Primary Air Fan fault warning based on universal gravitation neural network and deviation degree

Abstract

In view of the frequent occurrence of Primary Air Fan (PAF) failures in thermal power plants, we use universal gravitation neural network method to establish a PAF vibration fault diagnosis model based on the real-time data of PAF oil pump current, outlet pressure and air volume collected by the SIS system. First, we correlation analysis on the relevant parameters of PAF to determine the main variables which affect the vibration state of it, then estimate the vibration of PAF through the universal gravitation neural network model. Finally, we use the sliding window (deviation) method to realize the diagnosis and early warning of the PAF vibration signal failure. Using the data of a power plant's PAF for simulation, the result shows that the proposed PAF fault diagnosis model based on the universal gravitation neural network in essay has high accuracy and it is easy to implement in engineering.