of 43 h and 47 h, the same effect observed at 18 h can be seen in an attenuated form.
One explanation for the sudden increase in the RQ value is the composition of the
nutrient medium. Among other components, yeast extract was used as a nitrogen
source, which contains high amounts of both nitrogen and carbon. The fraction of
residual yeast extract in the medium was rather high, leading to an accumulation of
carbon sources and thus to an increasing RQ value due to the Crabtree effect. In the
Digital Twin model, the carbon component in the nitrogen sources was not considered, which is why this effect could only be recognised in the real experiment.
Despite this limitation of the Digital Twin model, an RQ feedback control could be
developed based on the Digital twin, leading to more than 50 g L
À1 dry biomass
concentration in the real process, with less than 10 g L
À1 ethanol produced within
48 h.
It took about 2 days to develop the RQ feedback control for the cultivation of
S. cerevisiae on the Digital Twin (simulations, controller adaptations). Real cultivation of 48 h in an STR, including preparation and evaluation, is expected to take
about 1 week. If, instead of the simulations on the Digital Twin, real cultivations had
to be carried out during the control strategy development process, the development
time would have been extended to up to 3 weeks. Besides the significant time
savings, the consumption of resources (nutrient media components, energy, etc.)
was also significantly reduced due to the reduced number of real cultivations.
4.2.3 Development of Open-Loop-Feedback-Optimal (OLFO) Control
for the Cultivation of S. cerevisiae
The principle of the OLFO control strategy has been described in Sect. 2.3.2. The
suitability of the “SSF-BC-Simulator” as a tool for the development of the OLFO
control strategy for the cultivation of S. cerevisiae was illustrated in Fig. 3, Sect. 4.1.
The core of the OLFO controller is a relatively simple mathematical model for the
cultivation of S. cerevisiae, which is different from the process model within the
presented Digital Twin. The controller model is limited to map the consumption of
glucose and nitrogen, the growth of S. cerevisiae and the formation of the side
product ethanol. The mathematical OLFO controller model was adapted based on
either measured (real process) or simulated (Digital Twin) concentrations of substrate (glucose), product (ethanol) and biomass density (S. cerevisiae). In the optimisation part of the OLFO controller, substrate feed rate trajectories were optimised
at several points during the real or simulated (Digital Twin) process using the
adapted mathematical process model, where the adaption was based on the data
available up to the actual processing time. The substrate feed rate trajectory yielding
the highest concentration of dry biomass at the end of the simulated cultivation
(OLFO process model) was transferred to the PCS at each time point of model
adaption and process optimisation.
During controller development using the Digital Twin, six simulations were
carried out in total. After each simulation, the simulated cultivation results were
evaluated and the control strategy was adjusted to approach the control target
86
C. Appl et al.
One explanation for the sudden increase in the RQ value is the composition of the
nutrient medium. Among other components, yeast extract was used as a nitrogen
source, which contains high amounts of both nitrogen and carbon. The fraction of
residual yeast extract in the medium was rather high, leading to an accumulation of
carbon sources and thus to an increasing RQ value due to the Crabtree effect. In the
Digital Twin model, the carbon component in the nitrogen sources was not considered, which is why this effect could only be recognised in the real experiment.
Despite this limitation of the Digital Twin model, an RQ feedback control could be
developed based on the Digital twin, leading to more than 50 g L
À1 dry biomass
concentration in the real process, with less than 10 g L
À1 ethanol produced within
48 h.
It took about 2 days to develop the RQ feedback control for the cultivation of
S. cerevisiae on the Digital Twin (simulations, controller adaptations). Real cultivation of 48 h in an STR, including preparation and evaluation, is expected to take
about 1 week. If, instead of the simulations on the Digital Twin, real cultivations had
to be carried out during the control strategy development process, the development
time would have been extended to up to 3 weeks. Besides the significant time
savings, the consumption of resources (nutrient media components, energy, etc.)
was also significantly reduced due to the reduced number of real cultivations.
4.2.3 Development of Open-Loop-Feedback-Optimal (OLFO) Control
for the Cultivation of S. cerevisiae
The principle of the OLFO control strategy has been described in Sect. 2.3.2. The
suitability of the “SSF-BC-Simulator” as a tool for the development of the OLFO
control strategy for the cultivation of S. cerevisiae was illustrated in Fig. 3, Sect. 4.1.
The core of the OLFO controller is a relatively simple mathematical model for the
cultivation of S. cerevisiae, which is different from the process model within the
presented Digital Twin. The controller model is limited to map the consumption of
glucose and nitrogen, the growth of S. cerevisiae and the formation of the side
product ethanol. The mathematical OLFO controller model was adapted based on
either measured (real process) or simulated (Digital Twin) concentrations of substrate (glucose), product (ethanol) and biomass density (S. cerevisiae). In the optimisation part of the OLFO controller, substrate feed rate trajectories were optimised
at several points during the real or simulated (Digital Twin) process using the
adapted mathematical process model, where the adaption was based on the data
available up to the actual processing time. The substrate feed rate trajectory yielding
the highest concentration of dry biomass at the end of the simulated cultivation
(OLFO process model) was transferred to the PCS at each time point of model
adaption and process optimisation.
During controller development using the Digital Twin, six simulations were
carried out in total. After each simulation, the simulated cultivation results were
evaluated and the control strategy was adjusted to approach the control target
86
C. Appl et al.