【作者】 宋晴；

【导师】 张国雄； 裘祖荣；

【作者基本信息】 天津大学 ， 测试计量技术及仪器， 2005， 博士

【摘要】 通过液滴分析技术和相关仪器可以获得被测液体的“液滴指纹图”,它揭示了经过液滴内部的光信号随液滴生长变化而变化的规律,液滴指纹图反映液体的多种特性且在一定条件下具有唯一性。本论文的主要目的是研究利用液滴指纹图对液体进行细微识别的方法。本文详细介绍了液滴分析仪,包括液滴传感器、信号处理电路、程控微量供液系统和计算机接口等各部分的工作原理和结构设计,并针对实验中发现的问题对传感器和供液泵作了实质性地改进,大大提高了信噪比。为了使得不同液体的液滴指纹图具有可比性,本文提出了归一化处理的思想和方法,归一化处理从数据形式上排除了供液速度变化对液滴指纹图的影响。本文提出了液滴指纹图特征提取的基本思路:用多个特征值组成的特征向量来描述图形,从而把液滴指纹图转变成多维空间中的一个点;并提出了“波形分析法”、“相关比较法”、“数据压缩法”和“多项式回归法”等对液滴指纹图进行特征提取的具体方法。本文定义了“分辨灵敏度”来定量地描述特征参数对液滴指纹图变化的反应能力。本文提出了液体识别的基本思路:比较被测液体与参考液体的液滴指纹图之间的“差异”,如果差异在设定的阈值范围内就判定为真,否则判定为假。根据“差异”的描述方法的不同,提出了“基于液滴指纹图图形的识别”、“基于液滴指纹图特征向量的识别”和“基于液滴指纹图特征值的识别”三种具体的识别方法,并提出了根据仪器测量不确定度和参考液体本身的分散性确定阈值的方法。通过典型样品的识别实验,研究了各种特征和识别方法适用的液体类型,对于常见液体推荐了适用的识别特征和识别方法。本文的分析证明:通过液滴指纹图对液体进行识别确实是一种新颖的、有效的、简单的且适应性广泛的研究方法,完全有可能应用于打假、液体生产工艺控制、环境监测和其它需要对液体进行细微鉴别的场合。更多还原

【Abstract】 The liquid drop fingerprint (LDF) can be obtained through drop analysis technology (DAT) and related instruments, which reveals the variation laws of the coupled light signal passing through the liquid drop along with the drop growth. LDF externalizes the overall properties of tested liquids and it is unique and definite for a certain liquid under certain testing conditions. This thesis aims to make a research into the method of liquid fine discrimination and liquid identification based on LDF.The working principle and structural design of the drop analyzer is introduced in detail, including the liquid drop sensor, signal processing circuits, program-controlled microflow-feeding pump and computer interface. The drop sensor and the pump are improved essentially to solve some problems found in experiments, so that the signal-to-noise ratio (SNR) is highly increased. An idea of normalization is put forward to make LDFs of different liquids comparable, and the specific processing algorithm is described. Normalization ensures the reproducibility of LDFs against the variation the feeding speed of the pump from the aspect of data format.A conception for feature extraction of LDF is presented, which converts the LDF into a point in the multidimensional space through constructing an eigen vector composed of multiple eigenvalues for LDF. Waveform analysis method, correlation comparison method, data compression method and polynomial regression method are developed to realize feature extraction. Resolving sensitivity (RS) is defined to describe the response of eigenvalues to the variation of LDF quantitatively.A general idea for liquid identification is presented, which compares LDF of the test liquid and that of the reference liquid. The identification result is true if the difference between two liquids falls within the threshold range, and is false if it falls outside. According to various descriptions of differences, three specific identification methods are developed, which are based on the graph study of LDF, based on the eigenvector of LDF and based on the eigenvalue of LDF. The threshold is determined according to the uncertainty of measurement and the dispersibility of reference liquid. Sampling experiments are carried out, and the effectiveness and applicability of feature extraction methods and liquid identification methods are studied. Some recommendations for common liquids are given.Analysis given in this thesis proves that liquid identification based on the liquid drop fingerprint is a simple novel effective method with a wide applicability. It can be used for attacking the counterfeits, liquid process control and environment monitoring and other fields where fine discrimination of liquid is needed.更多还原