【作者】 张晨；

【导师】 沈绪榜； 陈朝阳；

【作者基本信息】 华中科技大学 ， 模式识别与智能系统， 2005， 博士

【摘要】 从本质上讲,星跟踪器是一个带微处理器的电子照相机,它利用星光方向矢量来确定航天器姿态。本文以APS (Active Pixel Sensor)星跟踪器为研究背景,对基于星跟踪器的航天器姿态确定方法进行研究。论文由APS 性能分析和姿态确定算法两部分构成,其中以姿态确定算法为主。主要关注以下几个关键问题:如何选取导航星;如何设计快速、鲁棒的星图识别算法;如何在无姿态测量信息条件下预测航天器姿态。如何在无陀螺条件下获取航天器姿态和姿态角速率。主要研究工作为: 研究了APS 星跟踪器探测灵敏度,提出了一种计算APS 星跟踪器探测极限星等的简单方法。研究了APS 亚像素质心定位,根据星点模型和质心误差计算模型量化分析了窗大小、星点模型方差和APS 噪声源对亚像素质心定位的影响。对导航星选取算法进行了详细研究。分析了天文星表和天球中恒星的分布。指出了常规星等滤波算法的不足。提出了一种基于Boltzmann 熵的导航星选取算法,用该算法选出的导航星可以同时满足全天星图识别算法和姿态确定算法的需求。研究了角距、星等、星群几何结构等星模式识别SPR(Star Pattern Recognition)特征。为了提高SPR 概率,提出了将角距特征同星群几何特征相结合的SPR 算法。为了实现非等概率条件下快速查找星表,研究了角距分布概率,提出了一种基于导航星角距概率分布的快速查找算法。发展了一种不依赖恒星亮度特征的三角形识别算法,分析了匹配门限、星点质心偏差对该算法识别性能的影响。用实测星图和仿真星图验证了算法的有效性和可行性。证明了星模式矢量矩阵对应的Boltzmann熵在坐标旋转条件下是稳定不变的。为了提高星图识别算法的鲁棒性,提出了一种基于Boltzmann 熵的星图识别算法。该算法以星模式矢量矩阵的奇异值作为星模式识别特征,以最小Boltzmann 熵作为选取星群的准则。在无陀螺条件下,将基于最小模型误差MME(Minimum Model Error)准则的非线性滤波算法应用到星跟踪器姿态确定问题中,根据姿态运动学方程和动力学方程更多还原

【Abstract】 Basically, a star tracker is an electronic camera connected to a microcomputer. The orientation of the spacecraft can be determined by star light vector. The attitude determination algorithms for APS(Active Pixel Sensor) star tracker were investigated. This paper includes two parts: the performance analysis of APS for star tracker and the algorithm of attitude determination, the latter is the central content. Some key problems were considered in the field of spacecraft attitude control. Such as: how to select the guide stars, how to realize star recognition quickly and robustly, how to determine attitude without prior attitude information, how to acquire attitude and attitude rate of spacecraft without gyro. The main work is as following: The detection sensitivity of APS star tracker is analyzed. A simple calculation approach of stellar detection limit for APS star tracker is provided. The sub-pixel Centeroiding of APS is investigated. Based on the model of star point and error calculating mode, the biases of sub-pixel centroiding due to the size of window, the variance of spot and the noise of APS are analyzed theoretically. The selection algorithm for guide stars was studied in detail. The star catalog and the distribution of stellar in celestial sphere are analyzed. The disadvantage of magnitude filtering method was pointed out; a new Boltzmann entropy based guide star selection algorithm was put forward. The guide stars selected by this algorithm satisfy the demand of star recognition and attitude determination at the same time. The SPR(Star Pattern Recognition)features such as angular distance visual magnitude and configuration of star group were investigated. A star pattern identification algorithm, based on angular distance and geometry configuration of star group, was provided in order to increase the probability of star pattern identification. A new searching algorithm, based on the probability of angular distance distribution, was provided and the result of experiment was given with specific parameters. A triangle star identification algorithm independent of star brightness was developed. The influences of matching threshold, centroiding errors to identification algorithm were analyzed. The validity and feasibility of this algorithm were proved by actual star image and emulational star image. It was proved that Boltzmann entropy is invariant with respect to coordinate transformation and robust under uncertainty. A new Boltzmann entropy based star identification algorithm was put forward in order to increase the robust of star identification. The singular values of star pattern vector matrix were utilized as identification feature in star pattern matching, and the minimum Boltzmann entropy was used as criteria to select star group. The nonlinear predictive filter method with the MME(Minimum Model Error) criteria was investigated in the condition of gyroless. The single vector measure and mult-vector measure were analyzed respectively based on the attitude kinematics and dynamics equation. During some special period, star tracker cannot work normally. An attitude prediction method was investigated, which used the AR predictive model. The one-step prediction and mult-step prediction were tested based on the kinamic equation and the wheel angular momentum.更多还原