GENeric diagnosis Instrument for SOFC Systems (245128)

Philippe MOÇOTÉGUY EIFER/Project Manager

Genius Partnership & Budget

3 years collaboration project: 01-02-2010 to 31-01-2013

Total budget: 3928 k€ ; Total funding: 2068 k€

Participant Country Type

EIFER Germany R&D

CFCL England Industry / SME

EBZ Germany Industry / SME

FC LAB France University

Hexis Switzerland Industry / SME

HTc Switzerland Industry / SME

TOPSOE Denmark Industry / SME

UniGE Italy University

UniSA Italy University

VTT Finland R&D

Wärtsilä Finland Industry / SME

Inno France Industry / SME

Develop a “GENERIC” diagnosis approach that all SOFC developers could use and implement it in their systems according to their specific constraints:

only using process values (normal measurements and system control input parameters).

taking into account “on line” or "off-line“ situations.

Objectives & Outcomes

Steps:

Testing stacks & systems from 4 manufacturers, using commonly defined test plan based on “Design Of Experiment” methodology.

Developing a diagnostic protocol integrating the best algorithm(s).

Validate it (them) on 3 SOFC systems, either off-line or on-line.

4 academic institutions will evaluate 3 types of algorithms (based on signal treatment and on residuals determination from either black box or grey box models) to define the optimal tool(s) for fault detection and identification.

Iteratively adapting the defined test plan according to algorithm developers’ needs.

Diagnostic Methodology

Detection

Process

DEVICE

+ -

Estimated Output

Fault

Identification

Normal operation

Detection Algorithms

Pattern recognition

algorithms

Physical/GB/BB

Models

Residual based

algorithms

Indicators

Residuals

Raw Data

Controller

Corrective or

Mitigative

Actions

Faulty operation

Operating Parameters

&

Responses

Detection

Selection of N parameters to be studied and their

operating range

Selection of responses to be measured

DoE methodology

Reduced centred parameters

Determination of directly impacting parameters

(screening)

Determination of influent interactions between influent parameters

>3 experiments in the centre of the domain so as to:

o Evaluate measurement's reproducibility effect significance.

o Detect potential ageing during experiment.

N+1 ≤ nexp = 2p

-200

-150

-100

-50

0

50

100

coef

f P el [W

]

-100

0

100

200

300

400

500

600

700

coef

f P th

[W]

Results of DoE

PNG T Ustack CPO

3rd & 4th level interactions

System supplier : Hexis

Measurements: EIFER

WP 3: DoE analysis of stack voltage

(VTT experiments on HTc stack)

-600

-500

-400

-300

-200

-100

0

100

200

300

regre

ss c

oeff V

sta

ck [m

V]

Tfurnace FU

Tfurnace

* j

Tfurnace

*FU

Tfurnace

*CPO

j * FU

j*CPO

FU * CPOj

CPO

4th level

interaction

3rd level

interactions

Stack supplier : HTc

Measurements: VTT

Residual based diagnostic approach

Fault tree analysis

Inference

Stack fault

Electrolyte

electrode

delamination

Oxide layer

growth between

interconnect and

cathode

Rib

detachment

Seal failure

Increased thermal

gradient

Local shortage

of fuel

Anode

oxidation

Increase in cell

Ohmic resistanceOverheating

Temperature

controller failure

Increase

in stack

temperature

1 2

Decrease

in stack

voltage

Decrease

in stack

power

Temporany

increase in stack

temperature

Decrease

in stack

voltage

Decrease

in stack

power

Increase in air

temperature at

cathode outlet

Increase in gas

temperature at

anode outletTemp.

controller (1)

Post-

burner (1)

Temperature

controller

Increase in stack

temperature

Decrease in stack

temperature

Increase in

flow air mass

at compressor

outlet

Increase in

compressor

power

Decrease

in net stack

power

Increase of

air excess (λ)

at stack

Decrease in

flow air mass

at compressor

outlet

Increase in

compressor

power

Decrease

in net stack

power

Decrease of

air excess (λ)

at stack

1 2

Post

burner fault

(2)

Post

burner fault

(1)

Increase in

flow air

temperature at

pre-heater

outlet

Decrease in

flow air

temperature at

pre-heater

outlet

Fault

Symptoms (measured values)

s1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13 s14 s15 s16 s17

f1 0 1 1 0 1 1 1 1 0 1 0 0 0 0 0 1 1

f2 0 1 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0

f3 0 1 1 0 1 1 0 1 0 1 0 0 0 0 0 1 1

f4 0 1 1 0 0 0 0 1 0 1 0 0 0 0 0 0 0

f5 0 1 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0

f6 1 1 1 0 1 1 0 1 0 0 1 1 1 0 0 1 1 Threshold definition (e.g.: ± 2 , ± 3)

