PNNL

Abstract

During process monitoring applications, referenced optical spectroscopy, such as absorbance spectroscopy, can suffer from environmental and instrumental fluctuations which alter the intensity of irradiance reaching the spectrometer’s detector at each detected frequency. Temperature, vibration, light source aging, instrument damage, detector aging, detector registry shifts, sampling cell degradation, and similar perturbations create situations in which a previously collected reference spectrum may no longer be valid for the current state of the system. This can lead to the calculation of poor-quality absorbance spectra which are unsuitable for qualitative or quantitative analysis based on prior calibration models. The use of single-beam spectra in the creation of multivariate calibration models circumvents the need for collecting and maintaining a stable reference spectrum throughout an ongoing chemical process. However, unlike absorbance spectra which typically have a zero baseline, single-beam spectra contain a high background signal relative to analyte signal, and they may also contain intense peaks from the light source. Here, multivariate principal component analysis (PCA) and partial least squares (PLS) regression models are built using single-beam and absorbance spectra to compare the efficacy of both types of spectra for qualitative and quantitative analysis of lanthanide solutions. A multi-leg fiber optic UV-visible spectrometer is utilized to collect samples under three distinct wavelength registries, in three unique sampling cells, and under lighting conditions spanning 0.2 to 2.0 relative transmittance. Under these conditions, single-beam PCA models produced enhanced discrimination between sampling conditions, allowing spectra to be grouped by the instrumental conditions under which they were collected. Absorbance and single-beam PLS models produced equivalent quantitation of lanthanide concentrations.