1. Fault diagnosis of PEMFC based on fatal and recoverable failures using multi-scale convolutional neural networks.
- Author
-
Xu, Jiang-Hai, Zhang, Ben-Xi, Zhu, Kai-Qi, Zheng, Xiu-Yan, Zhang, Cong-Lei, Chen, Zhang-Liang, Yang, Yan-Ru, Huang, Tai-Ming, Bo, Zheng, Wan, Zhong-Min, Hsu, Shu-Han, Yan, Wei-Mon, and Wang, Xiao-Dong
- Subjects
- *
CONVOLUTIONAL neural networks , *PROTON exchange membrane fuel cells , *FAULT diagnosis , *SUPPORT vector machines - Abstract
This study introduces a multi-scale convolutional neural network (MCNN) for diagnosing various faults in proton exchange membrane fuel cell (PEMFC) systems. Using simulated fault data from a 240 kW PEMFC stack, the MCNN method is validated. Compared to traditional methods like backpropagation neural network (BP), support vector machine (SVM), and temporal convolutional neural network (TCNN), MCNN achieves an average accuracy of 96.31%, outperforming BP (88.88%), SVM (90.67%), and TCNN (93.80%). The study demonstrates MCNN's effectiveness in diagnosing recoverable faults, fatal faults, and multiple faults in PEMFC systems, achieving fault diagnosis accuracies of over 97%. By leveraging its inherent advantages such as multi-channel processing, automatic feature learning, and parameter sharing, MCNN provides an effective solution for accurately diagnosing faults in PEMFC. The utilization of these advantages has enabled MCNN to perform efficient and accurate fault diagnosis in PEMFC systems. • MCNN achieves 96.31% accuracy, surpassing BP, SVM, and TCNN in diagnosing faults. • MCNN achieves over 97% accuracy in categorizing recoverable, fatal, and multiple faults. • MCNN provides swift and accurate fault diagnosis in PEMFC systems. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF