【作者】 王剑；

【导师】 王新民；

【作者基本信息】 西北工业大学 ， 系统工程， 2020， 博士

【摘要】 舵机是飞机实现飞行姿态和飞行轨迹控制的关键控制部件,舵机一旦出现故障必将严重影响飞机的机动性、飞行安全性等性能。随着全电飞机计划的提出,电液舵机(EHA)和电动舵机(EMA)逐步被用于飞行控制系统来取代传统的液压舵机。虽然EMA在复杂性、重量、可靠性和维护要求等方面比EHA更具吸引力,但电动舵机具有结构复杂,且故障存在偶发性、不确定性、传播性等特点,将电动舵机引入到新的飞机执行器系统中,还需进一步完善故障诊断技术。目前故障诊断存在以下困难:一是突变性故障演变过程极为短暂,尽管故障特征明显,但是难以快速定位;二是渐变性故障演变过程时间很长,故障特征参数在短时间内变化很小,很难准确识别其中的故障信息;三是不同的故障模式可能具有相似的故障特征,故障诊断决策过程中难以处理这种相似故障特征造成的不确定性问题。因此,本文采用了基于人工智能的电动舵机故障诊断方法,借助神经网络、模糊规则推理、贝叶斯推理等方法将人类经验融入到故障诊断的决策过程中,以期准确判断当前可能存在故障的严重程度及潜在危害,提高电动舵机故障诊断的自动化程度,以及系统的可靠性和安全性。解决以上问题的过程中,本文的创新点如下:1)针对电动舵机的突变性故障种类多且故障持续时间短的特点,本文提出了基于交互式多模型—无迹卡尔曼滤波(IMM-UKF)的故障诊断算法。该方法通过引入主动匹配机制来改善故障模型与实际故障匹配度差的问题,主动发现故障或排除故障,改进现有IMM-UKF故障诊断算法采用的被动式贝叶斯推理诊断过程诊断效率低的问题,提高了故障诊断的准确性和快速性。2)针对电动舵机的渐变性故障持续时间长、故障特征参数在短时间内变化很小且容易受到外部干扰的影响、难以准确识别等特点,本文提出了基于动态小波神经网络(DWNN)的故障诊断算法。该方法通过解耦的无迹卡尔曼滤波(DUKF)算法对DWNN网络模型进行了解耦,降低了计算复杂度;利用小波分解算法去除了传感器测量信号中高频分量的影响,并利用反馈神经网络的记忆能力融合了过去的输入信息和预测信息,提高了对电动舵机渐变性故障诊断的准确性。3)针对电动舵机系统中各个部件或子系统之间故障传播的复杂性和不确定性,本文提出一种基于模糊规则推理的智能故障诊断方法。该方法利用模糊合成运算对反映的故障特征与故障模式之间的模糊规则进行综合决策,获得了某一组故障特征隶属于主要故障模式的程度,以及隶属于其它受影响故障模式的程度;然后通过在模糊合成运算中引入权值来修正诊断结果,并构建了与模糊规则推理相一致的模糊神经网络模型来训练权值,克服了人工设置权值会受限于经验和水平的问题,提高了对电动舵机多发性故障诊断的准确性。4)为了获得电动舵机某一故障现象对系统整体可靠性的影响和系统故障严重程度,本文提出一种模糊集—动态故障树分析(FS-DFTA)的系统可靠性分析方法,该方法利用模糊故障树处理电动舵机系统失效行为的随机性和不确定性,并将电动舵机故障树分解为静态和动态两部分;利用动态故障树分析(DFTA)模型描述电动舵机故障系统动态部分的失效概率,可以避免使用马尔可夫方法求解微分方程组的计算复杂性。更多还原

【Abstract】 The actuator is a critical component for aircraft to realize flight attitude and flight trajectory control.When the actuator fails,the performance of the aircraft will be affected,such as maneuverability,flight safety and so on.With the implementation of the all-electric aircraft plan,the traditional hydraulic actuator is replaced by the electro-hydraulic actuator(EHA)and the electro-hydraulic actuator(EMA).In terms of complexity,weight,reliability,and maintenance requirements,EMA is more attractive than EHA,but it is not necessary of fault diagnosis technology to introduce EMA into the new aircraft actuator system.There are many difficulties in fault diagnosis of EMA due to complex structure and specific fault characteristics such as contingency,uncertainty and propagation.First,the evolution process of abrupt fault is very short,although the fault characteristics are obvious,it is difficult to locate quickly Secondly,the evolution process of gradual fault is very long,and the fault characteristic parameters change little in a short time,so it is difficult to identify the fault information accurately.Third,different fault modes may have similar fault characteristics,so it is necessary to deal with the uncertainty caused by similar fault characteristics in fault diagnosis decision-making process.Therefore,artificial intelligence based fault diagnosis method is presented in this paper.With the help of neural network,fuzzy rule reasoning,Bayesian reasoning and other methods,human knowledge is integrated into the decision-making process of fault diagnosis,and the severity and potential hazards of the possible faults can be accurately judged,and the automation degree of EMA fault diagnosis,as well as the reliability and safety of the system are improved.Some innovative work in this paper are as follows:1)In view of the characteristics of multiple types and short fault duration in EMA,a fault diagnosis algorithm based on interactive multi model unscented Kalman filter(IMM-UKF)is proposed.This method improves the existing IMM-UKF fault diagnosis algorithm by introducing the active matching mechanism to improve the poor matching degree between the fault model and the actual fault.The accuracy and rapidity of fault diagnosis is improved by using the proposed method.2)In view of the characteristics of EMA,such as long duration of gradual fault,small change of fault characteristic parameters in a short time,and being easily affected by external interference,it is difficult to accurately identify the characteristics of EMA.A fault diagnosis algorithm based on dynamic wavelet neural network(DWNN)is proposed.Here the decoupling unscented Kalman filter(DUFK)algorithm is used to decouple the DWNN model,which reduces the computational complexity;the wavelet decomposition algorithm is used to remove the influence of high frequency components in the sensor measurement signal,and the memory ability of the feedback neural network is used to fuse the past input information and the past prediction information,so the accuracy is improved.3)Focusing on the complexity and uncertainty of fault propagation among components or subsystems in EMA,an intelligent fault diagnosis method based on fuzzy rule reasoning is proposed.Here fuzzy synthetic operation is used to make a comprehensive decision on the fuzzy rules between the reflected fault features and the fault modes,and the degree of fault features belonging to the fault mode is obtained.Then,the diagnosis results are modified by introducing weights into the fuzzy synthetic operation,and the reasoning of fuzzy rules is constructed.Here the fuzzy neural network model is used to train the weights,the limitation of setting the weights manually by experience is overcome and the accuracy of multiple fault diagnosis of EMA is improved.4)In order to obtain the severity of fault and its influence on the overall reliability of the system,a system reliability analysis method based on fuzzy set dynamic fault tree analysis(FS-DFTA)is proposed in this paper.The fuzzy fault tree,which is used to deal with the randomness and uncertainty of the failure behavior of EMA,is decomposed into static and dynamic parts.Here the dynamic fault tree analysis(DFTA)model is used to describe the failure probability of the dynamic part of EMA,which can avoid using Markov method to solve the differential equations and reduce the computational complexity.更多还原