CA2244808A1

CA2244808A1 - Method of monitoring co concentrations in hydrogen feed to a pem fuel cell

Info

Publication number: CA2244808A1
Application number: CA002244808A
Authority: CA
Inventors: Kirk Weisbrod; Rodney Lynn Borup; Jay Kevin Neutzler; Stephen Andreas Grot; Mark Alexander Meltser; Stanley Gutowski
Original assignee: University of California; Motors Liquidation Co
Current assignee: University of California; Motors Liquidation Co
Priority date: 1997-10-24
Filing date: 1998-08-11
Publication date: 1999-04-24
Also published as: US6063516A; US6265092B1; DE69819846D1; EP1267435A2; JP3357300B2; EP0911898B1; EP0911898A1; JPH11219717A; DE69819846T2

Abstract

The CO concentration in the H2 feed stream to a PEM fuel cell stack is monitored by measuring current and/or voltage behavior patterns from a PEM-probe communicating with the reformate feed stream. Pattern recognition software may be used to compare the current and voltage patterns from the PEM-probe to current and voltage telltale outputs determined from a reference cell similar to the PEM-probe and operated under controlled conditions over a wide range of CO concentrations in the H2 fuel stream. The PEM-probe is intermittently purged of any CO build-up on the anode catalyst (e.g., by (1) flushing the anode with air, (2) short circuiting the PEM-probe, or (3) reverse biasing the PEM-probe) to keep the PEM-probe at peak performance levels.

Description

METHOD OF MONITORING CO CONCENTRATIONS IN
HYDROGEN FEED TO A PEM FUEL CELL

The Government of the United States of America has rights in this invention pursuant to Agreement No. DE-AC02-9OCH10435 awarded by the U.S. Department of Energy.

TECHNICAL FIELD
The present invention relates to a carbon monoxide (CO) sensor, and a method for monitoring the CO concentration in the reformate fuel stream to a PEM fuel cell for controlling such concentration.
BACKGROUND OF THE INVENTION
Fuel cells have been proposed as a power source for many applications. So-called PEM (proton exchange m~"lblalle) fuel cells [a.k.a.
SPE (solid polymer electrolyte) fuel cells] potentially have high energy and 15 low weight, and accordingly are desirable for mobile applications (e.g., electric vehicles). PEM fuel cells are well known in the art, and include a "membrane electrode assembly" (a.k.a. MEA) col~lishlg a thin, proton trAn~mi~sive, solid polymer membrane-electrolyte having an anode on one of its faces and a cathode on the opposite face. The m~mb~le-electrode-20 assembly is sandwiched ~lweell a pair of electrically conductive el~mPnwhich serve as current collectors for the anode and cathode, and contain appropliate (~h~nnPls and/or openillgs therein for disllib.lling the fuel cell'sgaseous re~ct~nt~ over the surfaces of the respective anode and cathode catalysts. The channels/openings for the reacf~nt~ are often referred to as "flow channels". A plurality of individual cells are commonly bundled together to form a PEM fuel cell stack.

PEM fuel cells are typically H~-O2 fuel cells wherein hydrogen 5 is the anode reactant (i.e., fuel) and oxygen is the cathode reactant (i.e., oxidant). The oxygen can either be in a pure form (i.e., ~2)~ or air (i.e., ~2 admixed with N2). The solid polymer melllbldl1es are typically made from ion exchangc. resins such as perfluoronated sulfonic acid. One such resin is NAFI(~NTM sold by E. I. DuPont deNemeors & Co. Such membranes are 10 well known in the art and are described in U.S. Patent 5,272,017 and 3,134,697, and in Journal of Power Sources, Volume 29 (1990), pages 367-387, inter alia. The anode and cathode themselves typically comprise finely divided carbon particles, very finely divided catalytic particles supported on the internal and external surfaces of the carbon particles and proton 15 conductive resin h~tell~ lgled with the catalytic and carbon particles. One such membrane electrode assembly and fuel cell is described in U.S. Patent 5,272,017 issued December 21, 1993 and assigned to the assignee of the present invention.

The hydrogen used in the fuel cell can be derived from the reformation of m~th~nnl or other organics (e.g., hydrocarbons).
Unfol~una~ely, the lefor.nate exiting the ~fc,l.~er contains undesirably high concentrations of carbon monoxide which can quickly poison the catalyst of the fuel cell's anode, and accordingly must be removed. For example, in the - 25 m~th~nol reroalion process, m~th~nnl and water (as steam) are ideally reacted to generate hydrogen and carbon dioxide according to the reaction:

CH30H + H2O~CO2 + 3H2 This reaction is accomplished heterogeneously within a chemical reactor that provides the n~cess:lry thermal energy throughout a catalyst mass and actually yields a reformate gas comprising hydrogen, carbon dioxide, carbon monoxide, and water. One such reformer is described in U.S. Patent No.
4,650,727 to ~anderborgh. Carbon monoxide (i.e., about 1-3 mole %) is contained in the H2-rich refollllate/effluent exiting the refc,lllRr, and must be removed or reduced to very low nontoxic (i.e., to the anode) concentrations (i.e., less than about 20 ppm) to avoid poisoning of the anode by adsorption onto the anode catalyst. The unreacted water serves to hllmi-lify the fuel gas 10 and prevent drying of the MEA.

It is known that the carbon monoxide, CO, level of the reformate/effluent exiting a mPth~nnl reformer can be reduced by l~tili7.ing a so-called "shlft" reaction. In the shift reactor, water (i.e. steam) is injected15 into the methanol reformate/effluent exiting the reformer, in the presence of a suitable catalyst, to lower its telll~eld~ure, and increase the steam to carbon ratio therein. The higher steam to carbon ratio serves to lower the carbon monoxide content of the lefollllate according to the following ideal shift reaction:

CO+H20~C02+H2 Some CO survives the shift reaction and remains in the refolllldte. Depending upon the reformate flow rate and the steam injection rate, the carbon 25 monoxide content of the gas exiting the shift reactor can be as low as 0.5 mole %. Any residual ll.r-Lh~ l is converted to carbon dioxide and hydrogen in the shift reactor. Hence, shift reactor effluent comprises hydrogen, carbon dioxide, water and some carbon monoxide.

