c leyens (2000) effect of experimental procedures on the cyclic

8/13/2019 C Leyens (2000) Effect of Experimental Procedures on the Cyclic

1/22

Oxidation of Metals, Vol. 54, Nos. 3/4, 2000

Effect of Experimental Procedures on the Cyclic,

Hot-Corrosion Behavior of NiCoCrAlY-Type

Bondcoat Alloys

C. Leyens,* I. G. Wright, and B. A. Pint

Received October 18, 1999; revised February 17, 2000

A simplified test procedure was established to assess the hot-corrosion behaviorof MCrAlY-type nickel-base alloys under the influence of molten sodium sulf-ate as well as sodium sulfate/potassium sulfate salt blends. Salt-coated speci-mens were exposed to 1-hr thermal cycles at 950C in flowing oxygen forup to 500 cycles. Mass-change data of the specimens revealed a significantdependence of the corrosion attack not only on the average contaminant fluxrate, as expected, but also on the initial amount of salt deposited during eachrecoating cycle. Furthermore, deposit removal before salt recoating markedlyinfluenced the corrosion attack of the alloys. This was apparently related tochanges in salt chemistry by the dissolution of elements such as Cr from thealloy, which can shift the basicity of the salt and thus affect the extent of attack. Substituting Na for K in sodium sulfate/potassium sulfate salt blends

generally resulted in decreased attack. For K-containing salt deposits, increas-

ing the gross amount of alkali compared to sulfur resulted in increased sampleweight losses due to scale spallation. In contrast, decreasing the amount of sulfur in such deposits which contained exclusively Na as the alkali resulted ina significantly reduced corrosion attack compared to stoichiometric sodium

sulfate.

KEY WORDS: hot-corrosion testing; NiCoCrAlY-type bond coats; contaminant flux rate;biomass fuels.

*DLR-German Aerospace Center, Institute of Materials Research, 51170 Cologne, Germany.

Oak Ridge National Laboratory, P.O. Box 2008, Oak Ridge, Tennessee 37830-6156.At the time this work was performed, the author was on sabbatical at Oak Ridge NationalLaboratory.Author to whom all correspondence should be sent. email: [email protected](Christoph Leyens).

255

0030-770X001000-0255$18.000 2000 Plenum Publishing Corporation


2/22

256 Leyens, Wright, and Pint

INTRODUCTION

Although new materials for high-temperature applications mainly provide

high mechanical strength and creep resistance, it is mandatory to considertheir corrosion behavior in the anticipated environment early in their devel-

opment. Corrosion testing of materials is also of great importance when

well-established materials (or material systems including coatings) are to be

operated in environments where their behavior is not well known. As the

operating conditions for a specific material or material system can be quite

complex, the optimum test would be to run the actual component in an

existing facility or engine. However, such a test is generally extremely costly

and failure of the component may cause significant damage to the whole

operating device; thus, the component test is almost always the last step fora material to demonstrate its suitability for a specific application. Therefore,

simplified corrosion tests are extremely useful in research and development

activities in order to develop an improved understanding of the corrosion

mechanisms and in industrial applications to establish materials behavior in

a simulated environment. Unfortunately, no standardized test procedures

have yet been established, and most research laboratories, materials manu-

facturers, and their customers have developed their own tests.

Recently, the European Federation of Corrosion started an initiative

to develop guidelines and standards for high-temperature corrosion researchand testing.1 Although simplified tests cannot (and are not intended to)

simulate all aspects of actual operating conditions, standardized testing

methods would be a great help in generating comparable data. As for non-

standardized tests, there remains the question how well the results of simpli-fied tests correlate with the more complex service conditions of the realworld. The degree of correlation will, in many cases, depend on the sophis-tication of the test and on the proper control of the relevant test

parameters.2 In many cases, reasonable correlation between laboratory tests

and service experience has been achieved (for example, Ref. 3).Among laboratory tests, the discontinuous measurement of high-tem-

perature corrosion is widely used (for example, Refs. 4 and 5). The mostcommon methods are: (a) crucible tests, where the specimens are totally orpartly immersed in a molten salt; (b) salt-coating tests, where the specimens

are coated with salt prior to exposure to high temperatures, or continuouslycoated in a two-zone furnace arrangement6; and (c) burner-rig tests, wherethe specimens are exposed to a burner into which contaminants can beinjected.

Because of considerable differences in these test methods, it is not sur-prising that singular results obtained from one of the above tests methods

may differ significantly and sometimes may not even reflect the same per-formance ranking for a variety of materials. However, even when one par-ticular method is chosen, the results are strongly dependent on the exact


3/22


4/22


-phase containing precipitates of-phase, surrounded by (or mixed with)

regions of-phase containing precipitates of. Specimens ( 1.5 cmB0.1

0.15 cm) were cut from the castings and polished with SiC paper to 600 grit,such that preferred orientation of the grinding marks was avoided.

