Dual-frequency enhanced attention network for aircraft engine remaining useful life prediction

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Mechanical Engineering

Background:

• Accurate remaining useful life (RUL) prediction is vital for aircraft engine safety and maintenance.
• Existing methods may struggle with complex degradation patterns and sensor data redundancy.

Purpose of the Study:

• To propose a novel prediction framework for aircraft engine RUL.
• To enhance prediction accuracy and robustness using advanced deep learning techniques.

Main Methods:

• Developed a dual-frequency enhanced attention network architecture using separable convolutional neural networks.
• Introduced information volume criterion (IVC) and information content threshold (CIT) for sensor data quantification and redundancy removal.
• Incorporated trainable Fourier transform (FMB-f) and wavelet transform (FMB-w) modules for capturing degradation trends.
• Utilized an efficient channel attention block to model inter-sensor dependencies.

Main Results:

• The proposed framework demonstrated accurate RUL predictions in experimental evaluations.
• Frequency-enhanced modules effectively captured global trends and local details of degradation.
• Attention mechanism improved prediction stability and precision by considering sensor interdependencies.

Conclusions:

• The novel RUL prediction framework offers a significant advancement in aircraft engine prognostics.
• The integration of frequency-enhanced modules and attention mechanisms enhances prediction performance and robustness.