(233c) Data Driven Model for the Prediction of Particle Size Distribution from Measured Chord-Length Distribution: Model Extensions and Application to Population Balance Model Identification

A new data-driven model has been developed to determine 1D/2D particle size distribution (PSD) from measured FBRM chord-length distribution (CLD) data. The structure of the model consists of three steps: first, the measured CLDs are compressed down to a small set of parameters; second, these parameters are correlated with low order moments or a small number of percentiles of the PSD using regression models; third, the PSD low order moments are used as input variables, and a two layer-network sub-model is built to predict the PSD in a form of a histogram. In this work the model is further extended using geometric models and principal component analysis (PCA) as slave models for the first step. This modeling paradigm is utilized to convert in-line measured CLD to predicted PSD by laser diffraction of particles with an elongated morphology. The de-convoluted data is then used with population balance equations to identify growth and aggregation mechanisms. There is no special requirement for the preparation of the training set, and the model is robust to potential second order artifacts in the size distribution. Thus the modeling paradigm is very robust and economic in terms of data requirements. The model steps are physically intuitive allowing using this paradigm to model multiple types of systems.