The test procedure involved 1-hr thermal cycles at 950C performed in

an automated test rig in which the specimens were attached to alumina rods

by PtRh wire hooks in a vertical tube furnace for 1-hr and cooled to room

temperature for 10 min between cycles.

Dry oxygen flowed continuously into the bottom of the tube furnace

during the exposures. With respect to the mode of delivery of contaminants

from biomass-derived fuels, these will likely arrive directly in the combus-

tion gas, so that the way in which they impinge on the hot gas-path compo-nents has yet to be determined.9 In these tests, all deposits were applied by

the periodic coating method.

Prior to testing, all specimens were ultrasonically cleaned in acetone

and methanol. For hot-corrosion tests, the specimens were coated with

water solutions of different salt compositions (Table I). For the biomass

fuel-related deposits, the higher K content was simulated by three different

KNa ratios (KNaG0, 2, and 15) and the lower sulfur content was

approximated by three different sulfur levels (2S(NaCK )G1, 0.4, 0.2). A

stoichiometric sulfur content was represented by 2S(NaCK )G1. The saltsolutions were prepared using distilled water and high-purity Na2SO4(TmG

884C), K2SO4 (TmG1076C, and dewpoint lower than the melting point),

NaOH (TmG318C), and KOH (TmG360C ). The high basicity of the

water solution was neutralized to pH7 using HNO3 . The salt solutions were

deposited onto the specimens using a dropper pipette and the water was

evaporated by heating the specimens on a hot plate up to about 200C. In

order to provide a uniform salt coating at the beginning of the tests, the

Table I. KNa and 2S(NaCK) Ratios

for Different Salt Compositions

KNa 2S(NaCK )

0 1

0 0.4

0 0.2

2 1

2 0.4

2 0.215 1

15 0.4

15 0.2


5/22

Hot-Corrosion Behavior of NiCoCrAlY-Type Bondcoat Alloys 259

Table II. Amount of Salt, Recoating Frequency, Contaminant Flux Rate (CFR), and Washing

Procedure of Na2SO4-Coated Cast NiCoCrAlY

Amount of salt Recoating CFR Washing before

Designation [mg cm2] every [mg cm2 hr1] recoating

Sample A 1.016J0.062 100 hr 1B102 Yes

Sample B 1.061J0.072 100 hr 1B102 No

Sample C 0.262J0.050 20 hr 1.3B102 Yes

Sample D 0.264J0.043 20 hr 1.3B102 No

Sample E 1.059J0.101 20 hr 5.3B102 Yes

Sample F 1.051J0.156 20 hr 5.3B102 No

Sample G 4.746J0.383 20 hr 23.7B102 Yes

Sample H 0.283J0.085 100 hr 0.3B102 No

specimens were preoxidized for 1 hr at 950C (1 cycle) prior to coating,

rather than depositing salt on the bare metal surface.11

One set of NiCoCrAlY specimens was recoated with pure Na2SO4after

20 and 100, 1-hr cycle intervals (recoating frequency), respectively, apply-

ing different contaminant flux rates (CFRs) (Table II). In some cases, the

specimens were washed carefully in distilled water prior to recoating in

order to remove the existing deposits. A second set of NiCoCr

AlYHfSi specimens was coated with salts of different KNa ratios and sul-

fur levels (Table I) at an average CFR of 1.07 mg cm

2 hr

1. These specimenswere recoated every 100 cycles without deposit removal. All specimens were

weighed before and after coating with salt, after washing, and before and

after exposure. Mass change of the specimens was calculated taking into

account the evaporative mass loss of the PtRh hooks (approximately

2B104 mg cm2 hr1). Corrosion products were characterized using optical

microscopy, a field-emission gun scanning-electron microscope (FEG-SEM)

equipped with energy dispersive X-ray analysis (EDS), and a microprobe.

Specimens were sectioned and examined metallographically at the end of

the tests.

RESULTS AND DISCUSSION

Effect of Salt-Deposition Procedure

An overview of the data for mass gain vs. number of 1-hr cycles for

Na2SO4-coated NiCoCrAlY specimens tested at 950C is given in Fig. 1.

The CFR was varied by almost two orders of magnitude between 0.28B102

and 23.73B

10

2

mg cm

2

hr

1

and, accordingly, the extent of corrosionattack also differed widely. As expected, the highest mass loss was found

for the highest CFR. The mass-change data for most data lie in a scatter

band up to 350 1-hr cycles. However, one specimen exposed to the highest


6/22


Fig. 1. Sample mass change vs. number of 1-hr cycles for Na2SO4-coated cast NiCoCrAlY.Specimens were coated with different CFRs at different time intervals. Some specimens werewashed to remove salt deposits before recoating with salt. Specimen designation is accordingto Table II. The arrows indicate the range over which CFR, deposit removal, and recoatingfrequency affect the hot-corrosion attack of the alloy.

