Recent progress on machine learning techniques for flow field optimization toward performance and commercial scale proton exchange membrane fuel cell: A systematic review

• A systematic review of ML techniques for PEMFC flow field optimization is presented • Key challenges on flow field design optimization and performance trade-off are highlighted • Flow field optimization assisted by surrogate model and optimization algorithm is discussed • Directions toward performance and commercial scale PEMFC are identified Proton exchange membrane fuel cells (PEMFC) are promising technologies for sustainable energy conversion. One of the important factors governing its performance is the design of the flow field within bipolar plate, which plays a key role in ensuring efficient reactant distribution and water management in the membrane electrode assembly. This paper provides a systematic review of recent advancements in surrogate modelling and machine learning (ML)-based optimization techniques as applied to the design of flow fields in PEMFC. The review encompasses various flow field configurations, including both 2D and 3D designs that incorporate baffles, traps, and bio-inspired structures. The modelling and optimization strategies, either single- or multi-objective, are identified and examined, enabling clearer evaluation of each contribution and limitation. The predictive accuracies are reported. The principal factors affecting the performance of surrogate models are nonlinearity degree in the input–output relationships, dataset size, number and quality of input parameters, model architecture, training strategies, and hyperparameter tuning. While neural networks (NN), response surface methodology (RSM), support vector machines (SVM), Kriging models, and deep neural networks (DNN) are commonly utilized surrogate models, the NNs outperform other ML models. The other models, however, show more flexibility with limited datasets. In addition, genetic algorithm (GA) and non-dominated sorting genetic algorithm II (NSGA-II) are respectively the single- and multi-objective algorithms most frequently integrated with the surrogate models. The use of ML techniques leads to efficient approach greatly enhancing PEMFC flow field optimization and hold a promising solution toward performance and commercial scale PEMFC.