Skip to main content

DEGRADATION SIGNATURES IDENTIFICATION FOR STACK OPERATION DIAGNOSTICS

Final Report Summary - DESIGN (DEGRADATION SIGNATURES IDENTIFICATION FOR STACK OPERATION DIAGNOSTICS)

Executive Summary:
The overall objective of the DESIGN project was the development of a diagnostic methodology for insidious phenomena that slowly accelerate the degradation of SOFC stacks.
The project has studied the influence of abnormal high fuel utilisation on measures performed on the stack sub-components: Cells, Single Repeating Units (SRU) and small stacks. Specific testing protocols have been defined, shared and implemented by the consortium partners. Identification of characteristic signatures of associated degradation phenomena at the lower level has been done and tentatively transposed at the stack level, to pave the way for diagnosing slow degradation phenomena in commercial SOFC stack.
The basis for a new diagnostic methodology has been developed generating 2 patents (one pending) and 2 congress presentations (FDFC2013 and EFCF2014). Promising results have been obtained on single cells and single repeating units but there is a need for further validation and up scaling (within the project, TRL was raised from 1 to 3 (experimental proof of concept demonstrated)).
DESIGN direct output provides a solid basis for the development of a sound diagnostic for SOFC stationary applications.
Further details on the dissemination activity of the project are available on the web site: www.design-sofc-diagnosis.com

Project Context and Objectives:
One of the main condition for SOFC stationary systems competitiveness and massive market entry is to exceed a durability of 40 000 h. This objective has not been reached up to now except for very specific designs. A better understanding and detection of stack failure mechanisms and internal state of health would constitute a privileged way to reach this lifetime objective.
If, massive SOFC stack or auxiliary failure can be easily detected by the control system, it is not currently the case for insidious abnormal operating conditions that come either from the stack itself (e.g. a distribution channel blocked), or from minor system failures (e.g. abnormal decrease of performance of a fuel blower).
The overall objective of the DESIGN project was to provide a sound diagnostic method for these insidious phenomena that slowly accelerate the degradation of SOFC stacks, through the understanding of the local responses of sub-stack elements.
The project has studied the influence of selected slowly-damaging operation conditions on measures performed on the stack sub-components: Cells, Single Repeating Units (SRU) and small stacks. Identification of characteristic signatures of associated degradation phenomena at the lower level has been done and tentatively transposed at the stack level, to pave the way for diagnosing slow degradation phenomena in commercial SOFC stack as illustrated in Fig1.

Figure 1: Scheme of the Design project

The main targets of the project were:
• The identification of relevant signals to be monitored to diagnose full stack degradation phenomena;
• A data analysis methodology to be applied to measured signals;
• A set of characteristic signatures for abnormal high fuel utilisation operation condition at the local and stack level, to be used for diagnosing long-term degradation conditions;
• The recommendations for operation recovery, once a degradation condition is identified at the cell, SRU or stack level.

To reach these goals, the Design project was organized in three phases:
1. The identification and prioritization of the phenomena (operating conditions generating slow degradation) to be investigated in the project was done at the beginning of the project in link with relevant industry partners from the Genius consortium. A list of critical degradation mechanisms has been shared among the consortium and 3 of them have been selected for being studied within the project. On the basis of these selected degradation mechanisms, a test matrix and tests protocols have been defined by the Design partners;
2. Several iterations of experiments have been achieved in the second phase, on single cells, on Single Repeating Unit and on short stacks, with adapted measures including Electrochemical Impedance Spectroscopy (EIS). Experimental data produced have been analysed to identify degradation signatures related to selected phenomena. These local characteristics once identified served as a basis for proposing a diagnostic method also valid at the stack level, with limited instrumentation. Recovery recommendations have also been studied to correct the identified damaging situation;
3. Finally, in the last phase of the project, the specific signatures characterising slow degradation phenomena at the local level of single cells, Single Repeating Unit and small stacks were to be extended to the diagnostic of a simple and robust full-size commercial stack (30-40 cells), with limited instrumentation.
The Design project was a highly challenging project aiming at linking the understanding of local phenomena with the diagnosis of the operation of full stacks.
Scientific and technical challenges were numerous. In the first half of the project they were associated with the identification of specific signatures at the local cell / SRU / small stack level. In the second half of the project they consisted in assessing the signatures reproducibility under hydrogen and under reformate as well as in validating a diagnostic methodology derived from these signatures liable to serve as the basis for a diagnostic tool for commercial stacks.

