The goal of this research theme is to develop reliable multivariate calibration models from online measured data.
Multivariate Calibration of Spectroscopic Data
This ongoing research project initiated in 1996 is concerned with the development of novel multivariate calibration techniques for spectroscopic data. A novel way of calibrating multivariate models from reacting spectroscopic data was proposed, which is based on the reaction-variant (RV) part of the data, and led to advantages in terms of space-inclusion conditions over classical calibration procedures. Subspace correction methods have been developed to correct the effects of noise, bias, systematic drifts and disturbances caused by instrumental, process and operational conditions on calibration models. Recently, calibration models have been used in the extent-based incremental identification. [More]
Rank Deficiency and Rank Augmentation
In this ongoing research project, started in 1994, the problem of rank deficiency in the analysis of spectroscopic data by Factor Analysis was defined and rank augmentation solutions were formulated, such as the analysis of data from multiple batches or the addition of reactants/products during the reaction. Rank augmentation of Reaction-Variant (RV) and Reaction- and Mass-transfer-Variant (RMV) forms of spectroscopic data using calorimetric and gas consumption data was studied. Recently, rank augmentation of concentration data using calorimetry for the purpose of calculating extents of reaction and of mass transfer was investigated. [More]
Calibration and Data Reconciliation in Bioprocesses
This research project conducted in 2000-2002 and in 2005-2009 showed how calibration, monitoring and control of bioprocesses can be performed despite metabolism-induced correlations in concentrations and process/instrumental drifts. The concept of monitoring and controlling of bioreactors using calibration models based on mid-IR spectra and Partial Least Squares Regression (PLSR) was demonstrated. Real-time techniques able to maintain robust calibration models and to perform data reconciliation were elaborated. [More]
Factor Analytical Modeling of Biochemical Data
This research project directed in 1994-1995 focused on the use of Factor Analysis (FA) for the resolution of biochemical reaction networks. FA methods, which are well known in the chemometrics community, were adapted to the analysis of data from bioprocesses. A new technique involving the estimation of reaction stoichiometries and of reaction extents was also proposed. Theoretical concepts were applied to experimental data from the fed-batch production of yeast. [More]