The shift reaction is not enough to reduce the CO content of the reformate enough (i.e., to below about 20 ppm). Therefore, it is n~cess~ry to further remove carbon monoxide from the hydrogen-rich lefo""aLe stream exiting the shift reactor, and prior to supplying it the fuel cell. It is known to 5 further reduce the CO cont~nt of H2-rich l~fo""aLe exiting the shift reactor by a so-called "PROX" (i.e., ~rer~renLial oxidation) reaction effected in a suitable PROX reactor and can be either (1) adiabatic (i.e. where the temperature of the catalyst is allow d to rise during oxidation of the CO), or (2) isothermal (i.e. where the tem~e,aLu,e of the catalyst is m~int~in~d 10 substantially constant during oxidation of the CO). The PROX reactor comprises a catalyst bed operated at te"~l)e,aLu,es which promote the plefele~Lial oxidation of the CO by injecting controlled amounts of air into theeffluent from the shift reactor to consume the CO without co~ g/oxidizing substantial qll~ntities of the H2. The PROX reaction is 15 as follows: -CO+1/202-~ C~2 Desirably, the ~2 required for the PROX reaction will be about two times the 20 stoichiometric amount required to react the CO in the reformate. If the amount of ~2 iS substantially less than about two times the stoichiometric amount n~eded, insufficient CO oxidation will occur. On the other hand, if the amount Of ~2 exceeds about two times the stoichiometric amount n.oeded, excessive consu",~Lion of H2 results. Col~u~ Lion of the H2 raises the 25 temperature of the gas, which in turn causes the formation of CO by the reaction of H2 with CO2, known as the reverse gas-shift reaction. Hence, careful control of the amount of air injected in the PROX reaction is essential to control the CO content of the r~ro"nate feed stream to the fuel cell. The PROX process is described in a paper entitled "Methanol Fuel Processing for Low Temperature Fuel Cells" published in the Program and Abstracts of the 1988 Fuel Cell Seminar, October 23-26,1988, Long Beach, California, and in U.S. Patent Vanderborgh et al 5,271,916,intPr~

Whether an adiabatic or isothermal PROX reaction, a controlled amount of ~2 (i.e., as air), is mixed with the refolll.ate exiting the shift reactor, and the ll~i~Llulc passed through a suitable PROX catalyst bed known to those skilled in the art. To control t~le air input rate, the CO
concentration in the gas exiting either the shil~ reactor or the PROX reactor ismeasured, and based thereon, the ~2 collcelllldlion needed for the PROX
reaction adjusted. However, sensitive, real time, CO sensors have not heretofore been available, and accordingly system response to CO
concentration variations has been slow. This is particularly troublesome in dynamic systems where the flow rate, and CO content, of the H2-rich reformate vary continuously in response to variations in the power demands on the fuel cell system. Since the amount of ~2 (e.g., air) supplied to the PROX reactor must vary on a real time basis in order to accommodate the varying power clem~n-ls on the system, there is a need for a rapid response CO sensor to continuously monitor the CO in the lefolmate stream and thererlo.ll(1) m~int~in the proper oxygen-to-carbon monoxide concentration ratio in the PROX reactor, and/or (2) divert the lefo.-l~a~e stream away from the fuel cell until the CO content thereof falls within acceptable levels.

SUMMARY OF THE INVENTION

The present invention provides a sensitive CO sensor utili7.ing a mini PEM fuel cell as a probe, and a method for real time monitoring of the CO concentration in the reformate feed stream to a PEM fuel cell as a means to control the operation of the fuel cell system. In accordance with the present invention, the sensor is repeatedly refreshed by purging any CO
thererl~lll to m~int~in the CO sensitivity of the sensor. CO purging may be effected chemically or electrochemically as described hereinafter. The 5 invention is useful during system start-up to delelll~il~ when the CO level ofthe PROX effluent is sufficiently low that such effluent can be directed to the fuel cell without poisoning the anode catalyst. The invention is particularly useful for the real-time control of the amount Of ~2 (i.e., as air) supplied to the PROX reaction in response to the CO concellLl~lion il~ the H2 gas stream 10 exiting the PROX reactor so as to m~ximi7e CO col~um~lion while mi~ ing H2 col~ulllpLion in the PROX reactor. The CO concentration in the reformate may be measured at various locations in the lefolmate fuel stream to a fuel cell (e.g., after the reformer, shift or PROX reactions).

In accordance with a ~lefell~d embodiment of the present invention, there is provided a CO sensor comprising a PEM-probe, and a method of using the PEM-probe to m~int~in its sensitivity and provide real time control of the CO content of the refollllate fuel stream to a PEM, H2-O2 fuel cell stack. The PEM-probe is essenti~lly a mini PEM fuel cell which, 20 like the stack's cells, has an anode and cathode affixed to opposite sides of a proton exchange melll~ e and a hydrogen flow channel collrloll~ g the anode that receives hydrogen from the hydrogen-feed manifolds supplying the stack. The PEM-probe's anode will preferably have a smaller area and a lower catalyst loading (i.e., g/cm2) than the stack's cells for increased CO
25 sensitivity compared to that of the stack itself. Most preferably, the surface area of the PEM-probe's electrodes will be less than about 10% that of the stack's electrodes, and the catalyst loading will be about half the catalyst loading in the stack's cells. Moreover, in accordallce with the present invention, sensilivily of the PEM-probe is enh~n-~d even further by intermittently purging the probe's anode catalyst of any CO that might have become adsorbed thereon while monitoring the'reformate gas fed to the fuel cell. The frequency of purging is such as to m~int~in the catalyst in a substantially CO-free, or near CO-free, condition where the probe is most 5 effective and responsive in detecting CO buildup on its catalyst over short intervals. In this regard, the probe is quite effective/responsive during the early stages of co~ ion, but less so as the probe becomes more and more col,l~",i"~ted with CO. CO purging will preferably be effected b~-raising the anode potential sufficiently [(i.e., to at least 0.8 V measured 10 against a reversible hydrogen electrode (RHE)] to electroch~mi~-~lly oxidize any CO on the catalyst to CO2 by reaction with the water present in the fuel stream. This may be accomplished by reverse biasing or short ch~;uiLillg the PEM-probe, as described hereinafter. Alternatively, the probe may be flushed with ~2 (e.g., air) to chemically oxidize the CO.
The pl~felled CO sensor includes means for effecting the intermittent electrochemical purging of the PEM-probe to remove adsorbed CO. In one embodiment, the CO sensor comprises: a gas-moni~olhlg PEM-probe including a proton exchange membrane having an anode and a cathode 20 affixed to opposing first and second surfaces of said memb~ e; a first electrical current collector eng~ging the anode; a second electrical current collector eng~ging the cathode; an electrical discharge circuit connectable between the current collectors, wherein the discharge circuit has a first electrical lesis~llce valued for discharging the PEM-probe at a first rate - 25 selected to monitor the degrading output of the PEM-probe incident to CO
cont~min~tion of the anode; an electrical p~u~Sing circuit connectable between the current collectors, wherein the ~uiging circuit has a second electrical resistance which is less than the first electrical resistance such that upon - discharge of the PEM-probe through the second resistance the ~ole,llial of the anode is raised to at least 0.8 V (RHE) to effect electroch~ r~l oxidation of any CO adsorbed on the anode; and an electrical switch in electrical series connection between the current collectors and adapted to inte,n,il~cnlly, alternately electrically cormect the current collectors to the discharge and theS purging circuits. In this embo~iment, the sensor will also preferably include a motorized valve for shlltting off H2 flow to the PEM-prohe during the purging stage. Most preferably, the switch for switching between the discharge and the purging circuits will be built into the H2 shut-off valve for ~im~ n~ous stopping of the H2 flow to the probe and connPcting it to the purging circuit 10 during the purging cycle and vice versa during the discharge cycle.

