外测数据处理中小波滤波的鲁棒性算法研究

为消除外测数据处理中异常值和噪声信号对处理结果的影响,结合数据处理的实际,给出一种基于小波变换的鲁棒性滤波算法。首先用移动中值滤波算法剔除原始数据中的异常值,然后采用小波系数去噪算法并结合经验维纳阈值滤波算法,抑制数据中的噪声。仿真计算及实际工程应用表明,该算法在保留特征段及有用信息的同时,有效地剔除了异常值,抑制了噪声,具有很好的鲁棒性。     

基于UIO的航空发动机控制系统传感器故障诊断

研究了航空发动机控制系统传感器鲁棒故障检测与隔离问题,提出了一种克服不同干扰对控制系统诊断性能影响的方法.应用未知输入观测器(unknown input observer,简称UIO)理论来解耦航空发动机动态系统干扰,并针对控制系统传感器设计一族UIO,提取出一系列的传感器残差特征数据,通过分析残差队列的幅值特性,实现航空发动机控制系统传感器故障诊断.在高斯白噪声、模型工作点变化和非高斯噪声三类干扰下的数字仿真结果表明,不管何种干扰,UIO诊断方法均能检测和隔离出传感器故障,在诊断鲁棒性方面,要优于Kalman滤波器诊断算法.      

基于ResNet-LSTM的航空发动机性能异常检测方法

为了实现数据驱动的航空发动机性能异常的智能检测,提出了一种基于残差网络（ResNet）-长短期记忆网络（LSTM）的发动机性能异常检测方法。采用发动机性能数据图像化方法,在数据降维的同时,完备保留数据的关联特征和时序特征;以残差单元构建发动机性能异常检测模型,在加深网络结构的同时,消除深层网络梯度消失问题,提高发动机性能图像空间关联特征的提取能力。同时,引入LSTM,提出基于ResNet-LSTM的发动机性能异常检测模型,通过ResNet与LSTM的融合,强化异常检测模型对时序特征的提取,提升发动机性能异常检测的准确率;通过发动机运行数据进行验证。结果表明：在训练集上,该方法的异常检测准确率为94.95%,比基于ResNet18、ResNet34、ResNet50异常检测模型的分别提高10.87%、8.00%、3.23%;在测试集上,该方法的异常检测准确率为92.15%,比基于ResNet18、ResNet34、ResNet50异常检测模型的分别提高11.81%、9.45%、3.78%。     

基于深度学习的PIV流场图像修复技术

粒子图像测速技术（PIV）是空天动力装置研究中常用的流场测试方法。但对具有复杂流动特征的燃烧室,通过传统互相关算法处理得到的流场结果往往具有一定缺陷。本文将深度学习应用于PIV后处理中,以实现流场数据的异常检测和修复。在甲烷预混对冲火焰数据集上,将异常划分为两种类型,并搭建U-Net卷积神经网络架构。经过训练和优化,模型以较高置信水平识别两类异常并使用不同策略自适应修复,过滤噪声并保留原始正常数据。同时模型具有较好的可迁移性,可以为其它种类的流场数据修复提供参考。与POD迭代法和中值滤波相比,神经网络强大的非线性特征具有明显的优势,这种方法不仅修复率高,而且在不同工况下鲁棒性好。     

航空发动机气路故障诊断的SANNWA-PF算法

针对航空发动机非线性、非高斯的特点,提出一种用于航空发动机气路故障诊断的自适应神经网络权值调整粒子滤波（SANNWA PF）算法。该算法根据粒子分布情况确定分裂和调整的粒子数目,进而根据粒子权重采用正态分布的方式进行分裂,采用反向传插(BP)神经网络进行权值调整,缓解了粒子的退化和贫化,具有更强的自适应性能和跟踪能力。通过一维非线性跟踪模型和航空发动机气路故障诊断仿真研究表明:SANNWA PF算法具有良好的非高斯性能,相对粒子滤波一维非线性追踪模型估计精度提高约21%,航空发动机气路故障诊断在高斯噪声和非高斯噪声下分别提高约30%和26%,诊断速度分别提高约7倍和10倍。      