CFR (sample G) and one specimen exposed to a fresh salt coating every

20 hr at an intermediate CFR (sample E) started to lose mass rapidly after

200250, 1-hr cycles, indicating accelerated attack. It should be noted thatthe data represent sample mass change, i.e., the mass of the deposits that

possibly remained on the specimen surface was included in all data points.

Detailed comparison of mass-change data before and after sample washing,

i.e., with and without deposits included, revealed, however, that the course

of the mass-change curves was similar qualitatively, albeit the absolute mass

data were, of course, lower when deposits were removed.11

The qualitative influence on corrosion attack of the three experimental

parameters, contaminant flux rate, deposit removal or retention prior to salt

recoating, and recoating frequency, is indicated by arrows in Fig. 1. It isevident that, for the set of parameters investigated, the contaminant flux

rate (i.e., the amount of salt deposited on the specimen surface per unit time)

was the major factor determining the corrosion attack of the NiCoCrAlY


7/22


alloy. The effect of deposit removal became important with intermediate

CFRs on the order of 5.3B102 mg cm2 hr1, whereas, for CFRsG1

1.3B

10

2

mg cm

2

hr

1

, the effect was relatively minor. Recoating frequencydemonstrated some effect on the hot-corrosion attack of NiCoCrAlY.

The effect of these parameters will be discussed in more detail below.

Contaminant Flux Rate

The dependence of specimen mass loss on the contaminant flux rate of

Na2SO4-coated cast NiCoCrAlY after 350, 1-hr cycles at 950C is illustrated

in Fig. 2. The specimens (samples C, E, and G) were recoated every 20

cycles and the deposits were removed before recoating. Although the CFR

varied by a factor of approximately 20, the resulting specimen mass loss

Fig. 2. Mass loss vs. contaminant flux rate of Na2SO4-coated cast NiCoCrAlY after 350,1-hr cycles at 950C. As expected, lowest corrosion attack was caused by the lowest CFR

(a), whereas the highest CFR resulted in severe attack (c). The respective macrographs(ac) indicate that the alumina scale (light gray areas on the surface) was attacked on allspecimens. For intermediate and high CFRs the alumina-scale spinel phases (dark speci-men areas) were the dominant oxides formed. The specimen edges especially suffered fromsignificant metal loss due to repeated oxide scale spallation and reformation.


8/22


showed more than two orders of magnitude difference, indicating the influ-

ence of the CFR on the extent of corrosion attack. The greatest attack of

cast NiCoCrAlY was experienced with a CFR of 23.73B

10

2

mg cm

2

hr

1

.The specimen started to lose significant mass after about 200, 1-hr cycles

(Fig. 1) and was heavily corroded after the test period of 350 cycles (Fig.

2c). It should be noted that, for the experimental geometry, i.e., 1.5 cm-

diameter coupons, and their vertical suspension in the furnace, approxi-

mately 5 mg cm2 was the maximum amount of salt that could be coated

without losing salt due to droplet formation and subsequent dripping during

heating.

The optical macrographs in Fig. 2 also illustrate the significant increase

in attack with increasing CFR. With increasing CFR, spinel phases (darkgray surface areas) were the dominant oxide phases present for CFRG

5.3B102 and 23.7B102 mg cm2 hr1 after 350, 1-hr cycles, whereas for

CFRG1.3B102 mg cm2 hr1 quite a large portion of the initially formed

alumina scale (light gray phase on the surface) was still present. Mass loss

of the samples was caused by repeated spallation of the oxide scale (presum-

ably during cooling between cycles) and subsequent regrowth of the oxide

scale. The specimen edges especially showed significant loss of metal as a

result of this process.

Optical macrographs of the specimens exposed to CFRG

1.3B

10

2

mgcm2 hr1 (Fig. 3a) and CFRG23.7B102 mg cm2 hr1 (Fig. 3b) revealed

relatively non-uniform attack of the cast NiCoCrAlY alloy. For CFRG

1.3B102 mg cm2 hr1 the alumina scale was partly intact, although areas

of voluminous spinel phases were present, as shown in Fig. 2, indicating

degradation of the protective layer. For CFRG23.7B102 mg cm2 hr1,

Fig. 3. Optical micrographs of cast NiCoCrAlY specimens after 350, 1-hr cycles at 950

C.(a) For CFRG1.3B102 mg cm2 hr1 main parts of the specimen surface were covered witha relatively thin alumina scale, but in some spots voluminous spinel formed. (b) ForCFRG23.7B102 mg cm2 hr1 massive oxidation along the phase boundaries underneath thevoluminous, porous oxide scale occurred in many spots.


9/22


Fig. 4. Secondary-electron SEM images of the spinel-rich part of the oxide scale formed onNiCoCrAlY coated with Na2SO4 revealed crack formation of the specimen surface (a) andthrough the oxide scale parallel to the surface (b). The oxide scales typically consisted of amixture of alumina and spinel phases, but microprobe analysis revealed that chromia also hadformed.

voluminous oxide scales containing primarily spinel phase were formed.

