Date of Award
Summer 2019
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering
First Advisor
Lee, Chung, Hoon
Second Advisor
Richie, James E
Third Advisor
Medeiros, Henry
Abstract
A continuous, static, and non-interfering water contaminant detection method is presented to measure specific water contaminants (NaCl, MgCl2, and mixture of NaCl and MgCl2) using RF microwave principles. A coil is mounted on the surface of a glass tube and the liquid sample is placed inside of the tube. An external magnetic field generated by the coil continuously measures changes in radio frequency energy. The non-contact feature of the device allows a long sensor lifetime with high sensitivity for real-time measurements. The measurement parameter is reflection coefficient (S11) and the operating frequency is 10 MHz – 5 GHz. For NaCl and MgCl2, 11 different concentrations (1000 ppm – 400 ppb) liquid solutions are prepared. Amplitude changes and frequency shifts are noticeable among different materials and concentrations. Different test materials have different radio frequencies at which they undergo excitation and the responses are identified in S11 measurement. A machine learning algorithm is introduced to analyze the measured S11 data. A support vector regressor (SVR) model is trained using the measured data of various salt samples. The training data is constructed by concatenating the 20,000 amplitudes and 20,000 phase values from the measured S11 data. The hyperparameters of the SVR are optimized using 10-fold cross-validation method. Based on the trained model, the algorithm predicts the concentrations of the liquid samples. The experimental results indicate that the device can detect concentrations as low as 400 ppb with high accuracy.