Machine learning-based denoising of acoustic emission signals for fatigue crack growth monitoring in metallic piping materials

Developing effective denoising methods is essential for reliable acoustic emission (AE) monitoring-based fatigue damage assessment. In this study, several commonly used noise reduction techniques for AE-based fatigue damage monitoring are first reviewed, and their respective advantages and limitations are summarized. Subsequently, the AE characteristics during fatigue crack growth (FCG) in metallic pressure piping materials are investigated, with an emphasis on identifying crack-related signals from background noise. Specifically, a denoising framework integrating machine learning with frequency-domain AE features is proposed and validated through FCG experiments on Q235 steel and nickel-based Alloy 800H. The results show that the K-nearest neighbors (KNN) model achieves the highest classification accuracy of 93.5%, maintaining robust performance even during the early stage of crack growth, where noise interference is most severe. Moreover, the evolution of the identified FCG-related signals correlates well with the measured FCG behavior, and a strong positive correlation between amplitude and energy is observed in FCG-related signals for both materials. In addition, SEM analysis reveals that fatigue striations, secondary cracks, and tear ridges serve as the primary AE sources. These findings demonstrate the effectiveness of the proposed method in enhancing AE-based damage monitoring and provide insights into noise reduction strategies in complex noise environments.