In another, and most plerc"ed embo~imPnt, a CO sensor is provided that comprises: a gas-mt~ lo~hlg PEM-probe including a proton exchange membrane having an anode and a cathode affixed to opposing first 15 and second surfaces of said mcmblal~e; a first electrical current collector eng~ging the anode; a second electrical current collector eng~ging the cathode; an electrical discharge circuit connectable between the current collectors; the discharge circuit having a first electrical resislal1ce valued for discharging the PEM-probe at a rate selected to monitor the degrading output 20 of the PEM-probe incident to CO co"~;~"~ lion of the anode; an electrical purging circuit connectable between the current collectors and in~ ing a voltage source that imposes a reverse electrical bias on the PEM-probe sufficient to raise the potential of the anode to at least about 0.8 V (RHE) to effect electroch~ l oxidation of any CO adsorbed on the anode; and an 25 electrical switch in electrical series connection ~l~ n the current collectors and adapted to hlt.,.lllil~e..lly, alternately col-n~ the contacts to the discharge and purging circuits. This embodiment is seen to permit the qllickest and most controllable purging of the anode, without the need to shut off the H2 flow.

In accordance with the process of the present invention, the PEM-probe is intermittently purged of any CO buildup on its catalyst.
Between such purgings, the current and/or voltage outputs of the probe is/are monitored and compared to reference standards to deLe~ e the CO
concentration in the lefollllate (e.g., PROX effluent). More specifically, the process invention contemplates:

a. providing a CO sensor including a monitoring PEM-probe comprising a proton exchange membrane having an anode and a cathode affixed to opposing first and second surfaces of the membrane wherein the anode comprises a catalyst which is susceptible to poisoning incident to the adsorption of CO by the catalyst and consequent progressive degradation of the catalyst from a peak performance level in the early stages of CO adsorption to a poor pelrolm~lce level at later stages of such adsorption;

b. contacting the anode with a portion of the H2 feed stream to the fuel cell over a plurality of predetelll~il~ed time intervals;

c. cont~ting the cathode with oxygen;

d. dischal~ lg the PEM-probe during the time intervals;

e. monitoring the electrical output from the PEM-probe during the discharging to genelale an output signal having a behavioral pattern indicative of variations in the CO
concentration in the feed stream;

f. from a leference PEM-probe similar to the monitoring PEM-probe, clel~ . .,.il.;.. ~ a pluMlity of telltale electrical outputs which are correlated to known CO
concentrations in the feed stream;

g. storing the telltale electrical outputs in a readable memory;

h. comparing the output signal from the monitoring sensor to the telltale electrical outputs from the ~efeie"ce PEM-probe to identify a telltale electrical output that is subst~nti~lly similar to the behavioral pattern to ~e~e~ the CO
concentration in the feed stream; and i. periodically, ~ul~ing the catalyst of the CO between the time intervals to ll~ the catalyst at ~ubsL~llially its peak performance level.

Once the CO concentration has been de~- ---i--Pd, a ~I~L~ ion can be made as to what adjustments to the system are required. Hence for example, in one scenario, the ~2 injection rate to the PROX reactor may be varied, or in ~ 25 another scenario, the PROX effluent may be directed away from the fuel cell stack until its CO content falls within acceptable lirnits (i.e. below ca. 20 PPM) BRIEF DESCRIPTION OF THE DRAWINGS

The invention will be better understood when considered in the 5 light of the following detailed description thereof which is given helear~el in conjunction with the following drawings of which:

Figure 1 is a schematic illustration of a bipolar, PEM fuel cell stack, and a prefe~led CO sensor therefor in accordance with the present 10 invention;

Figure 2 illustrates an exploded, pelspecti./e view of a PEM-probe according to one embodiment of the present invention;

Figure 3 is a sectioned view of the PEM-probe of Figure 2; and Figure 4 sch~m~tically depicts another embodiment of a CO
sensor in accordance with the present invention.

