Real–Time Data on Manufacturing Bio–Processes with NIR spectroscopy

Related Vendors

Frewitt fabrique de machines SA

Sartorius AG

In this case, several variables (areas in the spectrum) are frequently used for the assessment, which is why it is also often referred to as a multivariate data assessment. It is particularly difficult to develop robust chemometric models for fermentation processes as, on the one hand, target parameters strongly correlate in such cases and on the other, variations between separate batches need to be taken into consideration.

Once identified, these hurdles can be overcome by setting up the calibration model using a solid data set. To do this, it is necessary to take between 10-20 samples during the processing of a single batch, so that the sample gives a sufficient indication of the variations that occur during the process.

How to Ensure a Safe and Sound Measuring

Furthermore, it is crucial that a certain number of batches are measured in order to ensure that even variations that occur between different process runs are included in the model. In order for this to happen, the data for the calibration and validation of the chemometric model must come from different batches. As a result, a practice emergency run will be carried out during the validation process and a completely new batch, which is not included in the calibration set as with cross validation, will be projected.

Experience shows that the models are already so sound after three to four runs that, in subsequent processes, they are able to predict multiple target parameters in real time and with a good level of accuracy. The elimination of correlations between various target parameters (e.g. glucose consumption and biomass structure) can, in most cases, only be achieved experimentally.

What Spiking Experiments can do for NIR

Spiking experiments whereby a sufficient quantity of analyte is added, for example at the end of the fermentation process, are particularly suitable for this task. Samples taken after the spiking will no longer show any correlation whatsoever to other analytes, thus ensuring that the only spectral changes that are incorporated into the model are those that are actually produced by concentration changes made to the target analyte.

(ID:42525331)