水电能源科学

针对泵站机组振动信号非平稳特性和传统故障诊断方法特征提取依赖经验的问题，提出一种基于连续小波变换(CWT)与卷积神经网络(CNN)融合的故障诊断模型，该模型以转子不对中、碰摩及其耦合故障为研究对象，通过CWT将振动信号转换为时频图像，利用CNN实现端到端的特征学习与分类。试验结果表明，该模型在泵站机组轴承故障数据集上的平均诊断准确率达98.7%,较传统支持向量机(SVM)、单一CNN模型分别提升13.6%、5.6%,且抗噪性能优良。模型创新性地集成小波基自适应选择、时频图生成和多尺度特征提取模块，能够显著提升复杂工况下泵站机组的故障识别能力，为水利工程机电设备的智能运维提供了有效解决方案。

To address the non-stationary characteristics of vibration signals in pumping units and the empirical dependence of traditional fault diagnosis methods in feature extraction, this study proposed a fault diagnosis model integrating Continuous Wavelet Transform(CWT) and Convolutional Neural Network(CNN). The research focused on three typical faults: rotor misalignment, rotor-stator rubbing, and their coupled faults. The vibration signals were converted into time-frequency images through CWT, and an end-to-end feature learning and classification was achieved using CNN. Experimental results demonstrate that the proposed model achieved an average diagnostic accuracy of 98.7% on the bearing fault dataset of pumping units, showing 13.6% and 5.6% improvements over traditional Support Vector Machine(SVM) and single CNN models respectively, with excellent noise immunity. The model incorporated adaptive wavelet basis selection, time-frequency image generation, and multi-scale feature extraction modules. The findings indicate that this approach significantly enhances fault identification capability under complex operating conditions, which provides an effective solution for intelligent maintenance of electromechanical equipment in hydraulic engineering.