Recent Advances in Food Drying Modeling: Empirical to Multiscale Physics-Informed Neural Networks.

Area of Science:

• Food Science
• Chemical Engineering
• Computational Modeling

Background:

• Food insecurity is a global issue, with food preservation techniques like drying crucial for enhancing food security and reducing waste.
• Drying removes significant water content from fruits and vegetables, but induces complex structural changes affecting stability and quality.
• Accurate modeling of these drying-induced changes is essential for optimizing the process.

Purpose of the Study:

• To conduct a comprehensive literature review of conventional drying modeling techniques.
• To explore the potential of Physics-Informed Neural Network (PINN) models for food drying.
• To identify strategies for overcoming limitations in current drying modeling approaches.

Main Methods:

• Literature review of empirical, physics-based computational, and data-driven machine learning models for food drying.
• Analysis of the strengths and weaknesses of each modeling approach.
• Exploration of the hybrid PINN approach, integrating physical principles with machine learning.

Main Results:

• Empirical models offer simplicity but lack generalizability and physical insight.
• Physics-based models provide high resolution but are computationally intensive.
• Purely data-driven models are less computationally demanding but struggle with sparse data.
• PINN models present a promising hybrid approach, merging physical laws with data-driven learning.

Conclusions:

• Conventional drying modeling techniques have inherent limitations in accuracy, generalizability, or computational cost.
• PINN models offer a significant advancement potential for food drying simulation and optimization.
• Further research into PINN applications can lead to improved food preservation strategies and reduced waste.