Briefly, the sensor of the present invention preferably monitors the current through, and voltage across, a col~lalll load conn~cted to the PEM-probe which is connected to the hydrogen fuel feed manifold to the fuel 25 cell stack for testing the gas therein. A voltage sensing device senses any voltage fluctuations across the co~ t load over a predelt, Illil-~d time interval and outputs a voltage signal which leplesell~ the behavior pattern of the voltage fluctuations over that interval. A current sensing device senses any current fluctuations through the constant load over a prede~ mil~ed time interval and outputs a signal which represents the behavior pattern of the current fluctuations over that interval. A first data processing device serves as a data acquisition unit, and, using conventional technology, samples the current and voltage signals, conditions the signals to filter out signal noise, 5 and converts them to digital data streams. A suitable memory device stores telltale voltage and current outputs which have been correlated to known CO
concentrations at various tempel~ules and ples~ules in a gas like that of the reformate. In this regard, the telltale outputs will have been previously generated empirically from a ,crclcllce cell which is similar to the PEM-probe 10 and has been discharged in a manner similar to the PEM-probe (e.g., has the same resistance thereacross as the load conn~cted to the PEM-probe). The reference cell is operated over a wide range of known CO-concentrations in the H2 feed stream to develop a library of telltale current and/or voltage outputs corresponding to dirÇtlcll~ CO-concentrations. Finally, a second data 15 processing device (e.g., a personal colll~ulel) receives the digital data streams, plots a curve of the behavior pattern of the voltage and current fluctuations from the PEM-probe over a given time interval, and compares those voltage and current behavior pallt~lllS to the telltale voltage and current outputs delelll~ined from the reference cell in order to match, or otherwise 20 identify, at least one of the telltale outputs that is subst~nti~lly similar to the behavior pattern being compared. Preferably, the comparison methodology and telltale outputs are those described in copendi~lg U.S. patent application U.S. Ser. No. 08/807,559 filed February 28, 1997 in the name of M.
Meltzer, and ~sign~-d to the ~csjgn~e of this invention, which is intenl1ed to 25 be herein incorporated by rcrerellce. .Al~ lively, rather than plotting the entire behavior pattern and telltale outputs, an abbreviated relationship between the behavior pall.,llls and the telltale outputs can be used. For example, the starting and ending voltages are determined for the beginning and end of a predel~ d incre,lRnt of time and the voltage changes over that increment assumed to vary linearly with time. The reference cell's telltale voltages corresponding to known CO-concentrations in the H2 stream are delelll~illed in the same manner. The slopes of the two curves are then compared. In either event, a substantial match belweell the PEM-probe's S output pattern and a telltale output from the refelel~ce cell in-1ic~tes the real time CO-concentration in the hydrogen-feed stream which is then used to trigger adjustments to the reformer, shift and/or PROX reactions to reduce the CO cc ltent of the H2 feed stream to the fuel cell stack, or to divert the reforfnate stream away from the fuel cell stack, ff n.ocess~ry.
More specifically, Figure 1 depicts a stack 2 of individual fuel cells 4 each comprising a MEA 6 having a proton conductive resin membrane 8 sandwiched between an anode 10 on one surface thereof and a cathode 12 on the opposite surface thereof. A cathode flow channel 16is provided adjacent the cathode 12 for flowing an oxygen-rich gas (i.e., preferably air) by, and into contact with, the cathode 12. Similarly, an anode flow channel 14is provided adjacent the anode 10 for flowing hydrogen fuel by, and into contact with, the anode 10. The lllelllbl~ne 8 will preferably comprise a perfluoronated sulfonic acid polymer such as NAFIONTM as is well known in the PEM fuel cell art. Each individual cell 4 is sepalaled from the next cell 4 in the stack by a bipolar plate 18, which is a conductive plate (e.g., metal, carbon, etc.) which sepal~t~s the several cells one from the next while con~ cting electrical current in electrical series directly from one cell to thenext. End plates 24 and 26l~ ule the stack 2 and define the respective - 25 cathode and anode flow channels for the end cells 28 and 30 of the stack 2.
An oxygen-feed manifold 32 supplies air to the several cathode flow channels 16. Similarly, an hydrogen-feed manifold 34 supplies hydrogen fuel to the several anode flow channels 14. An hydrogen exhaust manifold 36 collects anode exhaust gas from the several anode flow channels 14 for discharge from the stack. Similarly, a cathode exhaust gas manifold 38 collects exhaust gas from the cathode flow chaMels 16.

Stack performance degrades due to carbon monoxide poisoning 5 of the anode catalyst. Such poisoning is a potential problem when there is excess (i.e., more than about 20 PPM) CO in the hydrogen-feed stream which can result from inefficient m~th~nnl/hydrocarbon rcfolll~illg, shift and/or PROX reactions known to artesans skilled in this art. Accordingly, when the presence of excess CO in the H2 fuel stream is evident, efforts must be made 10 to correct the problem, preferably at its source. To this end, the present invention provides a sensitive, rapid response carbon monoxide sensor (CO-sensor) 40, and method of operating same, which senses CO concentration in the reformate fuel stream in the manifold 34. CO-sensor 40 includes a probe 41 (hereinafter PEM-probe) which is naught but a small (i.e., mini) PEM fuel 15 cell similar to the cells 4 in the stack 2, except for size and possibly catalyst loading. While monitoring the fuel stream in manifold 34, the PEM-probe 41 is discharged in such a maMer as to output an electrical signal whose behavioral pattern over time is dependent on the CO concentration in the rcfc.llllate fuel stream. The output signal behavior pattern is compared to 20 certain telltale outputs from a reference PEM probe identical to the gas-monitoring PEM-probe which have been correlated to known concentrations of CO in H2 at various tel~e,alu~es and p~s~u,.,s. Conventional pattern recognition technology is plefell.,d for reliably coll~ g the PEM-probe's 41 output(s) to the telltale output(s) of the ,cr. ,ence cell. However, less 25 sophisticated telltale outputs (e.g., appro~,mdte slope of voltage degradation curve) may also be used. Mol~ilolhlg the electrical pelrollnallce of the PEM-probe 41 and colll~alillg it to the expected pe,roll~lallce under known CO
concentrations conditions provided by the reference PEM probe provides a direct knowledge of the CO collcell~ tion in the lcfo~ ate feed stream to the fuel cells 4 comprising the stack 2. From this knowledge, needed corrections can be made to the reformer, shift or PROX reactions to bring the CO
concentration to within acceptable levels. AlLellla~ively, the fuel stream can be diverted away form the fuel cell stack until the CO content thereof is 5 corrected to within acceptable limits. To control the PROX reaction: (1) the CO concentration is measured at a given ~2 injection rate to the PROX
reactor; (2) the ~2 injection rate is increased and the CO col~cell~la~ion againdeteln~illed; and (3) if the CO ~oncentration goes down, too little ~2 iS being injected and if it goes up, too ~nuch ~2 iS being injected. The process is 10 repeated at dirrelenl ~2 injection rates until oylh~i~a~ion is achieved.

