AI optimization framework using digital layouts of array structures: A case study for fuel cells

[Display omitted] •A new method is proposed to engage AI in the layout optimization of energy devices.•Array layout is expressed in digital codes to enable AI in layout optimization.•The approach is validated by the layouts of baffles and modular channels in PEMFC.•Conditional Generative Adversarial Networks is used to expand the training dataset.•The results of numerical simulation have been verified by experiments. It is challenging for common artificial intelligence (AI) frameworks to optimize layouts of array structures, because AI is unable to recognize and adjust layouts. To address this challenge, a novel AI optimization framework has been proposed, where layouts of array structures are represented in digital codes to enable AI in layout optimizations. The optimizations of baffle layout and modular channel layout in proton exchange membrane fuel cells are used to validate the methodology. For the multi-objective optimization of baffle layout, although the channel with optimal baffle layout (OC) exhibits a little higher pressure drop than the conventional channel, it increases the net power density and oxygen uniformity by 2.4% and 1.5%, respectively. Consistently, for the multi-objective optimization of channel layout, the optimal flow field with an optimal channel layout (OFF-OC) shows a small increase of pressure drop that can be compensated by the much-enhanced output. So, the net power density of OFF-OC has been elevated by 7.0% in comparison with that of the conventional parallel flow field. Meanwhile, the oxygen uniformity of OFF-OC has also been improved by 4.0%. The simulation results have been verified experimentally. These examples have demonstrated the effectiveness of the AI optimization framework using digital layouts.