Assessment of near infrared and “software sensor” for biomass monitoring and control

Abstract

Spectroscopic instrumentation is often seen as promising for process analytical technology (PAT) to enhance control of manufacturing (bio)pharmaceuticals. The interpretation of near infrared spectra is challenging due to the large number of wavelengths recorded and the overlapping absorbance features of near infrared spectroscopy. This work applies a controlled random search procedure to select an optimal window of wavelengths giving a good calibration model for biomass concentrations during cultivation of Bordetella pertussis, the causative agent of whooping cough. The proposed wavelengths selection procedure outperforms the traditional calibration procedures. In the second half of the paper, the near infrared based predictions are compared with the estimations obtained from a software sensor for biomass and specific growth rate based in standard measurements of oxygen consumption. Both methods estimate the exponential biomass growth properly. The near infrared predictions depend on the quality of the training dataset, which needs to encompass all possible sources of temporary disturbances like pH and dissolved oxygen. If the training dataset does not comprise such disturbances, then the accuracy and robustness of the near infrared predictions are less favorable than those of the software sensor. Although near infrared has the potential to provide more information than just biomass, the software sensor is the preferred choice for feedback control of biomass and specific growth rate.