Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Chemical and Biomedical Engineering

Committee Chair

Fernando V. Lima

Committee Member

Debangsu Bhattacharyya

Committee Member

Stephen Zitney

Committee Member

Mario Perhinschi

Committee Member

Marcelo P. A. Ribeiro

Committee Member

Paolo Pezzini


One of the most critical aspects of any chemical process engineer is the ability to gather, analyze, and trust incoming process data as it is often required in control and process monitoring applications. In real processes, online data can be unreliable due to factors such as poor tuning, calibration drift, or mechanical drift. Outside of these sources of noise, it may not be economically viable to directly measure all process states of interest (e.g., component concentrations). While process models can help validate incoming process data, models are often subject to plant-model mismatches, unmodeled disturbances, or lack enough detail to track all process states (e.g., dissolved oxygen in a bioprocess). As a result, directly utilizing the process data or the process model exclusively in these applications is often not possible or simply results in suboptimal performance.

To address these challenges and achieve a higher level of confidence in the process states, estimation theory is used to blend online measurements and process models together to derive a series of state estimates. By utilizing both sources, it is possible to filter out the noise and derive a state estimate close to the true process conditions. This work deviates from the traditional state estimation field that mostly addresses continuous processes and examines how techniques such as extended Kalman Filter (EKF) and moving horizon estimation (MHE) can be applied to semi-batch processes. Additionally, this work considers how plant-model mismatches can be overcome through parameter-based estimation algorithms such as Dual EKF and a novel parameter-MHE (P-MHE) algorithm. A galacto-oligosaccharide (GOS) process is selected as the motivating example as some process states are unable to be independently measured online and require state estimation to be implemented. Moreover, this process is representative of the broader bioprocess field as it is subject to high amounts of noise, less rigorous models, and is traditionally operated using batch/semi-batch reactors.

In conjunction with employing estimation approaches, this work also explores how to effectively tune these algorithms. The estimation algorithms selected in this work require careful tuning of the model and measurement covariance matrices to balance the uncertainties between the process models and the incoming measurements. Traditionally, this is done via ad-hoc manual tuning from process control engineers. This work modifies and employs techniques such as direct optimization (DO) and autocovariance least-squares (ALS) to accurately estimate the covariance values. Poor approximation of the covariances often results in poor estimation of the states or drives the estimation algorithm to failure.

Finally, this work develops a semi-batch specific dynamic real-time optimization (DRTO) algorithm and poses a novel costing methodology for this specific type of problem. As part of this costing methodology, an enzyme specific cost scaling correlation is proposed to provide a realistic approximation of these costs in industrial contexts. This semi-batch DRTO is combined with the GOS process to provide an economic analysis using Kluyveromyces lactis (K. lactis) β-galactosidase enzyme. An extensive literature review is carried out to support the conclusions of the economic analysis and motivate application to other bioprocesses.