加权非负最小二乘光子相关光谱纳米颗粒粒径反演方法

为了降低采用光子相关光谱法进行纳米颗粒测量时噪音对颗粒粒径反演结果的影响,提出了一种基于加权非负最小二乘法的光子相关光谱纳米颗粒粒径计算方法.该方法以光子相关光谱自身作为权值,推导出反演算法的离散模型,避免了接近零点的测量数据波动对测量结果的影响.利用光子相关光谱纳米检测实验平台对90nm、190nm及混合的乳胶颗粒进行实验研究,并与传统非负最小二乘法反演结果进行了对比.60s测量时间的30次实验数据表明:对单峰颗粒群进行反演时,该方法多次测量结果与传统非负最小二乘法结果相近,但是多次重复测量的方差较小,证明该方法重复性较好;对多峰颗粒群进行反演时,该方法反演结果更接近颗粒的真实值,而非负最小二乘法其反演结果与真实值有较大偏离.在不同测量时间的实验数据表明:测量较短的情况下,该方法反演结果方差较小,能在更短的采样时间情况下,获得更准确的测量结果.     

基于无约束非线性优化算法的光强自相关函数拟合

在光子相关光谱(PCS)颗粒测量技术中非线性累积反演方法是一种重要的算法。采用无约束非线性优化算法对反演过程中相关曲线的拟合问题进行了分析和研究, 实现了从任意初始值开始获取最优非线性拟合解, 并给出了算法的实现步骤。采用此方法对50 nm, 100 nm, 200 nm和500 nm的标准聚苯乙烯乳胶颗粒的实测光强自相关函数数据进行了拟合, 并反演了颗粒粒径, 与其他曲线拟合算法相比, 本算法不但对初始值的依赖性小, 而且得到的颗粒粒径更接近乳胶颗粒的标称直径, 测量误差更小。     

用于光子相关光谱颗粒粒度测量的实数编码混合遗传算法

提出一种自动缩小反演范围的实数编码混合遗传算法,并用于光子相关光谱颗粒粒度测量中.该算法利用混沌遍历性质,融入模拟退火算法,具有较强的局部搜索能力.通过数值模拟,讨论在无噪声和多种噪声水平影响下单分散颗粒系的反演结果;分析90 nm聚苯乙烯标准颗粒样品的实验测量数据.结果表明,反演的粒径分布在峰值位置、峰宽度、峰对称性等方面很接近颗粒的真实分布,具有较强的抗噪声干扰能力.     

基于奇异值分解的光子相关光谱滤波方法研究

针对光子相关光谱颗粒测量法在测量超细纳米颗粒时,容易受噪声影响,导致拟合误差较大的问题,提出了一种基于奇异值分解的光子相关光谱滤波方法。其处理步骤为:利用颗粒系的光强自相关函数数据构造Hankel矩阵H;对矩阵进行奇异值分解;根据奇异值的大小分布,确定噪声级别和重建参数r;从重建矩阵H1中提取经滤波后的光强自相数据,再通过传统方法进行拟合,得到颗粒的粒径分布。实验中采用30nm标准乳胶球单分散颗粒系,以及30nm和100nm标准乳胶球双分散颗粒系进行实验对比。结果证明:基于奇异值分解的光子相关光谱滤波法有效地提高了测量准确性。     

基于奇异值分解的光子相关光谱滤波方法研究

针对光子相关光谱颗粒测量法在测量超细纳米颗粒时,容易受噪声影响,导致拟合误差较大的问题,提出了一种基于奇异值分解的光子相关光谱滤波方法。其处理步骤为:利用颗粒系的光强自相关函数数据构造Hankel矩阵H;对矩阵进行奇异值分解;根据奇异值的大小分布,确定噪声级别和重建参数r;从重建矩阵H1中提取经滤波后的光强自相数据,再通过传统方法进行拟合,得到颗粒的粒径分布。实验中采用30nm标准乳胶球单分散颗粒系,以及30nm和100nm标准乳胶球双分散颗粒系进行实验对比。结果证明:基于奇异值分解的光子相关光谱滤波法有效地提高了测量准确性。     

夫琅和费衍射颗粒粒度测量中的改进Chin-Shifrin反演算法

在基于线阵CCD的夫琅和费衍射颗粒粒度测量中,采用Chin-Shifrin积分变换反演算法使得反演的粒度分布出现假峰现象.为解决此问题,提出在该Chin-Shifrin积分变换反演算法中引入矩形窗函数,并在分析颗粒粒径与衍射光强导数最小值之间关系的基础上,确定矩形窗函数中心点位置及左右边界,利用该矩形窗函数对粒度分布进行截断处理,消除虚假峰,提高反演颗粒粒度分布的准确性.分别对两种标准颗粒进行了测量,并对不同算法的反演结果进行了对比.实验结果表明:引入矩形窗函数的改进Chin-Shifrin算法,能够有效排除粒度分布中的多假峰;粒度分布测量相对误差小于3%,重复性小于4%.     

