A CBAM-DRSNet method for vane pumps fault diagnosis based on acoustic-vibration synergistic multimodal information fusion

To address the issue where fault characteristics of vane pumps are obscured under high noise levels and single-sensor diagnostics fail to adequately characterise fault information, this paper proposes a CBAM-DRSNet method for vane pumps fault diagnosis based on acoustic-vibration synergistic multimodal information fusion. First, acquire raw acoustic-vibration signals from the vane pump failure test bench. Subsequently, employ a triangular topology aggregator optimizer (TTAO) to optimize variational modal decomposition (VMD) for signal denoising, effectively suppressing modal aliasing and significantly improving signal-to-noise ratio. Finally, the data-level fusion architecture CBAM-DRSNet is constructed. It employs the convolutional block attention module (CBAM) to adaptively weight multimodal features, enhancing the representation capability of critical fault information. Combined with the soft thresholding mechanism of the deep residual shrinkage network (DRSNet), it filters out noise interference. Experimental results show that TTAO-VMD effectively separates fault features in vane pumps, with a root mean square error of 1.5778, significantly improving upon PSO (1.8590) and GWO (1.7748); on a dataset of five fault states, the CBAM-DRSNet model achieves 100% accuracy, significantly outperforming ResNet (98%) and DRSNet (99.2%); Even in strong noise environments, it maintains an accuracy rate of 94.3%, validating the method’s effectiveness and robustness in complex noise conditions.