A deep-learning-boosted surrogate model of a metal foam based protonic ceramic electrolysis cell stack for uncertainty quantification

Abstract

Protonic ceramic electrolysis cells (PCECs) are widely considered as an emerging technology that shows capability for enabling environmentally friendly large-scale production of hydrogen. PCEC stacks’ performance is considerably lower compared to smaller, lab-scale button cells. This performance gap is largely attributed to the highly uneven distribution of reactants throughout the larger cell or stack, which presents a significant challenge to the development of PCEC stack. The issue of non-uniform reactant distribution in PCEC stacks can be effectively addressed by incorporating metal foam structures to replace the conventional rib design of the interconnect components. A 3D multiphysical model is developed to train a deep-learning-boosted surrogate model for PCEC with metal foam, which is subsequently used for parameter screening and sensitivity analysis. A collection of 12 inputs that encompass operating/structural/microstructural/material parameters are chosen, in which the operating parameters are identified as influential parameters. Three distribution uniformity indices maintain strong consistence to variations in the inputs, implying the positive role of metal foam. Faradaic efficiency (FE) is primarily influenced by the operating voltage with a total Sobol index of 0.6. The current density and FE are determined as the most stable performance indicators, exhibiting standard deviations of 0.15 and 0.81 respectively. According to a safety threshold of 10 K cm−1 for temperature gradient, the probability of the stack operating below this threshold is 88.61 %. Valuable insights into the design and performance of PCEC stacks that incorporate metal foam structures are provided. The workflow developed in this work can be further leveraged to explore and optimize PCECs.