| This thesis investigates the development and application of process systems engineering methods to a variety of medical and industrial problems. The projects fall into four categories: (a) dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for clinical diagnostics, (b) electrical impedance tomography (EIT), again as a means of clinical diagnostic imaging, (c) the use of passive acoustical sensors for on-line characterization of process or product states, (d) optimal experimental design for robust identification of chemical plants.; Two projects were tackled under the DCE-MRI topic. In the first a new method was developed to allow rapid (5 minutes), automated diagnostic imaging for acute heart-attack patients under free-breathing conditions. Continuous time-series modeling and multivariate analysis were employed to simultaneously extract localized tissue perfusion and viability information on a voxel-by-voxel basis in the image set. In support of this research a mechanistic model of the human cardiovascular system, and of heart tissue, was developed. In the second project a novel approach was developed to automate the characterization of tumors via DCE-MRI, with a specific clinical case study on prostate cancer. Principal Components Analysis (PCA) was used to model the structure in the dynamic data directly without a priori assumptions about model structure. This approach is expected to be more robust than alternatives such as ad hoc empirical modeling or pharmacokinetic modeling.; A new method was developed, using PCA, for automated analysis of multifrequency EIT images. This is shown to be a promising alternative to the traditional use of the Cole-Cole model. A novel experimental design approach for selecting an optimal set of electric current patterns was also developed. This is based on a D-optimal design based on a singular value decomposition (SVD) of the transimpedance matrix that relates the applied current patterns to the measured voltage patterns.; A novel on-line acoustic inferential sensor for snack chip texture was developed in collaboration with Frito-Lay Inc. based on experimental studies in one of their pilot plant facilities. A frequency-domain description of the acoustic signal generated by snack chips falling onto a metal surface in the process is used to predict the measured texture properties via a Partial Least Squares (PLS) model. In addition, a new approach to fusing the texture space measures into two robust and informative metrics was achieved using PCA.; Control relevant identification is a field that aims to develop methods for estimating a multivariable empirical model of a manufacturing process that promotes robust control on this process. This issue is particularly pertinent for processes that are inherently ill-conditioned. In this research project an existing, unconstrained robust identification formulation was refined and extended to explicitly handle all relevant linear process constraints, thus aiding the practical application of this technology. The approach applies a constrained D-optimal design to a SVD analysis of the process steady-state gain matrix.; The research presented in this thesis lends credence the notion that process systems engineering offers powerful tools for formulating and solving a diverse array of challenging real-world problems, for example, from manufacturing plants to the human body. |
