Aero Engine Instability Prediction and Detection Method Based on Gated Recurrent Neural Networks

Traditional methods for detecting surge in aero engines are typically based on bench testing to study the compressor surge mechanism and signal characteristics. However, these methods struggle to fully capture the complex and variable factors in engine operation. This paper proposes an aero engine surge prediction method based on Gated Recurrent Neural Networks. In practical application, compared with wavelet detection, the wavelet transform can predict surge instability 0.077s in advance, while setting the prediction sequence length of Gated Recurrent Neural Networks allows obtaining the compressor outlet pressure signal 0.28s later. Furthermore, unlike wavelet analysis which is limited to local analysis and basis function selection difficulty, Gated Recurrent Neural Networks can stably and accurately predict long-term data while retaining local surge signal features without loss, making it more suitable for long-term monitoring and use of aero engines.