Question 1

Which measurement technique suits which variable?

Accepted Answer

Vibrational spectroscopy suits dissolved substrates, metabolites and product where concentrations are high enough to leave a signal. Capacitance tracks viable biomass and distinguishes it from debris. Off-gas analysis gives respiration continuously and cheaply. Fluorescence reports certain intracellular states. Match the technique to the variable rather than adopting whichever instrument is available.

Question 2

How should a calibration model be built?

Accepted Answer

Collect reference samples spanning the full concentration range the process will reach, including deliberate excursions, since a model trained only on normal trajectories cannot handle the situations control is meant to correct. Use multivariate regression with careful preprocessing, and record the reference assay uncertainty, which sets the ceiling on achievable accuracy.

Question 3

Why is validation on independent runs essential?

Accepted Answer

Spectra collected minutes apart within one cultivation are nearly identical, so holding some out gives an optimistic error estimate that collapses on a new batch. Validate on cultivations never used in model building, ideally including different media lots and inoculum preparations, and report error on those runs rather than internal cross-validation figures.

Question 4

What practical problems affect probes in a vessel?

Accepted Answer

Probes must tolerate sterilisation cycles, resist fouling by cells and protein that attenuates signal over long runs, and cope with rising optical density. Bubbles crossing the measurement path add noise. Report probe position, cleaning regime and how signal quality was monitored across the cultivation rather than assuming stability.

Question 5

What is a soft sensor and when is one justified?

Accepted Answer

A soft sensor infers a variable that cannot be measured directly from ones that can, such as estimating growth rate from oxygen uptake and capacitance. It suits situations where no probe exists or where an instrument is unavailable, and it needs the same independent validation as any calibration, plus a stated range of conditions where it holds.

Question 6

What changes when a measurement drives control?

Accepted Answer

Feeding to a measured concentration rather than a fixed profile removes accumulation of excess substrate, but a faulty reading now moves the process rather than merely misreporting it. Build in plausibility checks, define fallback to a fixed profile when the signal is questionable, and demonstrate stable operation before claiming a control benefit.

Question 7

How is model maintenance handled over time?

Accepted Answer

Media lots, strain changes, probe replacement and scale differences all shift the relationship a model captured, so performance degrades unless monitored. Track prediction error against periodic reference samples, define a trigger for recalibration, and treat transfer between instruments as a task requiring its own validation rather than an assumption.

Biotechnology

Biotechnology

Research Frontiers in Bioprocess Analytics Real-Time Monitoring