Scholarlink

<p>Fault diagnosis of fuel cells often focuses on single faults, leading to lower accuracy in diagnosing simultaneous faults. This paper researches a data-driven diagnostic method for both single and simultaneous faults, aiming to establish an efficient online fault diagnosis approach. Firstly, a theoretical model of a proton exchange membrane fuel cell (PEMFC) system is established. Based on this, a radial basis function (RBF) neural network surrogate model is designed to improve computational efficiency. The average relative error across all features between the surrogate model and the theoretical model is below 1%. Subsequently, Sobol’s global sensitivity analysis is used to analyse the relationship between PEMFC system faults and various characteristic parameters during real-time operation. The sensitive feature set related to different faults in the PEMFC system is then identified. Finally, an adaptive diagnostic strategy is proposed, and a sensitivity-based diagnostic algorithm is established. Compared with other common single-label and multi-label diagnostic methods, the sensitivity-based diagnostic algorithm achieves an <i>F</i>1_Score of 99.1% on single-fault data, cutting training time by more than 80%. In scenarios with simultaneous faults and sparse data, the method achieves an accuracy of 92.5%, which is 7.5% higher than that achieved by the best multi-label method.</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

Enhanced fault detection in proton exchange membrane fuel cell via neural network model and sensitivity-based analysis

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • Xiuliang Zhao,
  • Weikun Huang,
  • Yinglong Zhou,
  • Bangxiong Pan,
  • Ruochen Wang,
  • Limei Wang,
  • Shaobo Ji

摘要

Fault diagnosis of fuel cells often focuses on single faults, leading to lower accuracy in diagnosing simultaneous faults. This paper researches a data-driven diagnostic method for both single and simultaneous faults, aiming to establish an efficient online fault diagnosis approach. Firstly, a theoretical model of a proton exchange membrane fuel cell (PEMFC) system is established. Based on this, a radial basis function (RBF) neural network surrogate model is designed to improve computational efficiency. The average relative error across all features between the surrogate model and the theoretical model is below 1%. Subsequently, Sobol’s global sensitivity analysis is used to analyse the relationship between PEMFC system faults and various characteristic parameters during real-time operation. The sensitive feature set related to different faults in the PEMFC system is then identified. Finally, an adaptive diagnostic strategy is proposed, and a sensitivity-based diagnostic algorithm is established. Compared with other common single-label and multi-label diagnostic methods, the sensitivity-based diagnostic algorithm achieves an F1_Score of 99.1% on single-fault data, cutting training time by more than 80%. In scenarios with simultaneous faults and sparse data, the method achieves an accuracy of 92.5%, which is 7.5% higher than that achieved by the best multi-label method.