(111a) Modeling Batch Pharmaceutical Reactions by Factor Analysis of Spectral Data

Authors: 
Afif, F., Tufts University


The availability of a variety of spectral measurement techniques (Raman, IR, NIR, et al) has enabled the collection of a plethora of data during the operation of batch reactions and many other processes. However, such spectral data is often influenced by several reaction components, the mixture temperature, and possibly the solvent. Consequently, a linear regressive model (e.g. via PLS), often called a Chemometric (CM) Model, is developed to relate spectral measurements to the respective mixture compositions. The development of such a model requires initial data for offline samples from the reacting mixture that includes spectral as well as composition information. The building of such a CM model is laborious and time consuming, especially when the temperature of the reactor is time-varying.

The present communication introduces a methodology that enables the modeling of the nonlinear time evolution of the reaction without the use of a CM model. Principal Component Analysis (PCA) of the spectral data at different times and from several batch runs reveals the time-invariant as well the time-variant characteristics of the reaction(s). Structured target factor analysis (Fotopoulos, Georgakis, & Stenger, 1994, 1998) of the time-invariant information helps define the possible reaction stoichiometry structure. The time-variant data is then used to develop the ?kinetic expression? of the model using the generalized tendency kinetic modeling approach presented earlier (Makrydaki & Georgakis, 2007). This model is seen primarily as an aid for the optimization of the process, rather than the complete understanding of the chemical phenomena involved. The approach is tested against experimental data from a model pharmaceutical reaction.

________________________________

Fotopoulos, J., Georgakis, C., & Stenger, H. G. (1994, June 28 - July 1). Structured target factor analysis for the stoichiometric modeling of batch reactors. Paper presented at the American Control Conference, Baltimore, MD.

Fotopoulos, J., Georgakis, C., & Stenger, H. G. (1998). Use of tendency models and their uncertainty in the design of state estimators for batch reactors. Chemical Engineering and Processing, 37(6), 545-558.

Makrydaki, F., & Georgakis, C. (2007). Multivariate Linear Regression as a Tool in Modeling the Kinetics of Complex Chemical Reactions, Annual AIChE meeting. Salt Lake City, UT.