Efficient Channel Attention-Gated Graph Transformer for Aero-Engine Remaining Useful Life Prediction

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Accurate Remaining Useful Life (RUL) estimation is crucial for aero-engine safety and maintenance.
• Current deep learning methods struggle with local sequential dependencies and long-term degradation patterns in sensor data.

Purpose of the Study:

• To develop an advanced predictive framework for robust RUL estimation in aero-engines.
• To address limitations in modeling local dependencies and progressive degradation behavior.

Main Methods:

• Introduced the Efficient Channel Attention (ECA)-gated Graph Transformer (EGG-Transformer).
• Combined Graph Convolutional Networks (GCNs) for local structure learning.
• Employed an ECA-gated fusion module for adaptive feature merging.
• Utilized a transformer-based encoder for long-range temporal pattern analysis.

Main Results:

• The EGG-Transformer demonstrated superior performance on the C-MAPSS dataset.
• Achieved a 22.85% reduction in mean Root Mean Square Error (RMSE).
• Reduced the Score index by 10.21% compared to leading methods.

Conclusions:

• The proposed EGG-Transformer framework offers improved accuracy and robustness for RUL forecasting.
• The method effectively models local and long-range dependencies in aero-engine sensor data.
• Highlights practical applicability for enhancing aero-engine maintenance and safety.