基于PCA-BP神经网络的光谱消光法颗粒粒径反演算法研究

光谱消光法广泛应用于颗粒粒径测量领域,在利用光谱消光法对颗粒粒径进行反演的过程中,由于颗粒的消光系数存在理论复杂、计算繁琐、收敛速度慢以及求解不稳定等问题,很大程度上影响了整个反演过程的快速性和准确性。且在众多波长的消光数据中,存在较多重复冗余的信息,也很大程度上增加了反演算法的时间。针对光谱消光法粒径反演算法计算繁琐、反演效率低的问题,提出了基于主成分分析（PCA）和BP神经网络的光谱消光颗粒粒径分析方法。基于Mie散射理论对不同粒径、不同波长下的光谱消光值进行了仿真计算,通过对光谱消光数据集的主成分分析及各个波长综合载荷系数的计算,实现了最优特征波长的选取,利用降维后的光谱消光数据训练了PCA-BP神经网络模型,并利用该网络模型计算了粒径颗粒分布。通过仿真计算,比较了PCA-BP神经网络模型与传统的BP神经网络模型的预测精度,并分析了波长数目对两种神经网络模型预测结果的影响。针对训练得到的PCA-BP神经网络模型开展光谱消光法粒径参数反演算法的验证实验,搭建了光谱消光法颗粒粒径参数测量实验系统,测量了粒径范围在0.5～9.7 μm内的6种不同粒径参数的聚苯乙烯标准颗粒。仿真和实验结果表明：基于主成分分析方法可确定各个波长向量之间的相关性,利用综合载荷系数选取最优特征波长对应的消光值对整体的光谱数据具有较好的代表性,可实现光谱数据的降维。相比传统的BP神经网络模型,基于PCA-BP神经网络模型的颗粒粒径分布的分析方法预测精度更高,对于较分散颗粒系的分布参数的预测有更加明显的优势。而且,被选取的波长数较少时,PCA-BP神经网络模型依然有较高的预测精度。利用训练好的PCA-BP神经网络模型对颗粒粒径参数进行实验验证,预测结果可瞬时输出,颗粒粒径分布误差在5%以内,验证了该算法的可行性。     

基于PCA-BP神经网络的光谱消光法颗粒粒径反演算法研究（英文）

光谱消光法广泛应用于颗粒粒径测量领域,在利用光谱消光法对颗粒粒径进行反演的过程中,由于颗粒的消光系数存在理论复杂、计算繁琐、收敛速度慢以及求解不稳定等问题,很大程度上影响了整个反演过程的快速性和准确性。且在众多波长的消光数据中,存在较多重复冗余的信息,也很大程度上增加了反演算法的时间。针对光谱消光法粒径反演算法计算繁琐、反演效率低的问题,提出了基于主成分分析(PCA)和BP神经网络的光谱消光颗粒粒径分析方法。基于Mie散射理论对不同粒径、不同波长下的光谱消光值进行了仿真计算,通过对光谱消光数据集的主成分分析及各个波长综合载荷系数的计算,实现了最优特征波长的选取,利用降维后的光谱消光数据训练了PCA-BP神经网络模型,并利用该网络模型计算了粒径颗粒分布。通过仿真计算,比较了PCA-BP神经网络模型与传统的BP神经网络模型的预测精度,并分析了波长数目对两种神经网络模型预测结果的影响。针对训练得到的PCA-BP神经网络模型开展光谱消光法粒径参数反演算法的验证实验,搭建了光谱消光法颗粒粒径参数测量实验系统,测量了粒径范围在0.5～9.7μm内的6种不同粒径参数的聚苯乙烯标准颗粒。仿真和实验结果表明:基于主成分分析方法可确定各个波长向量之间的相关性,利用综合载荷系数选取最优特征波长对应的消光值对整体的光谱数据具有较好的代表性,可实现光谱数据的降维。相比传统的BP神经网络模型,基于PCA-BP神经网络模型的颗粒粒径分布的分析方法预测精度更高,对于较分散颗粒系的分布参数的预测有更加明显的优势。而且,被选取的波长数较少时,PCA-BP神经网络模型依然有较高的预测精度。利用训练好的PCA-BP神经网络模型对颗粒粒径参数进行实验验证,预测结果可瞬时输出,颗粒粒径分布误差在5%以内,验证了该算法的可行性。     