Nonuniform attack of the alloy led to the formation of an oxide networkunderneath the oxide scale in several locations on the specimens, indicating

significant internal degradation.

As mentioned before, when it formed, voluminous oxide scale spalled

readily during cycling. The SEM micrograph in Fig. 4a shows a plan view

of the oxide scale typical of NiCoCrAlY specimens exposed to Na2SO4 . The

oxide scale was extensively cracked and part of the scale had spalled during

thermal cycling. A cross section through the oxide scale revealed massive

crack formation in the scale (Fig. 4b). Although such crack formation could

occur during metallographic preparation, the rounded appearance of somecrack tips suggested that most cracking actually occurred during testing.

Crack formation parallel to the specimen surface facilitated loss of part of

the oxide scale. The oxide scale was porous as a result of dissolution and

reformation during salt attack. In some areas, metallic -phase seemed tobe incorporated into the oxide scale. Microprobe analysis of this -phaserevealed that all the Al had been leached out during the molten-salt attack.

Deposit Removal or Retention Before Salt Recoating

According to the mass-change data summarized in Fig. 1, removal or

retention of deposits before salt recoating had a significant effect on the

corrosion attack of the NiCoCrAlY alloy. For a given recoating frequency,


10/22


Fig. 5. Influence of deposit removal before salt recoating on specimen mass loss for (a) CFRG1B102 mg cm2 hr1, recoated every 100 cycles; (b) CFRG1.3B102 mg cm2 hr1, recoatedevery 20 cycles; and (c) CFRG5.3B102 mg cm2 hr1, recoated every 20 cycles.

the mass loss of the specimens clearly depended on deposit removal or reten-

tion (Fig. 5). At lower CFRs, the effect was less pronounced, but still

measurable (Fig. 5a,b), than for higher CFRs (Fig. 5c). For CFRG

5.3B102 mg cm2 hr1, deposit removal resulted in a significant increase of

mass loss (by about one order of magnitude) compared to the specimen on

which the deposits were retained.

Based on the current understanding of the mechanisms of hot-cor-

rosion under salt deposits, fluxing of the alumina scale by a molten salt is

strongly dependent on the oxide-ion activity (aO2) of the salt melt.12 Basic

and acidic fluxing are associated with a high and low oxide ion activity,

respectively, thus any shift in the salt oxide ion activity will potentially

increase the solubility of the alumina scale.

13

Although Cr2O3is also solublein fused Na2SO4 ,14 dissolution of Cr has been reported to effectively buffer

the salt basicity near the level where alumina solubility is minimum and so

effectively retard accelerated attack of alumina scales.13,15 Similarly, chro-

mate fuel additives are known to effectively combat hot corrosion in many

practical applications. In the present study, the yellow color of the deposits

that were removed from the NiCoCrAlY specimens indicated the presence

of sodium chromate (Na2CrO4), i.e., Cr was leached out of the substrates

by the molten salt. In the case of salt deposition without deposit removal

before recoating, the fresh salt was blended with the salt that was alreadypresent on the surface for several cycles and that already contained a certain

amount of Cr. The basicity of fresh salt presumably would be less changed

due to the presence of the Cr buffer, compared to samples exposed to a


11/22


fresh salt on a deposit-free surface. Accordingly, specimens repeatedly

exposed to fresh salts were found to suffer stronger corrosion attack than

the unwashed specimens, which resulted in marked scale spallation (Fig.5c). For the lower CFRs this effect was reduced, presumably since the initial

amount of salt was much lower and the Cr supply from the substrate would

be expected to be sufficient to rapidly buffer the salt basicity even in the

case of a cleaned surface exposed to fresh salt. This mechanism will be

discussed in more detail below.

Recoating Frequency

While the effects of the contaminant flux rate and deposit removal or

retention on the hot-corrosion attack of cast NiCoCrAlY could be relativelywell described by either the macroscopic appearance of the specimens or

mass-change data obtained by thermal cyclic testing alone, additional met-

allographic examinations were necessary to characterize the influence of

recoating frequency. For either deposit removal or retention, mass-change

data for recoating frequencies of 20 and 100 cycles revealed only small differ-

ences, even for CFRs ranging from 0.1 to 1.1B102 mg cm2 hr1. Figure 6

Fig. 6. Effect of recoating frequency on the corrosion attack of cast NiCoCrAlY coated with

Na2SO4 and exposed at 950

C for 350, 1-hr cycles. (a), (b), (c), and (d) macrographs of speci-mens recoated every 100 and 20 cycles with CFRG1B102 and 1.3B102 mg cm2 hr1, respect-ively. The deposits were removed by washing or retained before salt recoating. Although mass-loss data revealed a fairly minor (but measurable) effect of deposit removal or retention,appearance of the specimens was significantly different.


