Intelligent warning method for fatigue crack propagation stages using acoustic emission and machine learning

Abstract

Acoustic emission identification of fatigue crack propagation stages in steel components is crucial for understanding fatigue damage evolution and critical warning. However, traditional statistical indicators rely heavily on empirical design and need help accurately identifying critical transition points under complex coupled factors. We propose a fatigue damage AE warning method based on improved Mel-frequency cepstral coefficients (MFCC) and Long Short-Term Memory (LSTM) networks. The study first employs the Sentry function to label AE signals during the fatigue fracture. Subsequently, it utilizes MFCC coefficients optimized with an improved frequency warping function and an LSTM model to establish pattern associations with critical AE signals. Real-time warning of damage stages is achieved through a sliding window approach. Validation experiments conducted on an AE monitoring dataset from cast steel specimens with fatigue cracks demonstrate that the cutoff frequency significantly influences the high-frequency representation of AE waveforms in MFCC coefficients. Compared to Bi-LSTM and GRU models, the LSTM model exhibits superior warning performance. The proposed improved MFCC-LSTM method accurately predicts critical zones of fatigue damage, providing reliable early warning technology for fatigue damage monitoring in steel structures.