Turbine Vibration Monitoring And Fault Diagnosis System Based On Virtual Instrumentation

Turbine generator is an important rotating machinery of power plant. Because of its highfailure rates and serious results, the research on its monitoring and diagnosis in operationalstatus has always been the hotspot of industrial diagnosis technology. As the vibration signalof turbine’s rotor contains plenty of time-domain and frequency-domain information,analyzing these signals becomes an important basis for vibration monitoring and faultclassification of the device, also constitutes the main contents of this dissertation.Firstly, after comparing the domestic and international research status and major methodsof the diagnosis of turbine system, we proposed a virtual instrument platform usingLabwindows/CVI system as the major develop tool and research approach. On the basis of thetheoretical analysis on the mechanism of turbine’s common faults, and the experiments on therotor experimental platform, we carried out the spectrums both on time-domain andfrequency-domain, and drew out the figures of the orbits of shaft center. These experimentsfocused on three typical faults: rotor imbalance, bearing misalignment and impact-rubbingbetween movable and stationary parts. Further we verified the vibration characteristics ofthese three faults. On the LabWindows/CVI platform, we developed a multi-channel dataacquisition and pre-processing software based on asynchronous timer mechanism. The systemcan analysis spectrums of real-time operation status.Secondly, after collecting the normal signals and three typical fault signals on the rotorexperimental platform, the dissertation adopted the optimal wavelet package treedecomposition and reconstruction algorithm to analyze the results, and successfully extractedthe features of the three faults. We also proposed to apply the energy spectrum analysis on thefeatures extraction, and ultimately we acquired80sets of sample data.Lastly, based on the Least Squares Support Vector Machine (LS-SVM) theory, whichadopted RBF kernel function, we designed a classifier of diagnosis system. It proposed theParticle Swarm Optimization (PSO) algorithm to optimize the two coefficients: penaltycoefficient C of LS-SVM model and parameter σ of RBF kernel function. The experimentsshow that the proposed method has high accuracy rate in fault diagnosis.