Compressor Flow Perception via Deep Learning Modeling with Multi-Source Dynamic Fusion of Temporal Features by...

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Fluid Dynamics

Background:

• Aeroengine operation stability and reliability are critical engineering challenges.
• Precise identification of aerodynamic instability is essential for safe and efficient operation.
• Traditional single-point monitoring methods suffer from information deficiency and incomplete representation.

Purpose of the Study:

• To propose a deep learning model for precise identification of aerodynamic instability in aeroengines.
• To enhance the operation stability and reliability of aeroengines through advanced monitoring.
• To overcome the limitations of traditional single-point detection methods.

Main Methods:

• Multi-dimensional feature extraction in time, frequency, and entropy domains from a high-speed multistage compressor.
• Development of a cross-attention and bidirectional Long Short-Term Memory (CA_BiLSTM) network for multi-source signal fusion.
• Integration of Variational Mode Decomposition (VMD) and Dung Beetle Optimizer (DBO) for nonlinear aerodynamic parameter prediction (VMD_DBO_LSTM).

Main Results:

• The proposed dual-channel fusion model based on the cross-attention mechanism precisely characterizes nonlinear flow features.
• The CA_BiLSTM model integrates complementary information from inlet and outlet signals, achieving collaborative signal characterization.
• The system demonstrates significantly superior anti-interference capability compared to single-point signals and detects instability 1580 r in advance.

Conclusions:

• The CA_BiLSTM model effectively addresses information deficiency and incomplete representation issues in traditional monitoring.
• The developed deep learning approach enhances the precision and reliability of aerodynamic instability detection.
• This method significantly improves the operational stability and safety of aeroengines.