Quantitative infrared spectroscopy in challenging environments: applications to passive remote sensing and process monitoring

Chemometrics is a discipline of chemistry which uses mathematical and statistical tools to help in the extraction of chemical information from measured data. With the assistance of chemometric methods, infrared (IR) spectroscopy has become a widely applied quantitative analysis tool. This dissertation explores two challenging applications of IR spectroscopy facilitated by chemometric methods: (1) passive Fourier transform (FT) remote sensing and (2) process monitoring by near-infrared (NIR) spectroscopy.

Passive FT-IR remote sensing offers a measurement method to detect gaseous species in the outdoor environment. Two major obstacles limit the application of this method in quantitative analysis: (1) the effect of both temperature and concentration on the measured spectral intensities and (2) the difficulty and cost of collecting reference data for use in calibration. To address these problems, a quantitative analysis protocol was designed based on the use of a radiance model to develop synthetic calibration data. The synthetic data served as the input to partial least-squares (PLS) regression in order to construct models for use in estimating ethanol and methanol concentrations. The methodology was tested with both laboratory and field remote sensing data.

Near-infrared spectroscopy has attracted significant interest in process monitoring because of the simplicity in sample preparation and the compatibility with aqueous solutions. For use in process monitoring, the need exists for robust calibrations. A challenge in the NIR region is that weak, broad and highly overlapped spectral bands make it difficult to extract useful chemical information from measured spectra. In this case, signal processing methods can be helpful in removing unwanted signals and thereby uncovering useful information. When applying signal processing as a spectral preprocessing tool and regression analysis for building a quantitative calibration model, optimizing the parameters that specify the details of the methods is crucial. In this research, particle swarm optimization, a population-based optimization method was applied. Digital filtering and wavelet processing methods were evaluated for their utility as spectral preprocessing tools.

Both a pump-controlled flowing system and bioreactor runs involving the yeast, Pichia pastoris, were studied in this work. In investigating the bioreactor runs, insufficient reference data resulted in difficulties in employing the PLS calibration method. Instead, the augmented classical least-squares modeling technique was applied since it requires only pure-component or composite spectra of the analyte and background matrix rather than a large set of mixture samples of known analyte concentration.