Like the cells that comprise the stack 2, the PEM-probe 41 includes an anode 42 and cathode 44 on the opposite surfaces of a proton exchange membrane 50 (see Figures 2 and 3). Conventional conductive diffusion 43 and 45 contact the anode 42 and cathode 44 lesye-;lively. Such material may comprise carbon paper, fine wire mesh, sill~eled porous metal (e.g., ~i~niulll or niobium). The PEM-probe 41 includes an anode flow channel 46 in the housing 54 which co~ ir~les with the hydrogen-feed manifold 34 via applopliate flow passages (e.g. inlet 49 and/or conduit 48), as 20 well as the hydrogen exhaust manifold 36 via outlet 51 and conduit 53. The cathode 44 is exposed to ambient air via openillg 52 in the PEM-probe's housing 54. Ambient air operation keeps the PEM probe l~ ,llyel~lul. low without external cooling, which increases the CO sel~i~ivi~y of the PEM-probe. A perforated metal current collector 49 contacts the carbon paper 45 25 and conducts current to terminal 47 thereof which exits the housing 54 through slot 55. Preferably, the PEM-probe 41 will have a lower catalyst loading than the stack cells 4 to increase its sensi~ivi~y to low CO-concentrations. Most preferably, the stack cells 4 will have anodes 10 and cathodes 12 whose surface areas are much greater (e.g., as much as 10 times greater) than the surface areas of the anode 42 and cathode 44 of the PEM-probe 41. This small area, coupled with lower catalyst loadings, provides a PEM-probe with heightened CO-concentration sensitivity. By way of example, an H2-O2 PEM fuel cell stack 2 having individual cells 4 with about 100 in2 of electrode area, can be effectively monitored with a PEM-probe 41 having an electrode area of about 1 in2 to 2 in2, and about one half the catalyst loading (i.e., g/cm2) of the stack cells 4. The conduits 48 and 53 may include valves 57 and 57' for isolating the probe 41 from the H2 manifolds 34 and 36 during purging, if desired. The conduits 48 and 53 may also include air inlet 10 59 and outlet 61 respectively with associated valves 63 and 64 for controlling the flow of purging air through the PEM-probe 41 (i.e., when the air purging embodiment is used).

Because of its small size and/or low catalyst loading, the anode 15 catalyst of the PEM-probe poisons at a faster rate then the fuel cell stack it is monitoring. Hence, the degradation rate of the electrical outputs of the PEM-probe is greater than that of the stack and provides a more demollstlali~e in-lir~tor of CO concentrations in the H2 fuel stream to the stack. However, the PEM-probe becomes progressively less sensitive to CO concentration 20 variations as it becomes more poisoned. In this regard, the peak pe,ro"~ ce level of the catalyst is at the point where it has substantially no, or very little, CO adsorbed therein, and the poorest pclro~ ce when a substantial amount of CO is adsorbed on the catalyst. In accordallce with the present invention, sensitivity of the PEM-probe is m~int~in~d near its peak pe,ro""ance level by 25 int~ ~iuently purging the PEM-probe's anode catalyst of any CO that might be adsorbed therein. Preferably, y~llgillg will be accomplished by raising the anode potential to a level sufficient to electroch~mirally oxidize the CO to CO2 in the presence of water. Typically, this requires raising the anode potential to at least 0.8 V as measured against a reversible hydrogen electrode (RHE), and may be accomplished by (1) periodically short cir.;uiling the PEM-probe as by discharging it through a low lesi~ r-e load, or (2) most preferably, by periodically reverse biasing the PEM-probe by means of a supplementary voltage source. Allclnalively, CO purging may be effected by S flushing the anode catalyst with oxygen (e.g. air) to ch.o.n~ir~lly oxidize the CO. The CO sensitivity of the PEM-probe may also be illclcased by cooling the H2 stream by means of a heat exchange (not shown) inserted in feed line 48 to the PEM-probe 41. Cooling to about 20~C-90~C is useful to condense out excess water which impedes sensitivity of the probe The most prcfcll~,d purging technique is to reverse bias the PEM-probe, as this technique is seen to be most easily controlled and performable without ~hll~ting off the flow of gas to the PEM-probe. To this end, a CO sensor 40 (see Figure 1) is provided which includes a PEM-probe 15 41, a voltage source 78 (e.g. a capacitor, or voltaic device such as a battery or connection to one or more cells 4 of the stack 2) in a purging circuit P, a motorized switch 80, and a discharge circuit D. The switch 80 will preferably be coupled to a timer or clock which periodically ~wilches the PEM-probe between a discharge mode through load L of the discharge circuit 20 D, and a reverse biased mode in the ~ulghlg circuit P. The ~ ging circuit P
also includes a small (e.g., about 0.5 ohm) resistor to avoid a high current surge upon switching between circuits D and P. More speciffcally, the PEM-probe 41 is coupled to a conslalll load L in a discharge circuit D (see Fig. 1).A voltage sensing device 65 (e.g., voltmeter) senses the voltage across the 25 load L while a current sensing device 67 (e.g., ~ lle~e ) senses the current flowing in the discharge circuit D. The PEM-probe 41 will typically operate with closed circuit voltages of about 0.4-0.9 volts and ~;Ull'e.llS densities ofabout 0.1 to 1.0 amps/cm2. The voltage sensing device 65 may be any such device as is well known in the art and is capable of uu~ a signal 58.

The current sensing device 67, on the other hand, may either be (1) a discrete such device as is well known in the art and is capable of uul~ullillg a signal 60, or (2) may be the voltage sensing device 65 from which current can autom~tir~lly be calculated using Ohm's law. The output signals 58 and 60 of 5 voltage sensing device 65 and current sensing device 67 respectively are inputted into a conventional high speed analog-to-digital converter 62 (i.e., data acquisition unit) which conditions the signal to elimin~te noise, and generates digital data streams 64 and 66. A prefclled such high speed converter useful with pattern recognition technology a SCU4 data ~cquisition 10 system sold by Generic Instruments and Systems Corporation (GenL~STM), as it is capable of reading inputted data, and making all needed com~ulations, in real time.

In accordance with another embodiment of the invention, the 15 PEM-probe is deprived of H2 and essentially short circuited or discharged through a relatively low resistance so as to raise the anode potential up to theoxidation potential of the CO (i.e., 0.8 V RHE). To this end, a CO sensor is provided, as shown in Figure 4, which comprises a PEM-probe 82 like that described above, a normal discharge circuit 84 discharging through load 86, a 20 short-circuiting purging circuit 88, a motorized switch 90 and a motorized valve 91 for cutting off the H2 during purging. The motorized switch 90 and motorized valve 91 will preferably be coupled to a timer or clock which periodically switches the PEM-probe bclw~ell (1) a normal CO-monitoring discharge mode through load 86 of the dischalge circuit 84, and (2) a rapid 25 discharge mode effected by short cil~;uiling the PEM-probe in the ~ulgillg circuit 88. Most preferably, the H2 cutoff valve 91 and the switch 90 will be integrated into the same structure/device such that H2 cutoff and switching between discharge and purging circuits are effected simlllt~n~ously.

