This paper proposes a novel approach for fault diagnosis of polymer electrolyte membrane fuel cell (PEMFC) with two-dimension (2D) image data, and investigates its effectiveness of identifying faults at different PEMFCs in terms of discrimination capacity and robustness. In the analysis, one-dimension (1D) voltage data from single cell is converted to corresponding 2D image using signal-to-image conversion technique. Various features are then extracted from the 2D image data, and optimal features are determined using Fisher discriminant analysis (FDA). Test data from PEMFC at different faulty states, including flooding and dehydration states, is collected for the analysis, and the effectiveness of optimal features in discriminating various states is investigated using K-means clustering method. Moreover, diagnostic performance with 2D image data is compared to those using 1D voltage signals, such that its effectiveness can be better highlighted. Furthermore, with test data collected from two different single cells, robustness of proposed method can be illustrated.

本文提出了一种利用二维（2D）图像数据诊断高分子电解质膜燃料电池（PEMFC）的新方法，并从判别能力和鲁棒性方面研究了其在不同PEMFC处识别故障的有效性。在分析中，使用信号到图像转换技术将来自单个单元的一维（1D）电压数据转换为相应的2D图像。然后从2D图像数据中提取各种特征，并使用Fisher判别分析（FDA）确定最佳特征。收集来自PEMFC处于不同故障状态（包括洪水状态和脱水状态）的测试数据进行分析，并使用K均值聚类方法研究区分各种状态的最佳特征的有效性。此外，将2D图像数据的诊断性能与使用1D电压信号的诊断性能进行比较，从而可以更好地突出其有效性。此外，利用从两个不同的单个单元格收集的测试数据，可以说明所提出方法的鲁棒性。