Gas Path Health Parameters Estimation For Turbofan Engines

Fault diagnosis for aero-engine is an important prerequisite to achieve condition basedmaintenance, reduce maintenance cost and ensure flight safety. Based on the nonlinear model of aturbofan engine, the method of gas path health parameters estimation is researched in this thesis.Improved data fitness method is proposed in this thesis, and the piece-wise state variable modelof turbofan engine is established. With the health parameters augmented into the state variable model,Kalman filter is designed and on-board self-tuning real-time model is constructed，then the estimationresults of gas path components’ health parameters are obtained.Considering the contradiction between fewer available sensors and more gas path healthparameters in an engine, a method of linear combination of health parameters is proposed. Atransformation matrix selected through minimizing the Kalman filter’s estimation error is used toconstruct a tuning parameter vector which is a linear combination of all health parameters, and itsdimension is equal to the number of available sensors. Estimation value of the health parameters isobtained through the tuning parameter vector, whose dimension is low enough to enable Kalman filterestimation. Simulations show that this method can solve the on-board real-time modeling problemwhere there are fewer available sensors.Because the difference between engine-to-engine and modeling error could enable the Kalmanfilter’s estimation inaccurate, a compensation module of least squares support vector regression isadded into the self-tuning model. Gaussian clustering method is used to cluster the flight data inreal-time. When the flight is over, new Gaussian clustering data are added into the compensationmodule’s training samples for training and updating off-line. Simulations show that it can increase thehealth parameters’ estimation accuracy of self-tuning model in full flight envelope.Combining the characteristic of the fault diagnosis method based on model and data, supportvector machine is used with Kalman filter and neural network, then a fault classification and healthparameters estimation method is proposed. Sensor measurements are inputted to the support vectormachine for fault classification, Kalman filter and neural network are only used to estimate the faultcomponents’ health parameters. This method can reduce the number of health parameters to beestimated, and yield a significant improvement in estimation speed.