Oxidation of Metals, Vol. 40, Nos. 5/6, 1993

Study of Cracking and Spalling of Cr203 Scale Formed on Ni-30Cr Alloy Yifan Zhang* and David A. Shores*

Received April 26, I993

Acoustic em&sion (AE) was used to monitor the scale cracking and spalling of Ni-3OCr alloy samples that had been oxidized at lO00~ for 2 hours or 20 hours then cooled slowly to room temperature in two different ways: natural furnace cooling or constant-rate cooling at a rate of about 19~ hour. The morphology of the remaining scale and spalled areas was observed by scanning electron mieroscope (SEM). The AE results show that the scale started to crack and spall continuously during cooling when a certain temperature was reached. The cracking and spalling continued over a temperature range, indicating that there is distribution of the critical-fracture stresses. The S E M analysis shows that the interface between the Cr203 scale and the Ni-3OCr alloy substrate is relatively weak. Under this condition, a mathematical model proposed by A. G. Evans can be used to estimate the size of the spalled areas and the critical size of the interfacial flaws required to produce buckling. Measurements of the diameter of spalled areas from S E M images yielded a range of values that were well described by a normal distribution. By applying the Evans model to the AE data, normal distributions of the diameter of spalIed areas were estimated, which were 2 - 3 times larger than the measured sizes. This lack of agreement may be because the model did not take into account stress-concentration effects of other dejects, such as oxide grain boundaries, small porosity in the scale, or small V-notches between individual oxide crystals at the scale surface. It is suggested that the distribution of the interfacial-flaw sizes causes the distribution of the criticalfracture stresses that are inferred from the AE experiments. KEY WORDS: acoustic emission; Ni-30Cr; scale buckling; spalling; flow size. *Corrosion Research Center, Department of Chemical Engineering, University of Minnesota, 112 Amundsen Street, 221 Church Street, S.E,, Minneapolis, Minnesota 55455.

529 0030-770X/93/1200-0529507.00/09 1993 Plenum Publishing Corporation

530

Zhang and Shores

INTRODUCTION The resistance of metals and alloys to high-temperature oxidation is due to the formation of a compact, adherent, protective oxide scale. The oxide scale can slow down the oxidation process by acting as a barrier to the diffusion of reacting species. The oxide scale will be subjected to growth stresses generated during the isothermal growth of the scale and to thermal stresses generated by cooling the oxidized sample. ~-4 Cracking and spalling of oxide scale may be induced by these stresses, resulting in accelerated oxidation. In most cases, the oxide stresses are compressive. H. E. Evans et al. 5,6 have proposed two routes for oxide-scale failure. The first one is that the strength of the metal/oxide interface is higher than the compressive fracture strength of the oxide. In this case, a shear crack will be formed within the oxide layer at an early stage. The crack is presumably nucleated at pre-existing defects, and it can penetrate to the metal surface. The formation of this shear crack can result in a geometry which permits tensile cracks to be wedged along the oxide/metal interface leading to spallation. Second, if the interface is relatively weak, Route II process of spallation will occur. A weak interface can result in buckling of the oxide layer with a later development of through-thickness cracks in tensile regions and subsequent spallation. The weakness of the metal/oxide interface can be intrinsic to the system, or arise from the segregation of elements to this region or growth of interfacial cavities. H. E. Evans' proposal qualitatively describes the processes of cracking and spalling of compressively-stressed oxide scale. Quantitative modeling of the oxide-scale spallation-process has been carried out by A. G. Evans et a/. 7-9 These results show that for flat scales the oxide-spalling process must be preceded by scale buckling, which occurs when an initial interfacial flaw exists with radius greater than a critical value ac : a~ > l.l h ( E ~ ~/2 \ao/

(1)

where h is the scale thickness, Eox is the Young's modulus of the oxide and ao is the biaxial stress 0-1~ or 0-22 (in the present case we assume that 0"ll = 0"22 ) applied to the scale. After buckling, the average compressive stress in the buckled region is reduced, but a stress concentration arises at the perimeter of the buckle. If the interfacial crack exceeds a critical radius, as, the crack will tend to deflect toward the surface at the crack tip, resulting in spalling, as is given by: a s

: 1.9h (E~ 20-0./

1/2

(2)

Cracking and Spalling of Cr203 Scale

531

In the work presented in this paper, the cracking and spalling process of Cr203 formed on a Ni 30Cr alloy has been studied. The major driving force for the cracking and spalling is the stress induced by differential thermal expansion upon cooling the oxidized sample to room temperature. An acoustic-emission technique (AE) was used to monitor the occurrence of scale cracking and spalling. Scanning electron microscopy (SEM) was used to determine the mechanism of scale spalling.

