Predictive maintenance programs for aircraft engines based on remaining useful life prediction.

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Mechanical Engineering

Background:

• Aero-engine condition monitoring data is complex, posing challenges for traditional maintenance strategies.
• Ensuring flight safety and operational efficiency necessitates accurate prediction of aero-engine remaining useful life (RUL).

Purpose of the Study:

• To develop a predictive maintenance planning framework for aero-engines based on RUL prediction.
• To design effective predictive maintenance strategies to enhance flight safety and reduce costs.

Main Methods:

• Proposed a deep learning integrated model, Transformer-Long Short-Term Memory (Trans-LSTM), combining Transformer and Long Short-Term Memory Network (LSTM).
• Utilized Bayesian optimization to fine-tune the hyperparameters of the Trans-LSTM model for improved predictive accuracy.
• Developed an engine alarm threshold based on predicted RUL data to trigger predictive maintenance tasks.

Main Results:

• The data-driven predictive maintenance strategy effectively monitors engine status in real-time and identifies potential failure risks.
• Significantly enhanced flight safety by reducing the risk of sudden engine failures compared to periodic maintenance.
• Achieved substantial cost reductions by avoiding unnecessary maintenance and optimizing resource allocation.

Conclusions:

• The proposed Trans-LSTM framework provides a valuable tool for proactive aero-engine maintenance.
• Accurate RUL prediction and optimized maintenance scheduling lead to improved engine availability and economic benefits for airlines.
• This approach demonstrates significant potential for practical application in air transportation, enhancing both safety and efficiency.