Technique Research On Health Monitoring For Turbine Hot Ends

Technique research on health monitoring for turbine hot ends has been the important research content in the field of signal processing.Its corresponding technology has continuous development and progress with the two directions.The first is the study of the feature extraction method.The second is the study of the pattern recognition algorithm,that is the design of the monitor model.The main task of feature extraction method is the study and selection of reflecting signal characteristics and reliable characteristic vector.And the main task of monitor model design is the research of all kinds of monitoring models' structure and algorithms,in order to train and monitor extract feature vector for achieving accurate monitoring purpose.As a result of the turbine blade working under the environment of high temperature and high pressure for a long time,its life directly affect the service life of gas turbine.The blade temperature as an important indicator of evaluation,blade quality is becoming more and more be taken seriously.So the health monitoring theory is used in the temperature monitoring of gas turbine blades.According to the characteristics of the turbine blade signal waveform,it is focused on the turbine blade feature extraction method and the design of the monitor model.According to the characteristics of the turbine blade waveform,this topic studies the characteristics of the three different extraction method to extract the feature information of turbine blade temperature signal.In time domain,through synthetic cosine function to match the turbine blade temperature signal can get the temperature signal amplitude domain,and calculate the dimensionless indicators.In time-frequency domain,on the hand,with the advantages of wavelet packet transform in time-frequency analysis,the theory and algorithm of wavelet packet transform are studied.An algorithm of extracting energy features based on wavelet packet transform is proposed.On the other hand,the Hilbert-Huang Transform(HHT)presents a completely new approach to the analysis of time series data.Based on wavelet analysis and HHT was proposed for feature extraction method.Through the above measures,HHT has higher resolution and accuracy in the aspect of health monitoring feature extraction.In order to improve the performance of monitoring model,this paper is designed and realized to predict the temperature signal of turbine blades based on the MIV,hybrid algorithm of genetic algorithm and adaptive mutation particle swarm algorithm in Elman neural network.The screening of input vector,the independent variable dimension reduction,the realization of Elman network to optimize the initial weights and improving the accuracy of prediction achieve the level of expert grade.This paper research results could be applied to turbine health monitoring areas,including the analysis of the turbine blade signal feature extraction,monitoring,prediction and other related applications.