Artificial-intelligence-guided design of ordered gas diffusion layers for high-performing fuel cells via Bayesian...

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Science

Background:

• Proton exchange membrane fuel cells (PEMFCs) require optimized gas diffusion layers (GDLs) for enhanced performance and cost reduction.
• Current GDL design relies on time-consuming trial-and-error methods, hindering rapid advancement.
• Developing advanced GDLs is crucial for improving energy conversion efficiency and fuel cell technology.

Purpose of the Study:

• To introduce a novel, closed-loop Bayesian machine learning workflow for the rational design of GDL structures.
• To accelerate the discovery of optimal GDL architectures that maximize PEMFC performance.
• To overcome the limitations of traditional, empirical GDL development approaches.

Main Methods:

• Utilized artificial neural networks to rapidly calculate anisotropic transport properties of reconstructed GDLs.
• Employed a Bayesian optimization algorithm to efficiently search for optimal GDL structures.
• Fabricated the computationally designed GDL structures using a controlled electrospinning technique.

Main Results:

• The Bayesian optimization identified optimal GDL structures in just 40 steps.
• The optimal GDL structure features highly oriented fibers with moderate diameters.
• Fabricated PEMFCs with the optimized GDL achieved a power density of 2.17 W cm⁻² and a limiting current density of ~7200 mA cm⁻².

Conclusions:

• The proposed machine learning approach significantly outperforms traditional methods in GDL design.
• The optimized GDL structure demonstrates superior performance, exceeding commercial GDLs by a substantial margin.
• This study validates the efficacy of integrating AI-driven design with advanced fabrication for next-generation fuel cells.