When using the reverse biasing embodiment, the anode catalyst will preferably comprise platinum black, and the diffusion layer will preferably comprise a porous metal in order to survive the reverse polarity reaction. For the other embodiments, carbon-supported pl~tinllm catalyst and 5 a carbon/graphite diffusion layer may be used.

The current and the voltage of the PEM-probe are preferably both sampled on a regular basis (e.g., every 10 to 100 milli~econds) during a specified discharge interval that can vary from about 100 milli~econds to 10 about 10,000 milliseconds. The reslllting signals 58 and 60 are conditioned by the converter 62, and the average voltage and current are plotted over that interval of time. These plots depict the behavior p~ lls for the voltage and the current outputs over that time interval. These behavior pall. l,ls are inputted as data streams 64 and 66 into the data processor 68 where they are 15 compared to predetermined reference current and/or voltage telltale outputs stored in memory 70. Operating conditions of the stack (stack operationals) such as fuel/air stream telllpela~ure and ples~ule (i.e., taken from sensors notshown) are also inputted to the data processor 68 to insure that the proper telltale voltage and/or current are selected from the library 70 for a given 20 voltage/current behavior ~ul~ulled from the sensor 40. The rcrerellce voltageand current telltale outputs are empilically dc~lll~illed before hand at various~elllpcl~lules and plessulcs from a lefelellce cell which (1) is similar to the PEM-probe 41, (2) is discharged through a constant load having the same value as the constant load L of the CO-sensor 40, and (3) is operated over a 25 wide range of carbon monoxide collce~ alions in the H2 feed stream. A large library of such telltale outputs is stored in the memory 70, and is available for the comparison to the voltage and current behavior patterns produced by the PEM-probe 41. The voltage behavior pattern and the current behavior pattern of the PEM-probe 41 are colllpaled to each of the many lcrel, nce voltage and current telltale outputs on file in the memory 70 until at least one of the reference current and/or voltage sign~lres closest to the behavior pattern of the PEM-probe's current is identified, and/or one of the refelence voltage signatures closest to the behavior pattern of the PEM-probe's voltage is 5 identified. Once a "match" is made between a rcr~lellce telltale output and a behavior pattern, the CO-concentration in the H2 feed stream is delelmil~ed from which adjustments can be made, as n~e~ed A perfect match between the behavior p~U~lllS and the telltale outputs is not n~cess~ry. Rather, a suitable match will be found if the telltale output is subst~nti~lly similar to the 10 behavior pattern with which it is being colll~a,ed. By "subst~nti~lly similar"
is meant a degree of similarity that falls within certain pattern recognition tolerances that the stack designer or operator can include in the pattern recognition software to be described hereafter. These tolerances permit a "match" to be made even though the signature and the pattern are not 15 identical.

The data processor 68 includes a common digital CO~ U~ with associated read-only memory (ROM), read-write random access memory (RAM), electrically prog,~lllmable read-only memory (EPROM), memory for 20 storing a library of predele~ d reference current and voltage ~ign~hlres for co~llpa~ g to voltage and current paU. llls produced by the PEM-probe 41, and input/output sections which interface with the A-D convellel 62 and the PROX control 72 that controls the air injection rate to the PROX reactor by means of control signal 74 to a controllable injector 76, or the like. The read-25 only memory (ROM) of the digital colll~ul~l contains the instructionsn~cess~ry to implement the basic input/output instructions. The electrically programmable read-only memory (EPROM) contains the instructions n.ocess~ry to implement the data processor's own internal control, data manipulation, and co..~"~ ion algorilllllls. The processor 68 communicates with the A-D converter 62 and the PROX control 72 by means of any applopliate co~ ic~tion network protocol, many of which are known in the art. A standard 486 or Pentium col"~ul~l with 16 meg of RAM, Running Windows~) 3.1 or Windows~ 95, and fitted with an ACB 530 bus S control board is adequate for this purpose. A specific program for carrying out the functions of the processor 68 may be accomplished by standard skill in the art using conventional i"rollllation processing languages.

Either the complete voltage and/or current pattern from the 10 PEM-probe 41 may be used, or an abridged pattern (i.e., approximate slope of degradation curve) characterized by (1) a current and/or voltage reading at the beginning of a discharge cycle and (2) a current and/or voltage reading at the end of a discharge cycle may be used. Preferably, the complete pattern will be used, and can be recognized using commercially available pattern 15 recognition programs. Pattern recognition programs are known in the art and have been used for numerous applications such as to (1) identify sea crea~ es from their acoustic patterns, (2) identify body hormonal changes from sensor measurements, (3) identify the fracture point in a tool using vibration patterns, (4) identify land vehicles from their acoustic and seismic sig~ es, (5) 20 identify wear patterns in materials from thirlrn.oss mea~ur~ll,t,ll~, (6) identify intruders in secure areas using microwave and IR mea~u,e.,,cll~ (7) identify automotive intrusion from shock and acoustic ~dllcllls, and (8) identify faulty power-seat assemblies from acoustic pa~ lS, inter alia. ~cÇ~ d pattern recognition software for the CO co,lce"l~dlion monilo,ii~g teçhniq~le of the 25 present invention is ess,onti~lly analog pattern recognition software which, based on current and voltage mea~u,~,n~"~ taken over-the ~cir,ed time intervals, is capable of creating voltage and current behavior pd~lllS that can be compared to reÇelellce current and voltage telltale outputs within a defined tolerance range. From such co"~l)a~isons, the carbon monoxide co~celll-:alion in the H2 feed stream to the stack can be determin~cl, and based thereon n~cess~ry adjustments to the reformer, shift and/or PR~X reactions made. A
efelled such pattern recognition software is coll.lllelcially available under the name Failure/Wear PredictorTM (FWP)TM cornmercially available from the S GeIlIASTM, supra. The FWPTM software has embedded therein GENMATCHTM software (also sold by GenIASTM), which is a pro~ lnable analog pattern recognition program which can ~imlllt~nrously measure an all~illal~ number of pattern r~ules, and includes three dirrelent tolerances for addressing several features of a pattern rather than just a single feature 10 (e.g., a peak) thereof. That software consists of a template-lllatching process based on a ~efelellce signature (i.e., telltale output) created in advance from a reference cell operated under controlled conditions. It is neither amplitude-sensitive nor time-sensitive in that input signals over wide dynamic ranges (e.g., microvolts to volts taking place over periods from nanoseconds to 15 lllhluLes) are norm~li7ed to just 600 dimensionless units in amplitude (Y axis) and 2000 dimensionless units in time (X axis). Following norm~li7~tion of the signals, an accl-mlll~ted slope, known as "angl~sllrn", is computed for each of the 2000 data points of the norm~li7Pd input data while traversing the signal contour. Anglesum is proportional to the accllmlll~ted slope of the 20 curve in such a way that as the curve increases along a positive slope the anglesum increases in m~gnit~de, and as the curve decreases along a negative slope the anglesum decreases in m~gnit~lde. The pattern recognition process utilizes the anglesum values, within defined tole.allces, as defined in the rer~rellce ~ign~t~lres. In this regard, all leferellce telltale outputs contain a 25 series of intervals wheleil1 anglesum values and tolerances are used to characteri_e each interval. These intervals are the discli.l.i..~ g factors usedfor signal recognition. If the intervals from the leferellce "match" (i.e., withconsideration of all tolerances) like intervals in the behavior pattern.c from the PEM-probe, a "matchn is declared and identifir~tion is complete. The CA oi244808 1998-08-11 program uses t~,vo interval types for its recognition process: so-called "key"
and "standard" intervals. The key intervals allow phase adjllstm~nt of the reference telltale outputs to the behavior patterns from the PEM-probe as well as a first pass discrimination by the recognition process. The standard 5 intervals are then used for the le.llAi~ g recognition process. Key intervals are selected for uniqueness and serve to ",il-il"i~e search/col~alison time through the reference telltale output tl~tAh~se as well as to phase align the reference intervals with the data being identified. Hence, key intervals allow the software to quickly ascertain whether the behavior pattern contains the 10 initial characteristics required by the lcferellce signature. If the characteristics of the key intervals are found in the PEM-probe pattern, a full comparison is initi~t~d using the rem~ining standard intervals. Standard intervals are, by definition, all intervals other than the key intervals. For the PEM-probe's behavior pattern to contain the characteristics of the reference key intervals, it 15 must satisfy two criteria. First, the anglesum values of the lerelel~ce telltale outputs must match corresponding anglesum values in the PEM-probe's pattern, within the same intervals. Second, the separation (llulll~er of data points apart) of the two intervals must be the same as that in the lefelellce telltale outputs. Hence, it is both the intervals and their separation which 20 de~ ille a match.

