【摘要】：通過提取光散射信號中顆粒粒徑和屬性的非線性特征向量,利用廣義神經(jīng)網(wǎng)絡(luò)(GRNN)同時解析顆粒粒徑和識別屬性。采用經(jīng)驗?zāi)B(tài)分解(EMD)方法分解顆粒物的光散射信號,提取三維能量分布,計算3種相同粒徑不同屬性顆粒的樣本熵,發(fā)現(xiàn)樣本熵能夠反映顆粒的屬性;為了消除粒徑和屬性對散射的影響,對散射信號進(jìn)行Hilbert變換,提取時頻域特征,與樣本熵結(jié)合組成高維特征集,通過局部線性嵌入(LLE)算法將特征集歸為6個特征向量,作為廣義神經(jīng)網(wǎng)絡(luò)的輸入層,解析粒徑和識別屬性;采用粒徑為0.11μm的二氧化硅顆粒、2μm和4μm的聚苯乙烯小球進(jìn)行實驗,結(jié)果表明,粒徑解析和屬性識別的正確率均在90%以上。
[Abstract]:By extracting the nonlinear eigenvector of particle size and attribute in the light scattering signal, the generalized neural network (GRNN) is used to analyze the particle size and the recognition attribute simultaneously. The empirical mode decomposition (EMD) method is used to decompose the light scattering signals of particles, extract the three-dimensional energy distribution, calculate the sample entropy of three kinds of particles with the same particle size and different attributes, and find that the sample entropy can reflect the properties of the particles. In order to eliminate the influence of particle size and attribute on scattering, the scattering signal is transformed into Hilbert transform, time and frequency domain features are extracted, and the high Viterbi feature set is formed by combining with sample entropy. The feature set is classified into six feature vectors by means of locally linear embedding (LLE) algorithm. As the input layer of the generalized neural network, the particle size and recognition properties are analyzed, and the experimental results show that the accuracy of particle size resolution and attribute recognition are above 90%. The experiments are carried out with silica particles of 0.11 渭 m and polystyrene pellets of 4 渭 m.
【作者單位】： 哈爾濱理工大學(xué)測控技術(shù)與通信工程學(xué)院;中興儀器(深圳)有限公司;
【基金】：國家科技重大專項(2016YFF0103000)
【分類號】：O436.2
，

本文編號：2162154