Hydrogen fueling stations are critical infrastructures for enabling large-scale hydrogen mobility. As station capabilities increase, particularly with cryogenic liquid hydrogen (LH₂) storage, reliable operation of LH₂ pumps becomes essential for safety, reliability, and overall station efficiency. The high-dimensional, time-dependent, and nonlinear nature of pump sensor data, combined with the need for explainable and actionable predictions, makes traditional fault detection approaches insufficient for timely and reliable maintenance decisions. Building on these challenges, this paper proposes an explainable AI–based decision support framework for reliability assessment and condition-based maintenance of LH₂ pumps. The framework processes multivariate time-series data using physics-informed feature extraction and degradation-sensitive indicators, feeding predictive fault classification models based on ensemble learning. SHapley Additive exPlanations (SHAP) is integrated for global and local feature attribution, linking explainable predictions directly to maintenance decisions. Feature attribution analysis reveals physically consistent degradation signatures aligned with known failure modes, supporting transparent, evidence-based decision-making under uncertainty. Overall, the framework advances trustworthy AI for safety-critical cryogenic systems and provides a scalable methodology for actionable predictive maintenance decisions in hydrogen infrastructure.

Keywords: Liquid hydrogen pumps, prognostics and health management, Explainable AI, fault detection and diagnosis.