Research On Soft Measurement Of Axial Force And Radial Force In Nuclear Pump Based On Deep Learning

Nuclear pump bearings are one of the key technologies for the development of nuclear power.The axial and radial forces on the pump shaft during operation will seriously affect the service life of the bearings,and even cause damage to the bearings in severe cases.The nuclear main pump system in service does not allow additional direct measurement devices for axial and radial forces.Therefore,a soft measurement method for axial and radial forces has been developed for the safety monitoring of the nuclear main pump in service It is of great significance to active maintenance.This paper takes the model pump of Shenyang Blower Group Nuclear Power Pump Co.,Ltd.as the test object,and uses the deep learning and soft measurement technology as the theoretical basis to design the soft measurement model of axial force and radial force respectively.This thesis first analyzes the causes of nuclear pump technology,axial force and radial force.Based on empirical formulas and consulting literature,the auxiliary variables that affect the axial force,such as medium temperature,flow rate,inlet and outlet pressure,etc.,are determined.Auxiliary variables such as flow,head,speed,etc.Completed the design of the mechanical structure test plan for axial force,radial force and auxiliary variables.Secondly,according to the characteristics of axial force and radial force,a simplified convolutional neural network soft sensor model structure of axial force and radial force was built respectively.The pooling layer,the model structure of the axial force and the radial force are composed of two continuous convolution layers and a fully connected layer.The two convolutional layers are 3*3*4 and 2*2*4 respectively.Due to the radial force The distribution is uneven.The output of the fully connected layer of the radial force model must have two neurons of radial force size and angle.Therefore,the fully connected layers of the axial force and radial force models are 4 *3 * 1 and 4 * 3 * 2 network structure,respectively.Completed the design of the measurement and control system for the axial force and radial force of the nuclear main pump based on LabVIEW,including the design of the test scheme of the axial force and radial force data and auxiliary variables;the design of the connection database;and the design of data software filtering.At the same time,the sensor model suitable for each auxiliary variable is selected and connected to the industrial control computer through the NI data acquisition card.Finally,the soft measurement model is simulated and verified.According to the error curve and calculation and analysis of commonly used evaluation indicators,the results show that the simplified convolutional neural network soft measurement model has higher accuracy in predicting axial and radial forces,and can be accurately The input variables predict the magnitude and angle of the axial force and radial force of the nuclear pump,which is of great significance for maintaining the safe operation of the nuclear pump.