Project Results:
The overall objective of the DESIGN project was to provide a sound diagnostic methodology for detecting insidious phenomena that slowly accelerate the degradation of SOFC stacks, through the understanding of the local responses of sub-stack elements.
To link the understanding of local phenomena with the diagnosis of full stack operation the project was organised according to the following main technical objectives:
• The identification and prioritization of dysfunctions and associated degradation mechanisms (operating conditions generating degradation) to be investigated in the project. This was to be done through the organisation of a common workshop between the project GENIUS and DESIGN,
• The establishment of a text matrix and the definition of common test protocols, including the specification of relevant sensors and signals to be monitored, to be implemented on single cells, single repeating units and short stacks,
• The implementation of several test campaign aiming at identifying some signals associated with one degradation mechanisms and validating it as a signature,
• The development of a data analysis methodology to be applied to measured signals,
• The identification and assessment of a set of characteristic signatures for the selected degradation phenomena at the local and short-stack level, to be used for diagnosing long-term degradation conditions of full stacks,
• The recommendations for operation recovery, once a degradation condition is identified at the cell, SRU or stack level.
2.1 Identification of most probable and critical system dysfunctions
SOFC cells and stacks dysfunctions are still not fully understood, especially when it occurs after long operation times. Among the several degradation phenomena identified, most relevant mechanisms examples linked to stationary operation are i) thermal gradients generating stresses that can cause contact losses between two adjacent layers, ii) growth of electrically less conductive oxide layer between the interconnect plate and the electrode, especially the cathode, iii) extended periods of operation at elevated temperatures favouring micro structural degradations inside porous electrodes, iv) electrode poisoning by contaminants like sulphur or chromium, v) carbon deposition in the anode or vi) high local fuel utilization liable to cause a re-oxidation of the anode.
Despite their difference in nature, the effect of these degradation modes on cell performance measured in long-term degradation testing is common to all of them: a decrease of available voltage at constant current load, or a drop in delivered current density at constant voltage. Therefore, it is not possible to identify a specific degradation mechanism or combination of mechanisms by observing only changes to the DC behaviour of the cell. Thus, there is a need to develop a diagnostic technique that allows the identification of specific degradation mechanisms of SOFCs in a minimally invasive way.
Although several cell and stack degradation mechanisms seem to be related to materials choice and qualities, production processes etc. that might be solved when the maturity of SOFC technology is increasing, system dysfunctions can also cause accelerated degradations. This is well illustrated by the degradation mechanisms listed above with 2 mechanisms typically related to material behaviour and 4 related to system related causes.
System dysfunctions can be classified in impurity based stack effects or in deviating operation conditions. These deviating operation conditions are for example too low CPOx air ratio or steam-to-carbon ratio in the reformer, inverter or fuel supply faults, increased temperatures, etc.
Furthermore, one has to distinguish between different operation-modes like normal operation and start-stop-cycles. The latter is especially critical for SOFC based power generators. However, it is not considered in the following in order to reduce the complexity of the investigations for allowing assessment of the methodology of degradation signature determination.
The investigation of system dysfunctions based upon a HAZOP (Hazards & Operability) study, a structured and systematic examination of a planned or existing process or operation in order to identify and evaluate problems that may represent risks to personnel or equipment, or prevent efficient operation. The HAZOP technique was developed to analyse chemical process systems, but has been assigned to other industrial systems and complex operations as well as to software systems. A HAZOP is a qualitative technique based on guide-words.
The HAZOP study in the current framework has allowed a systematic check of system dysfunctions and their consequences. For this purpose the SOFC system has been divided into functional groups. In the following, two different system layouts were investigated: CPOx or steam reforming based systems associated with a cathode air heater (heated by afterburner exhaust gas) and an inverter. As a consequence the four following functional groups have been considered:
- Steam reformer / evaporator unit
- CPOx reformer
- Cathode air heater (heated by afterburner exhaust gas)
- Inverter
These four units are at least partly affected by the control system. Conditions that cause system dysfunctions like drift of mass flow meters have been covered in the respective components investigation.
Table 1 summarizes the main impacts induced on stack by system dysfunctions.
Component Dysfunction Stack impact Recovery
Steam reformer / evaporator unit Deviation of S/C ratio Soot formation Depends on amount of soot deposited
Deactivation of catalyst; increased internal reforming rate High thermal gradients (contact loss, leakages) No recovery, but early detection
Sulphur breakthrough (after catalyst deactivation) At least partly recovery
Low fuel flow rate Local re-oxidation due to high fuel utilization Decrease of fuel utilization
CPOx reformer Deviation of air ratio Soot formation Depends on amount of soot deposited
Deactivation of catalyst; incomplete CPOx reforming Soot formation Depends on amount of soot deposited
High thermal gradients (contact loss, leakages) No recovery, but early detection
Low fuel flow rate Local re-oxidation Decrease of fuel utilization
Cathode air heater High-temperature Increased chromium poisoning No recovery, but adaptation of system parameters
Inverter Current ripples Change of microstructure No recovery, but maintenance possible
High current Local re-oxidation due to high fuel utilization Decrease of fuel utilization, maintenance
Table 1: Tentative analysis of induced degradation mechanisms, of possible detection and of potential mitigation or recovery
Thermal gradients can be caused by various mechanisms of system dysfunctions like reformer catalyst deactivation, where the stack is operated afterwards with an increased internal reforming rate that causes a dramatic temperature drop at stack inlet. Soot formation that is caused by sensor drifts or reformer catalyst deactivations is a second source of stack failures. High (local) fuel utilizations can be generated by sensor drift, inverter failure or a partly blocking of anode channels. It results in anode re-oxidation that can also cause stack degradation. Electrode poisoning classically by sulphur or by chromium can also be the consequence of a reformer catalyst deactivation or an increase of temperature of the cathode air heater.

