With the rapid advancement of artificial intelligence and the Industrial Internet of Things technologies, fault diagnosis of rolling bearings has become a critical area of research in the predictive maintenance of industrial equipment. Conventional preventive and periodic maintenance strategies often incur high costs and demonstrate low efficiency. In contrast, fault diagnosis can accurately identify fault locations in bearings, facilitating targeted repairs that ultimately lower costs, reduce labor demands, and enhance product quality. A significant challenge arises from the variations in data distributions across different fault datasets. Building on the QNN-Bi-LSTM architecture, we propose the NQNN-Attention-Bi-LSTM model, which integrates an attention mechanism and a noise-resilient module. The attention component improves the model by adaptively weighting different channels. To address the scarcity of industrial data, the noise-robust module simulates noisy industrial conditions, which not only expands the dataset but also enhances the model’s robustness. The proposed model effectively diagnoses damage across diverse fault datasets.