The Research Of Real-time Sensor Fault Diagnosis Method For Aero-engine

In the turbofan engine control system,the safe operation of the control system and the engine are affected directly by the sensors which work as measure elements.Therefore,to improve the reliability of the turbofan engine,three methods were researched to design the sensor fault diagnosis system,and simulations were carried out to verify these methods.First of all,the nonlinear mathematic model of turbofan engine and its linearization method were introduced in this dissertation.The Kalman filter was designed based on the linearization model.By designing the corresponding input information of each local filter,the sensor fault diagnosis system was established based on the joint Kalman filter.Secondly,based on Nonlinear Auto-Regressive Exogenous(NARX)neural network algorithm,the sensor fault diagnosis system was established.The feedback characteristic of NARX neural network was taken into account,and the outputs of past time were jointed into the inputs,which made use of the multi-step time delay information of the network input and output and improved the training precision of network.At the same time,the fault threshold was used to update the network parameters.According to the result of the diagnosis,the fault sensor signal was eliminated,and the network mapped value was used as the output of the system.Only the weights of hidden layer of the neural network was modified online to improve the real-time performance of the sensor fault diagnosis system.Finally,the studies were focus on the improving of neural network training algorithm and the design of redundancy fault diagnosis system.A new method of calculation the neural network hidden layer outputs based on combined excitation function was proposed.The neural network weights were updated by recursive least square method,which could improve the computational efficiency of the algorithm.Aiming at the redundant sensors fault diagnosis problems,a dual-channel sensor and the neural network were combined to form redundant fault diagnosis system,and information was fused by fuzzy logic.At the same time,the structure of the sensor fault diagnosis system was simplified by reducing the number of inputs and the hidden layer nodes.The advantage of this design is that the multi-sensors fault can be diagnosed effectively due to the less diagnosis input signals and the less signal correlation between the sensors to be diagnosed.The simulation results show that the proposed system can diagnose bias and drift faults of aero-engine sensors.