A coupled machine learning and genetic algorithm approach to the design of porous electrodes for redox flow batteries

Machine learning for flow batteries: opportunities and challenges

“…Wan et al combined a data generation method with ML algorithms to design porous electrodes with large specific surface area and high hydraulic permeability for FBs. 104 Stochastic reconstruction method, morphological algorithm and lattice Boltzmann method were adopted to construct the dataset, which contains 2275 fibrous structures (shown in Fig. 9 ).…”

“…Apart from predicting the properties of ORASs, ML can also be utilized in electrode design, 38,104 membrane design 105 and system optimization for FBs. 106 As the place in which electrochemical reaction occurs, the pore structure and specic surface area of the electrode will affect the efficiencies of the FB.…”

Machine learning-assisted design of flow fields for redox flow batteries

“…Though powerful and robust, machine learning-assisted screening remains restricted in terms of the inverse design with the goal of optimizing properties or performance objectives. 54 To overcome this difficulty, some studies have combined machine learning with optimization algorithms (e.g., genetic algorithm) to tackle material design problems related to polymers, 55 battery electrodes, 56 and optical glass. 57 Such methods, however, are infeasible when the solution set of the problem to be addressed is not continuous.…”

Starting from the bottom: Coupling a genetic algorithm and pore network model for porous electrode design

“…The integration of optimization tools within electrochemical numerical frameworks has been recently deployed to support the numerical identification of fitting parameters and find optimal operational conditions (33,(37)(38)(39), as well as to support the design of cell components (40)(41)(42)(43). Recently, Choi et al utilized a GA in combination with a two-dimensional model of an all-vanadium flow battery to identify the fluid parameters using a fitness function involving the mean square error of the voltage between the experimental and simulated data.…”

“…developed a coupled machine learning and GA data-driven approach to design porous electrodes for RFBs, which resulted in electrodes with larger specific surface area and high hydraulic permeability, but their design space was limited to fibrous structures (40). Inspired by these previous efforts, we aim to develop a computational framework that affords large versatility of design, robust physical representativeness, and maintains a low computational cost.…”