The lcrelellce output template consists of a series of signal intervals to which both X and Y tolerances are assigned. Each telltale output can be divided into as many as 2000 segments each of which is bounded by a 25 signal m~ximnm and ".;l~i",~l", The behavior of the signal between segment boundaries is modeled by a mea~ulclllclll including amplitude change, average rate of amplitude change, and i~ll~ eous rate of amplitude change.
Tolerances can be assigned in three areas, for each segment, i.e., so-called "anglesum tolerancen, "bit tolerance" and "m~C~in~ tolerance". Bit tolerance identifies the number of elements (points) beyond the start and end points of the specified reference interval within which the m~trhing process searches for an anglesum match. For example, consider a refelellce inteNal with start and end points at data elements 65 and 135, respectively, and a bit tolerance 5 of 5. The m~t~hing process will then look at angl~sllm values in the signal pattern with start and end points of (60, 130), (61, 131), (62, 132), (63, 133),(64, 134), (65, 135), (66, 136), (67, 137), (68, 138), (69, 139), and (70, 140), when trying to match with the lefelellce inteNal anglesum. If the bit tolerance = 0, then the anglesum of the coll~s~olld-llg interval in the data is 10 compared directly to the anglesum of the corresponding inteNal in the reference pattern. Anglesum tolerance provides an allowance for variation in the anglesum values being compared. This tolerance dictates the allowable error in anglesum values between an interval in a reference telltale output and a corresponding inteNal in the PEM-probe's data set. Consider a leferellce inteNal with start point at 65 and end point at 135 with an anglesum value =
100, bit tolerance = 0, and anglesum tolerance = 5. The intervals will match if the anglesum for the signal data interval starting at 65 and ending at 135 iswithin the range of 95 ~ signal ~nglPsllm < 105. Masking tolerance stipulates the number of non-matching intervals that can exist and still provide recognition. For example, consider a reference pattern with 30 intervals and a m~cking tolerance equal to 5. If the ~ er of .eferellce inteNals found to match corresponding inteNals in the signal data set is 2 25 there is a match.
Otherwise, the leferel~ce telltale output does not match the PEM-probe's behavior pattern.
- 25 During the m~trhing process, the software moves the leÇelel1ce telltale output segment (the template) back and forth along the X-axis within the limits set by the bit tolerance. The sorlw~,e looks for a match with a data segment from the PEM-probe's behavior p~ having an ~nglesllm between the selected upper and lower tolerance limits. Essentially then, the matching process is as follows: (1) a specifled width of voltage and/or current data is extracted from the PEM-probe; (2) this data is norm~li7ed to an anglesum of 600 points, and an element composition of 2000 points; (3) the lefel~ellce telltale output template is moved across the data set from the PEM-probe; (4) 5 when a match is found with certain key intervals, the template and PEM-probe data sets are locked in phase, and each data set is jittered in phase along the X-axis looking for the anglesum match; and (5) if the number of data seg-nents specified by the m~ing tolerance is met, the PEM-probe data set is co~lsidered to match the template. When such a match is made, the CO-10 concentration in the H2 feed stream is dele~ Pd The processor 68 is programmed to pelrollll the comparisonprocess. That is to say, digiti7Pd voltage and current values from the data acquisition unit 62 are fed to the processor 68 which calculates the behavior 15 patterns thereof as I = f (t) and/or V = f (t) over a predele. Il~ Pd increment of time. These behavior patterns are then collll~aled to the reÇ~ ;el~ce telltale outputs stored in memory 70. If a behavior pattern and a lefelence telltale output substantially match (as described above), a control signal 78 to the PROX control module 72 is issued to take corrective action (i.e., adjust air 20 injection rate).

While the invention has been disclosed in terms of a specific embodiment thereof it is not intended to be limited thereto, but rather only to the extent set forth hereafter in the claims which follow.

