Rapid package design using ai-driven bayesian optimisation

Overview

The proposed solution leverages an AI-driven Bayesian Optimization (BO) framework with Gaussian Process (GP) models to revolutionize package design. By integrating machine learning algorithms and human expertise, it aims to optimize preform designs and their outcomes through a systematic exploration and exploitation of the design space. This approach facilitates continuous improvement by learning from experimental data and recommending optimized designs for trial, significantly enhancing efficiency and effectiveness in package design.

Technical specifications

• Bayesian Optimization Framework: Utilizes Gaussian Process models to explore and exploit the design space of preforms.
• Integration of Human Expertise: Human-in-the-loop (HIL) designers contribute preferences, enhancing model accuracy and relevance.
• Geometric Characterization: Novel techniques for numerically defining preform designs, crucial for understanding design impact on outcomes.
• Multi-Objective Optimization: Balances multiple design objectives to achieve optimal container outcomes.
• Feedback and Iteration: Incorporates client feedback and experimental data to refine models continually.
• Prototype Development: Combines GP models, BO algorithms, and a HIL interface for practical testing and validation.

Technology readiness level

Currently at TRL 3, the solution is in the proof of concept stage. Initial development includes exploring client data and building a solution representation compatible with GP and BO algorithms. A prototype combining these models with a HIL interface is under development for testing and feedback integration.