【摘要】 为实现核电厂旋转机械轴承故障的有效诊断，提出了一种基于改进经验小波变换、自相关峭度和卷积神经网络的特征提取和智能诊断方法。首先，引入数学形态学改进经验小波变换，优化经验小波变换模态划分步骤，避免模态混叠现象。其次，通过改进经验小波变换得到多个模态分量，计算各模态分量的自相关峭度，提取周期性冲击特征，并构建特征向量。最后，搭建和训练卷积神经网络，得到智能诊断模型，实现轴承的智能故障诊断。利用核电厂实测轴承故障和正常信号进行测试，同时与基于经验模态分解和原始经验小波变换的智能诊断方法对比，结果表明，提出的方法平均准确率最高，准确率可达90.67%。更多还原

【Abstract】 To effectively diagnose bearing faults of rotating machinery in nuclear power plants, this paper proposes a feature extraction and intelligent diagnosis method based on improved empirical wavelet transform, autocorrelated kurtosis, and convolutional neural network. First, mathematical morphology is introduced to optimize the mode division steps of empirical wavelet transform, avoiding the phenomenon of mode aliasing. Second, multiple mode components are obtained by the improved empirical wavelet transform, the autocorrelated kurtosis of each mode component is calculated to extracted its periodic impact feature and construct the feature vector. Finally, a convolutional neural network is built and trained to obtain an intelligent diagnosis model, achieving the intelligent fault diagnosis of bearings. The test is conducted using the actual measured faulty and normal signals of bearings in power plants, and the proposed method is compared with the intelligent diagnosis method based on empirical mode decomposition and original empirical wavelet transform. The average accuracy of the proposed method is found to be 90.67%, indicating that it has the highest average accuracy.更多还原