Research on Compound Fault Diagnosis of Bearings Using an Improved DRSN-GRU Dual-Channel Model
Shuxin Yin, Zengxu Chen
Abstract
In practical engineering, noise often contaminates the fault signals of rolling bearings, making it difficult to accurately diagnose compound faults. To tackle this challenge, this article introduces a rolling bearing compound fault diagnosis model using an enhanced dual-channel deep residual shrinking network (DRSN)-GRU structure. The model improves the soft threshold function of the residual shrinkage building unit (RSBU), creating the progressive RSBU (PRSBU) module. It constructs a DRSN channel for initial feature extraction, while the gated recurrent unit (GRU) is integrated with convolutional pooling layers to form the GRU channel, designed for extracting linear features. By using a dual-channel connection approach, the model minimizes potential information loss or error accumulation that can occur in a single model structure. In the recognition module, a multilabel classification framework is established to identify compound faults. Experimental results show that, under strong noise conditions, the improved DRSN-GRU significantly outperforms the standard DRSN-GRU and other models, achieving 91.2% accuracy while effectively decoupling and recognizing compound faults.