Abstract <p>This study proposes a deep learning-based damage recognition method using vibration excitation of flange bolts, focusing on damage feature extraction and classification control of vibration signals, which holds significant value for health monitoring applications. To address noise and nonlinear features in acoustic emission signals, the Mel-frequency cepstral coefficient (MFCC) is used for feature extraction. The ResNet-50 model is improved by integrating the convolutional block attention module (CBAM) and squeeze-and-excitation (SE) modules to enhance recognition accuracy. This innovative approach effectively overcomes the limitations of traditional methods in complex signal processing, improving both the precision and reliability of the recognition process. Experimental results show that the method performs excellently in bolt damage identification with an accuracy rate exceeding 98%, validating its effectiveness and robustness in acoustic emission signal analysis and damage detection. Furthermore, the method demonstrates strong versatility, capable of adapting to damage monitoring tasks under various working conditions, providing reliable technical support for the engineering application of bolt health monitoring in complex environments.</p>