During the GENIUS-DESIGN workshop, a discussion has allowed the ranking of “most critical” degradation phenomena to be detected. Owing to their potential mitigation or recovery and to the experimental capacity to isolate and reproduce them, 3 degradation phenomena have been selected to be studied within the DESIGN project. They are:
1) Anode re-oxidation by locally increased fuel utilizations,
2) Carbon deposition and
3) Small leakage at anode side.
Small leakage at electrode side is a highly frequent dysfunction. The degradation it causes is irreversible and its detection remains a challenge. Increase of compression is not an option for in-situ recovery whereas a temperature excursion might be an option if the sealing is damaged.
High fuel utilization (fuel starvation) is at first a local phenomenon that is caused by cell / stack heterogeneities, gas channel blocking or deposition of carbon. It leads to a local anode oxidation that will affect the stack performance. A recovery is possible, with efficiency dependent on the cell type (ESC more than ASC), by a reduction of current. At stack level, local anode oxidation can easily be detected via single cell or block voltage measurement. However, single cell measurement will be too expensive at system level. At cell level, EIS (conversion impedance anode) can be used. Other tests like a variation of fuel utilization need to be investigated.
Carbon deposition occurs when the oxygen-to-carbon ratio is below a certain value at given temperature. The deposition of carbon is favoured at the anode due to its catalytic activity but might also be present in the gas supply pipes depending on temperature. Operation of the stack at high current / fuel utilization tends to mitigate carbon deposition due to the large amount of steam produced . Carbon deposition leads to block the active sites of the anode or the complete anode channel.

2.2 Definition of testing protocols and of a test matrix for selected degradation mechanisms
A test plan has been defined to simulate the faults associated with the first two mechanisms only. The test plan concerning the third mechanism has been postponed to allow further discussion on how to simulate accurately such a fault from a technical stand point. Some preliminary tests have been performed and published .
Each test on Single Cell (SC) and/or one Single Repeating Unit (SRU) has been dedicated to simulate only one fault. On the other hand, the same fault has been conducted on several SCs and SRUs to ensure reproducibility and reliability of detected signals.
2.2.1 High Fuel Utilisation
The fuel utilisation (FU) is related to the electrical current density (i), fuel composition and gas flow rates (Qi) through the following equation :
(1)
where ncell is the number of stack cells, F is the Faraday constant and Afc is the cell active area.
The high FU fault is simulated by either increasing the current density or decreasing the total gas flow rate with respect to nominal values defined to be in accordance with the nominal conditions defined by the stack producer (see Table 2). All the other control variables are kept at their nominal value.
Control Variable Nominal value
Current density [A/cm2] 0.30 (ESC)/ 0.50 (ASC)
Fuel Utilisation 0.6
Air Utilisation [%] 20
Furnace temperature [°C] 750
Table 2: Nominal operating conditions selected for the first test campaign