基于模式搜索的光谱消光粒径分布反演算法的研究

发展快速有效的反演算法用于粒径分布的重建是颗粒测量领域重要的研究课题之一。在光谱消光法粒径测量中,提出将模式搜索算法用于独立模式下粒径分布的重建,同时引入Tikhonov平滑泛函构建算法的目标函数,为保证搜索过程的快速性和准确性,设计了关于初始点的优选策略。利用该算法测量国家标准颗粒体积平均直径的相对误差为3.14%,不超过国家标准物质局给出的±8%的范围,且粒径分布宽度合理,没有明显展宽和伪峰现象。与Phillips-Twomey方法和遗传算法的对比结果表明,在综合考虑反演精度和反演时间条件下,该算法具有明显优势,更适合于快速准确的现场测量,具有较好的应用前景。     

前向光散射颗粒测量技术中遗传算法的应用

在前向光散射颗粒测试技术中,反演算法是影响测量结果的主要因素之一。遗传算法作为一种较新的自适应全局优化算法,其参数优化和目标函数的设计决定了颗粒粒径分布的反演效果。本文对遗传算法中相关参数的设置进行讨论,通过数值模拟验证了该算法对多分散颗粒系统粒径分布的反演效果,并对聚苯乙烯标准颗粒进行实测和反演计算。结果表明:遗传代数为3 200、种群尺度为50、交叉算子0.7、变异算子为0.002和光顺因子权重系数为1.5×10~(-5)的优化遗传算法在颗粒粒径分布反演时具有较好的稳定性和抗噪性能,能够有效重建颗粒粒径分布。     

Size measurement of nano-particles using self-mixing effect

In this letter, the technique of laser self-mixing effect is employed for nano-particle size analysis. In contrast to the photon correlation spectroscopy （PCS） and photon cross correlation spectroscopy （PCCS）, the main advantages of this technique are sensitive, compact, low-cost, and simple experimental setup etc. An improved Kaczmarz projection method is developed in the inversion problem to extract the particle size distribution. The experimental results prove that nano-particle size can be measured reasonably by using the self-mixing effect technique combined with the improved projection algorithm.     

Determination of Particle Size Distribution by the analysis of intensity-constrained multi-angle photon correlation spectroscopic data

Laser light scattering (LLS), especially dynamic laser light scattering (DLS), also known as photon correlation spectroscopy (PCS), is a well established method for particle size distribution analysis. It usually involves a Laplace inversion of the field autocorrelation function. However, the resolution is limited because of the ill-conditioned nature of this Laplace inversion. No unique solution exists when noise is present on the data. In contrast with this ill-conditioned nature, the angular dependence of scattered (static) intensities is precisely not ill-conditioned, which allows the resolution of the ill-conditioned inversion of DLS data to be improved. In order to characterize samples with more complicated size distributions, an intensityconstrained multi-angle PCS data analysis program has been developed, which is an alternative way of normalizing the field correlation function to that reported by Cummins and Staples [12]. In this program, the field autocorrelation function is normalized to the scattering intensity by using a predetermined coherent factor at each angle, which provides an additional constraint on the Laplace inversion of multi-angle PCS data analysis. The alternative analysis improves the resolution of PCS and provides a more reliable particle size distribution than single-angle data analysis. Both simulated and measured LLS data are used to illustrate its application, resolution and limitations.     

Optimal Filtering Applied to the Inversion of the Laplace Transform

Inversion of the Laplace transform, used in the laser scattering measurement of colloidal particle size distributions, presents severe numerical difficulties. In the presence of noise the variance of the inversion integral is infinite, indicating maximum uncertainty in the inversion. This paper applies the method of minimum variance, or “optimal”, filtering to the eigenfunction spectrum of the Laplace transform, giving an inversion which has finite variance. Spectral decomposition using the eigenfunctions of the Laplace transform gives a representation of the noise and desired signals analogous to the Fourier spectrum used in linear system theory. It is possible to obtain a filtered estimate of the unknown linewidth distribution. The requirement that the variance of this filtered estimate is minimum leads to a Wiener-Hopf integral equation defining the optimal filter. The results of this paper provide a basis of comparison of all methods of inversion of the Laplace transform, including the extensive literature of colloidal particle sizing by laser scattering or photon correlation.     

多角度动态光散射加权贝叶斯反演算法

研究加权贝叶斯算法在多角度动态光散射法测量单峰分布颗粒体系的颗粒粒度分布中的应用.采用颗粒粒度信息分布为底数、调节参数为指数的权重系数给各个角度下的光强自相关函数曲线加入不同的权重系数，再利用传统的贝叶斯算法反演.模拟与实验结果表明，加权后的贝叶斯算法能获得分布误差更小的反演结果，有效地抑制了数据噪声的影响，提高颗粒粒度分布反演的准确性.     

