Microwave Measurement Systems for Parameter Estimation and Classification

Abstract: Microwave measurement systems are attractive for diagnostics and monitoring purposes in a number of important applications. For example, the strong interaction between microwaves and water make microwaves well-suited for moisture measurements. Moreover, the power used in microwave measurements is often sufficiently low such that the measurement can be classified as non-destructive. As such, microwave measurements systems are appropriate for applications in, for example, biomedical imaging and monitoring of pharmaceutical processes.In this thesis, parameter estimation methods are employed for two microwave measurement systems with application in the pharmaceutical industry. Additionally, we present a numerical study of a simplified microwave measurement system for the localization of intracranial bleedings via classification. In order to achieve good agreement between measured and simulated data, we utilize accurate electromagnetic models by means of the finite element method and calibration methods using a reference case measurement. In addition, we utilize a priori information to mitigate problems associated with parameter ambiguity, where the a priori information may be incorporated by means of regularization. First, we consider a transmission/reflection tomography measurement system. Here, the parameter estimation method involves a goal function that corresponds to the misfit between the measured and simulated scattering data, where a non-linear gradient-based optimization method is used to determine the parameters. The gradients are computed by means of continuum sensitivity expressions based on an adjoint field problem. The tomography system is used to estimate the effective permittivity of densely packed microcrystalline cellulose (MCC) pellets and we find that the estimated permittivity depends on the moisture content of the MCC pellets. Second, we solve a minimization problem for resonance measurements in a pharmaceutical process vessel, which acts as a metal cavity. Here, we estimate parameters using a quadratic minimization problem with a regularization term, which incorporates a priori information provided from other sensors. The physical model is linearized and small perturbations of the resonant frequencies are related to small variations in the permittivity. During operation, the vessel is loaded with MCC pellets that are fluidized and circulated by injection of air, which yields a dilute MCC/air mixture. The measured resonant frequencies are used to estimate the effective complex permittivity of three different sub-regions inside the process vessel as a function of process time.