Browsing by Subject "Online state estimation"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Publication
    Online process state estimation for Hansenula polymorpha cultivation with 2D fluorescence spectra-based chemometric model calibrated from a theoretical model in place of offline measurements
    (2023) Babor, Majharulislam; Paquet-Durand, Olivier; Berg, Christoph; Büchs, Jochen; Hitzmann, Bernd
    The use of 2D fluorescence spectra is a powerful, instantaneous, and highly accurate method to estimate the state of bioprocesses. The conventional approach for calibrating a chemometric model from raw spectra needs a large number of offline measurements from numerous runs, which is tedious, time-consuming, and error-prone. In addition, many process variables lack direct signal responses, which forces chemometric models to make predictions based on indirect responses. In order to predict glycerol and biomass concentrations online in batch cultivation of Hansenula polymorpha, this study substituted offline measurements with simulated values. The only data from cultivations needed to generate the chemometric model were the 2D fluorescence spectra, with the presumption that they contain sufficient information to characterize the process state at a measurement point. The remainder of the evaluation was carried out with the aid of a mathematical process model that describes the theoretical interferences between process variables in the system. It is shown that the process model parameters, including microbial growth rate, the yield of biomass from glycerol, and lag time can be determined from only the spectra by employing a model-based calibration (MBC) approach. The prediction errors for glycerol and biomass concentrations were 8.6% and 5.7%, respectively. An improved model-based calibration (IMBC) approach is presented that calibrates a chemometric model for only biomass. Biomass was predicted from a 2D fluorescence spectrum in new cultivations, and glycerol concentration was estimated from the process model utilizing predicted biomass as an input. By using this method, the prediction errors for glycerol and biomass were reduced to 5.2% and 4.7%, respectively. The findings indicate that model-based calibration, which can be carried out with only 2D fluorescence spectra gathered from prior runs, is an effective method for estimating the process state online.