Gas Path Fault Diagnosis For Aeroengine Based On Fusion Technique

Aero-engine fault diagnosis plays an important role for flight safety and maintenance cost reducing. The design and development of turbofan engine gas path fault diagnosis based on information fusion approaches are focuses on in this dissertation. It covers gas path component health parameters estimation based on fusion filters, qualitative fusion for components fault diagnosis, components fault and sensors fault fusion diagnosis. The major work and contributions are as follows:Engine state variable model is established based on the component-level model and linear Kalman filter is designed to online estimate gas path component health parameters. The conventional Kalman filter with central architecture may lead to severe computational problem and poor fault tolerance. The fusing Kalman filter structure is proposed for gas path component health parameters estimation, and advantages of the fusion filtering approach are analyzed and proved.Since the strong nonlinear characteristics of aero-engine and strong coupling between health parameters, the estimation accuracy by linear Kalman filter to engine gas path component health state might be unsatisfactory. Extended Kalman filter(EKF), unscented Kalman filter(UKF) and particle filter(PF) with fusion filter architecture are separately introduced to formulate the fusion filtering architectures. In order to introduce the prior knowledge into gas path fault diagnosis, the fusing PF algorithm is modified by the truncated probability density function. To solve the problem of measurement noise uncertainty with regards to state estimation, the covariance of measurement noise is estimated by wavelet transform and utilized to update particle importance function of the fusing PF algorithm real time.Adaptive weight evidence theory is studied for aero-engine component fault qualitative fusion diagnosis. The fault fusion decision is made by adaptive weight D-S evidence theory bases on the model based and data driven diagnostic modules. In order to reduce the conflicts between local diagnostic results, the reliability of local diagnosis evidence is acquired with a confusion matrix to give different coefficient. The basic probability assignments of the D-S evidence theory from the local diagnostic results are processed weightily. According to the simulation for single and double component faults, the proposed method can effectively reduce the conflicts between local diagnoses and improve the accuracy of fault diagnosis.Components fault and sensors fault fusion diagnosis is researched. The forms and causes of normal sensor faults are analyzed. Based on the Kalman filter, fusing Kalman filter and Extended Kalman Filter(EKF)-Adaptive Genetic Algorithm(AGA) are separately proposed for components fault and sensors fault fusion diagnosis. The experiments show that the proposed methods meet a satisfactory performance for gas turbine performance monitoring no matter whether the sensor breaks down.