Analysis of transient chemical species in analytical chemistry using the Kalman filter.

摘要

Analytical techniques often require measuring transient chemical species. Here, transient chemical species are considered to arise from changes in concentration over time. Chemical kinetics, chromatography, and flow injection analysis, all make use of measurements of transient chemical species, by measuring chemical mixtures of rising or falling chemical concentration, as chemical reactions proceed or as the chemical species pass through a flow-through detector. The studies reported here show how chemical estimation is performed with a Kalman filter on multicomponent transient chemical species. A new approach to Kalman filter architecture, referred to as hierarchical Kalman filtering, makes use of the state space formulation, as it is applied to problems in analytical chemistry. Two kinds of Kalman filter architecture, described as master-slave and multiple-peer filters, are used to do multicomponent analysis of chromatographic, flow-injection and chemical kinetic data. The data analysis performed in the studies was applied to high information content diode-array spectrophotometric data.;Information-based application of the Kalman filter algorithm was a central theme of the work. In the first study, the information matrix determinant was evaluated to determine information content. By using these results, the extended Kalman filter was applied only to high information content data, resulting in reduced computational burden. In another study, new information arising from the evolving chemical estimates, while analyzing chromatographic data, was used in a dynamic filter model. This resulted in improved filter estimates by eliminating induced model errors. Another application to chromatographic data addressed a coeluting chromatographic pair and used a Kalman filter model that could be adjusted for subtle variations from baseline drift. Finally, flow injection data were analyzed by using a calculation of matrix condition number. In this way, system observability was evaluated for determining the choice of wavelengths of the Kalman filter models. The calculation burden was reduced by eliminating low information wavelengths from the filter model. In each study, filtering efficiency was addressed by considering information location and content or by incorporating new information that becomes available after the data analysis was initiated.