Gath Path Parameter Identification And Fault Diagnosis For Aeroengine

Aeroengine fault diagnosis technology is an important way to ensure flight safety and reduce maintenance costs.In this paper,the methodologies of stochastic process modeling and gas path fault diagnosis are studied for a turbofan engine.The main contents are as follows:This paper reports a self-tuning stochastic modeling to recognize the nonlinear model of engine by identification.A nonlinear static module and a linear dynamic module are designed to form the gray-box Wiener model.The nonlinear module in the gray box Wiener model describes the static mapping relationships between the engine parameters,which can be obtained from the aerodynamic thermodynamic relationships between the input and output experimental parameters.The linear model expresses the dynamic characteristics,where the time constant of the linear module is considered to change with the engine operating conditions.The self-tuning mechanism of time constant under different operating states is established off-line by fast leave one out cross validation kernel extreme learning machine.Compared to conventional Wiener model and neural network,the proposed gray-box Wiener model is in a favor of stochastic process modeling of engine aerodynamic parameters.The discrete and continuous hidden Markov models are separately employed to the time-series modeling method of the engine gas parameters in this paper.Then they are applied to fault diagnosis of the gas path for the engine,and the accuracy of the fault classification and localization is verified.The principal component analysis,kernel principal component analysis and dynamic principal component analysis are designed for the dimensionality reduction of gas path fault feature data,which are used in Markov fault diagnosis model.Three principal component analysis-Markov models are designed and compared in their effects on the fault diagnosis accuracy and computational time of the engine gas path.The result shows that principal component analysis can effectively reduce the online diagnostic time,while the kernel principal component analysis-Markov model method has the best trade-off between accuracy and computational efforts.Aiming at the health state prediction of aeroengine components,a method of combining the Particle Swarm Optimization(PSO)and hidden semi-Markov model(HSMM)is discussed.The PSO is introduced to the hidden semi-Markov model in order to avoid the HSMM training trapped into the local extreme value.The simulation experiments of gas path performance prediction revealed the effectiveness of PSO-HSMM.