Artificial intelligence-assisted design and CFD re-optimization of graded hybrid metal foams for faster hydrogen desorption in LaNi5 metal-hydride reactors

Abstract

Heat delivery limits the discharge rate of LaNi5 metal-hydride reactors. This study develops PG-SADE-GHF to co-design radial foam porosity, CuCr1Zr/AlSi10Mg allocation, and transient heat-transfer/pressure scheduling. The original MATLAB surrogate was used only for inexpensive screening. Its selected coordinates were then evaluated directly in OpenFOAM, and the same eight design variables were re-optimized within the original bounds, followed by an expanded-bound sensitivity study. Surrogate accuracy was assessed using 32 OpenFOAM development designs and 12 additional Sobol/maximin designs whose predictions were frozen before their CFD outcomes were computed. A CFD-corrected multifidelity model was then evaluated, and the optimized designs were verified with full-360-degree OpenFOAM calculations on a 196,608-cell mesh. The screening surrogate predicted td = 7.95 min, whereas direct OpenFOAM evaluation of the same coordinates gave 12.192 min. Re-optimization within the original bounds reduced td to 7.817 min, 21.83% faster than the 10-min CuCr1Zr-89% reference, while the expanded-bound sensitivity reached 6.442 min. On the locked held-out challenge, the original surrogate yielded RMSE = 3.354 min and MAE = 3.143 min; the CFD-corrected model achieved R2 = 0.982, RMSE = 0.159 min, MAE = 0.121 min, Spearman rho = 0.986, and 100% coverage of the nominal 95% prediction intervals. Full-360-degreeverification gave 7.8083 min for the original-bound design and 6.4417 min for the expanded design. Reliable acceleration of LaNi5 desorption requires simultaneous structure-material-operation co-design together with explicit high-fidelity discrepancy correction. Final optimum values are therefore based on direct OpenFOAM calculations rather than the screening surrogate. Experimental validation remains required.