Semester

Fall

Date of Graduation

2019

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Industrial and Managements Systems Engineering

Committee Chair

Feng Yang

Committee Member

Nianqiang Wu

Committee Member

Bhaskaran Gopalakrishnan

Committee Member

Xinjian He

Committee Member

Xi Chen

Abstract

Raman spectroscopy, recognized as a powerful analytical technique, has been widely employed in many fields especially in the detection of hazard material containing in food or water. However, multiplexed analyte quantification based on Raman spectra remains a challenge due to the difficulties in accurate and precise modeling of the relationship between multiple analyte concentrations and dense spectral data with noise.

In this work, a statistical procedure was developed to efficiently generate high-quality calibration models quantifying the analyte concentrations versus spectra relationship. The resulting calibration models are able to provide estimated concentration ranges (which reflect both point and uncertainty estimates) for the analytes of interest in an unknown sample based on its observed Raman spectrum. The calibration procedure integrates three unique methodology components. (i) Stochastic kriging with time-series errors was adapted to model Raman spectra as a function on analyte concentrations. (ii) Built on the kriging modeling, bootstrap resampling methods were adapted to quantify the uncertainty of analyte concentration estimates. (iii) Based on the uncertainty quantification capability, a two-stage experimental design method was developed for efficient sampling: How to use a minimum amount of experimental effort to achieve calibration models with desired uncertainty of analyte estimates?

Simulation studies were derived from laboratory experimental data, and used to demonstrate the efficiency of the calibration procedure over the methods.

Share

COinS