PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
October 5, 2025AIChE Journal4 citationsOpen Access

2D population balance modeling and ML‐based multi‐objective optimization for the crystallization process of resveratrol

View Full Paper
ÁOÁlmos OroszMNMonika NealRRRekha R. Rao

Key Points

  • The model predicts the crystallization behavior of resveratrol, improving purity and process efficiency.
  • Using multi-objective optimization, the study finds a balance between yield, batch time, aspect ratio, and median crystal size.
  • A hybrid approach of mechanistic modeling and machine learning significantly speeds up simulations for crystallization processes.
  • The findings offer a scalable strategy for optimizing complex processes in pharmaceutical manufacturing.

Abstract

Abstract Crystallization is critical in pharmaceutical manufacturing, influencing active pharmaceutical ingredient (API) purity and processability. This study models the cooling crystallization of resveratrol in a water‐ethanol solvent using a two‐dimensional population balance model (2D‐PBM). Experimental data from Focused Beam Reflectance Measurement (FBRM), UV/Vis spectroscopy, and microscopy supported model calibration via design of experiments. The well‐calibrated model enabled multi‐objective optimization (MOO) to (1) maximize yield and minimize batch time, and (2) explore the relationship between aspect ratio and median crystal size. While the first scenario showed minimal trade‐offs, the second revealed a balance between aspect ratio and size/yield. A hybrid approach combining mechanistic modeling with machine learning drastically accelerated simulations and enabled efficient prediction of Pareto‐optimal solutions. This integration offers a scalable and accurate optimization strategy for complex crystallization processes.

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Orosz et al. (2025) studied this question.

synapsesocial.com/papers/68e22da774308421369af0e1https://doi.org/10.1002/aic.70094
Ask AI
Helpful
Bookmark
Share
View Full Paper