12/22


summarizes the mass-loss data as a function of recoating frequency for

deposit removal and retention using the specimens subjected to CFRs of

1B

10

2

and 1.3B

10

2

mg cm

2

hr

1

for comparison. Despite the small dif-ferences in mass loss among the specimens, their macroscopic appearance

was significantly different. Whereas large parts of the specimens recoated

every 20 cycles contained alumina (light gray phases in Fig. 6c and d), the

oxide scale on specimens coated every 100 cycles contained mainly spinel

phases (dark gray phases in Fig. 6a and b), rather than alumina. As spinel

phases tended to spall more easily than alumina, it was concluded that the

mass-change data for these specimens was the net result of mass gain by

voluminous oxide scale formation and mass loss from subsequent spallation.

Therefore, mass-change data alone were obviously an insufficient measurefor determining the extent of corrosion attack of cast NiCoCrAlY, in these

cases. It is of note that the effective mass gain and mass loss for 20 and 100

cycle recoating frequencies, respectively, measured for specimens from

which the deposits were not removed from the surface (Fig. 6) was reversed

when the deposits were removed. This was a direct result of the loss of oxide

scale during deposit washing.

Figure 7ad shows representative cross sections of the respective speci-

mens illustrated in Fig. 6ad. For similar CFRs, the increased severity of

attack when a larger amount of salt was initially present on the specimensrecoated every 100 cycles (approximately 1.0B102 mg cm2 hr1 vs.

0.26B102 mg cm2 hr1 for those recoated every 20 cycles) is clearly evident.

This is presumably due to the longer time of continuous contact of the salt

with the specimen surface before removal by evaporation during 100-cycle

recoating tests, allowing dissolution of the protective scale and attack of the

substrate (see below for a discussion on evaporation). In the case of the

specimens recoated every 20 cycles, the shorter contact time with the molten

salt presumably resulted in diminished dissolution of scale, so that a protec-

tive scale reformed after the salt had evaporated. The cause of the increased

severity of attack of the specimens, which were not washed prior to recoat-

ing, is not obvious.

Based on the present results, a 20-hr salt-recoating frequency and an

average contaminant flux rate of the order of 1.0B102 mg cm2 hr1 with

deposit removal before recoating was considered to provide the most con-

sistent results regarding correlation between mass-change data and actual

corrosion attack of the specimens. However, the 20-hr salt-recoating pro-

cedure was very labor intensive. Therefore, for the sake of a reasonable

specimen throughput (100, 1-hr cycles per week) in the laboratory, a sim-

plified test procedure involving 1.0 mg cm2 of salt, recoated every 100 hr

without deposit removal before recoating, was chosen.


13/22


Fig.

7.

OpticalmicrographsofcastNiCoCrAlY

specimensteste

dunderdifferentexperimental

parameters.

Thecrosssections

in

dicatesignificanteffectoftheco

ntaminantfluxrate,recoatingf

requency,anddepositremovalorretentionuponhot-corrosion

at

tack.


14/22


Effect of Salt Evaporation

Comparison of mass-change data associated with recoating and deposit

removal revealed that, for small amounts of salt coated onto the specimens,mass loss by washing before recoating was fairly small (Fig. 8a). It should

be noted that mass loss, in general, included not only salt deposits, but also

loss of oxide during washing; however, this was found to be marginal in the

case of CFRG1.3B102 mg cm2 hr1. For larger amounts of salt deposited

on the surface (Fig. 8b), the absolute amount of salt present on the surface

before washing was much higher, indicating that the salt had not completely

evaporated within a period of 20, 1-hr cycles. However, after about 200

overall cycles, the mass loss after deposit removal started to increase. This

disproportionally high mass loss of the specimen was associated with partialloss of the oxide scale in addition to the mass loss due to deposit removal

during washing. Increasing amount of oxide loss during washing was corre-

lated with the onset of accelerated specimen mass loss (compare Fig. 8b

with Fig. 1). As a result of rapid scale growth and continuing attack by

Fig. 8. Relative mass change vs. number of 1-hr cycles fortwo NiCoCrAlY specimens Na2SO4-coated at different

CFRs. Different sets of data points were taken after differentnumbers of cycles. For example, between 280 and 300 cycles,data was recorded after 281, 283, 285, 290, 295, and 300 hr.Mass loss of the specimens was partly a result of salt evapor-ation during exposure (refer to text for details).


15/22


Fig. 9. Mass change vs. number of 1-hr cycles for two NiCoCrAlY specimens Na2SO4-coated at different CFRs. Mass

gain was attributed to the amount of salt added, whereasmass loss was mainly due to salt evaporation and loss ofoxide scale during cleaning.

fresh salt, the mechanical integrity of the oxide scale was significantly affec-

ted after longer times of exposure. Once part of the oxide scale spalled, the

next salt coating was deposited on either a thinner oxide scale or bare metal

(depending on the type of spallation), thus further facilitating attack.

