Machine Learning-Derived Correlations for Scale-Up and Technology Transfer of Primary Nucleation Kinetics.

Area of Science:

• Chemical Engineering
• Process Chemistry
• Crystallization Science

Background:

• Scaling up crystallization processes presents significant challenges due to the unpredictable nature of primary nucleation and scale-dependent mechanisms.
• Existing scale-up approaches are numerous, complicating the transfer of laboratory findings to industrial production.
• Understanding nucleation kinetics is crucial for successful and efficient crystallization scale-up.

Purpose of the Study:

• To investigate the relationship between hydrodynamic features of a vessel and the kinetic parameters of nucleation in crystallization.
• To develop predictive models for nucleation rate and growth time using machine learning techniques.
• To identify key hydrodynamic parameters influencing the scale-up of unseeded crystallization processes.

Main Methods:

• Performed isothermal induction time studies across varying vessel volumes, impeller types, and speeds.
• Estimated nucleation rate and growth time parameters using an induction time distribution model.
• Utilized computational fluid dynamics (CFD) to calculate vessel hydrodynamic features.
• Applied and compared 18 machine learning models to correlate hydrodynamic features with nucleation kinetics.

Main Results:

• Identified nonlinear random Forest and gradient boosting models as high-performing for predicting nucleation rate.
• A nonlinear gradient boosting model demonstrated superior performance in predicting growth time.
• Ensembled models accurately predicted nucleation probability directly from hydrodynamic features (RMSE of 0.16).

Conclusions:

• Machine learning effectively analyzes limited induction time data to reveal critical hydrodynamic parameters for crystallization scale-up.
• Hydrodynamic features are key predictors for the scale-up of unseeded crystallization processes.
• This approach offers a pathway to more reliable and predictable crystallization technology transfer.