Combining Statistical Methods and Fundamental Models for Polymerization and Pharmaceutical Processes
Loading...
Authors
Gibson, Lauren
Date
2024-08-01
Type
thesis
Language
eng
Keyword
Fundamental Models , Statistical Methods , Polyethylene Process , Bayesian Statistics , Model-Based Design of Experiments , Parameter Estimation , Pharmaceutical Model
Alternative Title
Abstract
Fundamental models and statistical methods are developed and applied for polymerization and pharmaceutical processes. First, two fundamental models for gas-phase ethylene/hexene copolymerization with a three-site metallocene catalyst are developed. The first model predicts the joint molecular weight and copolymer composition distribution for lab-scale copolymerization at temperatures between 60 and 85 oC. The second model predicts product properties obtained during continuous pilot-plant operation. Both polymerization models have many unknown rate constants and activation energies. Although large data sets are available for parameter estimation, only subsets of the model parameters (i.e., 53 out of 60 in the first modeling study and 34 out of 61 in the second) can be reliably estimated to avoid overfitting the data.
The second half of this thesis focuses on new Bayesian statistical methods that make it possible to estimate all of the model parameters using limited data. In the first statistical-method project, a new objective function for D-optimal model-based design of experiments (MBDoE) is derived and used to select experimental conditions for a pharmaceutical case study with a singular Fisher information matrix (FIM). This Bayesian objective function accounts for prior information about plausible values of the model parameters, resulting in an invertible augmented FIM. A minimum volume ellipsoid (MVE) methodology is proposed for comparing the effectiveness of designed experiments. Testing the proposed MBDoE method using MVEs confirms that the proposed D-optimal objective function leads to reliable parameter estimates that tend to occupy smaller ellipsoids than estimates from A-optimal experiments.
The second statistical-method project is concerned with Bayesian methods for estimating parameters in fundamental models of chemical processes. Simple objective functions are derived for situations involving truncated normal priors or uniform priors for model parameters. These types of priors are attractive because they are more consistent with hard parameter bounds enforced during parameter estimation than unbounded normal priors are. A new parametric bootstrapping approach that considers model nonlinearity and truncated normal priors is proposed to obtain information about uncertainties associated with estimated parameters. A pharmaceutical case study is used to illustrate the effectiveness of the proposed objective functions and bootstrapping method.
Description
Citation
Publisher
License
Queen's University's Thesis/Dissertation Non-Exclusive License for Deposit to QSpace and Library and Archives Canada
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.