Adaptively Lightweight Spatiotemporal Information-Extraction-Operator-Based DL Method for Aero-Engine RUL Prediction.

Area of Science:

• Aerospace Engineering
• Mechanical Engineering
• Artificial Intelligence

Background:

• Accurate Remaining Useful Life (RUL) prediction is critical for operational safety and economic efficiency in aerospace.
• Existing spatiotemporal models for RUL prediction often suffer from structural complexity and limited adaptive feature extraction.
• The need for efficient and adaptive methods to process complex degradation data in aero-engines is paramount.

Purpose of the Study:

• To propose a novel, lightweight deep learning framework for enhanced aero-engine RUL prediction.
• To introduce the Involution GRU (Inv-GRU) as an operator for adaptive spatiotemporal feature extraction.
• To improve the accuracy and reduce the computational burden of RUL prediction models.

Main Methods:

• Developed Involution GRU (Inv-GRU) by integrating an involution operator into a Gated Recurrent Unit (GRU) for adaptive spatiotemporal feature extraction.
• Constructed an Inv-GRU-based deep learning framework to process multi-raw aero-engine data, generating health indicators (HIs).
• Utilized fully connected layers for dimension reduction and RUL regression based on extracted HIs, validated on C-MAPSS datasets.

Main Results:

• The Inv-GRU-based framework successfully predicted aero-engine RUL on the C-MAPSS datasets.
• Demonstrated superior RUL prediction accuracy compared to existing methods.
• Showcased a significant reduction in computational burden, highlighting the model's efficiency.

Conclusions:

• The proposed Inv-GRU operator effectively extracts degradation information and enhances spatiotemporal processing for aero-engine RUL prediction.
• The Inv-GRU-based deep learning framework offers a computationally efficient and accurate solution for RUL prediction.
• This approach holds significant promise for improving the safety and reliability of aero-engine operations.