Fault diagnosis model for rotating machinery bearings combining 1D-GRU and 2D-ResNet-CBAM

This paper proposes a novel multimodal fault diagnosis method for rotating machinery based on a dual-branch 2D-ResNet-CBAM and 1D-GRU architecture. It simultaneously processes shaft trajectory images and vibration signals to leverage complementary spatial and temporal features effectively. The ResNetCBAM-GRU model was rigorously validated. Experimental results demonstrate superior performance, achieving 98.09% accuracy, 97.08% precision, 96.97% recall, and 97.86% F1-score, all surpassing existing methods. Ablation studies further confirmed the contributions: the dual-branch design achieved a 0.48% accuracy gain compared to the single 2D-ResNet-CBAM branch. This method provides robust and accurate support for rotating machinery fault diagnosis.