Therefore, the current density has to be increased stepwise from the nominal value corresponding to a FU=0.6 up to FU=0.9. Each step in current density should be gradual. Once the step value is reached, the operating condition should be kept for 20 to 24 hours. After such a period of time, the current density should be brought back to its nominal value and kept for 20 to 24 hours before a new step is made. At the beginning and end of each stabilisation plateau EIS spectra and polarization curves should be carried out.
The high FU fault is also simulated by decreasing the total inlet gas flow rate. This was done following exactly the same approach explained for the current density.
The test plan was to start running the SC SRU or short stack on pure hydrogen and wait long enough for the performance to stabilise. Once the performance would be stable, the SC or SRU should be brought to the nominal operating conditions and the protocol would start.
First experimental tests in these conditions with single cells, SRUs or short stacks failed to produce consistent behaviour. Despite several attempts by all testing partners, the fuel composition, although chosen to be consistent with Fuel Cell Systems Testing, Safety & Quality Assurance (FCTESQA) recommendations, appeared to generate a lot of discrepancies. It has then been decided to operate under humidified hydrogen and to concentrate on high fuel utilisation fault to ensure reproducibility before operating with reformate. In addition, it has also been decided to have all the partners working on the same anode supported cell produced by SOFCpower with the following composition (Ni-YSZ// 8YSZ //GDC-LSCF).
2.2.2 Carbon deposition
Test protocol has been defined with a special attention given to operating conditions enhancing carbon deposition and those inhibiting it and shared within the Design consortium. However, as it has not been used by the partners during the project implementation it is not reported here.
2.2.3 Variable to measure and raw data format
All variables to measure are gathered in Table 3. In order to exchange efficiently data within the consortium through the internal portal, specification for raw data formatting have also be produced.
Variable Description Symbol Sampling rate [Hz]
Control Variables Current Density i 1
Total inlet gas flow rate QINgas 1
Fuel composition xOUTgas
Furnace temperature Tfurn 1
Monitoring Variables Single Cell/Stack Voltage Vcell/Vstack 1
Inlet Temperature (fuel, air) TINf/ TINa 1
Outlet Temperature (fuel, air) TOUTf/ TOUTa 1
Cell/Stack Temperature Tcell/Tstack 1
Outlet Gases Flow Rates QOUTgas
Inlet Anode/Cathode Pressure pINan/ pINca 1
Outlet Gas Composition xOUTgas
Air Inlet/Outlet Relative Humidity φINair/ φOUTair
Anode Outlet Oxygen Content xOUTO2
Table 3: Variables to measure

2.3 Implementation of tests campaigns
To perform the test campaign test benches have been adapted so that EIS, lambda sensor, humidity sensors and gas sampling would be available and each test station. Special attention has been given to sealing in the case of SC tests.
Owing to major difficulties encountered to obtaining satisfactory reproducibility and good quality results for data analysis with reformate, it has been decided to focus on the identification of characteristic signatures of one dysfunction only, increased fuel utilization. Hydrogen was used as a fuel in these tests and reformate in the third test campaign. Both protocols, namely protocol 1 (increasing the current density at constant gas flow rate inlet,) or protocol 2 (decreasing gas flow rate inlet at constant current density) have been reproduced several times. Continuous measurements included standard voltage, pressure, and temperature measurements and also more advanced oxygen partial pressure measurements both on anode and cathode sides, water vapour pressure measurement. Polarisation curves and Electrochemical impedance spectroscopy was done on each stabilisation plateau as recommended.
A total of three test campaigns have been achieved during the project. Each test has been summarized in a memo and its measurement data transferred for in depth data analysis. Successful tests are summarized in Table 4:
Protocol 1: Increase of current density Protocol 2: Decrease of flow rate
H2 Reformate H2 Reformate
Single Cells 10 3 7 4
SRU 2 5 4 2
Short Stack 2 3 4 1
Full stack 2 stacks (25 cells)*
System 1 full system**
Table 4: Summary of the 3 test campaigns on high fuel utilization operation * test aborted before FU protocol could be achieved, ** protocol adapted to increase FU at system level