Black/grey box Model

Signature matrix

At least one

match

No match

Residual calculation

Uniquematch

Residual based algorithm performance

1000 2000 3000 4000 5000 6000 7000 800015

20

25

30

35

40

45

50

55

Sta

ck V

oltage [

V]

time [min]

measured

simulated

Stack supplier : TopSoE

Measurements: TopSoE

Simulations: Black box model (UniSA)

Simulations: Grey box model (UniGE)

SOFC System (BoP part)

SOFC stack as a sensor

Wavelet Based Diagnosis

SOFC failure

BoP Fault

No Fault

Bayesian Network Based

Diagnosis

Fault Type

Function:

1. Check the state of health of a “special” sensor – SOFC stack;

2. BoP fault detection.

Principle:

Decompose the SOFC voltage signal by wavelet transform method. The distribution of energies of the subsignals can indicate the state of health and the abnormal operating condition of the SOFC.

This algorithm has been validated on data from both test rounds at VTT on HTc stack (requested signal sampling rate is at least 1Hz)

Function: BoP fault identification.

Principle:

Use experimental data to train the Bayesian Network and obtain a meta-model of SOFC. Use this model to estimate the operating variables – Fuel & Air flow rates and utilisations, Furnace & Stack temperatures.

Fault to diagnose: Low current density and high FU.

Validation result: (VTT data)

67% of the1449 samples measured in such a faulty operating mode can be identified by the Bayesian Network model.

Pattern recognition based algorithm performance

"The aim will be to deliver (…) reliable control and diagnostics tools both at a component and at system level.“ "Applied research activities are directed towards developing components and sub-systems with improved performance, durability and cost (…)."

Main project objective: develop a Generic diagnostic algorithm integrated in a standard hardware equipment to detect faults and prevent failure before occurrence. better system reliability increase system lifetime.

Alignment to MAIP- AA3 & AIP 2008 – Topic 3.3

Alignment to MAIP- AA3 & AIP 2008 – Topic 3.3

“(…), substantial effort is needed to address lifetime requirements of 40,000 h for cell

and stack, as well as competitive costs..."

The approach will allow to detect faults and to prevent failure before their occurrence

better system reliability improve the competitiveness of fuel cell vs other µ-

CHP technologies by increasing availability ratio & system lifetime.

“Improved prediction and avoidance of failure mechanisms”.

“Development of strategies for recovery of cell and stack performance”

Experimental evaluation of failure mechanisms’ signatures & development of new data analysis methods identify the effect of each failure mechanism from normal base-line early detection system operating parameters optimisation degradation minimisation.

Organisation of workshops: • Workshop in Viterbo (M19) about degradation causes and effect. • Common workshop with D-CODE project (M28) in Belfort: 10 speakers, 50 attendees.

Cross-cutting issues: Education, Training & Dissemination

Publications & Communications: • 2 publications. • 3 presentations at conferences. • 2 presentations in workshops.

Website: http://genius-jti-project.eu/

Education & Training: • 4 PhD students working on the project, 2 of them spent/are spending time at

partners facilities. • 1 PhD defended (P. de VASCONCELLOS CARDONE, UniGE, 04/2012), 1 to be

defended (K. Wang, FC Lab-EIFER)

• Collaborations

Enhancing cooperation and future perspectives Technology transfer and collaboration

• Technology Interfaces

Enhanced interface between stack/system manufacturers and diagnosis algorithm developers

BoP

Stack

SRU

Cell

DESIGN

Object: SOFC (from SRU to stack)

Sensor: various levels

Outcome: Method and signatures as input for a diagnostic tool + recovery strategy recommandations

Object: SOFC system

Sensor: SOFC stack/system

Outcome: diagnosis algorithm integrated with hardware

GENIUS

Object: PEMFC system

Sensor: Electrochemical Impedance Spectroscopy made by the DC/AC inverter Outcome: diagnosis algorithm + Impedance measurement hardware

D-CODE

SAPPHIRE

Prognostics / adaptative control

AA 1

AA 4

Automotive Early Markets

PEM D-CODE

Electrolysis AA 2

GENIUS Solid oxyde

diagnostic

Stationary

Fuel cells AA 3

Future perspectives: Generalization to all FC and H2 technologies and AA