Generative AI Driven Process Calculations for Fuel Cells and Flow Batteries.

Area of Science:

• Computational science and engineering
• Electrochemical energy systems
• Artificial intelligence in scientific modeling

Background:

• Electrochemical systems like PEMFCs, SOFCs, and VRFBs involve complex, multi-scale transport-kinetics.
• Mechanistic solvers offer fidelity but are burdensome; data-driven models (ANNs, DRL) lack robustness.
• Existing methods struggle with the balance between physical accuracy and computational efficiency.

Purpose of the Study:

• To develop a Generative AI-assisted computational framework for electrochemical process calculations.
• To integrate Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG), physics-constrained prompting, and tool-integrated reasoning.
• To enhance the accuracy, robustness, and efficiency of modeling complex electrochemical energy systems.

Main Methods:

• Utilized a Generative AI framework orchestrated by LLMs.
• Employed Retrieval-Augmented Generation (RAG) for knowledge integration.
• Incorporated physics-constrained prompting and tool-integrated reasoning for calculations.
• Evaluated on synthetic and industrial Aspen Plus datasets for PEMFC and VRFB simulations.

Main Results:

• Achieved low RMSE for PEMFC polarization curve decomposition (9.6 mV synthetic, 7.8 mV Aspen data).
• Significantly reduced constraint violations in simulations (from 48%/42% to 1.2%/0.5%).
• Reached high energy efficiency for VRFB optimization (79.1% synthetic, 74.9% Aspen data) with reduced human effort (85%).

Conclusions:

• The Generative AI framework demonstrates robustness across diverse datasets and improves accuracy in electrochemical modeling.
• The approach offers a significant reduction in computational burden and human effort compared to traditional methods.
• This framework shows promise for integration with digital twins, fault detection, and advanced process control in electrochemical systems.