摘要
空化状态识别是离心泵状态监测的难点之一,为了提高空化状态识别的效果,提出了一种基于深度学习的离心泵空化状态识别方法。首先,采集了在3种工况下泵壳的振动信号,分别构建了振动信号的改进倍频带特征矩阵和时频特征矩阵;然后,基于自动编码器构建了深度学习网络,通过无监督训练自动学习输入数据的特征,利用监督训练对网络的参数进行了调整;最后,运用深度学习网络,对离心泵的4类空化状态进行了分类识别。研究表明,无论是基于改进倍频带特征矩阵还是基于时频特征矩阵,深度学习网络对4类空化状态都有很好的识别效果,尤其是对于弱空化状态,深度学习网络比BP神经网络更有效。
Cavitation state recognition is one of the difficulties in condition monitoring of centrifugal pump.A method for cavitation state recognition of centrifugal pump based on deep learning is developed.The vibration signals on pump casing under three conditions are collected,and the band matrixes of modified octave as well as matrixes of time-frequency features of vibration signals are established.The deep learning network is constructed based on autoencoder,the features of input data are learned automatically by unsupervised training,and the parameters of the network are adjusted by supervised training.Four cavitation states of centrifugal pump are recognized by deep learning network.It demonstrates that based on band matrixes of modified octave or matrixes of time-frequency features,the deep learning network is effective for recognizing four cavitation states and outperforms BP neural network,especially for slight cavitation state.
作者
曹玉良
明廷锋
贺国
苏永生
CAO Yuliang;MING Tingfeng;HE Guo;SU Yongsheng(College of Power Engineering, Naval University of Engineering, Wuhan 430033, China;Department of Management Science, Naval University of Engineering, Wuhan 430033, China)
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2017年第11期165-172,共8页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(51306205)
湖北省自然科学基金资助项目(2015CFB700)
海军工程大学博士生创新基金资助项目(4142C15K)
关键词
离心泵
空化状态识别
深度学习
自动编码器
神经网络
centrifugal pump
cavitation state recognition
deep learning
auto-encoder
neuralnetwork