Automatic Control of Bioprocesses - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 57

feeding control, no acetate was produced. The dissolved oxygen concentration was
adjusted to a set point of 30 % over 22 h while decreasing the temperature from
36 to 25 C.
For the probing feeding strategy, Velut et al. [ 66 ] examined the effect of reactor
scale as well as the influence of different medium types. They applied 1.5-min
glucose pulses for laboratory-scale fermenters and longer pulses of 3 min for
large-scale fermenters to compensate for their slower response. The probing
feeding strategy showed good results independent of the medium used. However,
the use of a complex medium leads to complications in the interpretation of the
pulse response. They determine that the pulsed feeding does not harm the pro-
ductivity and propose an optimized predetermined feeding trajectory with addi-
tional superimposed pulses only for monitoring purposes.
This control strategy was also employed by Xue and Fan [ 67 ] at laboratory and
pilot plant (500 L) scale for a recombinant E. coli strain producing human-like
collagen. For the laboratory-scale experiments, they obtained similar results to
previous optimized studies with 69.1 gL -1 dry cell weight and 13.1 gL -1 human-
like collagen. Compared with previous experiments, they observe a reduction of
the resulting dry cell weight when applied to the pilot plant. They assume that this
results from the different oxygen transfer capacity. However, the collagen pro-
duction of 9.6 gL -1 was a satisfying result. They therefore present a successful
application of the probing feeding strategy in a pilot-plant-scale fermentation
process.
3.8 Extremum-Seeking Control
Extremum-seeking control is a gradient method to determine online unknown
parameters through the analysis of measurement results as a response to a peri-
odical excitation signal called dither. Dochain et al. [ 68 ] present a survey on two
important classes of extremum-seeking control: the perturbation-based and the
model-based methods. They investigate the applicability to processes and reaction
systems using theoretical models and show the theoretical efficiency of this closed-
loop control algorithm. Cougnon et al. [ 69 ] carried out numerical simulation
studies on a fed-batch process model to illustrate their performance for closed-loop
control of bioprocesses. They present an adaptive extremum-seeking controller.
The controller drives the system to an unknown desired set-point in order to
maximize biomass production. In this contribution the authors assume that the
primary carbon source is measurable.
Dewasme et al. [ 70 ] note that model-based controls are subject to high
uncertainty. Therefore, they present a model-free extremum-seeking strategy in a
simulation study on S. cerevisiae. They present simulations where the tracking of
the critical substrate level (border between fermentative and reparative metabo-
lism) is correctly performed by two different gradient estimation procedures. The
input variables for the algorithm are OUR and CPR. The actuating variable is the