【作者】 刘洋；

【导师】 赵旭东；

【作者基本信息】 大连理工大学 ， 控制科学与工程， 2025， 硕士

【摘要】 航空发动机作为飞行器系统的核心部件,其工作环境极其复杂,长期处于高温、高压、高转速等严苛工况下,性能参数呈现显著变化,易引发发动机部件性能退化与故障的发生,其中传感器与执行器是故障率最高的两类部件。鉴于航空发动机的安全可靠运行直接关乎飞行器的整体安全,开展针对其传感器与执行器的容错控制研究具有重要的理论价值与工程意义。本文基于状态观测器、数据驱动和模型预测控制的方法,针对航空发动机传感器与执行器的故障容错问题展开深入研究,主要内容包括:(1)基于动力学与热力学原理,系统阐述了发动机核心部件的工作原理与数学建模方法,推导出发动机共同工作的非线性方程组。采用小扰动线性化方法,在发动机稳态工作点附近对非线性模型进行线性化处理,并通过仿真验证了线性模型的准确性,为后续容错控制器设计奠定了可靠的模型基础。(2)针对航空发动机传感器精度下降的故障问题,提出一种基于复合观测器的模型预测容错控制方法。该方法构建的复合观测器通过卡尔曼滤波器(Kalman Filter,KF)与扩张状态观测器(Extended State Observer,ESO)协同工作,其中KF用于抑制故障噪声信号对ESO状态估计的影响,并通过其估计状态动态替换ESO中的实际测量值,同时ESO实时更新的外部扰动信息被反馈至KF的发动机系统估计过程。在此基础上,通过将复合观测器结构嵌入模型预测控制框架,有效解决了传统复合观测器反馈控制难以处理约束条件的局限性。仿真结果表明,所设计的控制器在严格满足输入约束条件的同时,不仅能有效实现传感器精度下降的容错控制,还能在存在不可测扰动的情况下维持高压转子转速的稳定控制。(3)针对航空发动机执行器故障问题,提出两种基于不同方法的模型预测容错控制策略,旨在提高发动机实时在线模型的精度以确保模型预测容错控制的有效性。第一种方法采用基于n-step扩张状态观测器的模型预测容错控制,通过多步预估机制克服了传统单步ESO预估精度不足的缺陷,显著提高了系统状态和故障信号的估计精度;第二种方法采用基于在线窗口动态模态分解的模型预测控制,该方法无需额外引入故障检测和隔离单元。通过KF抑制过程噪声对数据采样的干扰,并融合窗口历史数据与实时测量数据实现故障信号的在线精确辨识。仿真结果表明,两种方法均能准确估计故障信号,并在模型预测控制框架下有效维持高压转子转速的稳定控制。更多还原

【Abstract】 The aero-engines,as the core component of aviation systems,operate in extremely com-plex environments characterized by prolonged exposure to high temperatures,high pressures,and high rotational speeds under demanding working conditions.These severe operational parameters exhibit significant variations that readily induce performance degradation and faults in engine components,with sensors and actuators being the two most fault-prone ele-ments.Given that the safe and reliable operation of aero-engines is directly critical to overall flight safety,research on fault-tolerant control for these sensors and actuators carries substan-tial theoretical significance and engineering value.This thesis conducts in-depth investigation into fault-tolerant solutions for aero-engine sensors and actuators by employing state observer,data-driven,and model predictive control.The main research contents include:(1)Based on the fundamental principles of dynamics and thermodynamics,this thesis systematically elaborates the working mechanisms and mathematical modeling approaches for core engine components,deriving the nonlinear equation set governing coordinated engine operation.Employing small-perturbation linearization methodology,the nonlinear model un-dergoes linearization treatment near engine steady-state operating points,with simulation ex-periments validating the linear model’s accuracy.This establishes a reliable modeling foun-dation for subsequent fault-tolerant controller design.(2)This thesis presents a composite observer-based model predictive fault-tolerant con-trol method to address sensor precision degradation in aero-engines.The proposed approach features a synergistic dual-observer architecture combining a Kalman Filter(KF)and Ex-tended State Observer(ESO),the KF is employed to suppress the interference of fault-induced noise on the ESO state estimation,while dynamically replacing the actual measured values in the ESO with its estimated states,and the ESO provides real-time disturbance feedback to enhance the KF’s estimation accuracy.By integrating this composite observer within a model predictive control framework,the solution overcomes traditional constrained-condition han-dling limitations of observer-based methods.The simulation results demonstrate that the de-signed controller strictly adheres to input constraints while effectively achieving fault-tolerant control against sensor precision degradation.Moreover,it maintains stable control of the high-pressure rotor speed even in the presence of unmeasurable disturbances.(3)This thesis proposes two model predictive fault-tolerant control strategies based on distinct methodologies for aero-engine faults,aiming to enhance the accuracy of real-time online engine models and ensure effective model predictive fault-tolerant control.The first method employs an n-step extended state observer-based model predictive fault-tolerant con-trol,which overcomes the limitation of conventional single-step ESO estimation accuracy through a multi-step prediction mechanism,significantly improving the estimation preci-sion of system states and fault signals.The second method utilizes an online windowed dynamic mode decomposition-based model predictive control that eliminates the need for additional fault detection and isolation units.KF is employed to suppress process noise in-terference in data sampling,while windowed historical data and real-time measurements are fused to achieve online accurate fault signal identification.Simulation results demonstrate that both methods can precisely estimate fault signals while effectively maintaining stable high-pressure rotor speed control within the model predictive control framework.更多还原