Time-varying speed fault diagnosis based on dual-channel parallel multi-scale information

As rolling bearings frequently operate at varying speeds, the time interval of the induced pulse in the measured vibration signal varies with the speed. This variability makes it difficult to capture the inherent time and frequency information, leading to poor diagnostic performance. In response, this study presents a model for time-varying speed fault diagnosis based on dual-channel parallel multi-scale information. This model incorporates multi-scale information learning into traditional convolutional neural networks (CNNs), employing multiple pairs of convolutional and pooling layers to establish a hierarchical learning structure. Additionally, the model integrates a gated recurrent unit (GRU) to account for time information and enhance the interdependence among time series data. Furthermore, a dual-channel parallel structure is devised to autonomously fuse source data collected by sensors. To validate its effectiveness, the proposed model is assessed using the time-varying speed-bearing dataset. This evaluation includes the introduction of Gaussian white noise and the simulation of experiments in noisy environments. Comparative analysis against advanced methods demonstrates the proposed model's strong discrimination and robustness under conditions of variable rotational speed and noise.