A more detailed analysis of the evaporation behavior of the salt

deposits is given in Fig. 9. The relative mass change of the specimensdenotes the mass of the specimen during exposure in relation to its starting

mass, including the fresh salt coating. It should be noted, of course, that

the specimen mass was affected by superposition of mass gain as a result

of corrosion and mass loss caused by salt evaporation and possibly oxide

spallation. It is apparent from Fig. 9 that sodium sulfate evaporated from

the NiCoCrAlY specimens at an almost linear rate, as observed for both

CFRG1.3B102 mg cm2 hr1 and CFRG5.3B102 mg cm2 hr1. The data

were determined at different times between different overall numbers of

cycles. Deviation from linear behavior was observed at higher numbers ofcycles and, to somewhat greater extent, for the higher CFRs. For higher

numbers of cycles, side effects like oxide spallation and accelerated scale

regrowth during the 20 cycle periods accounted for the deviation from the


16/22


linear behavior. For CFRG5.3B102 mg cm2 hr1 and more clearly, for

CFRG23.7B102 mg cm2 hr1 (not shown here), droplet formation and

subsequent loss by dripping caused disproportionally high mass loss afterthe first few cycles (Fig. 9b), but once the excess salt was lost, a linear

evaporation rate was reestablished (Fig. 9b), data points for 280300 hr). It

should be emphasized that these data did not accurately reflect the salt

evaporation behavior, because they included side effects related to oxide

spallation. This was clearly indicated by the significant scatter of the

data after 20 cycles, ranging from about zero to 40% relative mass change

(Fig. 9).

A more accurate determination of the salt evaporation rate would

demand a very corrosion-resistant alloy, which forms a slow-growing well-adherent oxide scale. Because of the significant attack of the NiCoCrAlY

alloy associated with scale spallation after extended exposure times, the

evaporation rate was calculated from only a few data points. From these

data, the evaporation rate for pure sodium sulfate at 950C in flowing oxy-

gen was calculated to be of the order of 0.01 mg cm2 hr1. However, the

salt evaporation rate was significantly dependent on the alloy chemistry,13,16

and the alloy composition, in turn, affected the deposit chemistry and, there-

fore, its vapor pressure.

Effect of Salt Composition

In order to assess the extent of corrosive attack likely to occur in the

presence of contaminants typical of combustion products of biomass-

derived fuels, cast NiCoCrAlYHfSi was exposed to different salt compo-

sitions representing KNa ratios of 0, 2, and 15, as well as different sulfur

levels. The simplified test procedure mentioned above was used involving

an average CFR of 1.0B102 mg cm2 hr1, with specimens coated every

100 hr without deposit removal before recoating. As expected, the mass

change vs. number of 1-hr cycles curves showed a significant effect of salt

composition on the corrosion behavior (Fig. 10). Pure sodium sulfate, as a

baseline contaminant, caused the most severe attack of all salt blends tested.

Repeated spallation of the oxide scale after a short period of initial mass

gain resulted in significant mass loss after 500, 1-hr cycles. Salt blends were

all less aggressive than sodium sulfate alone and, surprisingly, the two

sodium sulfate blends with a sulfur deficit demonstrated the largest differ-

ences in mass change compared to stoichiometric sodium sulfate. A moredetailed description of the data is given in Figs. 11 and 12, in which sample

mass change is summarized as a function of the KNa ratio (Fig. 11) and

sulfur content in the salt (Fig. 12).


17/22


Fig. 10. Mass change vs. number of 1-hr cycles at 950C of cast NiCoCrAlYHfSi coated withdifferent salt compositions. Salt blends were selected to represent contaminants relevant tobiomass-derived fuel combustion products. Pure sodium sulfate (filled circles) was included asa baseline deposit widely used in hot-corrosion studies.

For a given sulfur level in the salts, substituting K for Na generally

resulted in decreased mass loss of the specimens (Fig. 11ac). For a stoichio-

metric sulfur level, the addition of potassium of the order of KNaG2

resulted in formation of an oxide scale with a significantly decreased tend-

ency to spall; further increasing the amount of potassium to KNaG15

decreased the attack more significantly. These results are in good agreement

with earlier findings for a cast NiCoCrAlY alloy, which demonstrated a

significant dependence of hot-corrosion attack on KNa ratio.7 However,

unlike NiCoCrAlYHfSi, cast NiCoCrAlY gained mass during testing up to500, 1-hr cycles, indicating improved scale adherence over that of the Hf-and Si-modified version.

For a given KNa ratio, reducing the sulfur level of the sodiumpotass-ium salt blends seemed to slightly increase the hot-corrosion attack of castNiCoCrAlYHfSi, at least in terms of their tendency to promote scale spall-

ation (Fig. 12b,c). However, whereas changing the sulfur level in the saltmixtures showed little effect on overall weight change for the Na-containingsalt, increased attack (based on mass gain) was measured with decreasing

sulfur level, as shown in Fig. 12a.Although the extent of corrosion attack of cast NiCoCrAlYHfSi wasreduced when the relative amount of Na in the salt was decreased, metallo-

graphic investigations of the specimens still revealed intense corrosion


18/22


Fig. 11. Mass change vs. number of 1-hr cycles at 950C of cast NiCoCrAlYHfSi coated withdifferent salt compositions. Data were identical with those of Fig. 10, but plotted separatelyfor different sulfur levels [2S(NaCK )G1 is stoichiometric sulfates, whereas 2S(NaCK )F1represents S deficit relative to alkali].

