Principal component and neural network calibration of a microwave frequency composition measurement sensor.

Microwave sensors are becoming more prevalent throughout a variety of industries. While providing an effective form of measurement, microwave sensors are difficult to calibrate and provide results which can be difficult to interpret. An improved method for calibrating microwave sensors has been developed which transforms the waveform of a microwave spectrometer using principal component analysis and the results are used to train an artificial neural network to analyze a subject material. Broadband microwave spectrum calibration (BBMSC) is demonstrated using waveforms captured by a microwave spectrometer in a circular waveguide containing pulp stock slurry. This thesis provides a review of the general applications of microwave sensors, details state-of-the-art calibration methods, as well as providing an introduction to principal component analysis and neural networks. The thesis continues by presenting the BBMSC method in detail, as well as how this method is applied to a set of waveforms of pulp-stock data and concludes with a discussion of the potency of BBMSC and recommendations for the future.