中文摘要：

目前大部分大型水泵机组安装有状态监测系统,但如何从海量的状态监测数据中提取出机组故障特征仍是水泵机组故障诊断的一大难点和热点。提出了一种基于小波包和样本熵的水泵机组振动特征提取方法,该方法首先通过小波包变换对水泵机组振动信号进行分层分解,得到小波包频带系数,再结合样本熵算法对小波包频带系数进行重构,得到以各频带信号样本熵值为元素的反映机组故障信息的特征向量,最后采用LVQ神经网络对试验振动信号进行分类,验证结果表明：基于小波包变换与样本熵相结合的特征提取方法对水泵机组不同振动状态具有较好的区分度,是一种合适的水泵机组故障特征提取方法。

英文摘要：

Most large-scale pump units have installed condition monitoring systems currently. How to extract fauh features from the original data is the focus of the water pump faults diagnosis. The paper based on wavelet analysis, presents a method combining wavelet packet transform with sample entropy for signal feature extraction. Firstly, it decomposes the vibration signal by wavelet packet transform to obtain wavelet packet coefficients. Then, it reconstructs wavelet packet coefficients through the method of wavelet packet transform combining samples entro- py algorithm to obtain feature vectors consisting of each band signal sample entropy characteristic elements. Finally, the method combining wavelet packet transform with sample entropy is proved, by analyzing the vibration signal along with LVQ neural networks, to play a good performance in identifying different running conditions of water pump units. It is a suitable way to extract fault features from pump units.