基于出口压力脉动奇异值的离心泵早期汽蚀故障诊断

摘要

离心泵发生汽蚀故障时的空化流动会使内部流场发生改变，选择适当的算法对蕴含流场信息的出口压力脉动信号进行处理与分析，就可以判断离心泵是否发生汽蚀和汽蚀所处阶段。利用提升db4小波在频域与时域上优异的性能，将离心泵出口处的压力脉动信号分解到不同的时频子空间，通过构建小波重构系数矩阵保证时频信息的完备；随后对时频矩阵进行奇异值分解（ SVD）求取奇异值特征向量；再将所有得到的特征向量作为样本，对改进的BP神经网络进行训练，建立压力脉动信号到汽蚀不同阶段的映射。随机检测几种工况下的压力脉动信号，测试结果表明：对出口压力脉动信号进行小波奇异值分解，可以较好地识别离心泵的早期微弱汽蚀故障，抗干扰性和精确性优于传统诊断方法。%When the cavitations occur, the cavitating flow changes flow-field in the centrifugal pump.With the analysis of pressure fluctuation which containing the information of flow-field at the outlet, whether the cen-trifugal pump cavitation occurred can be judged and cavitation stages can be determined.In this paper, through utilizing the lift “db4” wavelet transform ’ s outstanding performances in the time-frequency domain and establishing wavelet reconstruction coefficient matrix, the pressure fluctuation signals at the pump’ s outlet were decomposed into different subspaces to ensure the integrity of the information; having the singular value decomposition(SVD) of the time-frequency matrix implemented to obtain the singular value’ s feature vector and then having this feature vector taken as the sample to train artificial neural network(ANN) improved to es-tablish the relations between features and stages of the cavitation.Testing some random signals show that the SVDs of pressure fluctuation at outlet can well diagnose the cavitations at early stage and it outperforms the tra-ditional methods in the anti-inference and accuracy.