Elaboration of a non intrusive diagnosis tool for the detection of - - PowerPoint PPT Presentation

EUROPEAN INSTITUTE FOR ENERGY RESEARCH EUROPÄISCHES INSTITUT FÜR ENERGIEFORSCHUNG INSTITUT EUROPEEN DE RECHERCHE SUR L’ENERGIE EUROPEAN INSTITUTE FOR ENERGY RESEARCH

Edition 2006

Elaboration of a non intrusive diagnosis tool for the detection of water management and CO poisoning defaults in PEMFC stacks

Pierre-Alexandre BLIMAN Mohamad SAFA Michel SORINE Sébastien WASTERLAIN Daniel HISSEL Denis CANDUSSO Fabien HAREL Xavier FRANCOIS

Philippe MOÇOTÉGUY

Nadia STEINER Vincent RAIMBAULT Alain GIRAUD Fabrice AUZANNEAU Sébastien ROSINI

SINTEF Conference – Trondheim – 24/06/2009 –- 2/20

Outline Outline

• Introduction
• Developped

measurements

• Stacks characterization
• Developped

model

• Conclusions and future work

SINTEF Conference – Trondheim – 24/06/2009 –- 3/20

• Fuel cell insufficiently mature, partly due to limited

Fuel cell insufficiently mature, partly due to limited lifet lifetime

ime ⇒ Need for diagnosis tools to detect and classify failures

• r faulty operation modes so as to prevent or limit

degradation.

• Important

portant causes of causes of degradations / degradations / failures: failures:

– Bad water management (flooding, drying): Bad water management (flooding, drying): usually reversible and usually reversible and quite easy uite easy to to co contro ntrol. l. – Poisoning: reversibility = f(pollut Poisoning: reversibility = f(pollutant ant natu ature, c e, concentratio ncentration), hardly n), hardly controllable for air pollution, more controllable for air pollution, more easily for fuel pollutant like CO. easily for fuel pollutant like CO. – Carbon corrosion, catalyst o Carbon corrosion, catalyst oxidation; idation; usually irreversible and impossible to usually irreversible and impossible to control, particularly at stack level. control, particularly at stack level.

⇒

focus on water management and CO poisoning issues.

Scope of the study Scope of the study

SINTEF Conference – Trondheim – 24/06/2009 –- 4/20

Basics on diagnostic Basics on diagnostic

Fuel Cell System regulation Raw Measurements Pre-treatment Residual Decision Fault identification Corrective actions Alarm Corrective actions determination Input variables Detection Model Experimental Output indicators Estimated Output indicators + -

OK

SINTEF Conference – Trondheim – 24/06/2009 –- 5/20

Outline Outline

• Introduction
• Developpement
• f new measurement

tools

New high New high power impedancemeter. power impedancemeter. Integrated Integrated acquisition cardboard. acquisition cardboard.

• Stacks characterization
• Developped

algorithm

• Conclusions and future work

SINTEF Conference – Trondheim – 24/06/2009 –- 6/20

• Previous

systems' limitations:

Many Many impedan impedancemeters emeters

• f the pubic market are

f the pubic market are limited to a limited to a few Volts with regard to few Volts with regard to the mea the measurement volta urement voltage. e.

⇒ Development

• f a new EIS system:

High resolution High resolution digital analogic digital analogic converte converter (26 bits). 26 bits). 32 acquis 32 acquisition channels (1 fo ition channels (1 for I + 31 for U up to 300V). r I + 31 for U up to 300V). Allows 2 simultaneous measurements (stack Allows 2 simultaneous measurements (stack + individual cells or groups of cells). + individual cells or groups of cells).

EIS measurement for high power stack

Stack impedance spectra are close and do not depend

• n time

Large dispersion in cell impedance spectra due to

• cell position in the stack,
• cell state of health.

SINTEF Conference – Trondheim – 24/06/2009 –- 7/20

Developped Developped acquisition tool acquisition tool principle principle

Data treatment cardboard (AMR 7)

• Data acquisition and treatment.software. (Labview)
• Control and reading of data coming from test bench.

