Scholarlink

<p>Accurate prediction of voltage and health index for Proton-Exchange Membrane (PEM) fuel cells is essential for maintaining reliable operation under varying load and environmental conditions. This work illustrates a thermo-fluid informed feature framework that captures the dynamic interaction between temperature, pressure, humidity and flow variables for real time fuel cell parameter and health index prediction. A comprehensive comparative analysis is performed using multiple machine-learning models trained separately on durability test data and embedded degradation data. This dual-dataset evaluation provides insight into how operational stress, thermal drift and long-term ageing influence predictive accuracy and model generalizability. The results demonstrate that thermo-fluid derived features significantly improve prediction fidelity, thereby reducing prediction error during transient and steady state operation. Furthermore, the proposed predictive scheme enables early identification of degradation patterns, facilitating proactive maintenance, minimising downtime and enhances the overall system reliability. Health indicators are formulated using weighted features of highly correlated Radviz plot parameters determined by explainable SHAP analysis. The results reveal the potential of data-driven degradation aware monitoring as a successful route toward life-extension and intelligent control strategies for PEM fuel-cell systems for an adaptive digital twin framework. This work directly contributes to the United Nations Sustainable Development Goals (SDG 7 and SDG 9) by enhancing the reliability, efficiency, and intelligent lifecycle management of clean hydrogen-based energy systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable AI assisted health index formulation and machine learning based thermo-fluid feature guided voltage prediction for PEM fuel cells

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月3日

查看全文
  • Asha Anu Kurian,
  • O. V. Gnana Swathika

摘要

Accurate prediction of voltage and health index for Proton-Exchange Membrane (PEM) fuel cells is essential for maintaining reliable operation under varying load and environmental conditions. This work illustrates a thermo-fluid informed feature framework that captures the dynamic interaction between temperature, pressure, humidity and flow variables for real time fuel cell parameter and health index prediction. A comprehensive comparative analysis is performed using multiple machine-learning models trained separately on durability test data and embedded degradation data. This dual-dataset evaluation provides insight into how operational stress, thermal drift and long-term ageing influence predictive accuracy and model generalizability. The results demonstrate that thermo-fluid derived features significantly improve prediction fidelity, thereby reducing prediction error during transient and steady state operation. Furthermore, the proposed predictive scheme enables early identification of degradation patterns, facilitating proactive maintenance, minimising downtime and enhances the overall system reliability. Health indicators are formulated using weighted features of highly correlated Radviz plot parameters determined by explainable SHAP analysis. The results reveal the potential of data-driven degradation aware monitoring as a successful route toward life-extension and intelligent control strategies for PEM fuel-cell systems for an adaptive digital twin framework. This work directly contributes to the United Nations Sustainable Development Goals (SDG 7 and SDG 9) by enhancing the reliability, efficiency, and intelligent lifecycle management of clean hydrogen-based energy systems.