【作者】 杨勇；

【导师】 万磊；

【作者基本信息】 哈尔滨工程大学 ， 船舶与海洋结构物设计制造， 2013， 硕士

【摘要】 智能水下机器人工作在一个复杂且不确定的海洋环境中，且受人工干预有限，容易出现各种故障，严重的故障会导致智能水下机器人的丢失。因此，对智能水下机器人进行故障诊断和容错控制研究是十分必要的。本文主要在智能水下机器人执行器和传感器的故障诊断与容错控制方面进行了研究，该项研究具有重要的工程应用价值和理论意义。首先，对智能水下机器人进行动力学和运动学分析，建立水动力方程和大地坐标系下的运动方程，在此基础上得到智能水下机器人运动的状态方程。根据状态方程给出该系统的执行器故障模型和传感器故障模型，为后续故障诊断打下基础。在执行器故障模型中，控制输入由期望的控制力(矩)和由于执行器故障导致的控制力(矩)损失两部分组成。将控制力(矩)损失与运动状态一起构成扩展状态，用基于高斯粒子滤波的非线性系统状态估计方法估计扩展状态，从而得到控制力(矩)损失的估计值；利用修正的贝叶斯(MB)方法分析控制力(矩)损失值的时间序列，检测故障；检测出故障后，用滑动窗口法估计控制力(矩)损失的幅值。根据执行器与控制力(矩)的关系，参考当前执行器的使用情况，将故障定位到执行器，完成执行器故障诊断。对水平运动面的执行器故障诊断进行了MATLAB仿真实验，结果表明该方法能够实现执行器的故障诊断。实际试验验证了该方法的可行性和有效性。在基于高斯粒子滤波的非线性系统扩展状态估计中，可以得到运动状态的先验估计值，从而得到输出的先验估计值，将此输出估计值与传感器的测量值比较得到残差，通过对残差序列的阈值分析来诊断传感器故障。通过MATLAB仿真实验检验了该方法的有效性，并且对执行器故障和环境噪声具有很强的鲁棒性。在分析了ZS智能水下机器人的推力分配方法后，在广义逆推力分配的基础上进行推力重分配，即在推力分配中改变推力器约束条件、推力器权值和推力器配置矩阵。在Simulink中，建立ZS智能水下机器人的推力器故障诊断与容错控制仿真系统，仿真结果表明故障诊断方法能够诊断出设置的故障，容错控制措施能有效地减轻推力器故障对ZS智能水下机器人运动控制效果的影响。更多还原

【Abstract】 The working condition of autonomous underwater vehicle (AUV) is uncertain andcomplex and the manual intervention to the vehicle is limited, then it’s unavoidable thatthere are all kinds of faults appearing in this system. What’s more, some serious faultsmay lead a lost of the vehicle. So it’s necessary to research fault detecting and diagnosis(FDD) and fault-tolerant control (FTC) system for autonomous underwater vehicle. Inthis paper, we have proposed and verified some methods to diagnosis actuator faults andsensor faults, and have research FTC measures overcome actuator faults. These methodsand measures are useful in engineering application and theoretical significant.Firstly, through dynamics and kinematics analysis of autonomous underwater vehicle,we make the hydro-dynamic model of this platform and found the kinematics equation inthe geodetic coordinate system. Then, get the state equation of motion for AUV. Basingon these works, we can build actuator fault models and sensor fault models.In the actuator fault models, the control inputs make up of two parts: one is theexpected control forces or moments and the other one is the control force or momentlosses, the results of actuator fault. We look on theses losses as extended states of theplatform and estimate them together with the velocity and position states by nonlinearsystem state estimation method based on the Gaussian particle filter. The sequences ofloss estimations are detected by Modified Bays method to find the fail and SlidingWindow method is used to estimate the amplitude of control force or moment losses.According to relationship between control forces or moments and actuators, Faults arepositioned and separated combining with actuators used by system at present moment. Asimulation of actuator FDD on the horizontal motion surface is conducted and the resultsshow that the method could realize the actuator fault diagnosis. Another trial proves thatthe method is feasible and effective.We can get the prior estimation of motion states through the nonlinear system stateestimation based on the Gaussian particle filter, and then the prior estimations of outputsis calculated. We compare the estimation of outputs with measured value to get theresidual sequence which will be detected by Threshold value analysis to diagnose thesensor faults. The effectiveness is verified by means of simulation, and the simulation results also show that the method is robust to actuator faults and ambient noise.We study the thrust distribution method of ZS, an autonomous underwater vehicle inmy Lab., then redistribute the control on the basis of generalized inverse thrustdistribution. The main strategies to redistribution are changing the bound value ofthrusters, the thruster weights and the matrix of thruster configuration. A simulationsystem of FDD and FTC to actuator faults is founded for ZS in MATLAB/Simulinkenvironment. The simulation results show that FDD method could diagnose the actuatorfaults added before and FTC measures could effectively reduce impact of actuator faultson performance of motion control.更多还原