Imput signal generation:

• Current steps.
• EIS.

Ucell Ustack

Cell nr

PAC

Stack

Acquisition cardboard

SINTEF Conference – Trondheim – 24/06/2009 –- 8/20

Acquisition cardboard Acquisition cardboard

• Basic principle:

Generation of a bias current:

Error < 1% (can be reduced but with sensitivity loss)

y = 0.0396x - 0.0066

• 0.08
• 0.06
• 0.04
• 0.02

0.02 0.04 0.06

• 2
• 1.5
• 1
• 0.5

0.5 1 1.5 2

UGMR (V) Ucell (V)

• GMR Performances:

GMR Performances:

Ustack

• r

Ucell

Ibias

Rbias Intrinsic galvanic insulation

B r

UGMR Usupply 0V

GMR cell

Measurement of UGMR similar with Wheatstone bridge principle

• Integration:

Amplifier & multiplexer

SINTEF Conference – Trondheim – 24/06/2009 –- 9/20

Outline Outline

• Introduction
• Developpement
• f new measurement

tools

• Stacks characterization
• Developped

algorithm

• Conclusions and future work

SINTEF Conference – Trondheim – 24/06/2009 –- 10/20

Experimentals Experimentals

• 3 stacks technologies:
• Design of experiment

methodology:

6 parameters: anodi 6 parameters: anodic and cathodi and cathodic overst

• verstoich
• ichiometric

iometric ratios, fuel a ratios, fuel and oxident d oxident rela relative humidities, fuel CO tive humidities, fuel CO con conten tent, stack temperature. t, stack temperature. 26-

6-2 2 (16

(16 experiments) design of experiments, experiments) design of experiments, with with aliases. iases.

• Characterisations:

Current Current steps teps profile: profile: Current + Individual and total stack voltages: 100 kHz during 5 to 10s. Process regulation parameters + pressure drops: 1 Hz. EIS. EIS.

3M

SINTEF Conference – Trondheim – 24/06/2009 –- 11/20

1 2 3 4 5 6 7 8 9 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4 4.1 4.2 time [s] Ustack [V]

ha = hc = 35 %, sa = 2.5, sc = 3, T = 80°C ha =35 %, hc = 75 %, sa = 1.5, sc = 3, T = 80°C ha = 75 %, hc = 35 %, sa = 2.5, sc = 3, T = 50°C ha = hc = 75 %, sa = 1.5, sc = 3, T = 50°C ha = hc = 75 %, sa = 2.5, sc = 1.5, T = 80°C ha = hc = 50 %, sa = 2, sc = 2.25, T = 65°C ha = hc = 50 %, sa = sc = 2, T = 80°C

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4 4.1 4.2 time [s] Ustack [V] 1 2 3 4 3.3 3.4 3.5 3.6 3.7 3.8 3.9 time [ms] Ustack [V]

Transient Transient behavior ehavior

• f CEA 5 cells
• f CEA 5 cells

stack during stack during a a current current step tep from rom 0.4 A/cm² .4 A/cm² to 0.2 A/cm²

• 0.2 A/cm²

SINTEF Conference – Trondheim – 24/06/2009 –- 12/20

5 cells 5 cells stack resistivity and individual stack resistivity and individual cell cell resistivity scatterin resistivity scattering

Manip 4 ha = hc = 35%, sa = 2.5, sc = 3, T = 80°C Manip 6 ha = 35%, hc = 75%, sa = 1.5, sc = 3, T = 80°C Manip 7 ha = 75%, hc = 35%, sa = sc = 1.8, T = 80°C Manip 12 ha = 75%, hc = 35%, sa = 2.5, sc = 3, T = 50°C Manip 14 ha = hc = 75%, sa = 1.5, sc = 3, T = 50°C Manip 15 ha = hc = 75%, sa = 2.5, sc = 1.5, T = 80°C Manip 0 ha = hc = 50%, sa = 2, sc = 2.25, T = 65°C Ref CEA ha = hc = 50%, sa = sc = 2, T = 80°C

