Proton exchange membrane water electrolysis (PEMWE) is restricted to high-demand iridium and faster voltage degradation during intermittent renewable operation which inhibits green hydrogen production. To solve this, we came up with a physics-aware machine-learning framework of degradation-aware operation of low-iridium PEMWE systems and demonstrated their validity. We combined literature-derived and experimental data at 30–90 o C, 0.2 -3 A cm-1, several Ir-based catalysts as the anode (Ir-black, IrO2, IrRuOx), and porous structures of the transport layer. Training Support vector regression using genetic-algorithm-tuned, gradient-boosting, and constrained neural networks were fitted to predict cell voltage, hydrogen-production efficiency and rate of degradation simultaneously. The monotonicity constraints of electrochemical models enhanced physical consistency and the SHAP-based interpretation found the dominant interactions between parameters. This was then followed by a multi-objective optimization layer to produce operating policies to maximize the hydrogen production with minimal degradation and iridium-specific cost of energy. The framework obtained errors of less than 2 percent on cell voltage and less than 10 percent on degradation pattern at unknown dynamic profiles. Quantitative data on catalyst-loading and pore-size threshold were identified in the analysis to provide a distinction between short-term and long-term gains and penalties in performance. Lastly, we built a functional decision map in the variable renewable inputs to select the set-point practically to achieve better lifetimeperformance ratio. These findings suggest a deployable pathway of reducing the cost and extending the lifetime of PEMWE operation and support in scalable green hydrogen production in carbon-limiting energy systems.