Date of Award
Summer 2014
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering
First Advisor
Josse, Fabien
Second Advisor
Yaz,Edwin
Third Advisor
Schneider, Susan C.
Abstract
Compact sensor systems for on-site monitoring of groundwater for trace organic compounds are currently under development. To permit near real-time analysis of samples containing multiple analytes, the present work investigates a sensor signal processing approach based on estimation theory, specifically using Kalman Filter and Extended Kalman Filter. As a first step towards the analysis of groundwater samples containing multiple compounds, the approach presented in this work permits estimation of analyte concentration(s) in binary mixtures and single analyte samples on-line, before the sensor response reaches steady-state. Sensor signals from binary mixtures and single analyte samples of BTEX compounds (benzene, toluene, ethylbenzene, and xylenes) were analyzed in this work because these compounds are good indicators of accidental release of fuel and oil into groundwater. Based on those previous experimental results, models for the sensor response to binary mixtures and single analyte samples are developed. These sensor response models were transformed into state-space representation so that estimation theory can be used to estimate the sensor parameters. For the case of the single analyte system, one state-space form was developed and for the case of the two-analyte system, two different state-space forms were developed. These state-space forms were tested using the available measured data, and the results indicate that relatively accurate estimates of analyte concentration(s) could be obtained within a relatively short period of time (six minutes or less for the tested sensor system) well before the sensor response reaches steady-state. Also presented in this work are new techniques that enable correcting for linear baseline drift and outlier points in the measured data on-line. The linear baseline drift correction technique uses estimation theory (particularly Kalman Filter) to rapidly perform linear extrapolation and linear interpolation. The elimination of the outlier points in the sensor data was performed by using a combination of discrete low pass filter and Kalman Filter (or Extended Kalman Filter depending on the state-space form). These techniques were tested on measured data with linear baseline drift and outlier points and the results obtained indicate that these sensor signal pre-processing techniques are indeed capable of correcting for linear baseline drift and outlier points in real-time.