EXPERIMENT

The N i - 3 0 C r alloy was cut into coupons of 9 mm x 9 mm x 1 mm and polished through several grades of SiC abrasion paper, ending at 1-/~m diamond paste on a napped cloth. A platinum wire of 0.5 mm in diameter was spot-welded on the sample surface to transfer the stress waves produced by the cracking and spalling of the scale to a piezoelectric AE sensor, which can transduce the elastic strain wave into electric signals. The oxidation experiments were conducted in pure oxygen flowing at 50 cc/min at 1000~ for 2 hours or 20 hours. Then samples were cooled to room temperature by either natural furnace cooling or constant-rate furnace cooling (20 hours oxidized sample only) at a rate of about 19~ The AE data were collected throughout the experiment by AET 5000 acousticemission equipment produced by the Acoustic Emission Technology Corporation. A schematic of the experimental setup is shown in Fig. 1. The AE sensor, located on the top of the reaction tube, has a sensitivity of - 70 dB referred to 1 volt per microbar. The transducer-output signals were sent to a 60 dB pre-amplifier first. The frequency response of the pre-amplifier is from 1 kHz to 2 MHz. The total amplification of the AET 5000 system was 95.6 dB. The acoustic-emission-data characterization is shown schematically in Fig. 2. The package of high-amplitude waves is called an event; typically an individual crack will produce an event comprised of several waves. The AET 5000 can discriminate and count the number of the events as a function of time. It can also measure the following parameters of an event: peak amplitude, ringdown count number, which is the number of individual peaks passing the threshold, the event duration and the rise time. The frequency can be calculated according to ringdown counts and event duration. The energy of an AE event is proportional to the square of the peak amplitude multiplied by the event duration. An amplitude threshold was set at the time of the experiment to filter out the background noise. Sometimes, the peak amplitude of noise events can be as high as the peak amplitude of the real events. In order to ensure that all the events recorded

532

Zhang and Shores

[ Pi Pre.

17 I I

I

3

I

I

AET 5000

I

.~

l

.~. 9 ~.~ =2 --=

1. 2. 3. 4. 5. 6.

AE sensor Pt waveguide Stainless steel flange Gas inlet Gas outlet Thermocouple connector (outside the reaction tube) 7. Quartz reaction tube 8. Sample 9. Furnace

I

Pt wire

spot weld

Fig. 1. The experimental setup.

came from the scale cracking and spalling processes, other criterion may also be set to discriminate the invalid events. Since most electrical noise events have a very short event duration, an event duration criterion of 20 #s was set for all the experiments. Only the events which last longer than 20 #s were counted as real events. After the experiment, the samples with remaining oxide scales were examined by SEM to provide information about the route of scale spalling.

Cracking and Spalling of

Cr203 Scale

533

Rise time

1i--AA,, Time < Noise

Fig. 2. The characterization of AE data.

RESULTS AND DISCUSSIONS

Acoustic-Emission-Data Analysis The oxidation at 1000~ in pure oxygen flow produced Cr203 scale on the Ni-30Cr alloy. The thicknesses of the scales are approximately 2 #m for 2 hr oxidation and 5 #m for 20 hr oxidation. Figures 3 and 4 show the acoustic-emission results of the experiments in which samples were oxidized for 2 hours or 20 hours, then naturally furnace cooled to room temperature. The figures show that in the isothermal oxidation stage, a small number of cracking events was collected. This indicates that some growth stress was generated during the isothermal-oxidation stage.* The isothermal-growth stress is expected to be relatively small, however, stress concentration can occur on sample edges and corners, causing cracks there under low average stress, as shown in the isothermal-oxidation stages of Figs. 3 and 4. These cracks can provide the through-thickness separation of the scale for later spallation process. During cooling to room temperature, an additional, and larger component of stress is induced by the different thermal-expansion coefficients of *In situ, oxide-strain measurements by X-ray diffraction H have shown that the compressive strain in the Cr203 scale formed on pure Ni-25Cr alloy by 8 hours isothermal oxidation at 950~ in pure oxygen is about 0.2%. A similar measurement with pure Cr at 940~ yielded an oxide growth strain of 0.13%. 12

534

Zhang and Shores . . . .

. . . .

. . . .

.

.

.

.

2 0 0

r

1000

8

soo 600

M

6

400

r

200

2 0

'"

A

10

15 20 Time (hours)

25

30

Fig. 3. The A E result of the experiment in which N i - 3 0 C r was oxidized for 2 hours and then naturally furnace cooled to room temperature.

the metallic substrate and oxide. Therefore the total stress applied to oxide scale will be the sum of the growth stress a ~ and the thermal stress O'thAccording to Tien and Davidson, 4 when the ratio o f the scale thickness to the sample thickness is very small, the average compressive thermal stress ath can be expressed as: E o x A T ( ~ A - aox)

1 -- o

O'th =

50

i

t

l

l

l

,

,

,

'i

. . . .

n . . . .

