AI-Driven Predictive Maintenance for Stationary Hydrogen Storage and Distribution Infrastructure: A Systematic Review of SCADA/PLC Integration and Implications for U.S. Pipeline Safety Policy

Abstract

The increasing deployment of hydrogen as a clean energy carrier has intensified the need for reliable, safe, and cost-effective maintenance strategies for stationary hydrogen storage and distribution infrastructure. This study systematically reviewed the application of artificial intelligence (AI)-driven predictive maintenance technologies in hydrogen pipelines, storage vessels, compressors, valves, sensors, and integrated SCADA/PLC environments, while also examining implications for infrastructure reliability and operational safety. A quantitative systematic review methodology was employed to synthesize evidence from 126 eligible studies published between 2014 and 2025. Data were extracted and analyzed using descriptive statistics, comparative performance analysis, correlation analysis, regression modeling, independent-sample t-tests, and one-way ANOVA. The review evaluated machine learning, deep learning, anomaly detection, and hybrid predictive maintenance frameworks using performance indicators including failure detection accuracy, remaining useful life estimation, downtime reduction, maintenance cost optimization, asset availability improvement, and anomaly detection reliability. The findings demonstrated that AI-driven predictive maintenance significantly improved infrastructure performance across all hydrogen asset categories. Compressor systems achieved the highest failure detection accuracy (94.6%), followed by pipelines (93.2%), while predictive maintenance reduced operational downtime by up to 34.7% and maintenance costs by as much as 29.8%. Hybrid analytical frameworks produced the strongest predictive performance, achieving a mean accuracy of 96.2%, followed by transformer-based models (95.4%) and long short-term memory networks (94.8%). Regression analysis indicated that SCADA- and PLC-derived variables explained approximately 73% of the variance in predictive maintenance outcomes (R² = 0.73, p < 0.001), with vibration monitoring quality (β = 0.43), pressure monitoring consistency (β = 0.38), and compressor current reliability (β = 0.35) emerging as the most influential predictors. Studies utilizing high-quality operational datasets achieved anomaly detection accuracies of 94.8%, compared with 86.9% for lower-quality datasets.