Aeroengine Fault Diagnostics Based On Fusion Technique

Aero-engine fault diagnosis is one of the essentials to advanced aerospace propulsion system. The effective diagnosis system is the important prerequisite to achieve the condition-based maintenance, reduce the operation cost, and ensure the flight safety. This dissertation focuses on the development of aero-engine fault diagnosis system based on information fusion technology. It mainly deals with gas-path components fault diagnosis, sensors fault fusion diagnosis, qualitative and quantitative fusion for components fault diagnosis, and cross-integration for components fault diagnosis. The major work and contributions are as follows:1. An on-board self-tuning aero-engine model is studied and the health parameters of aero-engine are analyzed. The large-deviation state-variable model is built from the nonlinear component-level model based on the small perturbation. The health parameters representing the health condition of gas-path components are estimated by kalman filter in the on-board self-tuning aero-engine model. The faults on components are simulated, and the relation between the measurements and the health parameters under gas-path components performance degradation is analyzed.2. The aero-engine component faults are researched based on the model based diagnosis method and data driven one separately. Single-output least square support vector regression (LSSVR) is extended to the multi-outputs one. Multi-outputs LSSVR (MO-LSSVR) algorithm is proposed and applied to the component fault diagnosis based on data driven method, and the simulation shows that MO-LSSVR could solve the multi-outputs parameters estimation problem, and simplify model structure. Parameters of LSSVR are optimized using adaptive genetic algorithm. LSSVR with adaptive genetic algorithm (AGA-LSSVR) is proposed and used to compensate the on-board self-tuning model. The method decreases the errors between the real engine and the model and improves the accuracy of fault diagnosis.3. The fusion method for aero-engine sensor fault diagnosis is studied. The Self-tuning particle swarm optimization (SPSO) algorithm is proposed to update the particle self-tuning velocity and location, which could expand the particle global search capability and improves the convergency. The SVR model with optimized hyperplane by SPSO algorithm is derived, which solves the high-order matrix problem encountered in the optimal SVR hyperplane based on Lagrange duality principle. The sensors fault diagnosis contains the function of monitoring, isolation, and recovering. The prediction model of sensor parameters is constructed in SPSO-SVR monitoring module, and the fault sensor signal is recovered under the SPSO-SVR signal recovery mechanism.4. Component fault diagnosis based on fuzzy decision fusion is proposed for aero-engine. The method improves the accuracy of component fault diagnosis, and reduces probability of false alarming and missing detection. The sensor outputs are sent to the model based and data driven diagnostic modules to estimate health parameters respectively. Fuzzy algorithm is operated to adjust the weights of two modules. The fault fusion decision is made by D-S theory. Simulation on the turbofan engine shows that compared to the two single-modules, the fusion one has better decision for fault diagnosis.5. Quantitative fusion in parallel is studied for aero-engine component fault diagnosis in characteristic level. The fusion strategy for component health parameters with self tuning weight optimized by quantum particle swarm (QPSO) is proposed, which could estimate the health parameters in continuous degradation space. The component quantitative fault diagnosis system is designed for abnormal detection and health parameters estimation. Evolutive SVR (ESVR) is applied to parallel fusion diagnosis in characteristic level. According to the simulation for single and double component faults, both QPSO and ESVR algorithms have better accuracy of fault diagnosis.6. The cross-integration technology for the fault diagnosis of aero-engine components is researched. The cross-integration fault diagnosis system is designed, which contains signal correlation knowledge, pattern recognition, information strengthening, and ESVR parameters estimation. The measure signals are analyzed by grey correlation for each fault pattern to obtain the signal correlation knowledge. Two-layer grey correlation net is set up for pattern recognition. The information coming from strong signal in data level and characteristic signal in characteristic level is crosswise integrated based on ESVR algorithm. The method is more conducive to meet the real-time requirement by eliminating redundant input signal.