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

M a n i p 4 M a n i p 6 M a n i p 7 M a n i p 1 2 M a n i p 1 4 M a n i p 1 5 M a n i p R e f C E A ( n e u f ) R e f C E A ( v i e i l l i )

ρstack (Ω.cm²)

de 0.5 A/cm² à 0.9 A/cm² de 0.9 A/cm² à 0.7 A/cm² de 0.7 A/cm² à 0.6 A/cm² de 0.6 A/cm² à 0.4 A/cm² de 0.4 A/cm² à 0.2 A/cm² de 0.2 A/cm² à 0.1 A/cm² de 0 à 0.5 A/cm²

0% 5% 10% 15% 20% 25% 30% 35% 40%

M a n i p 4 M a n i p 6 M a n i p 7 M a n i p 1 2 M a n i p 1 4 M a n i p 1 5 M a n i p R e f C E A ( n e u f ) R e f C E A ( v i e i l l i )

5*σ(ρ)/ρstack

SINTEF Conference – Trondheim – 24/06/2009 –- 13/20

2 4 6 8 0.52 0.54 0.56 0.58 0.6 0.62 0.64 0.66 0.68 0.7 time [s] voltage [V] 0,1 0,2 0,3 0,4 0.52 0.54 0.56 0.58 0.6 0.62 0.64 0.66 0.68 0.7 time [s] voltage [V]

ha = 50% ; hc = 50% ; sa = 2.4 ; sc = 2.5 ; T = 70°C current step: 0,5 to 0,7 A.cm-2

1 2 3 4 0,5 1,5 2,5 3,5 0.58 0.6 0.62 0.64 0.66 0.68 0.7 time [ms] voltage [V] Vcell1 Vcell2 Vcell3 Vcell4 Vcell5 Vcell6 Vcell7 Vcell8 Vcell9 Vcell10 Vcell11 Vcell12 Vcell13 Vcell14 Vcell15 Vcell16 Vcell17 Vcell18 Vcell19 Vcell20

Transient Transient behavior ehavior

• f 3M 20 cells
• f 3M 20 cells

stack during stack during a current a current step tep from rom 0.5 A/cm² .5 A/cm² to 0.7 A/cm²

• 0.7 A/cm²

SINTEF Conference – Trondheim – 24/06/2009 –- 14/20

Outline Outline

• Introduction
• Development
• f new measurement

tools

• Stacks characterization
• Developped

algorithms:

Physical Physical model based. model based. Black box model based. Black box model based.

• Conclusions and future work

SINTEF Conference – Trondheim – 24/06/2009 –- 15/20

Physical Physical model model

• Input variables:

– H2 O, H O, H2 , O , O2 an and CO partial pressures, H d CO partial pressures, H+

+ concentration.

concentration. – fraction fraction of

• f cata

catalytic lytic sit sites poisoned by s poisoned by CO. CO. – water content in membr water content in membrane and GDLs. ane and GDLs.

• 1D (⊥

to MEA plane) model taking into account:

– kinetics of electrochemical reactions. kinetics of electrochemical reactions. – diffusion-migration (mass conse diffusion-migration (mass conservation equation). vation equation). – water balance in each compartmen water balance in each compartment : GC t : GC, G , GDL, membrane,…(cf. Ben L, membrane,…(cf. Benziger iger et a et al.) .)

• Model simplification by :

– discretization discretization for a for approxima proximation tion of cons

• f conserva

ervation tion equation equations (via orthogona s (via orthogonal collocation meth llocation method).

• d).

– Analysis of the different Analysis of the different time time-s

• scales phenomena (in ad

cales phenomena (in adsorption/desorption, sorption/desorption, water water diffusion) diffusion) ⇒Reduced Reduced 0D model model describing escribing I-U rela

• U relation

tion in various in various

• perating

perating con

• ndition

ditions. s.

• Serie-parallel "assembly" of the model to

simulate a cell heterogeneity and a stack.

