基于深度学习的离心泵空化状态识别

发布时间：2018-05-20 20:00

本文选题：离心泵 + 空化状态识别　；参考：《西安交通大学学报》2017年11期

【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。
[Abstract]:The identification of cavitation state is one of the difficulties in the state monitoring of centrifugal pump. In order to improve the effect of cavitation state recognition, a method of identifying the cavitation state of centrifugal pump based on depth learning is proposed. First, the vibration signals of the pump shell are collected under 3 operating conditions, and the improved frequency doubler band characteristic matrix and the time frequency characteristic moment of the vibration signal are constructed respectively. Secondly, a deep learning network is constructed based on automatic coder, and the characteristics of the input data are automatically learned by unsupervised training, and the parameters of the network are adjusted by supervised training. Finally, the 4 types of cavitation state of the centrifugal pump are classified and identified by using the depth learning network. The eigenmatrix or time frequency characteristic matrix is based on the time frequency characteristic matrix, and the depth learning network has a good recognition effect for the 4 kinds of cavitation states. Especially for the weak cavitation state, the depth learning network is more effective than the BP neural network.
【作者单位】：海军工程大学动力工程学院;海军工程大学管理工程系;
【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)
【分类号】：TH311;TP18

【相似文献】