Session: 04-11-01 Posters

Paper Number: 195150

195150 - Physics‑informed Machine Learning to Determine Coaxial Extrusion Parameters for Precise Capsule Geometry in Personalized Medicine 

Abstract: 

Personalized medicine is a new patient care paradigm that calls for treating patients by delivering the right drugs at the appropriate dose to the right location at the right time, moving away from a ‘one-size-fits-all’ approach. It requires precise control over the dosage form's geometry and composition. Capsules as an oral dosage form are among the most common ways of drug delivery. 3D printing technologies have been used to fabricate personalized capsules in research settings. However, the throughput of current 3D printing processes is too low, making the technology not scalable for mass deployment. We propose developing a dynamic gas-assisted coextrusion process as a scalable freeform manufacturing platform to fabricate custom core-shell drugs for personalized medicine. By adjusting the flow-rate ratio, the core diameter and shell thickness can be varied, enabling the creation of capsules with custom dosages on demand within seconds. However, the nonlinear coupling between extrusion flow rates, viscosities, and interfacial forces makes it hard to produce capsules with the required diameter and wall thickness. Without a systematic way to navigate these coupled parameters, achieving consistent and reproducible geometry remains a critical unsolved challenge in personalized manufacturing. This study presents a physics-informed machine learning framework that links extrusion parameters to capsule geometry and enables the determination of the process settings for target capsule designs. The hydrogel capsules, both core and shell, are made of PF-127. The dataset is collected by measuring capsule diameter and wall thickness with different core and shell speeds. The controlling flow behavior within the coaxial extrusion device is described using dimensionless parameters, including the Reynolds number, capillary number, and flow-rate ratio. By integrating experimental data with physics-based feature construction, the approach captures the relationship between flow conditions and resulting capsule geometry. The results show that flow-structure interactions significantly influence capsule formation, enabling the predictive selection of extrusion parameters for specific designs. The model is validated by producing capsules with prescribed geometries using predicted operating conditions. The framework demonstrates how physics-guided learning can transform capsule fabrication into a predictive, controllable process, establishing a foundation for future studies with drug-loaded systems operating under long-term physiological conditions. 

Presenting Author: Chong Ma Washington State University

Presenting Author Biography: B.S. in Mechanical Engineering, Washington State University, 2024
Second-year Ph.D. student in Mechanical Engineering at Washington State University
Research interests in biomedical manufacturing and additive manufacturing

Authors:

Chong Ma Washington State University
Kuen-Ren Chen Washington State University
Andrew Lefors Washington State University