Crystallization is a fundamental unit operation in the pharmaceutical industry, particularly for the production of active pharmaceutical ingredients (APIs). Mathematical models could play a critical role to accelerate crystallization scale-up and optimization. However, their development presents significant challenges due to the complex and highly non-linear phenomena governing the process. Traditional first-principles models often struggle to capture the full scope of this phenomenological complexity, while purely data-driven approaches suffer from a lack of interpretability and limited generalization capabilities. To overcome these limitations, this thesis employs the recently developed HyMech software (Rossi et al., 2026, Comp. Chem. Eng., 210, 109634), an artificial intelligence-driven framework for discovering hybrid models composed of interpretable, mechanistic-like equations derived from existing first principles models and data. The framework is adapted specifically to the population balance modeling approach used to represent the crystallization process. The system investigated is based on real laboratory experimental data, precisely the batch crystallization of resveratrol in ethanol under two distinct cooling profiles: fast and slow. Each case includes online experimental concentration measurements. Under both scenarios, the current mechanistic model exhibits predictive discrepancies and a significant process-model mismatch. HyMech is therefore applied to address the limitations of the available model, to identify possible terms enhancing predictive accuracy while maintaining physical consistency. Results demonstrate the potential of the novel framework to successfully bridge data with first-principles knowledge. By employing a ‘‘human in the loop’’ approach, the methodology significantly improves the predictive performance of available mechanistic models, and paves the way towards more sophisticated approaches for model development and usage in crystallization applications for the pharmaceutical industry.
Crystallization is a fundamental unit operation in the pharmaceutical industry, particularly for the production of active pharmaceutical ingredients (APIs). Mathematical models could play a critical role to accelerate crystallization scale-up and optimization. However, their development presents significant challenges due to the complex and highly non-linear phenomena governing the process. Traditional first-principles models often struggle to capture the full scope of this phenomenological complexity, while purely data-driven approaches suffer from a lack of interpretability and limited generalization capabilities. To overcome these limitations, this thesis employs the recently developed HyMech software (Rossi et al., 2026, Comp. Chem. Eng., 210, 109634), an artificial intelligence-driven framework for discovering hybrid models composed of interpretable, mechanistic-like equations derived from existing first principles models and data. The framework is adapted specifically to the population balance modeling approach used to represent the crystallization process. The system investigated is based on real laboratory experimental data, precisely the batch crystallization of resveratrol in ethanol under two distinct cooling profiles: fast and slow. Each case includes online experimental concentration measurements. Under both scenarios, the current mechanistic model exhibits predictive discrepancies and a significant process-model mismatch. HyMech is therefore applied to address the limitations of the available model, to identify possible terms enhancing predictive accuracy while maintaining physical consistency. Results demonstrate the potential of the novel framework to successfully bridge data with first-principles knowledge. By employing a ‘‘human in the loop’’ approach, the methodology significantly improves the predictive performance of available mechanistic models, and paves the way towards more sophisticated approaches for model development and usage in crystallization applications for the pharmaceutical industry.
AI-supported hybrid modeling of crystallization processes in the pharmaceutical industry
SIMONCELLO, ALEX
2025/2026
Abstract
Crystallization is a fundamental unit operation in the pharmaceutical industry, particularly for the production of active pharmaceutical ingredients (APIs). Mathematical models could play a critical role to accelerate crystallization scale-up and optimization. However, their development presents significant challenges due to the complex and highly non-linear phenomena governing the process. Traditional first-principles models often struggle to capture the full scope of this phenomenological complexity, while purely data-driven approaches suffer from a lack of interpretability and limited generalization capabilities. To overcome these limitations, this thesis employs the recently developed HyMech software (Rossi et al., 2026, Comp. Chem. Eng., 210, 109634), an artificial intelligence-driven framework for discovering hybrid models composed of interpretable, mechanistic-like equations derived from existing first principles models and data. The framework is adapted specifically to the population balance modeling approach used to represent the crystallization process. The system investigated is based on real laboratory experimental data, precisely the batch crystallization of resveratrol in ethanol under two distinct cooling profiles: fast and slow. Each case includes online experimental concentration measurements. Under both scenarios, the current mechanistic model exhibits predictive discrepancies and a significant process-model mismatch. HyMech is therefore applied to address the limitations of the available model, to identify possible terms enhancing predictive accuracy while maintaining physical consistency. Results demonstrate the potential of the novel framework to successfully bridge data with first-principles knowledge. By employing a ‘‘human in the loop’’ approach, the methodology significantly improves the predictive performance of available mechanistic models, and paves the way towards more sophisticated approaches for model development and usage in crystallization applications for the pharmaceutical industry.| File | Dimensione | Formato | |
|---|---|---|---|
|
Simoncello_Alex.pdf
embargo fino al 09/10/2027
Dimensione
7.95 MB
Formato
Adobe PDF
|
7.95 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/116667