Machine Learning-aided Multi-objective Optimisation of Tesla Valve-based Membraneless Electrolyzer Efficiency

Hydrogen (H2) is an attractive fuel due to its high specific energy and zero direct carbon emissions. Membraneless electrolyzers (MEs) offer a lower-cost route to hydrogen production, but their operation is complex and current efficiencies are modest. Although multi-objective optimization is widely used, its heavy compute demands and weak integration with modern learning methods limit scalability and adaptability. We introduce a practical, ML-guided way to design Tesla-valve (TV) membraneless electrolyzers by building diodicity (Di) directly into the geometry search. Using multilayer-perceptron surrogates trained on 150 high-fidelity simulations (R2 > 0.95), we link four design knobs (We, Wc, Wd, Di) to pressure drop (Δp) and ohmic loss. A Genetic Algorithm (GA)-based multi-objective search over realistic ranges delivers 60 Pareto-optimal designs that make the Δp–ohmic trade-off explicit; TOPSIS then selects a balanced geometry (We = 1.708 mm, Wc = 0.200 mm, Wd = 1.012 mm, Di = 1.618) with ohmic loss 4.069 V and Δp 6.169 Pa. The approach delivers faster, lower-cost design maps and is supported by experimental checks, pointing to an actionable route for scalable, interpretable optimization of sustainable hydrogen production.