Research On Model-based Fault Diagnosis And Fault Tolerant Control Of Engine Control System

For modern aero-engine,digital control system is an important part which can improve performance and ensure safety and reliability of engines.In the engine control system,the sensors gathering the sensing signal are the basis of the control system operation,and the actuators executing control system command are the bridge between the controller and the engine.For engines operating in harsh environments with high temperatures,high pressures,and strong vibrations,it is important to ensure the reliability of the control system.The sensors and actuators are fault-prone components,therefore for the faults of the sensors and actuators of aero-engine control systems,the research of fault and fault-tolerant control has significant theoretical significance and engineering application value.In order to meet the requirements of fault diagnosis and fault-tolerant control simulation research of sensors and actuators in aero-engine control systems,a closed-loop simulation platform for turbofan engines was established.The general characteristic curve data of the core machines in GasTurb was scaled to obtain the characteristic data of a specific engine,then the engine common components library was designed by object-oriented programming technology,so that the component-level model of the two-spool mixed flow turbofan engine was established.On this basis,the regulation rules and PID control algorithm were designed to form a closed-loop simulation loop.The gas path component performance degeneration and engine fault simulation values were added to complete the establishment of a closed-loop simulation platform for a turbofan engine.The component-level model and the Extended Kalman Filter(EKF)algorithm were integrated to construct the on-board real-time model of the turbofan engine.In order to improve the adaptive tracking ability of the model,the improved non-dominated sorting multi-objective genetic algorithm was adopted to estimate the engine performance degeneration.Based on the real engine model and on-board model,a reduced deep kernel extreme learning machine(RDK-ELM)algorithm combining deep learning network structure,kernel function and reduced improvement method was proposed as the fault diagnosis algorithm.Simulation results show that the algorithm has high diagnosis accuracy as well as fast training speed.Multiple fault tolerant methods were proposed according to the fault type.The sensors with offset fault diagnosed by the fault diagnosis algorithm was replaced by the analytical redundancy margin provided by the on-board adaptive engine model,the offset fault of the main fuel valve actuator was automatically corrected by the controller with PID control rule because of integral effect,and for the stuck fault of the main fuel valve,active fault-tolerant control was tried by switching control laws.Digital simulation results show that the fault-tolerant control system works well.