摘要
针对神经网络结构复杂和训练时间长的问题,提出了一种基于核主元分析的反向传播神经网络齿轮泵故障诊断方法。使用经验模态分解对采集的齿轮泵振动信号进行特征分解形成原始特征参数集,利用核主元分析法提取信号的非线性特征,降低样本维数,并将结果作为神经网络的输入训练齿轮泵故障诊断模型,对测试样本进行诊断。实验结果表明,该方法对齿轮泵样本能够有效聚类,降低网络复杂度,减少网络训练时间和次数,并提高故障诊断的精度。
Aiming at the complex structures and time-consuming problem of neural network,this paper proposes a gear pump fault diagnosis method based on kernel principal component analysis (KPCA) and back propagation neural network (BPNN).Firstly, empiri- cal mode decomposition (EMD) is used to break down the acquired gear pump vibration signal characteristic to form the original char- acteristic parameter set.Secondly,KPCA is used to extract nonlinear feature of the signal and reduce the sample dimensions.Finally,the results can be used as the input of BPNN to train the gear pump fault diagnosis model for diagnosis of the test samples.The experimental results show that the method can effectively realize clustering of gear pump samples, reduce the network complexity, cut down the net- work training time and times, and improve the accuracy of fault diagnosis.
出处
《无线电工程》
2015年第9期72-76,共5页
Radio Engineering
关键词
核主元分析
BP神经网络
齿轮泵
特征提取
故障诊断
kernel principal component analysis
back propagation neural network
gear pump
feature extraction
fauh diagnosis