Automated Measurement and Monitoring of Bioprocesses: Key Elements of the M3C Strategy - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 3

a population changes, probably due to an overfeed of substrate. Exhaust gas is
usually analyzed in the exiting gas stream, outside the sterile barrier (behind the
off-gas filter) since all fermentation setups are open for the gas phase. Technical
terms such as ''at-line'' or ''inline'' are often used ambiguously and will be
avoided in this context.
Many in situ sensors deliver a continuous signal which is often gained in real
time and noninvasively—this is the monitoring part—while interfaced analytical
subsystems (such as chromatographs) deliver their results discretely and the time
steps between available data can be quite long—this is the measurement part. This
may be acceptable in slow bioprocesses but poses problems in highly dynamic
processes; in such cases, mathematical models are needed. Generally, sensors and
instruments with short, constant, and—very importantly—well-known response
behavior should be preferred.
Many sensors can be calibrated in usual units. One should take care of the
calibration status of the individual sensors, since systematic errors make the results
false. However, most of those sensors cannot be calibrated or recalibrated after
sterilization. This fact poses extra demands in terms of long-term stability, little
drift, and minimized effects due to sterilization.
Several sensors and process analyzers produce ''relative'' signals rather than
crisp concentration values. This is not a priori ''bad'' but has to be accounted for
accordingly. Especially in production processes, during which the design space
should not be exceeded, such signals can be compared automatically and in real
time with the respective time trajectories of historical (e.g., reference or valida-
tion) processes and, thus, give instantaneous information about deviations from
expected development, or are useful for fault detection. These relative signals are
also quite valuable in R&D environments because they make ''additional''
information available that would otherwise be skipped: modeling and chemometric
techniques can extract ''hidden'' information from such signals or can reveal
otherwise unseen, yet useful correlations. Modeling is an essential element in
bioprocess monitoring and state estimation [ 6 - 8 ]; however, this aspect will be left
out here since Chaps. 2 and 6 focus on this.
The structure of this chapter tries to depict the present situation in online
bioprocess monitoring: The state-of-routine section covers process variables that
are generally monitored and often also kept under closed-loop control. This holds
also for industrial production processes. Obviously, most of these variables are
physical and chemical variables rather than biological ones [ 9 ]. The state-of-the-
art section addresses all variables that can be accessed online, although they are,
with few exceptions, acquired in academic laboratories and in industrial R&D
laboratories and pilot plants only. A similar structure was used by Olsson et al.
[ 10 ], who discussed several advantages and disadvantages of the various methods
for academia and industry.
Variables are those properties of the system which vary in time, and whose
dynamic properties are therefore determined by the values of the parameters [ 11 ].
The parameters of a system are those properties which are inherent to the system.
What we can measure are the variables of a system; all bioprocesses are multivariate