Intelligent Monitoring for Fault Diagnosis of Proton Exchange Membrane Water Electrolyzer

Durability remains a key barrier to the widescale deployment of proton exchange membrane water electrolysis (PEMWE) for green hydrogen production, with membrane degradation significantly reducing efficiency and system lifetime. Traditional diagnostic methods often overlook early-stage degradation, underscoring the need for advanced detection strategies. This work develops a machine-learning driven framework for PEMWE fault detection, combining validated physics-based modeling with classification algorithms. A MATLAB Simscape electrochemical model, validated against experimental I V data (RMSE = 0.030, MAE = 0.025, R2 = 0.78), was used to simulate degradation by stepwise reductions in proton conductivity (90 50%). The resulting datasets trained a multi-layer perceptron (MLP) classifier, which successfully distinguished six operating states. The artificial neural network (ANN) achieved strong accuracy for moderate-To-severe degradation, while delivering perfect recall for normal operation and perfect precision across all fault classes. These findings demonstrate the potential of AI-based approaches to enable early fault detection, predictive maintenance, and improved PEMWE reliability for sustainable hydrogen production.