基于人工神经网络的离心泵空化故障诊断  

Centrifugal Pump Cavitation Fault Diagnosis Based on Artificial Neural Network

作　　者：刘朝玺 叶志锋[1] 王彬[1] 严社斌 LIU Chaoxi;YE Zhifeng;WANG Bin;YAN Shebin(Nanjing University of Aeronautics and Astronautics,Nanjing 2100016,China;AECC Guizhou Honglin Aro-engine Control Technology Co.,Ltd.,Guiyang 550009,China)

机构地区：[1]南京航空航天大学,江苏南京210016 [2]贵州红林航空动力控制科技有限公司,贵州贵阳550009

出　　处：《机械制造与自动化》2024年第4期151-153,共3页Machine Building & Automation

摘　　要：为了研究离心泵空化故障诊断问题,探究人工神经网络在该问题上的预测效果,通过数值仿真的方法对离心泵流场进行模拟,采集不同状态下流场内各点压力值及工作点作为输入特征,以旋转区域气体体积分数为标签特征,对离心泵空化状态进行神经网络建模。使用LSTM和一维卷积网络处理时序数据,并在特征提取阶段添加正则化损失函数以保证网络稀疏性。最终模型在测试集上的分类任务准确率达到95%以上,能够有效地对离心泵空化程度进行诊断。In order to study the cavitation fault diagnosis of centrifugal pump and explore the prediction effect of artificial neural network concerning the fault diagnosis,the flow field of centrifugal pump was simulated by numerical simulation,the pressure values and working points of various points in the downstream field in different states were collected as the input characteristics and the volume fraction of the gas in the rotating region was taken as the label characteristic to conduct the neural network modeling for the cavitation state of the centrifugal pump.The LSTM and one-dimensional convolutional network were used to process the time series data,and the regularization loss function was added in feature extraction stage to ensure network sparsity.The accuracy rate of classification task of the trained model on the test set exceeded 95%,which can effectively diagnose the cavitation degree of centrifugal pump.

关 键 词：离心泵 空化 人工神经网络 自编码器 

分 类 号：TP277.3[自动化与计算机技术—检测技术与自动化装置]