needs update from Glenn's thesis

Optimization of PEMFC Channel Configuration Using A Genetic Algorithm

Using hydrogen to power the next generation of portable electronics and vehicles is a task that requires contributions from many areas of science. Problems must be solved in areas ranging from hydrogen production, distribution, and storage to the production of materials and devices that can use hydrogen efficiently. One way to make hydrogen more effective as an energy carrier is to create devices that can use it more efficiently. For that reason, this research will focus on optimizing the internal geometry of a polymer electrolyte membrane (PEM) fuel cell (FC) to maximize power density.

A hydrogen fuel cell is divided into 7 layers which include bi-polar plates, gas diffusion layers, membrane, and catalyst layers. This research focuses on optimizing the parameters that describe the serpentine channel configuration for the highest power output. The important parameters are channel width, height, and length, rib width, and gas diffusion layer thickness. Varying the dimensions for these parameters creates different pressure distributions within the cell, which results in varied power outputs from the cell. Figure1 shows an example of a fuel cell model with the serpenine channel configuration.

Single Serpentine FC Model

Each simulation is constructed in GAMBIT© as a solid model and is divided into an appropriate mesh. The mesh is imported into FLUENT© with the PEMFC Add-on Module, boundary conditions and material properties are specified, and the problem is initialized. This module is capable of modeling the fluid dynamics, heat transfer, and electrochemical reactions that occur in a fuel cell. Each simulation results in a unique output that is used to evaluate cell performance.

A genetic algorithm is a practical and effective way to search a relatively large search space for an optimum solution. This project uses the Genetic Algorithm (GA) Toolbox available as a MatLab© add-on module to control the optimization process. The module is a population and mutation based optimization process that is capable of looking for a maximum fitness value. For this research, an optimum solution will be a maximum current density at a given operating current density. A script has been written that calls the GA, which returns an initial set of individuals. Each individual is a set of numbers that describe a unique channel width, height, length, rib width, and GDL thickness. The script then calls GAMBIT© to create the model and mesh for an individual, sends the mesh to FLUENT© to run the fuel cell simulation, and receives an output of current density of the cell. A unique simulation is completed for each of the individuals, and the first set of individuals is known as the first generation. The first generation contains a set of randomly created fuel cells and their respective fitness values. The parameters of individuals that perform well are crossed to try to create better-performing individuals. The new set of individuals is called the first generation and each individual is evaluated for a new fitness value. This process is repeated until the population cannot create better performing individuals.

Flow chart of the cycle of the genetic algorithm.

This process produces the optimum set of parameters that define the serpentine channel fuel cell at a given set of operating conditions.

Absolute pressure in the cathode GDL mid-plane for the best (left) and worst (right) performing individuals found.