Chemometric Studies for Temperature Dependent Near Infrared Spectroscopy

Near-infrared (NIR) spectra are sensitive to the variation of experimental conditions, such as temperature. Fluctuation of temperature causes the changes of NIR spectra. Therefore, the effects of temperature variation on NIR spectrum have been well studied. In recent studies, structural analysis of proteins was performed based on the effect of temperature on NIR spectra, and the hydrogen bond interactions in aqueous solutions of alcohols were studied based on the temperature-induced spectral variations. On the other hand, the temperature-induced spectral variations will affect the predictive ability of the multivariate calibration models in quantitative analyses. Much effort has been made to correct the effect of temperature variation on NIR spectra for standardization of the spectra at different temperatures and correcting temperature effects. If the effects of temperature on NIR spectra are considered as useful information, however, quantitative or structural information may be obtainable from the spectra measured at different temperature. For examples, a parallel factor (PARAFAC) model has been used to extract the relevant sources of information about the physical and chemical changes in mixture system, and aquaphotomics was proposed for NIR spectral analysis using the effects of temperature on the spectrum of water.
In our study, the equipment was constructed using a temperature control unit and a spectrometer. Premium heated immersion circulator (Thermo Fisher Scientific, New Hampshire, USA) was used for temperature control. The precision of the equipment for temperature control is ± 0.01 oC and the temperature can be changed according to a program. Spectral measurements were obtained with a Vertex 70 spectrometer (Bruker Optics Inc, Ettlingen, Germany) furnished with a transmittance optical fiber probe. The spectra at different temperature were measured from 4000 to 12000 cm-1.
Chemometric methods were studied for extracting information from the spectral matrix. At first, partial least squares (PLS) regression was employed to model the the relationship between NIR absorption spectra and temperature. For the solvents such as water and ethanol, a quantitative spectra-temperature relationship (QSTR) model between NIR spectra and temperature can be established, and the difference between the models of different solutions is a quantitative reflection of concentration. Therefore, quantitative analysis can be achieved using the QSTR models. Further studies show that the QSTR model can also be obtained using multilevel simultaneous component analysis (MSCA). A between-temperature model describing the effect of temperature and a within-temperature model describing the variation of concentration were obtained. The former is a kind of QSTR model, and the latter can be used as the calibration equation for quantitative analysis. QSTR models of commonly used solvents have been investigated and quantitative analysis of binary (water-ethanol) and ternary (water-ethanol-isopropanol) mixtures. Recently, alternating trilinear decomposition (ATLD) was found to be more convenient for building the QSTR and calibration model from the three dimensional data of the spectra measured at different temperature for the samples with different concentration.