Claims

1. A method of monitoring the CO concentration in the reformate feed stream to a PEM fuel cell comprising the steps of:

a. providing a CO-sensor including a monitoring PEM-probe comprising a proton exchange membrane having an anode and a cathode affixed to opposing first and second surfaces of said membrane, said anode confronting an anode chamber and comprising a catalyst susceptible to poisoning by said CO over time incident to the adsorption of CO by said catalyst and consequent progressive degradation of said catalyst from a peak performance level in the early stages of said CO adsorption to a poor performance level at later stages of said adsorption;

b. contacting said anode with a portion of said stream over a plurality of predetermined time intervals;

c. contacting the cathode with oxygen;

d. discharging said PEM-probe during said intervals;

e. monitoring the electrical output from said PEM-probe during said discharging to generate an output signal having a behavioral pattern indicative of variations in the CO content of said stream;

f. from a reference PEM-probe similar to said monitoring PEM-probe, determining a plurality of telltale electrical outputs which are correlated to known CO concentrations in said stream;

g. storing said telltale electrical outputs in a readable memory;

h. comparing said output signal from said monitoring PEM-probe to said telltale electrical outputs from said reference PEM-probe to identify a telltale electrical output that is substantially similar to said behavioral pattern to determine the CO concentration in said stream; and i. intermittently, purging said catalyst of said CO between said time intervals to maintain said catalyst at substantially said peak performance level to provide an accurate, real time determination of the CO concentration in said stream.

2. The method according to claim 1 including the step of flushing said anode chamber with oxygen between said intervals to chemically oxidize said carbon monoxide on said catalyst.

3. The method according to claim 1 including the step of raising the potential of the anode of said PEM-probe between said intervals to at least 0.8 V (RHE) to electrochemically oxidize said carbon monoxide on said catalyst.

4. The method according to claim 3 including the step of substantially depleting said anode chamber of reformate and short circuiting said probe to raise said potential.

5. The method according to claim 3 including the step of imposing a reverse bias on said PEM-probe to raise said potential.

6. The method according to claim 1 including discharging said PEM-probe under a substantially constant current and monitoring the voltage from said PEM-probe during said interval.

7. The method according to claim 6 wherein said voltage is the voltage of said anode as measured against a reference electrode electrically isolated from said cathode.

8. The method according to claim 1 including discharging said PEM-probe under a substantially constant voltage and monitoring the current output from said monitoring PEM-probe during said interval.

9. The method according to claim 1 including discharging said PEM-probe through a constant electrical load and monitoring both the current and voltage output from said PEM-probe during said interval.

10. The method according to claim 1 wherein said catalyst comprises platinum.

11. In a method of operating a H2-O2 fuel cell system comprising the principal steps of (1) providing a hydrogen-rich fuel gas having a first CO content sufficient to poison the fuel cell's anode, and (2) injecting controlled amounts of air into said fuel gas upstream of said fuel cell to selectively oxidize said CO with O2 in the presence of said hydrogen to produce a fuel stream for said fuel cell which has a second CO content less than said first content, the improvement comprising controlling the rate of injection of said air into said fuel gas so as to optimize the consumption of said CO from said fuel gas while minimizing the consumption of H2 therefrom by:

a. providing a CO-sensor including a monitoring PEM-probe comprising a proton exchange membrane having an anode and a cathode affixed to opposing first and second surfaces of said membrane, said anode confronting an anode chamber and comprising a catalyst susceptible to poisoning by said CO over time incident to the adsorption of CO by said catalyst and consequent progressive degradation of said catalyst from a peak performance level in the early stages of said adsorption to a poor performance level at later stages of said adsorption;

b. contacting said anode with a portion of said feed stream over a plurality of predetermined time intervals;

c. contacting the cathode with oxygen;
d. discharging said PEM-probe during each of said intervals;

e. monitoring the electrical output from said PEM-probe during said discharging to generate an output signal having a behavioral pattern indicative of variations in the CO content of said stream;

f. from a reference PEM-probe similar to said monitoring PEM-probe, determining a plurality of telltale electrical outputs which are correlated to known CO concentrations in said stream;

g. storing said telltale electrical outputs in a readable memory;

h. comparing said-output signal from said monitoring PEM-probe to said telltale electrical outputs from said reference PEM-probe to identify a telltale electrical output that is substantially similar to said behavioral pattern to determine the CO concentration in said stream, and initiate such adjustment to said concentration as might be warranted; and i. intermittently purging said catalyst of said CO between said intervals to maintain said catalyst at substantially said peak performance level to provide an accurate, real time determination of the CO concentration in said stream.

12. The method according to claim 11 wherein said adjustment includes increasing the rate of said injecting into said fuel gas to increase the rate of concentration of CO from said fuel gas by said O2.

13. The method according to claim 1 wherein said adjustment includes reducing the rate of said injecting into said fuel gas to reduce the consumption of H2 from said fuel gas by said O2.

14. A method of controlling the operation of a H2-O2 PEM
fuel cell system by monitoring the CO concentration in the reformate feed stream thereto comprising the steps of:

a. providing a CO-sensor including a monitoring PEM-probe comprising a proton exchange membrane having an anode and a cathode affixed to opposing first and second surfaces of said membrane, said anode confronting an anode chamber and comprising a catalyst susceptible to poisoning by said CO over time incident to the adsorption of CO by said catalyst and consequent progressive degradation of said catalyst from a peak performance level in the early stages of said adsorption to a poor performance level at later stages of said adsorption;

b. contacting said anode with a portion of said stream over a plurality of predetermined time intervals;

c. contacting the cathode with oxygen;

d. discharging said PEM-probe during said intervals;

e. monitoring the electrical output from said PEM-probe during said discharging to generate an output signal having a behavioral pattern indicative of variations in the CO content of said stream;

f. from a reference PEM-probe similar to said monitoring PEM-probe, determining a plurality of telltale electrical outputs which are correlated to known CO concentrations in said feed stream;

g. storing said telltale electrical outputs in a readable memory;

h. comparing said output signal from said monitoring PEM-probe to said telltale electrical outputs from said reference PEM-probe to identify a telltale electrical output that is substantially similar to said behavioral pattern to determine the CO concentration in said stream, and initiate whatever corrective action to the operation of the fuel cell that might be warranted by such concentration; and i. intermittently, purging said catalyst of said CO between said time intervals to maintain said catalyst at substantially said peak performance level to provide an accurate, real time determination of the CO concentration in said stream.

15. The method according to claim 14 including the step of diverting said stream from said fuel cell until the CO content of said stream iswithin acceptable limits.

16. The method according to claim 14 including the step of shutting said system down if warranted by said concentration.