On-Line Particle Size Analysis Using Ultrasonic Velocity Spectrometry

A new method is described for the determination of particle size distribution of slurries based on ultrasonic velocity spectrometry combined with gamma-ray transmission. This method shares the advantages of ultrasonic attenuation spectroscopy of being capable of analyzing highly concentrated samples without dilution. However the ultrasonic velocity method is better suited to fine particles of diameter from about 0.1 to 30 μm, a greater volume of slurry is analysed and therefore sampling errors are reduced, and precise theoretical models are readily available to permit the accurate determination of size distribution by inversion of ultrasonic velocity measurements. The method can also be used to accurately determine particle size cut points by linear correlation. Using either inversion or correlation methods, the accuracy of particle size information from ultrasonic velocity spectroscopy is significantly enhanced by the independent measurement of solids loading by gamma-ray transmission. In addition, larger sizes can be measured by combining the ultrasonic velocity method with ultrasonic attenuation measurements. The method has been tested in the laboratory on a wide variety of mineral and paint slurries. The method determined the size distribution of single component silica and alumina samples in water in agreement with laser diffraction measurements and the method successfully distinguished well and poorly dispersed TiO₂ suspensions. For composite samples the method discriminated separate TiO₂ and CaCO₃, components and determined their proportions to within 0.25% volume. In addition the method, in combination with ultrasonic attenuation measurements, determined the size fractions of iron ore slurries below 10 and 30 μm to within 1.3% and 1.0% relative respectively, when compared with laser diffraction measurement of particle size. The CSIRO is presently designing an industrial gauge which will be manufactured and installed in an industrial slurry stream in 1997.     

Wavelets-regularization method for particle size inversion in photon correlation spectroscopy

For ill-posed inversion problem of photon correlation spectroscopy (PCS), a wavelet-regularization inversion method (WRIM) which combines wavelet multiscale inversion strategy with classical regularization inversion method (CRIM) was proposed. By using this method, the original inversion problem is decomposed into several subproblems on different multiscale spaces. As a result, we can successively obtain solution of original inversion problem according to the particle sizes inverted from the coarsest scale to the finest scale. The simulation and experimental data was respectively inverted by two methods. The inversion results demonstrate that WRIM has better global convergence, higher accuracy and more strong noise immunity than CRIM.     

表征超声衰减谱粒度的改进和声搜索算法

针对超声衰减谱法颗粒粒径测量,提出一种改进和声搜索算法,在和声搜索算法迭代达到200次时,加入拟牛顿算法,调节和声搜索算法全局和局部搜索能力,提高算法的收敛精度.对服从3种典型粒径分布函数颗粒系进行数值模拟,改进算法反演得出罗辛-拉姆勒(Rosin-Rammler)分布参数值相对误差小于8％,正态分布参数值相对误差小于...     

Calculation of Particle Size Distributions from Ultrasonic Measurements: A Mineral Slurry Case Study

Ultrasonic transmission measurements have been used extensively to determine the particle size of solids in slurries. This case study examines the application of mathematical inversion techniques to the determination of the particle size distribution of a mineral slurry from data collected at a minerals processing plant. A new mathematical inversion technique, based on an extension of modified Chahine iteration combined with the principle of maximum entropy has been developed. Four algorithms were constructed and used to calculate particle size distributions from synthetic and raw plant data. These incorporated modified Chahine iteration and its extension, together with two different approaches to applying a density measurement constraint on the particle size distribution. In general the algorithms performed well with regression errors below 3 %. The correlation coefficients and slopes for this technique were 0.86 and 1.35 for the weight fraction of particles less than 75 microns when compared with the laser diffraction analysis. A better match was obtained for the plant data by using the new inversion technique, into which the principle of maximum entropy has been incorporated whereas this was not the case with the synthetic data, illustrating the need to match the inversion technique to the problem.     

Determination of Particle Size Distribution in Fluids using photon correlation spectroscopy

Photon correlation spectroscopy (PCS) has been applied to various systems of particles suspended in a fluid. Computer simulations were used to analyse CONTIN, a program for the numerical inversion of the experimental data. Measurements were performed on a spectrometer equipped with a digital correlator. Various suspensions of polysterene latex spheres differing both in size and size distribution were prepared in double distilled water. Further some samples provided by an external institution without prior specification were analysed. Diffusivity measurements on submicron DES (di-2-ethylhexyl sebacate) particles in nitrogen were also performed. The results obtained demonstrate the strengths and the limitations of PCS as a method of determining particle size distributions in particle fluid systems.