Intellegent Fault-Tolerant Control System Design And Hardware-In-The-Loop Simulation Validation Of A Turbofan Engine

Fault diagnosis and fault-tolerant control technology is one of the most important methods to improve the safety and reliability of an aero engine control system. This paper considers a component level model of a turbofan engine as the study object. For the different malfunctions existed in actuators and sensors of the control system, researches on intelligent-algorithm-based fault diagnosis and fault-tolerant control methods are carried out. A hardware-in-the-loop simulation test is also conducted with engineering feasibility.Firstly, a nonlinear component level model above the idle stage of the certain turbofan engine is constructed, functioning as a research object in this paper. Then, an on-board real-time model isbuilt based on data collected from the component level model. The hardware-in-the-loop simulation testing platform, which is to validate the proposed method, consists of a hardware-in-the-loop tester, a fuel regulator and a rapid prototyping controller. The closed-loops of actuator and speed for the turbofan engine are built on the base of rapid prototyping controller.Secondly, a fault diagnosis method is proposed based on the actuator mathematic model and the inverse mapping fuel model in order to detect and isolate the fault of actuator and LVDT sensor in the fuel actuator closed-loop. The aero-engine fuel inverse mapping model is established using extreme learning machine(ELM) so as to estimate the fuel flow precisely and timely. The hardware-in-the-loop simulation experiment results show that the fault diagnosis method can detect and distinguish actuator fault and LVDT sensor fault with amplitude more than 2% accurately and quickly.Next, the sensor fault diagnosis technology is also studied. An algorithm of Selectively-update regularized online sequential extreme learning machine(SROS-ELM) is put forward to train the model online and estimate sensor measurements. The presented algorithm tackles the problems of singularity and ill-posedness by regularization, which minimize the risk of structure and experience. A mechanism of selecting is used to update output weight of neural networks according to the prediction accuracy and norm of output weight vector, and adopt a dual activation function in the hidden nodes combing neural and wavelet theory to enhance prediction capability. The results of the experiment using regression dataset verify the good generalization performance of SROS-ELM. Finally, SROS-ELM is applied to sensor fault detection and isolation of areo-engine. The results show that the method with this algorithm is effective and feasible.At last, fault diagnosis unit of actuators is isolated from that of sensors. The signal from malfunctioning sensors is re-constructed accurately by using adaptive weighted method to combine the output from on-board real-time model and the estimated sensor signal through SROS-ELM algorithm. An improved global fast non-singular Terminal sliding mode controller is proposed based on previous research to fulfill the design of the intelligent fault-tolerant control system. As shown in the results of semi-physical simulation test, the proposed fault-tolerant control system can swiftly and effectively realize the fault-tolerant control in aero engine control system, improving the reliability.