On-line estimation of biomass in an E. coli fed-batch fermentation

Abstract

In this work, an Extended Kalman Observer is applied to the on-line determination of biomass concentration in a high-cell density fed-batch fermentation of Escherichia coli. Although the importance of this fermentation process for the biopharmaceutical industry is widely recognized, there are still several difficulties associated with the design of monitoring and control algorithms that could improve the performance of the process by decreasing the production costs and increasing the yield. In this process, biomass concentration has an important role for model predictive control, estimation of specific growth rates, prevention of acetate accumulation and optimization of the production of recombinant proteins (regarding both productivity and moment of induction). However, nowadays it is still determined using off-line laboratory analysis, making it of limited use for control purposes. For the development of the Extended Kalman Observer, a dynamical mathematical model of the process was used, which includes balance equations for the main state variables (biomass, glucose, acetate, dissolved oxygen and carbon dioxide concentrations) together with a complex kinetic model describing the 3 main metabolic pathways of Escherichia coli. The observer applied in this work requires the on-line measurement of a subset of state variables (dissolved oxygen and carbon dioxide concentrations) together with broth weight and gaseous mass transfer rates. State-of-the-art sensors were used for measuring dissolved oxygen and carbon dioxide concentrations and gaseous transfer rates were determined on-line using commercial gas analysers. The calculations were performed on-line in a developed LabVIEW data acquisition and control system. The extended Kalman observer exhibited a good convergence to the real values of biomass concentration, with a very low quadratic difference between experimental and estimated data. Also, the sampling frequency for the measured variables is compatible with the existing experimental data.