Ai-enhanced digital twins for food process optimization

Overview

The proposed solution involves the creation of AI-enhanced digital twins for key steps in food manufacturing processes. By integrating physics-based models with machine learning (ML) techniques, this approach enables high-fidelity simulations that can predict key quality attributes, facilitate real-time process adjustments, and conduct 'what-if' scenario testing. This innovation is designed to optimize food process design and operations, providing manufacturers with the tools for improved quality control and efficiency.

Technical specifications

The digital twins are developed using established platforms like gPROMS FormulatedProducts, Dyssol, and PharmaPy to generate detailed mechanistic simulation data. This data is then used to train ML models, including physics-informed neural networks, to create fast and accurate surrogate models. These models serve as digital twins for individual unit operations, enabling soft sensing and real-time process optimization. Key features include:

• High-throughput mechanistic simulation
• Fast surrogate ML models for real-time prediction
• Integration of physics-based and data-driven approaches for enhanced accuracy

Technology readiness level

The solution is currently at Technology Readiness Level 5, indicating that it has been validated in a relevant environment. Future validation plans include the prioritization of high-impact unit operations, hybrid digital twin development, and comprehensive 'what-if' analysis to further refine and optimize the models.