Rationalising crystal nucleation of organic molecules in solution using artificial neural networks

Timothy Hjorth,Åke C Rasmuson,Michael Svärd

doi:10.1039/c8ce01576g

Timothy Hjorth, Åke C Rasmuson + Show 1 more

Open Access

https://doi.org/10.1039/c8ce01576g

Copy DOI

Journal: CrystEngComm	Publication Date: Jan 1, 2019
Citations: 3	License type: CC BY 3.0

Affiliation: KTH Royal Institute of Technology

Abstract

Artificial neural network modelling is used to analyse and predict primary nucleation based on various physicochemical solute and solvent parameters.

Highlights

Nucleation is an important first step of many crystallisation processes, both natural and industrial, with a direct impact on several properties of the final crystal products
The results indicate partial qualitative consistency between the artificial neural networks (ANNs) models and the classical nucleation theory (CNT), with the nucleation difficulty increasing with an increase in mass transport resistance and a reduction in solubility
Some parameters not included in CNT, including solute molecule bond rotational flexibility, the entropy of melting of the solute, and intermolecular interactions, exhibit explanatory importance and significant qualitative effect relationships

Summary

Introduction

Nucleation is an important first step of many crystallisation processes, both natural and industrial, with a direct impact on several properties of the final crystal products. The present understanding is limited as regards the governing mechanisms and the predictability of nucleation from solute and solvent properties. Industrial crystallisation processes including a nucleation step are often designed on an empirical basis, with limited control of important governing variables. Crystal nucleation is a complex and stochastic process, sensitive to minor changes in key conditions.[1,2] Accurate prediction of nucleation behaviour is difficult through ordinary regression analysis and related methods, and ANNs could provide a new path towards analysing this complex process and provide important predictive capability. ANN modelling does not explicitly require assumptions regarding the functional relationships of the modelled process: the functional mapping between the input and output is produced during calibration to a supplied training dataset.[6]

Objectives

Methods

Results

Conclusion