First experimental iteration has been run on single cells, SRUs and short stacks to study the signature of high FU under hydrogen. It was restricted to one type of cell (ASC from SOFCpower). Major trends are summarized hereafter:
• The fuel utilization rate was found to clearly affect the steady state voltage as well as the IV- and EIS-characteristics.
• Results achieved with single cells, SRUs and short stacks showed similar behaviour when FU was increased, i.e. low frequency arc in EIS-spectra increased and mass transport losses decreased cell voltage in IV-curves.
• EIS-curves measured with SRUs and short stacks were similar at low fuel utilizations (0.6 and 0.7) indicating a good reproducibility. Higher FU-values were more challenging for short stacks, and FU=0.9 was not reached with a short stack. This is typically an indication of uneven flow distribution.
The second test campaign aimed at ensuring reproducibility of first identified signatures. Similar results could be obtained for single cells, SRUs and short stacks. On single cells, operation up to FU=0.9 appeared to have no damaging impact (irreversible degradation). Thanks to monitored data analysis using both frequencies and time scales, a possible signature of high FU operation could be confirmed.
A typical single cell voltage evolution with time during the two protocols simulating increasing FU is presented in Figure 2. Operation temperature is 750°C. As fuel utilisation is increased, the voltage drops. This voltage drop is more pronounced when the current density is increased maintaining the fuel flow constant than when decreasing the fuel flow maintaining the current density constant.

Figure 2: Example of single cell voltage evolution with time during the two protocols for simulating increasing FU.
Synthetic reformate is used.

The third test campaign was achieved under synthetic reformate (CH4=0.15 H2O=0.3 H2=0.29 CO2=0.26). Some tests at increased temperature (800°C) compared to nominal (750°C) have been achieved and some commercial cells have been tested in addition to those of SOFCpower. These tests were decided to extend the application domain of identified signatures.
Under reformate some difficulties have been encountered regarding reproducibility between single cells, SRUs and short stacks. Operation up to FU=0.8 appeared to have no damaging impact. After operation at FU=0.9 a slight degradation has been detected on single cells. Such high FU could not be easily reached with SRUs or short stacks. Nevertheless, similar type of signal as previously identified could be detected and analysed.
Experiments on different full size stacks and on one commercial system have been achieved. Increasing fuel utilisation on both stacks and systems in order to highlight a potential signature appeared to be rather challenging. In the case of stacks, all tests were aborted before reaching high fuel utilisation. It is worth noting that these stacks either had been designed to operate under specific gas mixture and did not withstand pure hydrogen, or were already “aged” stack that did not withstand the transient operations required by the protocol. In the case of the commercial system, the protocol had to be adapted but some interesting signals could be identified when FU was increased that have been analysed.
To complement the high FU simulation test campaigns, preliminary tests simulating gas leakage have been successfully performed on single cells. A dedicated setup has been implemented for single cell operation with four capillary tubes allowing simulating controlled leakage both in term of flux and in term of localisation. A controlled and localised O2/N2 mixture could be introduced on the surface of the cermet. Hence, its impact has been qualified and quantified on OCV and on the performances of the cell .
A clear effect of the leakage on the OCV and on the limiting current density could be highlighted. Moreover the closer the leakage to the gas entrance the bigger the effect.
Recovery strategy recommendations have also been studied in order to propose corrective measures of high FU degradation phenomena and to maximize the cell/stack lifetime. High FU can induce anode re-oxidation and generate irreversible damage. An easy way to decrease the risk of local high FU is to decrease current or to supply more fuel. However it has been shown that single cell and SRU can withstand FU up to 0.9 without any damage when back to nominal conditions. More severe high FU conditions have been carried out to reach irreversible damages and to highlight local anode re-oxidation phenomenon. This dedicated protocol could be useful to detect anode re-oxidation and to define the threshold of acceptable high FU, i.e. the maximum FU that can be applied before re-oxidizing the cermet in a given configuration. Such an approach could not be implemented in the time frame of the project.

