An improved Cox proportional hazards model for reliability analysis of aviation gas turbines considering varying environmental conditions and operational settings
| dc.contributor.author | Cui, Bo | |
| dc.contributor.examiningcommittee | Wang, Jay (Mechanical Engineering) | |
| dc.contributor.examiningcommittee | Chen, Yuejian (Mechanical Engineering) | |
| dc.contributor.supervisor | Liang, Xihui | |
| dc.date.accessioned | 2025-02-21T16:28:10Z | |
| dc.date.available | 2025-02-21T16:28:10Z | |
| dc.date.issued | 2025-02-18 | |
| dc.date.submitted | 2025-02-19T02:15:56Z | en_US |
| dc.degree.discipline | Mechanical Engineering | |
| dc.degree.level | Master of Science (M.Sc.) | |
| dc.description.abstract | Aviation gas turbines are critical components in the aerospace industry, where reliability directly influences maintenance costs, operational efficiency, and safety. The Cox proportional hazards model (Cox model) is crucial in reliability analysis, traditionally employed to evaluate the impact of multiple covariates on the likelihood of failure. However, its application in aviation gas turbines has revealed limitations, particularly when working conditions and gas path parameters are used as direct covariates. These factors, while relevant, often fail to fully capture the degradation process due to their sensitivity to varying working conditions. In recent years, the integration of machine learning techniques with the Cox model has gained traction, offering a potential pathway to enhance the accuracy of reliability predictions. Despite this progress, existing approaches still predominantly rely on direct covariates that may not accurately reflect the underlying degradation of gas turbines. This thesis proposes a novel method that leverages machine learning to generate new, more robust covariates for the Cox model. By first modeling a healthy gas turbine, the method produces covariates that are less influenced by the variability in working conditions, leading to a more precise representation of turbine degradation over time. The thesis explores the theoretical foundations of this approach, detailing how machine learning algorithms can be employed to model a healthy gas turbine and subsequently generate new covariates that better align with the actual degradation process. This method addresses the inherent limitations of conventional covariates, providing a more comprehensive method for reliability analysis under varying operational scenarios. The effectiveness of the proposed method is validated through extensive experiments using the NASA C-MAPSS dataset, which includes data from turbofan engines operating under diverse conditions. Comparative analysis demonstrates that the improved Cox model, enhanced with machine learning-derived covariates, offers better accuracy in predicting gas turbine reliability compared to the traditional model. The results demonstrate the potential of this approach to significantly improve reliability assessments in the aerospace industry, ultimately contributing to safer and more efficient operations. This research not only advances the application of the Cox model in the context of aviation gas turbines but also opens new avenues for integrating machine learning to generate new covariates with traditional reliability analysis techniques. The findings have broader implications for the aerospace industry and could inform future developments in turbine maintenance strategies and operational practices. | |
| dc.description.note | May 2025 | |
| dc.identifier.uri | http://hdl.handle.net/1993/38895 | |
| dc.language.iso | eng | |
| dc.subject | Reliability Analysis | |
| dc.subject | Cox Proportional Hazards Model | |
| dc.subject | Machine Learning | |
| dc.subject | Gas Turbine | |
| dc.subject | Environmental Conditions and Operational Settings | |
| dc.title | An improved Cox proportional hazards model for reliability analysis of aviation gas turbines considering varying environmental conditions and operational settings | |
| local.subject.manitoba | no |