Multi-Fidelity Bayesian Optimization in Chemical Engineering

Abstract

This presentation introduces two chemical engineering applications that utilize Bayesian optimization, showcasing their potential and benefits. The first application addresses the challenge of plant-model mismatch in uncertain processes through real-time optimization. By integrating concepts from Bayesian optimization and derivative-free optimization, this approach combines a physical model with trust-region ideas to minimize risk during exploration while employing Bayesian optimization techniques. The second application focuses on multi-fidelity data-driven reactor design. A framework is presented that leverages different computational fluid dynamic simulation fidelities. Gaussian processes are utilized to adaptively learn a multi-fidelity model of reactor simulations across various continuous mesh fidelities. The search space of reactor geometries is explored using a combination of different fidelity simulations, selected for evaluation based on a weighted acquisition function that balances information gain with the cost of simulation. To validate the results, the optimal reactor geometry is 3D printed and experimentally tested, confirming its improved mixing performance. Overall, these applications demonstrate the potential of Bayesian optimization in chemical engineering, offering solutions to plant-model mismatch and enhancing reactor design through the integration of multi-fidelity simulations.

Notes

• References:

  • Multi-Fidelity Data-Driven Design and Analysis of Reactor and Tube Simulations: https://arxiv.org/abs/2305.00710

  • Safe Real-Time Optimization using Multi-Fidelity Gaussian Processes: https://ieeexplore.ieee.org/abstract/document/9683599

  • Real-time optimization meets Bayesian optimization and derivative-free optimization: A tale of modifier adaptation: https://www.sciencedirect.com/science/article/abs/pii/S0098135421000272

  • Modifier-Adaptation Schemes Employing Gaussian Processes and Trust Regions for Real-Time Optimization: https://www.sciencedirect.com/science/article/pii/S2405896319301211

• Personal website can be found here .