Promoteur : Prof. Alain VANDE WOUWER

Summary

This work thesis presents research results in the field of modeling, state estimation and real time optimization of Escherichia coli fed-batch cultures.
Escherichia coli is one of the most popular host microorganisms for the production of several compounds relevant for the industry, such as recombinant proteins, insulin, hormones, among others. Escherichia coli offer a number of advantageous features such as a rapid growth rate, high cell densities, a genome sequence which is completely known and simple nutritional requirements.
To reach high cell density, a rich medium supplemented with a high glucose concentration is used. However, an excessive presence of this carbon source can lead, under aerobic conditions, to acetate production which inhibits cells growth. This phenomenon is known as « Overflow metabolism ».
The objective of this thesis is the optimization of fed-batch cultures of E. coli. The first step of this study is to derive a mathematical model of E. coli cultures which is able to describe the overflow metabolism while being as simple as possible. The second step is the design of a state estimator to compensate for the lack of on-line measurements, and of different controllers making use of partial measurement and model information.
In this connection, the first part of the thesis proposes an original sequential parameter identification method of a dynamic model of Escherichia coli BL21(DE3) fed-batch cultures. The proposed macroscopic model is based on the overflow metabolism assumption of Sonnleitner and Kappeli, suggesting two metabolic pathways (respirative and respiro-fermentative), and consists of a set of nonlinear mass balance differential equations. Model unknown parameters are estimated from experimental data collected from dedicated experiments with a 5-liter pilot bioreactor. Experiments are designed in order to force switches from one metabolic pathway to another. A sequential identification procedure, based on a specific data partitioning, is achieved. The resulting dynamic model is in good agreement with the experimental data as shown in validation tests.
The second part of the thesis focuses on monitoring and optimization of the Escherichia coli fed-batch cultures. First, an extended Kalman filter is used to estimate the unmeasured key components, glucose and acetate, from the measurements of biomass only. The observer is validated using the experimental data base. Then, real time optimization techniques are used for the optimization of the fed-batch cultures. The on-line output signal is based on the on-line measurements of the gas stream and some pseudo-stoichiometric coefficients. Three different approaches are studied, including an extremum seeking controller based on a bank of filters and the extended Kalman filter, a PI controller which uses an event detection algorithm and a two-level controller which is based on the switches around the optimal input. The controllers are compared in terms of on-line measurement requirements, model-free or model-based algorithms, performance and practical implementation.