Multi-scale investigation of SSBA-15/sulfonated polysulfone composites for fuel cells: Integrating machine learning predictions and quantum chemical validation

Developing high-performance proton exchange membranes (PEMs) requires a clear understanding of how macroscopic physical characteristics relate to microscopic interfacial interactions. In this study, a new multi-scale simulation method was established to predict and analyze the performance of sulfonated polysulfone membranes modified with sulfonated mesoporous silica (SSBA-15). Using Orange Data Mining, a rapid data-driven materials-screening surrogate was developed to accelerate alternative membrane candidate selection prior to device-scale multi-physics revision optimization. The machine learning (ML) work flow employed descriptors from experimental measurements to successfully predict fuel cell power. In the top-performing architectures (Gradient Boosting and Random Forest), a correlation coefficient of (r = 0.99) in a leave-one-out cross-validation (LOOCV) parity plot, accompanied by a high R 2 = 0.950, confirms high-quality predictions. Global sensitivity analysis (RReliefF) identified proton conductivity as the most critical material descriptor, followed by water uptake and ion-exchange capacity (IEC). Complementary quantum chemical validation at the PM6 semi-empirical level (Gaussian 09) revealed strong interfacial binding (∆G = - 46.71 kcal mol −1 ) and an increased dipole moment (6.48 D), proving a highly polarized, thermodynamically stabilized interface. This atomistic route uncovers a specialized interaction channel where unique interfacial binding drives optimal membrane morphology, effectively bridging molecular-level interactions to device performance via an efficient pre-screening pathway.