Hybrid modeling for industrial fermentation processes with an "Intra-Batch Experimental Design"

Area of Science:

• Biotechnology
• Chemical Engineering
• Process Systems Engineering

Background:

• Accurate modeling of oxygen transfer rate is crucial for fungal fermentation optimization.
• Traditional viscosity models are complex, requiring extensive offline measurements.
• Digital twins/shadows for fermentation require high-fidelity process models.

Purpose of the Study:

• To develop a novel, cost-effective framework for modeling fungal fermentation viscosity.
• To reduce reliance on tedious offline rheological measurements.
• To integrate online sensor data for improved predictive modeling.

Main Methods:

• Developed a model for oxygen mass transfer coefficient (kLa) using reduced fermentation runs.
• Evaluated three machine learning algorithms as soft sensors for online viscosity prediction.
• Created a hybrid model combining mechanistic and data-driven approaches.

Main Results:

• Achieved R² of 0.92 with online data vs. 0.67 with offline data for kLa modeling.
• Reduced experimental effort for kLa model development from nine to two fermentations.
• Hybrid model accurately predicted fermentation dynamics across various conditions with modest accuracy improvements.

Conclusions:

• Online sensor integration significantly enhances mathematical model development.
• The proposed framework lowers costs and effort for model development.
• This work advances automated model development for industrial fermentation processes.