This paper presents a prognostic approach based on an ensemble of two degradation indicators for the prediction of the Remaining Useful Life (RUL) of a Proton Exchange Membrane Fuel Cell (PEMFC) stack. When the fuel cell stack experiences variable operating conditions, degradation indicators, such as the stack voltage and the stack State Of Health (SOH), are not able to individually provide precise and robust RUL predictions. The stack voltage does not directly measure the component degradation, as it is only related to degradation symptoms, which are significantly affected by operating conditions. The SOH provides aging information but it can only be measured at low frequency in industrial applications. The objective of this work is to combine the two indicators, leveraging their strengths and overcoming their drawbacks. Two different physics-based models are used to this aim: the first model receives a signal directly observable and related to the stack voltage, which can be frequently and easily measured; the second model is fed by periodic measurements from the physical characterization of the stack, which gives reliable information on the SOH evolution. The prognostic procedure is implemented using Particle Filtering (PF), and the outcomes of the two prognostic filters are aggregated to obtain the ensemble predictions. The ensemble-based approach employs a local aggregation technique that combines the outcomes of two prognostic models by assigning to each model a weight and a bias correction, which are obtained considering the individual models’ local performances. The dependence between the two indicators is also accounted for, by dependent Gamma processes. The results obtained show that the accuracy of the RUL predictions obtained by the proposed ensemble-based method outperforms that obtained by the individual models.

An ensemble of models for integrating dependent sources of information for the prognosis of the remaining useful life of Proton Exchange Membrane Fuel Cells

Zhang, D.;Baraldi, P.;Zio, E.
2019-01-01

Abstract

This paper presents a prognostic approach based on an ensemble of two degradation indicators for the prediction of the Remaining Useful Life (RUL) of a Proton Exchange Membrane Fuel Cell (PEMFC) stack. When the fuel cell stack experiences variable operating conditions, degradation indicators, such as the stack voltage and the stack State Of Health (SOH), are not able to individually provide precise and robust RUL predictions. The stack voltage does not directly measure the component degradation, as it is only related to degradation symptoms, which are significantly affected by operating conditions. The SOH provides aging information but it can only be measured at low frequency in industrial applications. The objective of this work is to combine the two indicators, leveraging their strengths and overcoming their drawbacks. Two different physics-based models are used to this aim: the first model receives a signal directly observable and related to the stack voltage, which can be frequently and easily measured; the second model is fed by periodic measurements from the physical characterization of the stack, which gives reliable information on the SOH evolution. The prognostic procedure is implemented using Particle Filtering (PF), and the outcomes of the two prognostic filters are aggregated to obtain the ensemble predictions. The ensemble-based approach employs a local aggregation technique that combines the outcomes of two prognostic models by assigning to each model a weight and a bias correction, which are obtained considering the individual models’ local performances. The dependence between the two indicators is also accounted for, by dependent Gamma processes. The results obtained show that the accuracy of the RUL predictions obtained by the proposed ensemble-based method outperforms that obtained by the individual models.
2019
Dependent processes; Ensemble; Particle Filtering (PF); Prognostics; Proton Exchange Membrane Fuel Cell (PEMFC); Remaining Useful Life (RUL); Control and Systems Engineering; Signal Processing; Civil and Structural Engineering; Aerospace Engineering; Mechanical Engineering; Computer Science Applications1707 Computer Vision and Pattern Recognition
File in questo prodotto:
File Dimensione Formato  
2019_MSSP_Zhang_Baraldi_PEM_Prognistics.pdf

accesso aperto

: Pre-Print (o Pre-Refereeing)
Dimensione 1.71 MB
Formato Adobe PDF
1.71 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1077904
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 56
  • ???jsp.display-item.citation.isi??? 48
social impact