Fig. 12. Mass change vs. number of 1-hr cycles at 950C of cast NiCoCrAlYHfSi coated withdifferent salt compositions. Data were identical with that of Fig. 10, but plotted separately fordifferent KNa ratios.

attack (Fig. 13a,b). The cast alloy was preferentially attacked in all cases

and the oxide scales consisted of a mixture of alumina and spinel phases.

Microprobe analysis revealed that chromia had also been formed to a cer-

tain extent. For lower sulfur levels, the general trend of decreased attack

with a reduced amount of Na was identical with that of stoichiometric sulfurcontent but, again, significant preferential attack occurred (Fig. 13c,d).

Notably, salts with Na as the only alkali did not follow this trend. For a

given sulfur level (2S(NaCK )G0.4 and 0.2), mass gain rather than mass


19/22


Fig. 13. Optical micrographs of cross sections through specimens coated with different saltblends after 500, 1-hr cycles at 950C (a) Pure sodium sulfate [KNaG0, 2S(NaCK )G1];(b) KNaG0, 2S(NaCK )G0.2; (c) KNaG15, 2S(NaCK )G1 ; and (d) KNaG15,2S(NaCK )G0.2.

loss was observed after an incubation period (Fig. 12b,c), which was in

contrast to stoichiometric sodium sulfate which caused significant scale

spallation. However, mass-gain curves indicated nonprotective oxide scale

formation, and postoxidation SEMEDS investigations of the oxide scales

revealed that on all NiCoCrAlYHfSi specimens, spinels were the dominantoxide phases in the scale rather than alumina. Therefore, the general hot-

corrosion mechanism of alumina-scale fluxing appeared to be identical for

all salt blends tested in this study.

The surfaces of NiCoCrAlYHfSi specimens exhibited notably different

appearances (Fig. 14). For sodium sulfate alone, back-scattered SEM micro-

graphs indicated a rough surface topography caused by partly spalled oxide

scale and the growth of voluminous spinel clusters (Fig. 14a). In contrast,

the K-containing salt blends maintained a relatively smooth surface (Fig.

14b) and spinel clusters were rarely found on the surfaces of thesespecimens.

The present results suggested that the role of sulfur might have been

indirect with regard to alloy attack. As mentioned earlier, no additional SOx


20/22


Fig. 14. Back-scattered SEM surface image of cast NiCoCrAlY after 500, 1-hr cycles at 950Cdeposited with (a) pure sodium sulfate and (b) potassiumsodium salt blend [KNaG15,2S(NaCK )G0.2]. The oxide scale formed under sodium sulfate alone was much more proneto spallation than under salt blends.

was used for the experiments to simulate the salt impact delivery scenario.

Therefore, salt decomposition to Na2O and SOx , as well as evaporation,

might have had significant effect. Increasing the Na2O level in the saltchanged its basicity and accordingly the extent of attack of the alumina

scale. At this time, no further data on basicity of the different salt blends is

available. On the other hand, as a result of a sulfur deficit in the deposit

simulated in this study, free excess alkali might have been present on the

specimen surfaces, resulting in a much higher vapor pressure than that of

the respective sulfates at the test temperature of 950C. Therefore, part of

the alkali was assumed to evaporate from the surface, thus reducing the

amount of alkali available and producing conditions comparable to a

reduced contaminant flux rate. However, this expectation was only noticedfor the change of corrosion attack for Na-containing salts shown in Fig.

12a. When potassium was added to the salt blends, a similar mechanism

was expected, since the vapor pressure of potassium hydroxidenitrate is

also higher than that of the sulfate at the test temperature. However, as the

mass-change data indicated, the differences of corrosion attack induced by

different sulfur levels were not as great as for only Na-containing salts (e.g.,

Fig.12b,c). These results suggest that K plays a role in either affecting the

deposit chemistry itself such that its vapor pressure is not sufficiently high

to evaporate notable amounts of salt, or that K directly alters the extent ofcorrosion attack, e.g., by altering the salt basicity. Additional microana-

lytical investigations and information on the effect of salt composition on

its basicity are needed to clarify the potentially important role of potassium


21/22


on the hot-corrosion behavior of MCrAlY-based alloys for applications in

biomass-derived fuel-fired gas turbines.

