【作者】 赵志刚；

【导师】 赵伟；

【作者基本信息】 清华大学 ， 电气工程， 2009， 博士

【摘要】 较之静态测量系统,动态测量系统更具普遍意义。但长期以来,对于动态测量系统测量不确定度的评定大多采用静态化处理的方法。随着测量系统实时性、动态性要求的提高,这种做法已越来越不适用。而现有动态测量不确定度理论在理论体系和实践方法等都存在有待完善的地方。本文正是着眼这方面问题,从动态测量不确定度的评定、离群值的剔除、测量结果真值的估计、多观测器情况的处理以及多维情况下的处理等对动态测量不确定度理论进行研究,尝试建立一套较实用的动态测量结果测量不确定度评定的理论和方法。首先探讨动态测量数据的组成特点,进而采用建模的方法对动态测量数据的各种成分进行描述和不确定度评定。以评定结果为基础,通过定义置信距离和求解一致性矩阵,实现了对动态测量数据中离群值的剔除。针对动态测量通常无法通过时间上的重复测量来提高可靠性的情况,给出了多观测器空间冗余结构模型和数据处理方法。为适应多维测量结果测量不确定度评定的应用需求,推导了多维情况下显式和隐式映射的多维不确定度传播律公式,给出了多维测量不确定度的几何意义和不同的多维测量结果的质量比较方法,并对多观测器情况下多维测量结果真值最佳估计方法及其测量不确定度评定进行了研究。针对动态测量不确定度在数字计算条件下,计算机处理方法存在较多问题这一情况,本文在每一个环节都给出了算法实现的具体流程,并对最大一致测量数据组的求取、舍入误差引起的不确定度的处理等问题进行了重点研究,提出了采用“逻辑与”运算搜索最大全通图以及分级合并算法等具体解决方案。鉴于测量不确定度评定更多处于附属地位的现状,提出了以测量不确定度参数为核心的系统构建方法。结合理论研究,构建了滤波、表面增强拉曼散射光谱信号处理、传感器失效检测、数据融合以及多平台协同定位等一系列算法,并通过仿真及应用实例的检验,证明了这些算法的可行性。这些方法也为测量不确定度理论在更广范围内的应用和推广提供了一条较为有效的思路。更多还原

【Abstract】 Compared with static measurement systems, dynamic measurement systems are more widely used. But under most circumstance, the measurement uncertainty evaluations of dynamic measurement systems are staticize as static measurement systems. This kind of method is more and more inapplicable for the dynamic measurement systems whose need more and more dynamic and real time performances. That is because the dynamic uncertainty has a lot of deficiencies in the theory and application methods. This thesis aims to do researches in this area, including dynamic uncertainty evaluation, outlier data elimination, true value evaluation, multi-observers and multi-dimensions processing method, etc.The characters of dynamic measurement data are discussed first, and then some models are selected to describe the components of dynamic measurement data and evaluate the dynamic measurement uncertainty. Based on the evaluation result, a confidence distance is defined to describe the consistency of the data from different observers. After that consistency matrix can be got, and the outlier data can be detected by solving the matrix. A spatial redundancy model and its data processing method are set up to resolve the reliability problem of the dynamic measurement system whose reliability usually cannot be improved by time repeat measurement. To evaluate the measurement uncertainty of multi-dimension measurement results, the multi-dimension measurement theory are also studied, including multi-dimension measurement transfer rule, geometric meaning of multi-dimension measurement result, the comparison method of different multi-dimension result, and the true value evaluation method of multi multi-dimension observers.To provide more processing methods for dynamic measurement uncertainty, the applicable algorithms of every theory in this thesis are also provided in every part. Especially, some key technique, such as picking up method of the maximum consistency measurement data set, and reduction method of rounding off errors, are discussed detailedly. Some algorithms, such as searching maximum consistency measurement data set by graph theory and logical AND operation, and hierarchical combine method are set up to solve the problem. As the measurement uncertainty theory and its evaluating are usually not treated as a necessary step, a method using measurement uncertainty as a key parameter to build systems and algorithms are discussed in this thesis. Based on this method, a set of algorithms, such as filtering method, surface enhanced Raman scattering signal processing algorithm, sensor failure detective method, data fusion algorithm, and multi-platform locating method, are set up. After simulation and using in real system, these methods are proved to have good performances. The using methods of measurement uncertainty also supply a new way to spread measurement uncertainty theory.更多还原