Abstract
To address the uneven distribution of fault categories in data sets for deep learning-based fault diagnosis, we propose a fault diagnosis framework combining an improved Variational Autoencoder (Similarity-Aware VAE) with a Wavelet-Convolutional 1D-CNN. The Similarity-Aware VAE employs a novel similarity loss function for data augmentation, measuring feature distances in high-dimensional space while automatically adjusting training parameters and weights through an enhanced attention mechanism to balance the dataset.The Wavelet-Convolutional 1D-CNN replaces the first convolutional layer of CNN with a Wavelet-Convolutional layer based on continuous wavelet transform, enabling multi-scale feature extraction for fault data analysis. Experimental validation using public datasets demonstrates that this method effectively enhances data quality while maintaining robust diagnostic performance, offering practical implications for industrial fault diagnosis.