Proton exchange membrane fuel cell failure mode early diagnosis with wavelet analysis of electrochemical noise. (7th September 2016)

• Record Type: Journal Article
• Authors: Rubio, M.A.; Bethune, K.; Urquia, A.; St-Pierre, J.
• Is Part Of: International journal of hydrogen energy. Volume 41:Number 33(2016)
• Page Start: 14991
• Page End: 15001
• Publication Date: 2016-09-07
• Subjects: Electrochemical noise -- Electrochemical impedance spectroscopy -- Diagnosis -- Wavelet transform -- PEM fuel cell; Hydrogen as fuel -- Periodicals; Hydrogène (Combustible) -- Périodiques; Hydrogen as fuel; Periodicals; 665.81
• DOI: 10.1016/j.ijhydene.2016.05.292
• Languages: English
• ISSNs: 0360-3199

Abstract

Abstract: A diagnostic method for the performance degradation of low temperature proton exchange membrane fuel cells is proposed. The method is based on the analysis of the cell electrochemical noise. Experimental noise data were collected for a range of air relative humidities and stoichiometries including conditions leading to water flooding, membrane dehydration and air starvation failure modes. Data were converted with a Fourier transform (frequency window averaging of the amplitude) and a wavelet transform (coefficients standard deviation). Data were compared to impedance spectroscopy results. The method based on the wavelet transform was more sensitive. Cell states labeled by their air relative humidity and stoichiometry were correctly identified using a brute force algorithm by minimizing the Chebyshev distance between the actual and the calculated states. Independent and uniformly distributed random variations were added to experimental wavelet coefficients' standard deviations to define the calculated states. Highlights: Electrochemical noise is used to diagnose proton exchange membrane fuel cells. Noise data were collected for different operating condition based faults. Data were analyzed with Fourier and wavelet transforms. A wavelet method is more sensitive than a Fourier method and impedance spectroscopy. Cell states were correctly identified using a brute force search algorithm.