Impact of Pellet Thickness on Quantitative Terahertz Spectroscopy of Solid Samples in a Polyethylene Matrix

Abstract

Pellets composed of different weight-percent (wt-%) of lactose within a polyethylene (PE) matrix are used to examine how the physical thickness of solid samples impact analytical measurements performed over terahertz (THz) frequencies when using time-domain THz spectroscopy. Results indicate that the thickness of each pellet depends on the mass and physical properties of the individual components that comprise the pellet. Thickness of mixture pellets depends on the porosity of the individual pellet components. Porosity measurements presented here for PE and lactose give values of 25.6 +/- 0.3 and 14.5 +/- 0.1, respectively, which indicate that more air is trapped within the compressed PE matrix compared to that for lactose. This difference in porosity creates different pellet thicknesses for pellets of the same nominal mass but with different relative amounts of PE and lactose. For this binary matrix, the thickness of each pellet is found to be a linear combination of the compressed densities of the individual components. Analysis of the time-domain THz spectra reveals that thinner samples are confounded by a fringe pattern observed in the frequency-domain spectra. This fringe pattern is created by an etalon corresponding to the air/pellet interfaces for the sample in the optical path. Spectra collected from thicker pellets are confounded by a sloping baseline caused by scattering effects within the pellet matrix. The quantitative impact of pellet thickness is determined by comparing the mean standard error of calibration (MSEC) and mean standard error of prediction (MSEP) for a set of leave-three-out cross validation multivariate calibration models based on the partial least-squares (PLS) algorithm. Results indicate that PLS models are capable of analytical measurements with MSEC and MSEP values between 0.04 and 0.20 wt-%. Analysis of spectral variance captured within the corresponding spectral loadings for each model indicates that spectral variance is lowest for the 300 mg samples where the impact of scattering is minimal under conditions when the sample etalon is nonexistent.