2.4 Data Analysis Results and signature identification
During the three experimental campaigns, data sets from single cells (SCs), single repeating units (SRUs), and short stacks (SSs) fuelled either by pure hydrogen or realistic synthetic gas reformate have been analysed for high fuel utilisation detection. Moreover, the effects of varying cell technology and temperatures have been analysed, showing that degradation signatures seem to not be strongly influenced by different cell technology or small furnace temperature variations, whenever these signatures are available. Also some commercial stacks have been tested having their data analysed.
It was assumed that, after a short stabilisation period (few hours), the system (SC, SRU, SS, commercial stack) under test could be considered as in a (quasi) “steady-state” working condition for the remaining period when FU was kept constant, allowing some data analysis.
In every data set, the observed signal is the output voltage, for which different strategies of detecting degradation signatures have been implemented and tested. Three possible degradation signatures have been identified, namely the sample voltage standard deviation, the peaks magnitude of the power spectrum (by means of spectral analysis) of the output voltage signal and the continuous wavelet transform methodology applied to the output voltage signal.
2.4.1 Analysis using the peaks magnitude of the power spectrum of the output voltage signal
The signatures for high FU faults of fuel cells derived by a spectral analysis of the output voltage are identified in many data sets, where some peaks in Power Spectral Density (PSD) of the voltage segments appear, and whose magnitude is related to the FU value. For some data sets, the acquisition interval voltage signal has been measured not uniformly in time. In order to avoid problems related to the computation of periodograms (PSD estimators) of not-equally sampled continuous signals, a spline interpolation has been adopted for the time series, and thus obtaining approximated functions. These approximating functions have been uniformly sampled. Then, all the PSDs have been computed to the time series obtained by the said spline-based approximation of the signal. A validity check of the approach based on spline approximation has been performed: it has been shown that the actual voltage and the spline approximation time series are very close to each other and the periodograms of the actual (non-equally spaced) time series and the spline-approximation-based PSD, both computed by using the Lomb algorithm , are very similar and contain the same features. Thus the spline approximations of all the measured time series have been analysed for those data sets.
Every segment of each voltage time series has been analysed by estimating the Power Spectral Density (PSD) of the signal obtained by the difference between the voltage segment and its (linear) regression model. For most experimental data, a periodic signal is found to increase with increased FU (this result has been checked with 3 different math algorithms).
A patent is pending on this diagnosis methodology.
2.4.2 Analysis using the voltage standard deviation
Another signature to reveal high FU faults is related to the variability of the measured voltage that increases as FU increases in most analysed data sets provided by partners, including those from the tests of commercial systems. Then, the sample standard deviation can be considered a good signature of FU faults and can be easily derived by a regression analysis of the voltage signal. Indeed, it increases as the FU value increases. In order to highlight this behaviour by comparing the data sets of all the Design experimental campaigns, a normalized standard deviation is introduced. For each data set, the normalized standard deviation is defined as the ratio of the sample standard deviation (measured at a given FU value) and the “mean” standard deviation when the system is in nominal conditions (i.e. obtained as the mean value of the sample standard deviation computed for FU=0.6 in the same data set). The scatter plot of the normalized standard deviation for the available FU values derived for all data available in the Design project is shown in Fig. 3, together with a power-law fitting curve. By observing Fig. 3, the influence of the FU on the sample standard deviation is evident.

Figure 3: Scatter plot of the normalized standard deviation for different FU values. A power-law fitting curve has been added (solid line).

2.4.3 Analysis using the continuous wavelet transform of the output voltage signal
The data collected from the testing partners in the three test rounds were also analysed applying the Continuous Wavelet Transform (CWT) methodology. The only analysed signal was the voltage and the objective was to detect a signature between some reference operating conditions (FU=0.6) and purposely induced faulty conditions (FU=0.7 0.8 and 0.9).

The CWT was applied to the voltage signal for each step in FU, as shown in Figure 4 and to verify:
1. Qualitative difference between the signals at different FU for the same tests;
2. Correspondence between the same conditions for different tests.
Thereafter, a statistical study was done to quantify the observed difference and to find out the correspondence between CWT coefficients and operating conditions.
The qualitative analysis of the CWT coefficients was done comparing the CWT surfaces for consecutive conditions within the same test (Figure 4). The objective was to find any signature, i.e. difference between the diagrams (for the different reference conditions segments within a single test and for all the similar increased FU among the different tests). The dimensions of each CWT surface are length(a) x length(time), where a is the vector of the scale values (50 values from 4 to 1000) while the time vector refers to the analysed voltage segment.

Figure 4: Example of comparison between consecutive CWT surfaces related to consecutive segments of the signal in different FU conditions. ‘a’ is a parameter.

