A stable PLS calibration for solute concentration measurement in batch cooling crystallization processes is presented. Two example cases are included: Sulfathiazole dissolved in five different mixtures of water and 1-propanol and C15 dissolved in toluene. The data to be used in calibration are validated using Multivariate Statistical Process Control (MSPC) and sensitivity analyses criteria. Two Orthogonal Signal Correction (OSC) filters are compared as preprocessing techniques to enhance the model performance. The inherent criteria for selecting the best performing model was RMSEV of the test set. Models are validated with two different external test sets. One measured in conditions similar to calibration set and another measured in conditions close to true crystallization measurement conditions. The latter one tests the flexibility of the model to be used in future predictions. Results show that careful data and model validation steps clearly improve the modeling procedure. OSC filtering clearly simplified the model and also enhanced the performance of the PLS model. The models have successfully been used in several crystallization experiments.
- Cooling crystallization
- Multivariate calibration