An integrated XAI-enhanced framework for multi-scale optimization of proton exchange membrane systems

Proton Exchange Membrane (PEM) technologies face optimization challenges due to complex multiphysics interactions and prolonged development cycles. This study presents an integrated framework combining high-fidelity multiphysics modeling, explainable AI (XAI), multi-objective optimization, and multi-criteria decision making (MCDM) for accelerated PEM system design. Validated 3D two-phase computational fluid dynamics models generate comprehensive datasets for eleven design parameters. These train interpretable surrogate models with exceptional accuracy (R² = 0.9986) and > 1000 × computational speedup. XAI techniques provide physically interpretable insights into non-linear parameter interactions. The surrogate models enable efficient multi-objective optimization, generating Pareto-optimal designs. For PEM fuel cells, optimized membrane electrode assemblies achieve 28% higher power density with 40% reduced platinum loading. For PEM electrolyzers, graded-porosity transport layers improve high-current-density performance by 15% while reducing auxiliary power. An MCDM process integrates technical, economic, and durability criteria for design selection. The framework serves as an intermediate layer in a hierarchical AI-driven ecosystem, connecting generative material discovery with system-level renewable energy integration. This approach enables rapid exploration of design spaces while maintaining engineering insights, accelerating PEM technology deployment for sustainable energy systems.