Skip to main content
18.97.14.81

A framework for model reliability and estimability analysis of crystallization processes with multi-impurity multi-dimensional population balance models

By Fysikopoulos, D; Benyahia, B Borsos, A Nagy, Z K; Rielly, C D

Published on

Abstract

The development of reliable mathematical models for crystallization processes may be very challenging due the complexity of the underlying phenomena, the inherent Population Balance Models (PBMs) and the large number of parameters that need to be identified from experimental data. Due to the poor information content of the experiments, the structure of the model itself and correlation between model parameters, the mathematical model may contain more parameters than can be accurately and reliably identified from the available experimental data. A novel framework for parameter estimability for guaranteed optimal model reliability is proposed then validated by a complex crystallization process. The latter is described by a differential algebraic system which involves a multi-dimensional population balance model that accounts for the combined effects of different crystal growth modifiers/impurities on the crystal size and shape distribution of needle-like crystals. Two estimability methods were combined: the first is based on a sequential orthogonalization of the local sensitivity matrix and the second is Sobol, a variance-based global sensitivities technic. The framework provides a systematic way to assess the quality of two nominal sets of parameters: one obtained from prior knowledge and the second obtained by simultaneous identification using global optimization. A cut-off value was identified from an incremental least square optimization procedure for both estimability methods, providing the required optimal subset of model parameters. The implemented methodology showed that, although noisy aspect ratio data were used, the 8 most influential and least correlated parameters could be reliably identified out of twenty-three, leading to a crystallization model with enhanced prediction capability.

Journal

Computers & Chemical Engineering. Volume 122,, 2019, 275-292

DOI

10.1016/j.compchemeng.2018.09.007

Type of publication

Peer-reviewed journal

Affiliations

  • Loughborough University

Article Classification

Research Article

Classification Areas

  • Modeling

Tags