Autoencoder-based inverse design and surrogate-based optimization of an integrated wet granulation manufacturing...

Area of Science:

• Pharmaceutical Manufacturing
• Chemical Engineering
• Process Systems Engineering

Background:

• Model-based design and optimization accelerate pharmaceutical process development.
• Wet granulation is a key unit operation in solid dosage form manufacturing.
• Integrating advanced modeling techniques can improve process efficiency and product quality.

Purpose of the Study:

• To explore Machine Learning (ML) as a surrogate model for optimizing a wet granulation flowsheet.
• To compare an autoencoder-based inverse design with surrogate-based forward optimization.
• To identify optimal granulation and milling parameters for maximizing dissolution time and product yield.

Main Methods:

• Developed a reduced representation of a wet granulation flowsheet model.
• Incorporated a novel dissolution model considering particle size, porosity, and microstructure.
• Implemented and compared autoencoder-based inverse design and surrogate-based forward optimization for bi-objective optimization.

Main Results:

• Both optimization approaches were effective, achieving results in under 4 seconds.
• The autoencoder approach provided significant dimensionality reduction, unlike surrogate-based optimization.
• Dimensionality reduction enhanced process understanding and visualization of the design space.

Conclusions:

• AI and ML, specifically autoencoder-based inverse design, offer powerful tools for pharmaceutical process development.
• This approach can enhance efficiency and product quality in complex manufacturing scenarios.
• The method facilitates improved process understanding and feasibility studies for complex designs.