Rolling Bearing Fault Diagnosis Based on CEEMDAN-IWT Denoising and Multiscale Feature Fusion

To address the issue of insufficient fault characteristics in rotating machinery such as bearings operating under complex conditions, where environmental noise interference makes accurate diagnosis difficult, a fault diagnosis approach for rolling bearings is presented, utilizing CEEMDAN-IWT denoising and a multiscale feature fusion convolutional network. First, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method is utilized to break down the noisy vibration signals, obtaining intrinsic mode function (IMF) components at various levels. Key IMF components are selected for screening, and redundant noise is further removed from the selected IMF components using an improved wavelet thresholding algorithm, achieving dual denoising of the noisy signals. The denoised signals are then transformed into two-dimensional time-frequency images via wavelet transform and fed into a multiscale feature fusion convolutional neural network (MFCNN). This model is capable of cascading convolutional layers at different scales to extract multiscale feature information, thus achieving effective fusion of multiscale features. Finally, the viability and effectiveness of the suggested approach were validated using an open-source dataset. Experimental results demonstrate that the method proposed in this paper achieves better denoising performance and higher fault diagnosis accuracy compared to other methods.