航天器下行数据异变检测算法研究

针对航天器下行遥测数据故障检测问题,建立了相应的非平稳+异常分量模型,并借鉴采样数据新息增量过程的回归系数有界影响辨识方法,提出了航天器下行数据异常突变在线检测算法,该算法具有简捷的递推关系和良好的容错能力。实测数据结果表明,该算法可以有效地检测出航天器异常数据,并能克服异常数据的不利影响,提高在轨航天器测控过程的可靠性。     

基于RS-CART决策树的航空发动机小样本故障诊断

针对CART(classification and regression tree)分类决策树构建过程中由于小样本集特征维数高及噪声等造成的过拟合问题,在CART决策树算法训练过程中引入基于互信息的粗糙集(rough sets,RS)属性约简,考虑信息熵与基尼(GINI)系数刻画样本集“纯净度”的相似关系,对历史故障数据进行属性约简,降低属性维度以优化训练集,在此基础上构建分类决策树,可视化输出规则。实验结果表明:将改进的CART决策树算法应用于某型航空发动机油液故障诊断,提取的规则可解释性强,能够减小冗余属性及噪声对决策的影响,与常用故障诊断算法相比,该模型的诊断准确率提升20%左右,AUC(area under curve)值高达92%,可以有效处理高维离散型航空发动机小样本故障问题。     

基于多模型的航空发动机传感器混合故障诊断方法

本文在多模型架构下,提出一种航空发动机传感器在线混合故障检测与隔离算法。利用长短期记忆网络逼近航空发动机建模误差、健康参数变化、过程噪声和测量噪声等不确定性源引起的真实发动机与机载模型之间的偏差。将传感器测量输出与不确定性值的偏差用于一种基于多模型的混合卡尔曼滤波器组算法中,利用贝叶斯方法计算每个传感器在健康模式和不同故障模式下的条件概率,然后根据最大概率准则进行传感器故障检测与隔离,克服了阈值难以选取的问题。针对某型涡扇发动机传感器发生偏置故障、漂移故障和间歇性故障的情形进行仿真验证,并对比了不同传感器之间的检测与隔离精度。结果表明：所提出的方法可以在更高水平的退化下诊断出发动机传感器常见的故障,混合方法对不同不确定性源具有鲁棒性。     

基于新息异常检测的改进抗差自适应卡尔曼滤波算法

在工程应用中,量测异常及量测噪声统计特性的时变是引起标准卡尔曼滤波振荡甚至发散的主要原因。经典抗差Sage-Husa自适应滤波方案,对量测中的孤立型异常有所抵抗,并可在线估计量测噪声统计特性改善滤波效果,但当连续型异常值出现时,其滤波效果不佳。针对现有抗差Sage-Husa自适应滤波方案的不足,提出了新的改进滤波方法。在改进算法中,当检测到量测异常时采用模值更大的先验预测方差阵代替原算法中的后验估计方差阵,在估计量测噪声方差时起到放大作用,以降低异常量测权重,提高滤波精度;采用IGG方案构造了新的权函数,可在抑制异常影响的同时调节估计方差阵,以免连续异常时新息持续置零引起的滤波发散;采用标准卡尔曼滤波新息辅助异常检测的双重检测策略,避免了因量测噪声方差阵的调节引起检测阈值变化而导致的漏检率增高。仿真实验表明,与常规抗差自适应滤波算法相比,该方案可更加有效地抑制量测异常值的影响。     

面向空中小目标检测任务的YOLOv7改进模型

针对空中小目标检测任务,本文进行了基于改进的YOLOv7模型的空中小目标检测算法的研究,解决了当前空中小目标检测算法推理速度慢和检测精度不足的问题。首先,建立了涵盖多种目标尺度、姿态和天气条件的飞机目标基准数据集;其次,在YOLOv7基准模型的基础上,提出了一种基于广义特征金字塔网络和Wasserstein度量的目标检测方法;最后,在公开数据集与自建数据集上对该方法与其他主流算法进行了对比试验,结果表明,相较于原始YOLOv7模型,改进模型在自建数据集上对小目标检测的平均精确率提高了7.3%,并且推理速度高于大部分主流检测算法。本文研究为空中小目标检测任务提供了一种快速且高精度的检测算法,对于相关算法在航空航天领域的进一步工程应用具有重要的推动作用。     

