Thermal Performance Monitoring And Fault Diagnosis For Large Centrifugal Compressor

Centrifugal compressor is one of the most important mechanisms in chemical plants, and it's necessary for us to monitor its working conditions. By now, most of the chemical plants have equipped with online monitoring systems for centrifugal compressors, but few of those systems have fault diagnosis functions. Considerable attention has been devoted to the study of intelligent fault diagnosis for large rotating mechanisms in recent years, and all kinds of advanced intelligent diagnosis theories have been applied to this research. Especially the fashionable method of artificial neural networks, which has powerful abilities of function approximation and pattern recognition, has been widely used in fields of nonstationary time series forecasting and fault diagnosis.Forecasting parameters that reflecting the equipment state, and diagnosis for some probable faults are two most important parts ins ystem of state monitoring and fault diagnosis. In this paper, application of RBF(Radial Basis Function) neural network and Adaline(Adaptive Linear Element) neural network for nonstationary time series forecasting is discussed, and they have been successfully applied to the vibration forecasting of centrifugal compressor. On the other hand, fault diagnosis for centrifugal compressor based on wavelet transform and artificial neural network is also studied in this paper. An experiment system of state monitoring and fault diagnosis for centrifugal compressor is developed by our project group, and the mainly job for the author is to compile software modules, including thermal performance monitoring module and fault diagnosis module. A universal and modularized program is also developed based on theoretical research and experimental results, which is suitable for digital signal processing and wavelet transform and artificial neural networks analysis.Diagnostics for rolling element bearing is also investigated in this paper. Fault detection at the early stage of failure development can be seen as fault prediction to a certain extent. Vibration models of one-point defect and multi-points defects for rolling element bearings are established based on the method of demodulated resonance technique (DRT), and one-point defect vibration model is testified to be a so sensitive and reliable method that it could find the fault position exactly. In addition, fault diagnosis for rolling element bearing using RBF(Radial Basis Function) neural networks is also discussed, and simulation result of the network is very good.