Multi-fidelity bayesian optimization for premium ice cream formulation

Overview

This innovative solution revolves around a multi-fidelity Bayesian optimization (MFBO) framework designed to perfect ice cream formulation and processing. By integrating low-cost simulations like computational fluid dynamics (CFD) with high-fidelity data from production lines, this framework aims to optimize the process with fewer costly experimental trials. The approach not only models subprocesses individually but also combines them for comprehensive optimization objectives, ultimately enhancing the sensory quality of premium ice cream products.

Technical specifications

Key features:

• Multi-fidelity Bayesian Optimization (MFBO): Utilizes both low and high-fidelity data to refine and optimize ice cream production processes.
• Subprocess Modeling: Decomposes the production process into individual subprocesses for targeted optimization.
• Soft Sensors: Predicts sensory attributes by translating process and formulation data, enhanced by novel sensing technologies.
• Integration of Novel Sensing Technologies: Where feasible, incorporates in-line or at-line sensing to improve real-time predictions.

This framework combines mechanistic models with experimental data, reducing the need for extensive plant tests and accelerating scenario evaluations. It ensures adaptability and accelerates product development while maintaining high sensory standards.

Technology readiness level

The technology is currently at TRL 5, indicating that the concept has been validated in a relevant environment. The ongoing development includes refining low-fidelity models using historical data, integrating subprocess models into a modular optimization framework, and developing soft sensors for robust sensory prediction. The future steps include enhancing real-time predictions through novel sensing technologies, setting the stage for practical deployment in industrial settings.