Aeroengine Fault Diagnostics Based On Kalman Filter

Aeroengine diagnostics is the core of modern propulsion system health management technology. The effective diagnostic system is the essential foundation for achieving the condition-based maintenance, reducing the equipment operation cost and improving the flight safety. This dissertation focuses on the development of aeroengine diagnostic system. Kalman filter and model-based fault diagnostic method are used to build up both sensor and component fault diagnostic system.Firstly, an on-board self-tuning linear engine model is developed based on the nonlinear model. An optimal approximate method is proposed to improve the modeling efficiency and accuracy. Gas path components health parameters are defined to represent the health condition of components and are introduced into the linear model. The degraded health parameters are estimated by Kalman filter. Through updating the health parameters, the linear model is able to track the outputs of both nominal and degrading aeroengines.An engine fault diagnostic system is build up based on the linear model and Kalman filter. Several hypotheses are proposed to estimate the component health parameters for both gradual and abrupt fault by using linear Kalman filter. Through the innovation sequence characteristic analysis, a robust Kalman filtering algorithm is worked out to eliminate the sensors disturbance and outliers'effect on the estimation results. Simulation results prove efficiency of the developed robust Kalman filter.The aeroengine sensor diagnostic system is set up based on a bank of Kalman filter. The sensor fault detection, isolation and reconstruction algorithm is studied to realize both single and double sensor fault diagnosis. A component fault diagnosis module is added to set up an enhanced aeroengine fault diagnostic system for both sensor and component fault diagnosis. Digital simulation results show the enhanced diagnostic system could easily distinguish sensor or component fault in the case of either component or sensor fault. However, when both sensor and component fault taking place, the system can only detect abnormal but can not isolate the fault.A nonlinear model based aeroengine diagnostic system is developed. The unscented Kalman filter is chosen as the health parameter estimator for its high accuracy and easy implementation for nonlinear estimation. The system and measurement noise covariance calculation method is studied to improve the accuracy and convergent speed. Theδ-c function is referred to improve the UKF robustness against the measurement outliers or sensor fault while another dual filtering method is proposed to achieve both the sensor and component fault diagnosis. Digital simulation results show that the robust and dual filtering algorithm work effectively in resolving the component fault diagnosis issues when sensor fault or disturbance existing.In order to address the defect of original UKF, a spherical square root UKF (SSUKF) is proposed, which could guarantee the same accuracy of nonlinear parameter estimation with better stability and less computational cost than original UKF. Using SSUKF, aeroengine performance monitoring and diagnosis are achieved based on transient measurements, which would be great helpful to realize online real-time engine monitoring and diagnosis system.Aeroengine algorithm fusion fault diagnostic algorithm is studied via integrating neural network and Kalman filter techniques. The auto-associative neural network is used as the sensor validation function module to detect sensor fault and reconstruct the biased measurements. The probabilistic neural network works as classifier to detect which component may be faulty. Based on the fault diagnostic conclusion of neural networks, the Kalman filer estimates the health parameters of only the faulty component to attain the assessment of the aeroengine health condition. Digital simulation results highlights efficiency of the proposed algorithm fusion aeroengine diagnostic system.