• Output: polarization and EIS curves, are

determined analytically

SINTEF Conference – Trondheim – 24/06/2009 –- 16/20

Flooding diagnosis algorithm

Experimental Experimental parameters parameters Diagnosis Diagnosis decision decision Threshold Threshold function function Model output Model output in case of no in case of no flooding flooding

• Flooded

Flooded (1) (1)

• Not

Not flooded flooded (0) (0)

Residual Residual calculation calculation Model inputs Model inputs

Fuel Cell NN

Black-Box model

+

• exp

P Δ

calc

P Δ

exp exp

P P Pcalc Δ Δ Δ −

[ ]

C Tdwpt °

[ ]

A I

[ ]

C T °

[ ]

1

min .

−

Nl Q

Flooded cell Non flooded cell

s (No flooding)

SINTEF Conference – Trondheim – 24/06/2009 –- 17/20

• Definitions:

Definitions:

Neuron Neuron = succession of = succession of 2 2 mathemati mathematical funct al functions: ions: multipar multiparameter ameter linear combination + linear combination +

• ther
• ther (e.g. identity, sigmoid, linear,…)

Layer = group of unconnected Layer = group of unconnected ne neurons. urons.

• How is it build (

How is it build (3 steps) ? steps) ?

Archite Architecture d cture definitio finition:

Inputs = experimental parameters. Number of layers ≥ 2. Number of neuron/layer ⇔ compromise risks of overlearning and underlearning.

w6 w7

Ustack ΔP

I Tdew T Q w1 w4 w9 w5 w8 w3 w2 w0

wi ⇔ coefficients of multiparameter linear combination function

Neural network ?... Neural network ?...

Database random spliting: Database random spliting:

20% 70% 10% Learning Validation Test

Learnin Learning + Va + Valida lidation tion:

determination of wi and b and bi by iterative interpolation.

• ptimization of iteration number on learning:

Test Test ⇔ the network ability to predict the output

l underlearning

• verlearning

Prediction error nIterration = f((w0 * I) + (w1 * Tdew ) + b0 + b1 ) = f((w9 * Q) + b9)

SINTEF Conference – Trondheim – 24/06/2009 –- 18/20

Results: Neural network build-up

T [°C] ∈ [35-40] Tdwpt [°C] ∈ [25-50] I [A] ∈ [0-35] Q [Nl.min-1] ∈ [30-55]

Database: Learning on DP = f(t):

Threshold definition: s = 3*|σ(residual)|

Test:

Not flooded cell, data not previously seen by the Neural Network One punctual wrong alarm

SINTEF Conference – Trondheim – 24/06/2009 –- 19/20

Results: Model application to flooding diagnosis

σ x 2

Flooding detection: Detection

• f flooding

and recovery:

s

SINTEF Conference – Trondheim – 24/06/2009 –- 20/20

Conclusions & next Conclusions & next steps steps

• Main achievements:

Developments of: Developments of: A diagnosis model for water management issues. A new EIS system for operation at high stack voltages (up to 300 V). A hardware for acquisition, treatment and storage of system data during

• peration.
• Next steps:

Design of exper Design of experiment analysis on diffe iment analysis on different rent 20 20 cells stacks by EIS and cells stacks by EIS and current steps current steps

(in progress).

Exten Extend diagnosis model to diagnosis model to CO poisoning detection CO poisoning detection (in progress). Generalize the diagnosis model to Generalize the diagnosis model to diffe different PEMFC stack te ent PEMFC stack technologies chnologies (in progress). Interface the di Interface the diagnosis model agnosis model with the hardware in a with the hardware in a diagn diagnosis tool to be validated

• sis tool to be validated
• n a 20 cells stack.
• n a 20 cells stack.

Export the methodology to develop a to Export the methodology to develop a tool for other fuel cells technologies: for other fuel cells technologies: FCH-JU JTI CP 2008 " FCH-JU JTI CP 2008 "GENIUS" project for SOFCs. " project for SOFCs.