航天器在轨故障特征数据生成方法

针对故障特征数据模拟生成算法与航天器的原理、模型关联度高,航天器状态描述模型不成熟、难以应用于工程实践的现状,根据反映航天器在轨故障的遥测数据表现形式和变化时间长短,将航天器故障进行分类。从遥测数据变化的角度,针对每类故障遥测数据的变化特点,研究了符合每类故障数据变化规律的故障特征数据生成算法及基于WEB服务的故障特征数据生成原型系统实现方式。生成的故障特征数据在故障检测、诊断与预警等方面得到成功应用,使航天器对象、状态演化规则和测控信息产生等故障输入数据更接近真实的动态特征。     

The Pioneer Anomaly in the Light of New Data

The radio-metric tracking data received from the Pioneer 10 and 11 spacecraft from the distances between 20–70 astronomical units from the Sun has consistently indicated the presence of a small, anomalous, blue-shifted Doppler frequency drift that limited the accuracy of the orbit reconstruction for these vehicles. This drift was interpreted as a sunward acceleration of a _P =(8.74±1.33)×10^?10 m/s² for each particular spacecraft. This signal has become known as the Pioneer anomaly; the nature of this anomaly is still being investigated. Recently new Pioneer 10 and 11 radio-metric Doppler and flight telemetry data became available. The newly available Doppler data set is much larger when compared to the data used in previous investigations and is the primary source for new investigation of the anomaly. In addition, the flight telemetry files, original project documentation, and newly developed software tools are now used to reconstruct the engineering history of spacecraft. With the help of this information, a thermal model of the Pioneers was developed to study possible contribution of thermal recoil force acting on the spacecraft. The goal of the ongoing efforts is to evaluate the effect of on-board systems on the spacecrafts’ trajectories and possibly identify the nature of this anomaly. Techniques developed for the investigation of the Pioneer anomaly are applicable to the New Horizons mission. Analysis shows that anisotropic thermal radiation from on-board sources will accelerate this spacecraft by ～41×10^?10 m/s². We discuss the lessons learned from the study of the Pioneer anomaly for the New Horizons spacecraft.     

A method for satellite time series anomaly detection based on fast-DTW and improved-KNN

In satellite anomaly detection, there are some problems such as unbalanced sample distribution, fewer fault samples, and unobvious anomaly characteristics. These problems cause the extisted anomaly detection methods are difficult to train accurate classification model, and the accuracy of anomaly detection is hard to improve. At the same time, the monitoring data of satellite has high dimension and is difficult to extract effective features. Based on the DTW over-sampling method, this paper realizes the over-sampling of fault samples in satellite time series, and constructs a distributed and balanced time series data set. The Fast-DTW method is applied to calculate the distance between different time series, which can improve the speed of similarity calculation. KNN (K-Nearest Neighbor) method is applied for classification and the best classification result is obtained by search the optimal hyper-parameters k. The results show that the proposed method has high anomaly detection accuracy and consumes short calculation time.     

航天器故障征兆识别机制研究

为克服针对二值逻辑型阈值比较方法的固有缺陷,本文基于航天器状态划分故障征兆的思想,提出了利用遥测数据的数值型特征甄别遥测野值并获取故障征兆可信度的规则推理新方法。经验证,该机制不仅能够有效避免遥测野值导致的诊断误报现象,而且能够充分体现遥测数据的变化趋势和满足判断条件。     

False Alarm Rate and False Alarm Number for Discrete and Continuous Time Sampling

A comparison of the false alarm rates and false alarm numbers for discrete and continuous time sampling is given for different noise spectrums (rectangular, Hamming, Gaussian, and cosine noise voltage spectrums) and for different sampling rates for the case of discrete time sampling. Also given is the increase in the false alarm rate when going from discrete to continuous time sampling when the detection threshold is held fixed. The amount that the threshold has to be increased for continuous sampling to achieve the same false alarm rate obtained with discrete time sampling is also given.     

