近红外光谱快速定量分析天然纤维的研究

发布时间：2018-06-15 15:14

本文选题：近红外光谱法 + 洗净绒　；参考：《北京化工大学》2016年硕士论文

【摘要】：天然纤维是纺织工业的重要材料来源。近年来,面对相关产业结构的调整优化,以及企业降低成本和保证产品质量的巨大需求,亟待解决纤维的现场或在线快速检测技术。但在应用近红外技术进行现场和在线分析时,光散射以及湿度引起的样品水分变化等因素会影响模预测性能,从而影响现场和在线测量准确度和精度。此外,市场上还缺乏某些纤维的现场和在线快速分析系统。因此,本论文研究了水和光散射对山羊绒光谱的影响以及光谱处理方法,并建立了山羊绒净绒率快速分析方法；研制了一套天然纤维素纤维浆粕性质在线分析系统,并建立了在线分析方法。(1)本文研究了水分对山羊绒洗净绒纤维漫反射近红外光谱的影响。发现光谱受到水分的严重影响。提出了一种算法—扣水算法来减小水分的影响。该算法基于向量-子空间夹角判据算法,通过从样品光谱中扣除水的光谱信息来消除水的影响。分别利用未扣水的光谱和利用该方法预处理后的光谱与净绒率相结合,利用偏最小二乘法建立洗净绒的定量模型。未扣水模型的预测决定系数为0.87,预测均方根误差(SEP)为8.53%,而扣水模型的预测决定系数为0.94, SEP为5.28%。结果表明,扣水算法可明显提高模型的预测性能。(2)本文研究了洗净绒中主要组成羊绒和粗毛的漫反射近红外光谱特征。研究发现由化学组成等化学信息引起的羊绒和粗毛的光谱差异极小,而差异主要来源于物理性质例如直径不同带来的光散射效应。研究了多元散射校正(MSC)对消除散射效应的校正效果,并对MSC处理后的光谱残差进行了主成分分析,研究发现在MSC光谱残差中包含对洗净绒定量校正有用的物理信息。为充分利用样品的化学和物理信息,本文提出光谱重建方法。该算法利用主成分分析分解残差光谱,然后根据主成分光谱与净绒率的相关系数曲线,来选择与净绒率相关关系高的主成分光谱,并将其添加到MSC校正后的光谱,构建新的用来建模的样品光谱。利用该方法建立的PLS定量模型,预测决定系数为0.92,SEP为6.10。(3)本文研究了扣水算法和光谱重构方法联合使用对光谱与净绒率相关关系以及模型性能的影响。研究发现,两种方法联用可有效提高光谱与净绒率的相关系数,并获得最佳的模型,模型预测决定系数为0.95,SEP为5.18%,可满足净绒率的检测要求,从而实现洗净绒净绒率的快速分析。(4)本文研制了一种快速在线分析浆粕性质的近红外光谱仪。利用该在线分析仪动态采集86个浆粕的漫反射光谱图,利用PLS方法结合S-G导数、均值中心化和MSC预处理方法,分别建立α-纤维素含量和聚合度的在线分析模型。其模型的决定系数分别0.89和0.98,SEP为0.94和25.1,且具有良好的预测重复性。研究结果可实现浆粕连续生产过程的在线监控。
[Abstract]:Natural fiber is an important material source of textile industry. In recent years, in the face of the adjustment and optimization of related industrial structure, as well as the huge demand of enterprises to reduce the cost and ensure the quality of products, the field or on-line rapid detection technology of fiber urgently needs to be solved. However, in the field and on-line analysis using near-infrared technique, light scattering and moisture change of sample caused by humidity will affect the prediction performance of the model, thus affecting the accuracy and accuracy of field and on-line measurement. In addition, the market also lacks some fiber field and on-line rapid analysis system. Therefore, the effect of water and light scattering on cashmere spectrum and the method of spectrum processing were studied in this paper, and a rapid analysis method of cashmere net velvet was established, and a set of on-line analysis system for the properties of natural cellulose fiber pulp was developed. The effect of water content on the diffuse reflectance near infrared spectroscopy of cashmere washed-down fiber was studied. It was found that the spectrum was seriously affected by moisture. In this paper, an algorithm, the water impingement algorithm, is proposed to reduce the influence of water content. Based on the vector-subspace angle criterion algorithm, the influence of water is eliminated by deducting the spectral information of water from the spectrum of the sample. The quantitative model of washed velvet was established by partial least square method by combining the spectrum of uncapped water and the spectrum after pretreatment with the net velvet ratio. The predictive decision coefficient of unincorporated water model is 0.87, the root mean square error (SEP) of prediction is 8.53, while the predictive decision coefficient of water model is 0.94 and SEP is 5.28. The results show that the water buckling algorithm can improve the predictive performance of the model. (2) the diffuse reflectance near infrared spectra of cashmere and coarse wool are studied in this paper. It is found that the spectral difference between cashmere and coarse wool caused by chemical information such as chemical composition is very small and the difference is mainly due to the light scattering effect caused by physical properties such as different diameters. The correction effect of multivariate scattering correction (MSCI) on the elimination of scattering effect was studied. The spectral residuals after MSC treatment were analyzed by principal component analysis. It was found that the spectral residuals of MSC contained physical information useful for the quantitative correction of washed velvet. In order to make full use of the chemical and physical information of the sample, a spectral reconstruction method is proposed in this paper. In this algorithm, principal component analysis (PCA) is used to decompose the residual spectrum, and then according to the correlation coefficient curve between principal component spectrum and net velvet ratio, the principal component spectrum with high correlation with net velvet ratio is selected and added to the corrected spectrum of MSC. A new sample spectrum for modeling was constructed. Using the PLS quantitative model established by this method, the predictive decision coefficient is 0.92g SEP is 6.10.3.In this paper, the influence of the combination of water blocking algorithm and spectral reconstruction method on the correlation between the spectrum and the net velvet ratio and the performance of the model is studied. It is found that the two methods can effectively improve the correlation coefficient between the spectrum and the net velvet rate, and obtain the best model. The prediction decision coefficient of the model is 0.95g SEP 5.18, which can meet the requirements of the net velvet ratio detection. In this paper, a near infrared spectrometer for fast on-line analysis of pulp properties has been developed. The diffuse reflectance spectra of 86 pulps were dynamically collected by the on-line analyzer. The on-line analytical models of 伪 -cellulose content and degree of polymerization were established by PLS method combined with S-G derivative, mean centralization and MSC pretreatment. The determination coefficients of the model are 0.89 and 0.98SEP 0.94 and 25.1respectively, and the model has good prediction repeatability. The results can realize the on-line monitoring of continuous pulp production process.
【学位授予单位】：北京化工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TS102;O657.33

【参考文献】