As illustrated in Figure 4, it is shown that the CWT surface for the reference segment FU=0.6 is different from the segment FU=0.9 and from the next reference segment FU=0.6 too. This means that the CWT result highlights different conditions, i.e. different fuel utilization. Therefore, in principle, it is possible to build a diagnostic tool based on this type of calculation to detect faults during actual operation.
In order to explore the possibility to develop a diagnostic tool based on the CWT approach, a statistical study of results obtained on single cells (SCs) was carried out. The CWT matrix coefficients were compared for the different segments j of the voltage signal for the different tests. The vectors maxj and minj were computed for each CWT matrix. These vectors contain the maximum and the minimum coefficients for each row, respectively, that means for each value of a. Thus, the vectors Wj=|maxj – minj| were calculated. Finally two analyses were done:
1. calculating the mean values for each Wj (that means for each FU segment) to avoid the dependence on the a scale; then, these values were statistically compared (for the different SC tests);
2. fixing three values of a and averaging the values of Wj over the tests. Three ranges of membership were defined for the CWT coefficients and the percentage of coefficients that falls within each range by varying the analysed segment was evaluated. The middle range was chosen around zero based on the observed values, which mainly belong to this range. The low and the high range were selected accordingly to the middle one. The results are shown in Figure 5, where only the low and high range are plotted, since the complementary percentage is the middle range and is also much higher than the other two. It can be seen that the reference segments (ref 99, 0-1, 0-2, 0-3 and 0-4) are quite similar (percentage of membership between 5 and 8% for both low and high range) except for the segment 0-3 for which the percentage is higher than 14%. Increasing the FU from 0.6 to 0.9 results in a significant increase of the percentage of membership (from 8 at FU=0.7 (1), 15 at FU=0.8 (2) to 44 at FU=0.9 (3)).
This last approach appears to be the most promising for diagnosing high FU as illustrated in Figure 5 in the case of single cell tests. Some further work is however required to validate this approach on all experimental data produced by the Design project.

Figure 5: Percentage of membership of CWT coefficients to fixed ranges.

2.5 Scaling-up and tests on stacks

As final step of the DESIGN project, tests on full stacks were carried out to verify the relevance of the degradation indicators on full stacks. High fuel utilization was taken as the critical process to be considered at stack level in agreement with the entire project implementation. The same test protocols on the variation of fuel utilization were to be applied on the full stacks, similarly the measurements already achieved on single cells, single repeating units and short stacks in WP2.
To ensure a generic dimension to the degradation indicator, several stacks have been considered from 3 different suppliers: i) Sunfire GmbH, Germany, ii) Elcogen Oy, Finland and iii) HTceramix, Switzerland.
The stacks under observation were:
- A 64 cells stack with a nominal stack power of 1.4 kW. It was equipped with block voltage monitoring as well as a full stack voltage monitoring.
- A 30 cells stack with nominal electrical power of 500 W and with full stack voltage monitoring.
- A 33 cells stack with single cell voltage monitoring a nominal stack power of 0.8 kW
- A 40 cells stack with single cell voltage monitoring a nominal stack power of 1 kW.

All experiments showed problems either with the stacks or with the test rigs:
- A problem of massive degradation with a rate of 50%/1.000 hours was faced caused by impurities in a fuel delivery pipeline.
- A test was aborted due to failure in the air feed.
- A test was aborted due to anode leakages leading to Nickel oxidations.
- A stack was completely damaged during the experiments.
The high degradation rates obtained in this project could be most of the time attributed to the specific operation conditions that were recommended, different from those specified by the manufacturers.
Some preliminary analysis of the measurement data could be achieved however. They tend to confirm at the stack level that voltage standard deviation would be linked to the change of the fuel utilization. However, the extent of degradation occurring while running the experimentations makes such an observation very preliminary and requiring additional confirmation.

Potential Impact:
3.1 Dissemination of project result

Web site: design-sofc-diagnosis.com

Publications and oral presentations:
- Reliability Engineering and System Safety 140 (2015) pp. 88–98. M. Guida, F. Postiglione, G. Pulcini, A random-effects model for long-term degradation analysis of solid oxide fuel cells
- The 5th International Conference on Fundamentals & Development of Fuel Cells - April 16-19 2013 in Karlsruhe, Germany. B. Morel, B. Sommacal, F. Lefebvre-Joud, Towards local high Fuel Utilization diagnosis in SOFCs: Experimental approach
- 11th European SOFC & SOE Forum - 1– 4 July 2014, Luzern, Switzerland. B. Morel, A. Moutte, M. Reytier, Experimental evaluation of controlled gas leakages effects in SOFC and SOE modes
Patents:
- One patent application published: Application reference: Brevet n°E.N. 14 51708 « Procédé de détection de l’oxydation d’une anode métallique d’un dispositif électrochimique » - Applicants: CEA
- One patent application pending: Application reference: pending “Methodology based on data analysis for signal signature” - Applicants: Universita Degli Studi di Salerno
Workshop organization:
One workshop has been organised jointly with the GENIUS project consortium on in February 2011 in Brussels allowing to validate most critical “insidious” phenomena leading to irreversible degradation of SOFC stack and to select those to be studied in the frame of the Design project.

