k-Nearest Neighbours Fault Diagnosis of Proton Exchange Membrane Fuel Cell - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2014

k-Nearest Neighbours Fault Diagnosis of Proton Exchange Membrane Fuel Cell

Résumé

In this paper, a fault diagnosis of a Proton Exchange Membrane Fuel Cell (PEMFC) is presented. The aim of this tool is to avoid non-reversible degradations of a fuel cell system linked to fault occurrence. The damage incurred by the fuel cell depends on the nature of the fault and its duration time. Some faults are non-reversibles for the fuel cell after few seconds. Some experiments are performed on an in-lab test bench considering a forty-cell stack. The electrode has an active area of 220cm² and a nominal power of 3.8 kW. The aim is to reproduce different operating conditions and especially possible fault states. The fuel cell system was operated at different conditions such as oxygen starvation, electrical short-circuit at the output of the fuel cell stack and heating of the cooling system. All these states will be compared to the nominal operating conditions defined, beforehand, by test specifications and are used as database for the algorithm inputs. The purpose is to be able to detect if the fuel cell is in abnormal operating conditions and also to isolate the fault. The diagnosis is here based on Electrochemical Impedance Spectroscopy (EIS) that measures the impedance of the stack over a large band of frequencies from 100 mHz to 5 kHz. Some relevant points are extracted from the EIS in order to be used as inputs of the classification algorithm. Depending on the application, different constraints have to be taken into account when building a diagnosis tool of fuel cell systems. One of the most constraining limitations is to be as least intrusive as possible. Indeed, it is preferable to use as less costly sensors as possible when considering industrial implementation. The second constraint is linked to the computation time of the diagnosis approach which should be fast enough to allow the implementation of an on-line diagnosis of fuel cell systems. These two factors motivate the choice of the k-NN algorithm, which is well known for its efficiency and simplicity of implementation. Results will be detailed in the final presentation.
Fichier non déposé

Dates et versions

hal-02300468 , version 1 (29-09-2019)

Identifiants

  • HAL Id : hal-02300468 , version 1

Citer

Elodie Pahon, Samir Jemei, Fabien Harel, Rosa Elvira Silva Sanchez, Latifa Oukhellou, et al.. k-Nearest Neighbours Fault Diagnosis of Proton Exchange Membrane Fuel Cell. Conference Internationale Discussion on Hydrogen Energy and Applications, May 2014, Nantes, France. ⟨hal-02300468⟩
61 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More