基于小波包与模糊聚类算法的声波法检测流化床内颗粒的平均粒度
陈惜明1,2，陈德钊1
(1．浙江大学化学与生物工程学系，浙江，杭州，310027；2．淮北煤炭师范学院化学系，安徽．淮北，235000)
摘 要：平均粒径是气固流化床反应器运行时需要监控的重要参数之一，利用声波信号检测床内颗粒平均粒度的方法能克服传统方法不能实时在线测量的缺陷，安全环保不侵入流场。先用Db5小波包将声发射信号3尺度分解，求出各细节信号小波系数的绝对值加和，构成声信号的能量模式，标准化之后经主成分分析得出主成分，再用模糊均值聚类方法分类。由于不同粒度的声波信号经小波包分解后，其小波系数绝对值加和具有特定的模式，因而，这种方法分类准确性达98％以上。
关键词：多尺度；离散小波变换；小波包；声测量；模糊聚类；聚类
中图分类号：TQ051.9
文献标识码：A
文章编号：1001-4160(2008)06-689-692
Abstract: Average particle size is one of the key parameters need to be supervisory controlled for fluidized bed reactore. Measuring the average particle size by acoustic emission ( AE) signal is superior to traditional methods because it is safety , environmental protection ,non-invasive, and the average particle size can be measured at real time and online. Original AE signals received from a detector were firstly decomposed by the wavelet packet with Daubechies level five wavelet ( db5). The absolute wavelet coefficients were summed andthe summations were known as energy pattern , which used for classification. Principal Component Analygis ( PCA) was applied to the summations after they had been normalized. Some of these principal components were classified through fuzzy cluster algorithm. Due toacoustic emission signals originated from different particle sizes are different from each other; the energy patterns originated from those AE signals are distinctly different, too. Measuring the average particle size in the fluidized bed by AE signal got over 98% of accuracy.
Key words: multi-scale, discrete wavelet transform ( DWT) , wavelet packet, acoustic measuring, fuzzy cluster
Chen XM and Chen DZ. Average particle size measuring by acoustic signal for fluidized bed based on wavelet pack-et and fuzzy cluster algorithm. Computers and Applied Chemistry, 2008 , 25(6) ;689-692.
1 引言
气固流化床反应器中，多相反应是典型的多尺度问题[1]。利用声发射信号获得床内物料的运行状态，进而在宏尺度下控制生产过程，近几年已有较为广泛的研究。声波测量法在线监控流化床反应过程具有实时、无损的特点，但声信号有突发瞬态性及多尺度特征，并且往往有噪声干扰，因此，如何从声信号中提取特征信息是急需解决的问题[2-4]。一般说来，声信号与床层高度、物料组成、温度、空床气速等多种因素有关[2]，但稳定运行的流化床，声信号则主要受粒度及粒度分布的影响。因此，根据声信号获取床内颗粒的粒度信息，在生产中有其实际意义。
频谱分析[5]、小波分析[6-7]、小波包分析[8]、分形特征分析与复杂性分析[9]等都可以用于分析声信号以便了解床内物料的运行状态。文献[10,11]以声信号的频谱作为依据，用分类的方法建立了颗粒粒度与声信号的关系，但是，床内颗粒组成复杂，其频谱信号亦然；并且信号中存在大量高频噪声，利用小波去噪技术虽然能滤去一部分，但仍会导致信号失真，难以彻底消除。小波分析将声信号分解为低频概貌信号和高频细节信号，虽然对分析非线性非平稳脉动信号有显著的优势，但是对渐变信号，却不如傅里叶分析或加窗傅里
叶分析有效。因此，建立更加合理可行的模型实时预测和监控床层内的平均粒度有其实际意义。本文以小波包分析和模糊聚类分析为工具，建立了声信号与颗粒平均粒度的量化关系，实践表明，该方法能根据声信号对床内颗粒的平均粒度进行检测，对于需要控制产品平均粒度的场合，它具有特别重要的意义。
