Research On Fault Diagnosis For Turbines Based On Manifold Learning Algorithm

With the rapid development of power industry, the high demands are putforward for the condition monitoring and fault diagnosis of turbo-generator. Thepaper made an intensive and valuable study on several key technologies associatedwith the simulation of vibration fault, feature extraction of vibration fault signal andclassification of fault types.Characteristics of vibration fault and fault symptoms of turbine are studied andseveral types of typical vibration fault are simulated on the ZT-3vibration test-bed,the vibration fault signal is collected and the methods of feature extraction isdiscussed as well, LLE, ISOMAP and LPP et al algorithms are adopted to extract thefeature of vibration signal, which consists of misalignment, misalignment with looseand imbalance of rotor.Artificial Neural Network and Support Vector Machines have been used todetect the type of vibration fault, to make the diagnosis results intelligently.Based on analysis of the principle of LLE, ISOMAP and LPP et al algorithms,the feasibility of feature extraction based on manifold learning has been discussed,this method solves dimensionality problem.This paper adopts the fault diagnosis model of “vibration signal-highdimensional data-feature extracting-pattern recognition” to accomplish faultdiagnosis; feature of fault is extracted by manifold learning, using cloud neuralnetwork classifier and support vector classifier to recognize the fault type. Thispaper proposed fault diagnosis method based on LLE and cloud neural network,method based on ISOMAP and SVM, method based on LPP and SVM. Simulationresult has proved this approach is useful to turbine vibration fault diagnosis, and thelast method provides best diagnosis result.A fault diagnosis method of turbine based on manifold and cloud neural network and support vector machine is proposed in this paper, the results indicatethis method is available to fault diagnosis of rotor.