SUMMARY AND CONCLUSIONS

The hot-corrosion behavior of NiCoCrAlY-type alloys under thermal

cycling conditions was found to depend significantly on the testing pro-

cedures. Contaminant flux rate alone, as an average measure of the amount

of deposits applied onto the specimens, was not a sufficient predictor of the

corrosion attack when the contaminants are not applied continuously dur-

ing exposure as in, for example, burner-rig tests. For intermittent depositionof contaminants onto the specimens, a significant dependence of corrosion

on recoating frequency was evident. Even when the same average contami-

nant flux rate was applied, mass-change data revealed that the corrosion

attack was less severe for those specimens coated more frequently but with

a lower amount of salt in each recoating step. Therefore, it appears that the

residence time of the deposit on the surface is an important consideration.

Hence, it is suggested that for hot-corrosion tests with intermittent salt

deposition, the recoating frequency must be always considered along with

the average contaminant flux rate. Furthermore, the effect of depositremoval should be taken into account when data from different corrosion

tests are compared. Depending on the contaminant flux rate, deposit

removal (washing) before recoating may result in an accelerated attack,

because of a change of the basicity of the molten salt compared to deposits,

which were present on the surface for longer exposure times.

For different KNa ratios and sulfur levels considered to be relevant

to combustion products obtained from biomass-derived fuels, significantly

different corrosion attack was observed, showing that seemingly minor

changes in the deposit chemistry may cause important differences in the

extent of corrosion attack. With increasing K content in the deposits, the

amount of corrosion, relative to pure sodium sulfate, seemed to be reduced.

Reducing sulfur in the salts relative to their alkali content reduced corrosion

for salt deposits containing Na alone, whereas in the presence of potassium,

corrosion was slightly increased. The role of K and its possibly synergistic

effect with sulfur on the hot-corrosion behavior of candidate coating mater-

ials for application in biomass-derived fuel-related gas turbine applications

has yet to be determined.As a final note, it should also be emphasized that, while measurement

of specimen weight change is a convenient technique for following trends in

laboratory experiments, given all the caveats mentioned regarding effects of


22/22


salt evaporation and scale loss, confirmation of such trends by metallo-

graphic measurements is essential. Further, such measurements are neces-

sary to determine the maximum depth of attack required for quantitativecomparison of alloy performance.

ACKNOWLEDGMENTS

Research sponsored by the German Aerospace Center and the U.S.

Department of Energy, Assistant Secretary for Energy Efficiency and

Renewable Energy, Office of Industrial Technologies, as part of the

Advanced Turbine Systems Program under contract DE-AC05-96OR22464

with Lockheed Martin Energy Research Corporation. P. F. Tortorelli,D. F. Wilson, and J. R. DiStefano at ORNL provided comments on the

manuscript.

REFERENCES

1. H. J. Grabke and D. B. Meadowcroft,Guidelines for Methods of Testing and Research inHigh Temperature Corrosion (The Institute of Materials, London, U.K., 1995).

2. P. Hancock, Corros. Sci. 22, 51 (1982).3. S. R. J. Saunders, Mat. Sci. Technol. 2, 282 (1986).4. P. Steinmetz, C. Duret, and R. Morbioli,Mater. Sci. Technol. 2, 262 (1986).

5. J. R. Nicholls, in Guidelines for Methods of Testing and Research in High TemperatureCorrosion, H. J. Grabke and D. B. Meadowcroft, eds. (The Institute of Materials, London,U.K., 1995), p. 11

6. A. V. Dean,Investigation into the Resistance of Various Ni- and Co-Based Alloys to Sea SaltCorrosion at Elevated Temperatures, National Gas Turbine Establishment, Report (1964).

7. C. Leyens, I. G. Wright, and B. A. Pint, in Elevated Temperature Coatings: Science andTechnology III, J. M. Hampikian and N. B. Dahotre, eds. (TMS, Warrendale, PA, 1999),p. 79.

8. N. S. Bornstein and W. P. Allen,Mater. Sci. Forum 251254, 127 (1997).9. I. G. Wright, C. Leyens, and B. A. Pint, International Gas Turbine and Aeroengine Con-

gress, Munchen, Germany, 811 May 2000, paper 2000-GT-0019, submitted.10. D. K. Gupta and D. S. Duvall,A Silicon and Hafnium Modified Plasma Sprayed MCrAlY

Coating for Single Crystal Superalloys (TMS, Warrendale, PA, 1984), p. 711.11. C. Leyens, I. G. Wright, B. A. Pint, and P. F. Tortorelli, in Cyclic Oxidation of High

Temperature Materials, M. Schutze and W. J. Quadakkers, eds. (The Institute of Materials,London, U.K., 1999), p. 169.

12. R. A. Rapp and Y. S. Zhang, JOM, 47 (1994).13. J. A. Goebel, F. S. Pettit, and G. W. Goward, Metall. Trans. 4, 261 (1973).14. Y. S. Zhang,J. Electrochem. Soc. 133, 655 (1986).15. R. A. Rapp,Mat. Sci. Eng. 87, 319 (1987).16. C. Leyens, unpublished research (1998).

c leyens (2000) effect of experimental procedures on the cyclic

Documents