动态数据建模中时间序列采样间隔的研究

动态数据建模认为,平稳连续随机过程是白噪声激励相应的线性时不变连续时间系统的输出,为便于计算机处理,需要确定对连续随机过程进行采样的采样间隔,通常人们采用仙农采样定理来对待。文中阐述了对动态数据建模中时间序列采样的问题,用多项式插值的采样理论进行了研究,指出在动态数据建模中,若要使从连续时间随机过程到离散时间序列的统计特性得到保持,采样频率应大于系统最大极点自然频率的6倍。     

Tracking and data acquisition for space exploration

The tracking and data acquisition systems provide the key link between the remote spacecraft and the scientific experimenter on the ground. The operation of the space experiment takes place through the links of command, telemetry and tracking. The evolution from the early very simple spacecraft missions toward more complex and sophisticated missions has been paralleled by a similar evolution in the tracking and data acquisition systems. The early Minitrack interferometer tracking system still carries the major tracking workload for space missions; however greater tracking accuracy requirements for more recent missions, such as the Orbiting Geophysical Observatory and the Apollo mission, have brought about the development of unified tracking and data acquisition systems which utilize hybrid pseudo-random code/sidetone ranging techniques. The data acquisition has evolved from analog telemetry systems to the present day heavy use of PCM digital telemetry. Likewise the command systems have evolved from early simple on/off command systems into PCM digital command data systems. The trend is toward greater real time control of more complex functions on board the spacecraft. Newer spacecraft are incorporating computer-type systems in the spacecraft which require programming and memory load through the ground command link. The most attractive concept for the next generation network for tracking and data acquisition is a network consisting of synchronous-orbit Tracking and Data Relay Satellites for covering launches and low-orbit earth satellites plus a few selected ground stations for supporting spacecraft in high earth orbit and lunar orbit.     

CLUSTER MISSION OPERATIONS

The Cluster ground segment design and mission operations concept have been defined according to the basic mission requirements, namely, to allow the transfer of the four spacecraft from the initial geostationary transfer orbit achieved at separation from the launcher into the final highly elliptical polar orbits, such that in the areas of scientific interest along their orbits, the four spacecraft will form a tetrahedral configuration with pre-defined separation distances, to be changed every six months during the mission. The Cluster mission operations will be carried out by ESA from its European Space Operations Centre; the task of merging the Principal Investigators' requests into coordinated, regular scientific mission planning inputs to ESOC will be undertaken by the Joint Science Operations Centre. The mission products will be distributed to the scientific community regularly in form of CD-ROMs. Principal Investigators will also have access to quick-look science, housekeeping telemetry and auxiliary data via an electronic network.     

基于深度学习的ADS-B异常数据检测模型

广播式自动相关监视（ADS-B）是下一代空中交通运输系统的重要组成部分，是新航行系统中非常重要的通信和监视技术，但其协议没有提供相关的信息认证和数据加密，因此极其容易受到欺骗干扰的影响。针对ADS-B报文数据特点，采用深度学习的seq2seq模型对ADS-B报文数据进行重构，通过重构误差来检测异常，并对数据进行特征扩展，使模型能更好的捕捉数据的时间依赖性。实验结果表明，所采用的方法优于传统的机器学习方法，且在数据特征扩展后，模型检测效果提升。相比于现有的欺骗干扰检测方法，该方法不需要改变ADS-B系统的协议，也不需要额外的节点或传感器参与，具有一定的适应性和灵活性。     

A New Sampling Procedure for the Synthesis of Linear Transversal Filters

In linear transversal filters for signal processing, the impulse response of the filter is syntheszed in a sampled form by means of a tapped delay line with weighted and summed taps. The synthesis procedure consists of sampling the specified impulse response at appropriate intervals and using the sample values as the tap weights. A nonuniform sampling procedure for the synthesis of the impulse response is introduced, which can produce results more conveniently than the classical Shannon sampling procedure because of the simplicity of the transversal filter implementation. According to the new procedure, in the time domain the samples are taken at points where the instantaneous phase of the impulse response is a multiple of ir radians. The spectral distribution which ensues in the frequency domain consists of a series of spectra all having the same fractional bandwidth. The transversal filters synthesized by this sampling criterion are referred to as phase-sampled impulse response (PSIR) filters. An important application of PSI R filters is to the generation and matched filtering of a large class of AM and FM waveforms.