- 195 views
- 238 downloads
Machine Learning for Error Estimation and Compensation for Microwave Sensors
-
- Author / Creator
- Nazli Kazemi
-
Microwave planar resonators have been widely used in material characterization, environmental monitoring, proximity sensing, mechanical motion detection, etc. In general, non-contact sensing, real-time measurement, and CMOS compatibility, makes them interesting for chemical sensing, especially in harsh environments. However, the planar nature leaves them vulnerable to the ambient changes with potential impact on the perception of the material under test. Temperature, as one of the most important factors that affects the sensor response, is considered an undesired environmental impact and relevant methods are introduced to tackle it. Even though many types of human- or environment-centric errors such as material displacement, uncontrolled pressure, and relative humidity could be easily removed from the sensor response using conventional methods; compensation of temperature is complicated enough to look for machine learning algorithms as the main focus of this thesis. In this study, a split ring resonator (SRR) operating @ 1 GHz is designed to measure methanol/acetone content level in water. The planar sensor holding the material inside a tube and a commercial temperature sensor are located inside a chamber. LabView is employed to record scattering parameters during a temperature cycle from room temperature up to 50 ℃.Simulation results suggest the dependency of both material- and sensor-properties on the temperature. It is also shown that an uncontrolled environmental situation can deviate the sensor response into incorrect or meaningless results, based on which our decision is made in industrial application. Transmission/Reflection response of the sensor with all features including resonance frequency, magnitude, and quality factor are examined with the aim of reliable network input data. Various machine learning algorithms, including Decision Tree, KNN, Logistic Regression, and MLP are examined to remove the effect of temperature on sensor response, whose performance comparison for classifying 10 materials concluded MLP as the best classifier with 100 % accuracy.But for all that, eliciting environmental status, besides recognizing the material type, is lucrative information to be utilized in further data-processing. Temperature of the measurement, imbued in the recorded transmission profiles, is extracted using a novel technique of cascading a primary classifier with linear regression.Finally, in pursuit of improving the limit of detection in the microwave sensor, MLP algorithm is significantly scrutinized with the help of hyperparameter optimization, wherein concentrations of methanol-in-water are discriminated with increments as low as 1 % with accuracy of >95%
-
- Subjects / Keywords
-
- Graduation date
- Spring 2021
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.