多角度光散射颗粒的粒径解析和属性识别

Size Detection and Attribute Recognition of Particles by Multi-Angle Light Scattering

通过提取光散射信号中颗粒粒径和属性的非线性特征向量,利用广义神经网络(GRNN)同时解析颗粒粒径和识别属性。采用经验模态分解(EMD)方法分解颗粒物的光散射信号,提取三维能量分布,计算3种相同粒径不同属性颗粒的样本熵,发现样本熵能够反映颗粒的属性;为了消除粒径和属性对散射的影响,对散射信号进行Hilbert变换,提取时频域特征,与样本熵结合组成高维特征集,通过局部线性嵌入(LLE)算法将特征集归为6个特征向量,作为广义神经网络的输入层,解析粒径和识别属性;采用粒径为0.11μm的二氧化硅颗粒、2μm和4μm的聚苯乙烯小球进行实验,结果表明,粒径解析和属性识别的正确率均在90%以上。

This study aims to detect the size and attribute of particles simultaneously by extracting nonlinear eigenvector of size and attribute in light scattering signals and using general regression neural network （GRNN）. The scattering signals are decomposed by the method of empirical mode decomposition （EMD）, and the three dimensional energy distribution is extracted. Sample entropies of three kinds of particles with same attribute and different sizes are calculated. It is found that the sample entropy can identify the attribute of particles. In order to eliminate the influence of particle size and attribute on the scattering, the Hilbert transform is used for the light scattering signals, and time-frequency domain eigenvectors are extracted, which form a high-dimensional eigenvectors set together with the sample entropy. The eigenvectors set is summed up into six eigenvectors by the local linear embedding （LLE） algorithm and used as the input layer of the GRNN to identify the particle size and attribute. Finally, an experiment is conducted to test the 0.11 μm SiO2 particles, 2 μm and 4 μm polystyrene pellets. The results show that the accuracy of particle size detection and attribute recognition exceeds 90%.