An artificial neural network ensemble method for fault diagnosis of proton exchange membrane fuel cell system

• Published in: Energy (Oxford) Vol. 67; pp. 268 - 275
• Main Authors: Shao, Meng, Zhu, Xin-Jian, Cao, Hong-Fei, Shen, Hai-Feng
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01.04.2014; Elsevier
• Subjects: Applied sciences; Artificial neural network ensemble; Dynamic model; Dynamical systems; Dynamics; Energy; Energy. Thermal use of fuels; Equipments for energy generation and conversion: thermal, electrical, mechanical energy, etc; Exact sciences and technology; Exchange; Fault diagnosis; Fuel cells; Learning theory; Membranes; Neural networks; PEMFC (proton exchange membrane fuel cell) system; Fault diagnosis; Artificial neural network ensemble; Dynamic model; PEMFC (proton exchange membrane fuel cell) system; Neural network; Proton exchange membrane fuel cells; Fault diagnostic; Fuel cell
• ISSN: 0360-5442
• DOI: 10.1016/j.energy.2014.01.079

The commercial viability of PEMFC (proton exchange membrane fuel cell) systems depends on using effective fault diagnosis technologies in PEMFC systems. However, many researchers have experimentally studied PEMFC (proton exchange membrane fuel cell) systems without considering certain fault conditions. In this paper, an ANN (artificial neural network) ensemble method is presented that improves the stability and reliability of the PEMFC systems. In the first part, a transient model giving it flexibility in application to some exceptional conditions is built. The PEMFC dynamic model is built and simulated using MATLAB. In the second, using this model and experiments, the mechanisms of four different faults in PEMFC systems are analyzed in detail. Third, the ANN ensemble for the fault diagnosis is built and modeled. This model is trained and tested by the data. The test result shows that, compared with the previous method for fault diagnosis of PEMFC systems, the proposed fault diagnosis method has higher diagnostic rate and generalization ability. Moreover, the partial structure of this method can be altered easily, along with the change of the PEMFC systems. In general, this method for diagnosis of PEMFC has value for certain applications. •We analyze the principles and mechanisms of the four faults in PEMFC (proton exchange membrane fuel cell) system.•We design and model an ANN (artificial neural network) ensemble method for the fault diagnosis of PEMFC system.•This method has high diagnostic rate and strong generalization ability.