Publication Date

2014-08-24

Availability

Open access

Embargo Period

2014-08-24

Degree Name

Master of Engineering (ME)

Department

Electrical and Computer Engineering (Engineering)

Date of Defense

2012-11-12

First Committee Member

Kamal Premaratne

Second Committee Member

Manohar N. Murthi

Third Committee Member

James N. Wilson

Abstract

The Environmental Protection Agency (EPA) has set forth guidelines for drinking water to ensure the safety of the public from harmful contaminants and pollutants. Current standardized methods for the detection of water borne toxins and pollutants are expensive, and vague in their analysis: qualitative and quantitatively. We introduce a data fusion model using Dempster-Shafer Theory to qualitatively detect multiple combinations of analytes suspended in Toluene. The benefits of data fusion model are its ability to be extended for additional sources of evidence such as pH, turbidity, and electrical conductivity and its ability to handle epistemic uncertainty. In addition, we develop a method of modeling spectroscopy data and an ability to synthetically add spectroscopy noise and perturbations to the signal. This novel chemometric detection method that is introduced has reported 99% detection under the most extreme noise condition of η=2.0, using cepstral coefficients as an evidence source when fused over all the simulated spectra data. This was an increase in the averaged recognition using correlation coefficients by 46.3 percent.

Keywords

Dempster-Shafer; Detection of Analytes; Water Toxins; Generating a Synthetic Spectroscopy Database; Synthetic Spectroscopy Noise; Chemometrics

Share

COinS