Failure mode diagnosis in proton exchange membrane fuel cells using local electrochemical noise

Kurzfassung

Early diagnosis of fuel cell failure modes is a very active research topic, as it improves robustness and durability of fuel cells used in commercial applications. The diagnosis method should be suited for being applied in real time, without interfering with the fuel cell operation, and it should be implemented using inexpensive hardware and light equipment. A novel method of failure diagnosis in PEM fuel cells, based on the analysis of local electrochemical noise, is proposed. Seven electrochemical noise signals are acquired in different parts of the cell, significantly increasing the information for an effective diagnosis, since previous studies have only analyzed a single signal from the electrochemical noise in the cell. Each electrochemical noise signal is frequency decomposed using wavelet transform to create a characteristic pattern. These patterns are used in a deep learning neural network to perform the cell state classification. The proposed method has been successfully applied to the classification of 26 different states achieved in experiments where the following factors have been varied: (1) average current density; (2) airflow; (3) drying; and (4) air pressure. The mean successful identification rate of the 26 states is above 85%. The proposed diagnosis method is well-suited for real-time diagnosis, and it can be implemented using lightweight and inexpensive hardware