Skip to Main content Skip to Navigation

Diagnostic robuste de pile à combustible PEM par modélisation physique et mesures d’impédance : prise en compte de conditions dynamiques et du vieillissement

Abstract : The PEMFC fuel cell is an electrochemical generator that has interesting potential for automotive applications and which use could help to meet pollution challenges. Poor management of system auxiliaries or malfunctions can place the fuel cell under operating conditions that accelerate degradation processes and shorten its useful life. The- operating conditions of the fuel cell core (temperature, humidity and partial pressures) must be monitored to identify as soon as possible and without any error abnormal situations, which is particularly difficult in dynamic operating conditions and during ageing.The aim of this thesis is to provide solutions to this problem. To that end, a robust diagnostic approach of operating conditions without direct measurement, in a dynamic environment and taking ageing into account has been developed.In order to characterize the fuel cell, a campaign of experimental tests on a test bench was carried out during 1000 hours of operation, with and without faults. This test campaign also allowed to verify to what extent the easily accessible polarization curves and impedance spectroscopy depend on the internal operating conditions.The approach developed is based on one hand on the use of a physical fuel cell model that capture its behaviour for given operating conditions and on the other hand on easy-access current, voltage and impedance measurements. Thus, this allows the development of an embedded solution that minimizes the number of sensors required.The differences between the experimental measurements and the outputs computed by the physical fuel cell model – called residuals – are indicators which are sensitive to faults in operating conditions, and insensitive to usual operating dynamic conditions. Two residuals, generated from fuel cell output voltage and high frequency impedance, are used to detect abnormal operating conditions thanks to threshold detection. The choice of the detection threshold levels allows to set the detection performance in terms of good detection and false alarm probabilities.In order to take ageing into account, a degradation module computes the decrease of fuel cell voltage with time so that ageing is taken explicitly into account by residuals.Going beyond detection alone, a method to class the operating conditions faults has also been proposed. It uses a database of residuals from various known faults to train a K-nearest-neighbour classifier, so that faults can be identified and classified.The model developed in the CEA was compared with experiments carried out on the test bench. An experimental determination of the model constants was carried out using electrochemical methods (cyclic voltammetry...) and numerical ones (linear regression). It appears that the model correctly computes voltage and high-frequency impedance, confirming the possible use of this specific model for diagnostic purpose. The method has been tested with optimal thresholds that have been empirically determined. The detection score obtained is 80%. The false alarm rate is less than 5% during the test.The K-NN classifier was then validated on experimental data. The classification score during the 1000h test is around 60% with large disparities depending on the faults. This score is more than 99% for two of the studied faults (high pressures and low humidity), 63% for low pressures but only 20% for a temperature drop or humidity increase.This work concluded that the approach using a physical model diagnosed most faults with a low level of false alarms during 1000 hours of ageing. The search of new measurements to increase the score of poorly diagnosed faults thus improving diagnostic performance is a main perspective.
Document type :
Complete list of metadata

Cited literature [124 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Wednesday, October 23, 2019 - 10:19:09 AM
Last modification on : Thursday, February 11, 2021 - 3:28:46 AM
Long-term archiving on: : Friday, January 24, 2020 - 2:36:43 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02190669, version 2


Gauthier Jullian. Diagnostic robuste de pile à combustible PEM par modélisation physique et mesures d’impédance : prise en compte de conditions dynamiques et du vieillissement. Automatique. Université Grenoble Alpes, 2019. Français. ⟨NNT : 2019GREAT009⟩. ⟨tel-02190669v2⟩



Record views


Files downloads