibi, Saeid","Ziada, Youssef"]},{"key":"dc:contributor.department","label":"Department","values":["Mechanical Engineering"]},{"key":"dc:creator","label":"Author","values":["Zhang, Yu"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-07-23T15:15:55Z"]},{"key":"dc:date.issued","label":"Date","values":["2026"]},{"key":"dc:type","label":"Dc Type","values":["Dissertation","Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Hairpin winding","fault detection","machine learning","model-based"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution-NonCommercial-NoDerivatives 4.0 International"]},{"key":"dc:rights.uri","label":"Rights URI","values":["http://creativecommons.org/licenses/by-nc-nd/4.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/11375/33400","https://doi.org/10.71548/1065"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Electric vehicles have been flourishing owing to the increasing demand for decarbonization and low emissions. Electric motors with hairpin windings are replacing conventional vehicle motors with random-wound wire windings due to hairpin windings presenting a high level of power density, energy efficiency, and production automation. The complex manufacturing processes involved in hairpin welding and insulation introduce subtle defects, which can evolve into faults under thermal, electrical, and mechanical stresses. These faults ultimately result in reduced motor lifetime or complete breakdowns, critically affecting production reliability and economic cost. Therefore, it is critically important to detect incipient welding and insulation faults during the manufacturing stage. Various Fault Detection and Diagnosis (FDD) methods have been explored to assess the health condition of hairpin windings. However, they are subject to environmental constraints and additional testing-induced stresses, typically requiring specialized testing equipment, as well as suffering from susceptibility to noise and limited robustness in detecting incipient faults. This thesis proposes an effective and efficient impedance-based methodology utilizing impedance measurements as the primary source of information for assessing the health condition of hairpin windings. This non-destructive and low-stress methodology is highly sensitive to the incipient faults. It establishes mappings between impedance information and incipient faults, along with facilitating the integration of advanced Artificial Intelligence (AI) methods. Furthermore, the approach employs simple, compact, and cost-effective instrumentation, making it fully compatible with the end-of-line mass production testing. Various diagnostic methods are employed based on collected impedance measurements to extract the underlying patterns of fault conditions. A statistical method is employed first to develop a similarity indicator for the detection of incipient welding faults. The developed similarity indicator is used as a metric to establish the threshold, effectively identifying abnormal hairpin stators and locating faulty phase windings. In addition, a data-driven method based on machine learning is used to detect the insulation faults related to epoxy. With a minimal amount of collected impedance measurement, a machine learning approach based on Support Vector Machine (SVM) establishes the decision boundaries to determine the health condition of hairpin windings. Further, a model-based method based on the Equivalent Circuit Model (ECM) is built to characterize the frequency response of hairpin windings. It reflects the health status of hairpin windings through the variation of the extracted parameters in the distinctive range, providing the interpretation and insights into the failure mechanism. A hybrid diagnostic method is subsequently proposed, integrating both model-based and data-driven methods to detect epoxy faults, leveraging the strengths of both methodologies. Specifically, the model-based method is applied to extract features by identifying parameters within the ECM using system identification techniques, and the data-driven method is adopted for fault classification by employing a machine learning system incorporating Principal Component Analysis (PCA) and SVM. The results achieve higher detection accuracy over pure data-driven methodologies. To overcome the limited data availability during the early stage of manufacturing processes, a data augmentation technique is employed to generate large-scale, high-fidelity, and diverse synthetic data using a physics-informed model-based method. With a substantial amount of generated synthetic data, advanced Deep Neural Networks (DNN) are deployed to further improve FDD performance. Specifically, advanced Convolutional Neural Networks (CNN) and state-of-the-art Transformer networks are modified to adapt to hairpin winding applications, outperforming conventional machine learning methodologies. Overall, the research conducted is essential to enable robust quality assurance for hairpin winding stator production and meet the increasing reliability and safety demands of modern EVs. Moreover, this thesis enables the application of FDD for hairpin winding stators anomaly detection, pushing the boundaries of the research areas in model-based, data-driven, and hybrid diagnostic methods, as well as advanced DNN learning system deployment for diagnostic applications."