Genetic algorithm prospects, to guide the manufacturing of bottom-up designed electrodes

In our developed genetic algorithm (GA), optimized porous electrodes are computed from the bottom-up by natural selection of the fittest structures driven by the theory of evolution. A GA has a high degree of evolutionary freedom and can easily be adapted and expanded depending on the operating conditions (potential, flow rate), network properties (artificial, extracted), system configuration (flow field, electrolyte), and desired optimization functions (mutation, migration). The framework was built to speed up and guide the design of electrodes for electrochemical applications, with a focus on redox flow batteries. As a greater scientific impact can be made by sharing research, the modeling framework with all added extensions is made available on GitHub as the presented GA is a proof-of-concept and should be further developed in future work. Possible extensions or improvements of the modeling framework include:

(1) The selection of an appropriate fitness function and geometrical definitions. In the presented model, both the pumping power and electrical power are optimized to an equal degree. The current fitness function must be analyzed in depth and a suitable definition should be used depending on the desired electrode optimization. Moreover, the geometrical definitions of OpenPNM are currently employed, which significantly oversimplify and underestimate the electrode performance, affecting the optimization of the electrodes by e.g., an inaccurate internal surface area definition. Moreover, with the manufacturing technique of choice in mind, other geometrical shapes beyond spheres and cylinders could be incorporated to better represent the to-be-manufactured optimized electrode structure. For example, Misaghian et al. used pyramids to represent the pore space, which they claimed is a better representation of fibrous materials with binder [1].

(2) The implementation of manufacturing constraints. During the structure evolution, the electrodes could be screened for dead-end pores, interconnection of the solid phase, and a minimum size/length of the void- and solid phases. Moreover, additional manufacturing constraints, concerning the manufacturing technique of choice, could be incorporated. For example, when using 3D printing as a manufacturing technique, the printing resolution and layer thickness could be implemented in the GA, together with a shrinkage factor when a carbonization step is required. Moreover, the void space should be converted into a solid structure as is generally necessary for electrode fabrication. One possible method could be the manufacturing of the inverted void space, as was presented in Section A4.7.2, which could be used in combination with 3D printing. Another option would be to develop an “inverse” SNOW algorithm [2] that can translate the void space back into a fibrous material.

(3) The extension of the evolutionary freedom of the GA. Examples include the addition of pore migration and the coupled optimization of the electrode structure with the flow field design. Furthermore, the optimization could be performed for full-cell designs for which both the cathode and anode can be independently optimized, considering e.g., species crossover. In addition, the full parallelization of the GA should be investigated. In the current GA version, only the performance evaluation of the structures is parallelized to the number of individuals. The mutation, crossover, volume scaling, and merging and splitting steps could also be parallelized to further reduce the computational time and enable the optimization of larger electrode structures.

(4) The adaptation of the GA approach for the optimization of other electrochemical technologies such as hydrogen fuel cells, water electrolyzers, and CO2 electrolyzers. Moreover, the PNM implemented in the GA could be replaced by another pore-scale model, and/or can be coupled to additional modeling frameworks to increase the accuracy of the optimization and/or decrease the computational costs.

References:
[1] N. Misaghian, M. A. Sadeghi, K. Lee, E. Roberts, J. Gostick, J. Electrochem. Soc. 170, 070520 (2023).
[2] J. T. Gostick, Phys. Rev. E. 96, 1–15 (2017).