This study presents a machine learning model based on Support Vector Regression (SVR) to predict freeze-dryer chamber pressure and generate equipment capability curves for different machine configurations. The model is trained using Computational Fluid Dynamics (CFD) data that incorporate various geometric parameters of three freeze-dryer configurations: side condenser (type 1), bottom condenser (type 2), and bottom integral condenser (type 3). These configurations are characterized by variations in duct diameter, valve opening, and chamber volume. The results indicate a significant influence of duct diameter for type 1 and 2 geometry, and valve opening for type 3 geometry, on the discharge coefficient. Understanding of these nonlinear relationships between parameters can help guide future freeze-dryer design and optimization strategies, improving performance and efficiency. Overall, the SVR model demonstrates its potential as a powerful tool for accurately predicting chamber pressure and optimizing the geometries of freeze-dryer machines. These findings pave the way for exploring the model's application in broader optimization efforts across the entire freeze-drying process, from laboratory scale setups to large-scale production machines. By leveraging this model, future research and development could lead to more effective, cost efficient, and reliable freeze-drying systems.
The model's performance is compared to that of polynomial regression, Reynolds based model and a compact model from the literature, showing better accuracy in capability curve predictions, with errors below 10%.