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Doctor of Philosophy (PhD)"]},{"key":"dc:title","label":"Title","values":["Fault Detection and Diagnosis for Hairpin Windings in Electric Motors"]},{"key":"others:access_status","label":"Others Access Status","values":["embargo"]}]}],"canonical_facts":{"dc:contributor.advisor":["Habibi, Saeid","Ziada, Youssef"],"dc:contributor.department":["Mechanical Engineering"],"dc:creator":["Zhang, Yu"],"dc:date.accessioned":["2026-07-23T15:15:55Z"],"dc:date.issued":["2026"],"dc:description.abstract":["Electric vehicles have been flourishing owing to the increasing demand for decarbonization and low emissions. Electric motors with hairpin windings are replacing conventional vehicle motors with random-wound wire windings due to hairpin windings presenting a high level of power density, energy efficiency, and production automation. The complex manufacturing processes involved in hairpin welding and insulation introduce subtle defects, which can evolve into faults under thermal, electrical, and mechanical stresses. These faults ultimately result in reduced motor lifetime or complete breakdowns, critically affecting production reliability and economic cost. Therefore, it is critically important to detect incipient welding and insulation faults during the manufacturing stage. Various Fault Detection and Diagnosis (FDD) methods have been explored to assess the health condition of hairpin windings. However, they are subject to environmental constraints and additional testing-induced stresses, typically requiring specialized testing equipment, as well as suffering from susceptibility to noise and limited robustness in detecting incipient faults. This thesis proposes an effective and efficient impedance-based methodology utilizing impedance measurements as the primary source of information for assessing the health condition of hairpin windings. This non-destructive and low-stress methodology is highly sensitive to the incipient faults. It establishes mappings between impedance information and incipient faults, along with facilitating the integration of advanced Artificial Intelligence (AI) methods. Furthermore, the approach employs simple, compact, and cost-effective instrumentation, making it fully compatible with the end-of-line mass production testing. Various diagnostic methods are employed based on collected impedance measurements to extract the underlying patterns of fault conditions. A statistical method is employed first to develop a similarity indicator for the detection of incipient welding faults. The developed similarity indicator is used as a metric to establish the threshold, effectively identifying abnormal hairpin stators and locating faulty phase windings. In addition, a data-driven method based on machine learning is used to detect the insulation faults related to epoxy. With a minimal amount of collected impedance measurement, a machine learning approach based on Support Vector Machine (SVM) establishes the decision boundaries to determine the health condition of hairpin windings. Further, a model-based method based on the Equivalent Circuit Model (ECM) is built to characterize the frequency response of hairpin windings. It reflects the health status of hairpin windings through the variation of the extracted parameters in the distinctive range, providing the interpretation and insights into the failure mechanism. A hybrid diagnostic method is subsequently proposed, integrating both model-based and data-driven methods to detect epoxy faults, leveraging the strengths of both methodologies. Specifically, the model-based method is applied to extract features by identifying parameters within the ECM using system identification techniques, and the data-driven method is adopted for fault classification by employing a machine learning system incorporating Principal Component Analysis (PCA) and SVM. The results achieve higher detection accuracy over pure data-driven methodologies. To overcome the limited data availability during the early stage of manufacturing processes, a data augmentation technique is employed to generate large-scale, high-fidelity, and diverse synthetic data using a physics-informed model-based method. With a substantial amount of generated synthetic data, advanced Deep Neural Networks (DNN) are deployed to further improve FDD performance. Specifically, advanced Convolutional Neural Networks (CNN) and state-of-the-art Transformer networks are modified to adapt to hairpin winding applications, outperforming conventional machine learning methodologies. Overall, the research conducted is essential to enable robust quality assurance for hairpin winding stator production and meet the increasing reliability and safety demands of modern EVs. Moreover, this thesis enables the application of FDD for hairpin winding stators anomaly detection, pushing the boundaries of the research areas in model-based, data-driven, and hybrid diagnostic methods, as well as advanced DNN learning system deployment for diagnostic applications."],"dc:description.degree":["Doctor of Philosophy (PhD)"],"dc:identifier.uri":["https://hdl.handle.net/11375/33400","https://doi.org/10.71548/1065"],"dc:language.iso":["en"],"dc:rights":["Attribution-NonCommercial-NoDerivatives 4.0 International"],"dc:rights.uri":["http://creativecommons.org/licenses/by-nc-nd/4.0/"],"dc:subject":["Hairpin winding","fault detection","machine learning","model-based"],"dc:title":["Fault Detection and Diagnosis for Hairpin Windings in Electric Motors"],"dc:type":["Dissertation","Thesis"],"others:access_status":["embargo"]},"updated_at":"2026-11-18T19:30:39Z"}