Reconstructing acoustic emission of mechanical seals using PILFN with soft measurement strategy

Mechanical seals are essential for ensuring the reliability of rotating machinery. However, direct acquisition of acoustic emission (AE) signals at the seal interface is infeasible due to structural constraints, and indirect measurements are often employed that result in significant attenuation, leading to significant attenuation, especially in high-frequency components that are critical for condition monitoring. To address this, a novel soft measurement method based on a hybrid deep learning framework called PILFN was proposed. The LSTM module captures the temporal dynamics of the AE signal, while the Fourier feature layer extracts frequency features. Additionally, physical knowledge is integrated via a modified power-law attenuation formula, guiding the model to recover pre-attenuation signal characteristics. Experiment results demonstrate that the PILFN model achieves better reconstruction accuracy, with an average RMSE of 0.0062, compared to 0.2 obtained by the LSTM-only model. This improvement allows the reconstructed signals to better represent the tribological information of the seal interface, enhancing the accuracy and reliability of mechanical seal condition monitoring.