Due to the often complex and harsh working environments of pump equipment, the signals collected are frequently marred by high-noise interference. Traditional Convolutional Neural Network (CNN) models struggle to adequately mine fault features under such interference, leading to subpar diagnostic performance. To address this issue, this paper proposes a pump equipment fault diagnosis method based on an improved Densely connected convolutional networks (DenseNet) model. This model incorporates the Depthwise Separable Convolution (DSC) algorithm, Self-Attention mechanism modules, the Gaussian Error Linear Unit (GELU) activation function, and Global Average Pooling (GAP). These enhancements are applied to the foundational DenseNet model to improve its diagnostic capabilities. Experiments conducted with pump equipment data collected by the authors show that the improved DenseNet model achieves a classification accuracy of 99.58% after sample denoising. Additionally, noise simulation experiments further demonstrate that the model proposed in this paper can achieve more effective pump equipment fault diagnosis compared to traditional DenseNet models.