Larrechi M S, Rius F X
Departament de Química Analítica i Química Orgànica, Universitat Rovira i Virgili, Pl. Imperial Tàrraco, 1, 43005 Tarragona, Spain.
Appl Spectrosc. 2004 Jan;58(1):47-53. doi: 10.1366/000370204322729469.
When applied to near-infrared (NIR) data, multivariate curve resolution methods, in particular alternating least squares (ALS), make it possible to calculate the concentration profiles and the spectra of all species involved in the reaction of curing epoxy resins. In this paper, the model reaction between phenyl glicidyl ether and aniline (2:1) was studied at 95 degrees C. A NIR spectrum was recorded every five minutes throughout the eight-hour reaction process. The data display rank deficiency. This problem was overcome by supplying additional information to the system in the form of known spectra of some reactants. The recovered spectra and concentration profiles satisfactorily reproduced the experimental data. In this way, 99.99% of the variance associated with the experimental matrix was reproduced. A value of 0.87% was obtained for lack of fit while the similarity coefficient r between the spectra recovered and the spectra corresponding to the three pure species involved in the reaction were PGE (r = 0.994), aniline (r = 0.994), and tertiary amine (r = 0.999). The maximum and minimum limits associated with the ALS solutions were calculated, which made it possible to limit to a considerable extent the ambiguity that is characteristic of these curve resolution methods.
当应用于近红外（NIR）数据时，多元曲线分辨方法，特别是交替最小二乘法（ALS），使得计算固化环氧树脂反应中所有参与物种的浓度分布和光谱成为可能。本文研究了苯基缩水甘油醚与苯胺（2:1）在95℃下的模型反应。在整个八小时的反应过程中，每五分钟记录一次近红外光谱。数据显示存在秩亏问题。通过以某些反应物的已知光谱形式向系统提供额外信息，克服了这个问题。恢复的光谱和浓度分布令人满意地再现了实验数据。通过这种方式，与实验矩阵相关的99.99%的方差被再现。失拟值为0.87%，而恢复的光谱与反应中涉及的三种纯物质对应的光谱之间的相似系数r分别为苯基缩水甘油醚（r = 0.994）、苯胺（r = 0.994）和叔胺（r = 0.999）。计算了与ALS解相关的最大和最小限值，这使得在很大程度上限制了这些曲线分辨方法所特有的模糊性。
