Nonlinear state estimation as tool for online monitoring and adaptive feed in high-throughput cultivations

Abstract

Increasing the comparability to large scale fermentations is a constant aim during scale-down, including growth-limiting feeding of the carbon source. Minibioreactor facilities greatly increase the throughput and offer many advantages. However, online measurements in this small scale often do not include critical process variables, necessary to control the feed rate. State estimators utilize mathematical models to estimate non-measurable states applying existing information about the inputs, outputs, and measurement uncertainties. Though, existing applications focus on bench-top or production-scale applications, where advanced process analytical technologies such as spectroscopic methods are available. Here, we present a concept for model-based nonlinear state estimation in a high-throughput platform with 24 parallel experiments in the mL-scale. An extended and a sigma-point Kalman filter are implemented based on online measurements only. The dissolved oxygen concentration and a relation of biomass growth and nitrogen consumption are used as measurement input. The current state of the cultivation is estimated iteratively throughout Escherichia coli cultivations and used to maintain predefined growth setpoints by adapting the feed rate. With this, we bring a further level of monitoring and process understanding to minibioreactor systems, and therefore enable automatization and optimization of miniaturized bioprocesses under controlled cultivation conditions.

High‐throughput screening of optimal process conditions using model predictive control

“…Other implementations are suggested, as for example, also consider a Kalman Filter (Kemmer et al, 2022), to deal with these uncertainties and plant-model mismatches to ensure a sufficiently accurate parameter estimation and optimal control.…”

“…However, the current framework is limited by the uncertainties in the parameters, the model structure, and the time evolution of the system dynamics. Other implementations are suggested, as for example, also consider a Kalman Filter (Kemmer et al, 2022), to deal with these uncertainties and plant‐model mismatches to ensure a sufficiently accurate parameter estimation and optimal control.…”

Modeling of enzyme-mediated glucose release to facilitate continuous feed in miniaturized cultivations

“…After the batch phase of approx. 9 h, the "control pulse" cultivations were supplied with bolus additions of a 377.3 mg L -1 glucose feed by the robotic pipetting tips (Kemmer et al 2022). For the remaining MBRs, the glucose release, and thus the glucose feed, was initiated and controlled by repeated enzyme additions by the liquid handler.…”

“…The cells were separated from the supernatant by centrifugation at 15000𝑔 for 10 min, and the glucose, acetate and magnesium concentrations were determined in the supernatant using the Cedex Bio HT Analyzer. An additional estimate of the biomass concentration 𝑋 𝑁𝐻 3 was obtained from the added ammonia as described by Kemmer et al (2022).…”

“…After each of these actions, the simulation is stopped, the volume-dependent states are recalculated, and the simulation is restarted. The values of the model parameters were estimated as described by Kemmer et al (2022).…”