Electrode informatics accelerated the optimization of key catalyst layer parameters in direct methanol fuel cells

Area of Science:

• Electrochemistry
• Materials Science
• Computational Modeling

Background:

• The catalyst layer is crucial for direct methanol fuel cells (DMFCs), enabling species, proton, and electron transport.
• Optimizing DMFC catalyst layer performance is complex, costly, and time-consuming due to extensive experimental requirements.

Purpose of the Study:

• To accelerate power density prediction in DMFCs.
• To evaluate the influence of catalyst layer parameters on maximum power density.
• To reduce the experimental cost and time for DMFC optimization.

Main Methods:

• Developed a finite element simulation model for DMFCs.
• Created a database of over 200 sets of 19 eigenvalues from simulation data.
• Employed machine learning models for training, prediction, and parameter importance ranking.
• Utilized sequential model-based algorithm configuration for high-throughput screening of 200,000 parameter combinations.

Main Results:

• Identified the importance of 19 characteristic parameters using tree-integration methods.
• Screened 200,000 parameter combinations, selecting the top 10 based on expected improvement.
• Numerical simulations using top parameter combinations yielded polarization curves exceeding the original database's maximum power density.

Conclusions:

• The combined finite element simulation and machine learning approach effectively accelerates DMFC power density prediction and parameter optimization.
• This methodology significantly reduces the experimental burden for optimizing DMFC catalyst layers.
• The findings demonstrate a faster pathway to achieving higher DMFC performance.