Font Size: 

In the process of structural damage identification, the structural signals collected by sensors will inevitably be affected by noise generated from environmental and human factors. To prevent weak damage features from being submerged by external noise, enable convolutional neural network (CNN) to accurately extract key features, and improve the damage recognition rate, this study proposes a noise reduction method that combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and an improved wavelet thresholding technique. The effectiveness and practicality of the proposed method are verified through numerical simulations and vibration experiments. The results indicate that the combination of CEEMDAN and the improved wavelet thresholding method yields the best denoising performance, effectively removing noise from structural signals. The average accuracy of CNN-based damage recognition exceeds 95%, which is significantly higher than the damage recognition accuracies obtained by using single denoising methods or CNN recognition without prior denoising.

Structural damage detection, Vibration signal denoising, Empirical modal decomposition, Wavelet threshold decomposition, Convolutional neural network