Loading...

Improving Engine Design to Reduce Maintenance Costs using Machine Learning Algorithms

Mohebbi, Hossein | 2025

11 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58829 (45)
  4. University: Sharif University of Technology
  5. Department: Aerospace Engineering
  6. Advisor(s): Mazaheri, Karim
  7. Abstract:
  8. The aviation industry is rapidly growing, and with the increasing number of aircraft, maintenance, repair, and overhaul (MRO) activities have become more complex and costly. Airline operators seek to improve performance, enhance safety, and reduce maintenance costs and unexpected failures. Improving engine design and reducing maintenance costs are fundamental requirements in the aerospace, transportation, and energy industries. The aim of this thesis is to develop a machine learning model to predict the life of an Auxiliary Power Unit (APU) based on the lifetimes of its subsystems and variations in flow parameters through the system, as well as to redesign one of its components in order to increase both the component life and the overall system life. Estimating the APU lifetime is of great importance for predicting failures before they occur and preventing damage to other subsystems. The component life data used to train the model are obtained from industrial sources and published literature. In addition, gas path flow data are generated by monitoring changes in flow parameters in similar turbine systems throughout the operational process until failure. Therefore, the data used are synthetic but are generated to closely resemble real operational data of the APU system. To this end, the generated data are based on flow data at the operating point of the APU. A Long Short-Term Memory (LSTM) recurrent neural network model is used to predict the remaining useful life of the APU. The data are split into training and testing sets with an 80/20 ratio, and 90% of the total data are used the aerospace, transportation, and energy industries. The aim of this thesis is to develop a machine learning model to predict the life of an Auxiliary Power Unit (APU) based on the lifetimes of its subsystems and variations in flow parameters through the system, as well as to redesign one of its components in order to increase both the component life and the overall system life. Estimating the APU lifetime is of great importance for predicting failures before they occur and preventing damage to other subsystems. The component life data used to train the model are obtained from industrial sources and published literature. In addition, gas path flow data are generated by monitoring changes in flow parameters in similar turbine systems throughout the operational process until failure. Therefore, the data used are synthetic but are generated to closely resemble real operational data of the APU system. To this end, the generated data are based on flow data at the operating point of the APU. A Long Short-Term Memory (LSTM) recurrent neural network model is used to predict the remaining useful life of the APU. The data are split into training and testing sets with an 80/20 ratio, and 90% of the total data are used for model training and testing, while the remaining 10% are reserved for model validation. Using a machine learning model to predict the remaining useful life of this system prevents unforeseen costs caused by premature failures. Accordingly, by applying the approach presented in this thesis and utilizing real operational data from the system, many maintenance-related costs can be avoided. In this system, one of the most critical components of the gas path subsystem is the turbine wheel. To increase the overall lifetime of the APU, improvements in the design of this component are pursued. The turbine wheel is subjected to aerodynamic analysis while considering structural constraints, and recommendations are made regarding modifications to the material used in this component. Based on suggestions from the literature, in order to maintain turbine efficiency while increasing its lifetime, a reduction in the number of blades and an increase in blade thickness within an allowable range are investigated. Numerical simulations are performed using ANSYS CFX. The turbine wheel geometry is redesigned, and the number of blades is reduced to 16, then 14, and finally 12 blades. Simultaneously, the blade thickness is varied within an allowable range. All simulations are conducted in three dimensions. The generated computational meshes contain approximately 9 million elements, and all analyses are performed at the operating point of the system. Finally, based on the evaluation of flow parameter variations and efficiency, the optimal blade thickness range and appropriate number of turbine blades are reported. Reducing the number of turbine blades to 12 and increasing their thickness to 1.6 to 1.8 times the original thickness results in a 36% increase in the turbine wheel lifetime.
  9. Keywords:
  10. Aircraft Engine ; Machine Learning ; Turbomachinery Blade ; Numerical Simulation ; CFX Code ; Predictive Maintenance ; Maintenance Cost Reduction

 Digital Object List

 Bookmark

No TOC