D2FNet: dual-domain fusion network for enhanced bearing fault diagnosis

Abstract Accurate bearing fault diagnosis remains a challenging task, particularly in early stages and under low signal-to-noise conditions. In this study, we introduce D2FNet, a novel dual-domain fusion network designed to enhance fault diagnosis capabilities by integrating time and time–frequency domain features. D2FNet leverages input feature mappings to convert one-dimensional time domain data into two-dimensional images, facilitating the extraction of fault-aware features. Furthermore, a contrastive learning-based Vision Transformer is employed in the time–frequency domain to capture long-range dependencies and boost the separability, significantly improving diagnostic accuracy. Experiments conducted on the challenging Paderborn University dataset demonstrate that D2FNet can achieves superior performance over state-of-the-art competitors, effectively learning discriminative features from both time and time–frequency domains.