AI-aided optimization of droplet layout in distribution zone for commercial bipolar plates of PEMFC

The distribution zone of flow fields directly impacts pressure drop and oxygen uniformity of proton exchange membrane fuel cells (PEMFCs). Herein, a layout of droplet units is introduced into the distribution zone to reduce pressure drop and improve oxygen uniformity. Then a multi-objective optimization (MOO) framework combining surrogate models of artificial neural network and non-dominated sorting genetic algorithm is established to tackle this complex problem of layout optimization. Notably, the layouts have to be represented in digital codes for MOO to recognize and manipulate. So, a previously proposed ‘digital bridge’ strategy is used with an added trick of ‘translocation’, which moves the binary codes in some rows to other rows. The top three layouts from the Pareto solutions of MOO are compared with the base model (D180°-S0°) by simulation and experiments. It is found that the orientation of droplet units significantly influences air flow resistance and D180° layout with the spherical head pointing right at the inlet achieves both lower flow resistance and higher electrochemical reaction efficiency. Interestingly, artificial intelligence (AI) is capable of performing layout optimization based on the ‘translocated’ codes, although AI might not know the patterns before translocation. Both the predicted values and trends from the surrogate models align well with simulation and experimental results, demonstrating the reliability of the AI-aided approach. More importantly, AI can generate optimal layouts that surpass human imagination. Indeed, the top-ranking layout achieves a 6.7% higher net power density than the base model (D180°-S0°).