Research On Early Warning And Diagnosis Method Of Gas Turbine Blade Fault Based On Gas Path Performance Hybrid Model

Blade is an important part of gas turbine,working at high speed,temperature,pressure and load for a long time.It is polluted and corroded by impurity in the air,and the probability of failure is very high.The failure modes include scaling,wear,corrosion and injury.Blade failure seriously affects the stability,economy and safety of gas turbine operation.Therefore,it is necessary to study the fault diagnosis of gas turbine blades.Based on the method of gas path performance diagnosis,this paper studies some key problems in gas turbine blade fault early warning and diagnosis based on hybrid model: 1)the personalized difference of different gas turbines of the same type affects the simulation accuracy of the gas path performance mechanism model;2)Mechanism-only modeling method is difficult to adapt to residual individual difference of gas turbine performance;3)the gas turbine blade fault early warning with single parameter and fixed threshold has high false alarm rate and missed alarm rate;4)There are some problems in the process of blade fault diagnosis,such as limited model accuracy and easy to fall into local optimum.The main work is as follows:The personalized gas path performance mechanism model of gas turbine was developed.In view of the personalized differences in the component performance for different gas turbines of the same type,the existing generalized analytical solutions of components are improved.At the same time,a performance adaptive method based on particle swarm optimization algorithm is proposed.By defining the update factor,the target control of the component characteristic curve shape is realized,and then the accurate matching between the component analytical solution and the actual component characteristics is realized.Aiming at the personalized difference between gas turbine cycle design point and cycle reference point,a cycle reference point tuning method based on backward iteration and genetic algorithm is proposed,which realizes the precise tuning of cycle reference point and improves the accuracy of gas path performance mechanism model.The adaptive adjustment of component characteristic curve and cycle reference point significantly reduces the personalized difference between the actual performance of gas turbine and the mechanism model of gas path performance.The effectiveness of the method is verified by field data of gas turbine.Two type of hybrid drive model construction method for gas path performance of gas turbines is proposed.Aiming at the problem that it is difficult to obtain the characteristic curves and cycle reference points for some gas turbines,a hybrid model construction method of gas path performance combined with gas turbine mechanism is proposed and defined as Type ?hybrid model.This method constructs the Type ? hybrid model based on UnitOriented theory.The selection of neural network structure,neuron number and activation function refer to the modular division of gas turbine,the number of thermal parameters of section and the degree of nonlinearity of components.In view of the fact that the cycle reference point and the characteristic curve of the components are available,but there are still residual personalized differences between the gas turbine mechanism model and the measured data,a hybrid model based on the error compensation of the RBF neural network is proposed and defined as Type ? hybrid model.Based on the mechanism model,this method compensates the error caused by the residual personalization difference through the radial basis function neural network.The effectiveness of the method is verified by the measured data of gas turbine in service.A gas turbine blade fault warning method based on broadband vibration and hybrid model is established.Aiming at the problems of false alarm and missing alarm in the fault warning mode with single parameter and fixed threshold,a multi-parameter based fault warning approach for gas turbine blade fault under varying condition is proposed.Firstly,the deviation characteristic parameters are extracted based on the broadband vibration signal,and the downgrade characteristic parameters are extracted based on the gas path performance signal.Secondly,the threshold setting method of characteristic parameters is studied.Considering the influence of variable working conditions on threshold setting,a three-level early warning rule for blade fault is established.Finally,the effectiveness of the method is verified by the actual fault case data of gas turbine.The gas turbine blade fault diagnosis method based on the hybrid model is studied.Aiming at the problem that the optimization algorithm of nonlinear gas path fault diagnosis is easy to fall into the local optimum,a nonlinear blade fault diagnosis model based on the enhanced particle swarm algorithm and the hybrid model is established.Taking the measured data as the target,the component performance degradation parameter is determined through the adaptation of the gas path performance hybrid model,and then the failure mode of the gas turbine blade is identified.Aiming at the situation that the characteristic curves and cycle reference points of gas turbine components are difficult to obtain,the blade fault diagnosis is carried out based on the measured parameters.However,this method only has high accuracy for single blade fault and low accuracy for multiple blade faults.Therefore,a single blade fault diagnosis method based on SVM and Type ? hybrid model is studied.Based on the above model,an automatic blade diagnosis method based on improved similarity algorithm is proposed,which can automatically identify the type of blade fault.The effectiveness of the method is verified by the measured data of gas turbine blade failure.The research results of this paper can supplement and expand the current blade fault early warning and diagnosis theory,and provide reference for the application of relevant theories in engineering practice.