The design, fabrication and implementation of point detection and chemical imaging sensors utilizing optical and digital regression techniques

This dissertation describes state-of-the-art optical computing applied predominantly to spectroscopic applications with interference filters called Multivariate Optical Elements (MOEs) as well as modified traditional instrumentation. Optical computing (also referred to as optical regression) allows simple instrumentation to realize the advantages of multivariate calibration and pattern recognition without the need for post data analysis. Such instruments are intended to be placed in the hands of non-experts for applications ranging from pharmaceutical content uniformity analysis to remote chemical imaging. Optical computing instrumentation is also rugged, fast, inexpensive, contain little to no moving parts and yield higher signal-to-noise measurements compared to traditional spectroscopic instrumentation.; The discussion of MOEs is separated in to three categories: design, fabrication and implementation. For MOE design, cluster computing schemes are explored for optimizing the performance of point detection and chemical imaging MOEs through the use of a novel Matlab computer cluster. A MOE design method is developed which is a minimization procedure using optimal seed point conditions based on interference filter spectral resolution and selectivity enhancements of spectroscopic optical pre-filtering. A method of refining the fabrication process by removing the dependence on a thin film matrix model, which we have named Bootstrapping is then discussed. In situ optical spectra are used to modify the Fresnel reflectance amplitude of the thin film/substrate interface which becomes a pseudo substrate for modeling all future thin film layers. Finally, two applications of chemical imaging MOEs in the UV-Vis and instrumentation considerations are described to explain the direction of future optical computing modalities.; The final topic of this dissertation explores the use of current wide-field chemical imaging instrumentation to perform optical regressions. Weighted co-additions are designed based on digital multivariate regression techniques like Principle Component Regression (PCR) to collect and predict an analyte species concentration in a single step. Measurement precision considerations are discussed for the modification of current instrumentation and software for developing optical computing alternatives.