Date of Award
Spring 1999
Document Type
Thesis - Restricted
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering
First Advisor
Heinen, James
Abstract
Acoustic wave sensors are used for chemical sensor applications such as the classification, identification, and estimation of chemical analytes in both gas and liquid media. The parameters of interest used to rate the performance of the sensors are sensitivity and selectivity. While the issue of sensitivity can be partially solved by selecting the appropriate sensor device, selectivity still remains a major concern. This is primarily because the analytes under consideration belong to the same group/class of chemicals. In such cases, the sensor output by itself does not provide enough information to reliably identify, estimate and/or classify the analytes being investigated. Hence a need arises for the use of data analysis techniques to process the sensor signal data so that the sensing system can be made more selective. A new approach to analyze sensor signal data using statistical pattern recognition techniques such as principal component analysis and nearest neighbor algorithm is presented. In this study, sensor signals such as frequency and attenuation data were collected for single component and binary solutions of alkali metal ions (cesium, lithium, potassium, and sodium) using acoustic plate mode (APM) sensor devices. The analysis of principal components of the attenuation data is used to identify an unknown sample of single component metal ion solution and the distance measurement of the frequency data utilizing the nearest neighbor algorithm is used to estimate the concentration of the identified solution. The proposed algorithms perform the identification and estimation of an unknown sample of a dilute metal ion solution with very high accuracy and very few errors and exhibit little sensitivity to the sampling rate of the measured data. An attempt is also made to quantify binary mixtures of alkali metal ion solutions using the APM sensor devices. The proposed algorithm for this quantification is based on the distance comparison between frequency and attenuation data of the binary solutions and similar data measured for the respective single component solutions. The test results indicate that it is possible to estimate the individual proportions of the constituent metal ions as well as the concentration of the binary solution, provided that the sensor response to the mixture is measured with little error. Such errors are often caused by local changes in concentration of the mixture.
Recommended Citation
Shah, Sejal, "Principal Component Analysis of APM Sensor Signals for Identification of Dilute Metal Ion Solutions" (1999). Master's Theses (1922-2009) Access restricted to Marquette Campus. 4385.
https://epublications.marquette.edu/theses/4385