3.2 Main outcome of the project
The direct output of DESIGN is a signature identified to be specific from abnormal high fuel utilization liable to produce damaging operating conditions in SOFC systems. This signature is the first step towards the establishment of efficient diagnostic methodology and contributes to narrowing the gap between laboratory test and pre-commercial systems.
To pursue this approach and implement a diagnosis tool the testing protocols developed within the DESIGN project have been shared with the consortia of two new FCH JU 2013 projects “ENDURANCE” and “DIAMOND”. Moreover, the latter is a direct continuation of the DESIGN project as it aims at implementing innovative strategies for on-board diagnosis and control of solid oxide fuel cells (SOFCs) for CHP applications.
Techno-economical barriers for wide-spread market penetrations of the stationary SOFC systems are mainly linked to:
1. Degradation (insufficient lifetime)
2. High cost of the systems
DESIGN outcomes are fully in line with the priority research activities for stationary applications envisioned until 2030 (Tools for in-situ diagnostics and operation control; Improvement of durability, reliability and lifetime of SOFC systems with the objective to reach the 40 000 h.
The accordance of the project with the MAIP and AIP targets are detailed in the following Table 5:

Programme Objective/ Quantitative target Corresponding project objectives/ targets Achievements
MAIP objectives
Achieve the principal technical and economic specifications necessary for stationary fuel cell systems to compete with existing and future energy conversion technologies Development of diagnostic tool for in situ fine tuning of cell/stack operation conditions will pave the way to SOFC stack reliability and durability as already observed with other mature competing technologies Diagnostic tools developed at single cell, SRU and short stack level
appearing to be promising but still deserving further validation
not applicable at full size stack yet
Deliver reliable control and diagnostics tools both at a component and at system level The main target of the project is to deliver a diagnostic methodology liable to serve as the basis for a diagnostic tool at the stack level Diagnostic methodology developed at component level
Include the use of multiple fuels Hydrogen and synthetic reformate are tested in the project 2 test campaigns have been performed under hydrogen and one under synthetic reformate following FCTesQA recommendations
Include a lifetime increase up to 40,000 h Early detection of identified degradation mechanisms, will allow avoiding failure by modifying operation parameters thereby increasing substantially cell/stack lifetime. In addition, recommendations will be provided for recovery strategies for later detection. Project has focused on the detection and the recovery of High FU induced degradation. The local conditions for damage appearance have been quantified and the way to recover this damage studied and discussed. A patent has been produced.
AIP objectives
Novel diagnostics to identify potential failures, including in-operation diagnostic tools for cell/stack Identification of relevant sensors and signals to be monitored to diagnose full stack degradation phenomena;
A data analysis methodology to be applied to measured signals; A new diagnostic methodology has been developed for one selected harsh operating condition (high FU) generating 2 patents (one pending) and 2 congress presentations (FDFC2013 and EFCF2014)
Table 5: MAIP and AIP objectives.

Programme Objective/ Quantitative target Corresponding project objectives/ targets Achievements
Improved prediction and avoidance of failure mechanisms A data analysis methodology to be applied to measured signals and diagnose degradation mechanisms The developed methodology appears to be most promising but has not been evaluated as a prediction tool yet.
Targets extending the DESIGN project have been included in the project “DIAMOND”

Tools for improved diagnostics and services A set of characteristic signatures for the different degradation phenomena at the local and stack level, to be compared with the actual sensor signal to diagnose long-term degradation conditions;
Development of strategies for recovery of cell and stack performance Recommendations for recovery strategies once a degradation condition is identified at the cell, SRU or stack level are an outcome of the project. As a focus has been given in the project on high FU induced degradation and recovery strategies have been tested at single cell level that remain to be up scaled

List of Websites:
www.design-sofc-diagnosis.com