(3)

u . . . .

1200

I0~~

1000

soo

40

600 400

30

200

o

20

L gfl

10

"~

o

g

0

10

20

Time

30

41)

50

(hours)

Fig. 4. The AE result of the experiment in which N i - 3 0 C r was oxidized for 20 hours and then naturally furnace cooled to room temperature.

Cracking and Spailing of Cr203 Scale

535

where AT is the cooling amount and ( " A - aox) is the thermal-expansion difference between oxide scale and the substrate alloy, o is the Poisson's ratio. These stresses will be compressive in the oxide and tensile in the metal. For example, upon cooling from 1000~ to room temperature, O'th (oxide) would be about - 4 5 0 0 M P a (equivalent to an elastic strain of about 1.54%) if no creep relaxation occurred during cooling. 1~ The total stress ao in Eqs. (1) and (2) is: a o = ag w + E~

(4)

-- a ~ 1--~

Therefore, the stress will increase during cooling in proportion to AT. Figures 3 and 4 also show that continuous cracking and/or spalling of Cr203 scale started when the sample was cooled to a certain temperature (about 500~ indicating that a minimum stress is required to initiate the cracking/spalling process. The extensive cracking and spalling process continued over a temperature range, indicating that the scale does not fracture at a single value of stress. From Fig. 4 it can also be found that the peak of the time-to-cracking distribution of the 20-hr oxidation experiment is skewed toward the left. This is because the cooling rate was decreasing with cooling time. When the sample is cooled at a controlled constant rate, the peak was symmetrical or skewed slightly toward the right, as shown in Fig. 5. Three parameters of the AE events, i.e., mean frequency per event, mean peak amplitude per event and mean duration time per event, of the above three experiments are shown in Figs. 6-14. These figures show that

,2 =

t'2~176 1000 ,00 400

8

r

200

6

0

o

20

40

60

80

100

Time (hours) Fig. 5. The A E result of the experiment in which N i - 3 0 C r was oxidized for 20 hours and then cooled to room temperature at constant rate.

536

Zhang and S h o r e s 400

. . . . . . . .

i ....

I ....

i ....

r ....

1200

300 ~,~

800 600 400 200

250 200 ~-

.q f~

r

,% r

150

100 o

50 0

E- 9

k

~

0

.~

"

;

_~L

~

,

5

,

* ' ,"' s ' *~ . .

.*

"..s

,

I

. . . .

10

I

" ' ". " ,

,

,9

15 Time

"~ : . * .9"

"s". . . . .

9

~

20

~

I.,

,

25

,

,

r

30

(hours)

Fig. 6. T h e f r e q u e n c y s p e c t r u m o f A E signals f r o m the e x p e r i m e n t in w h i c h N i - 3 0 C r w a s oxidized f o r 2 h o u r s a n d t h e n n a t u r a l l y f u r n a c e c o o l e d to r o o m temperature (time increment was 6 minutes).

in all the experiments, the events generated during the isothermal-oxidation stages and the events collected during the early cooling stage (prior to extensive cracking) have higher frequency and peak amplitude and much lower event duration, compared with the events produced later in the extensive cracking and spalling process. Jha e t al. 13 reported that the crack-nucleation results in higher AE-signal frequency since the nature of

80

~

~ ....

i ....

i ....

i ....

1200

60

~

40

~

20

~

o

e~

. . . . . . . .

5

I 0

I 5

Time

21)

25

soo 600

I'D

400 200 0

e~ o

"-"

30

thours)

Fig. 7. T h e p e a k - a m p l i t u d e s p e c t r u m o f A E signals f r o m the e x p e r i m e n t in w h i c h N i - 3 0 C r w a s oxidized f o r 2 h o u r s a n d t h e n n a t u r a l l y f u r n a c e c o o l e d to room temperature (time increment was 6 minutes).

Cracking and Spalling of Cr203 Scale 3000

537

i ........

I ....

2500

I ....

1200

'

800

600 2000 L

1500

E

400 t

' '.

~

~

0

5

10

200 0

~

1000 500

15

Time

20

25

30

(hours)

Fig. 8. The event-duration spectrum of AE signals from the experiment in which N i - 3 0 C r was oxidized for 2 hours and then naturally furnace cooled to room temperature (time increment was 6 minutes).

the crack nucleation is more transient. The internal-crack propagation and scale spalling, on the other hand, will give a low-frequency AE signal due to a cleavage type of crack propagation, a less transient process. Christi et al. 14 found that the through-thickness cracking of oxide scale leads to AE events with very high amplitude, whereas buckling and delamination generate low-amplitude AE signals9

400

.... 1000~

!

. . . .

I

. . . .

I

'

'

'

'

,,,

L

1200 1000

350 =

800 600

300

400

250

200

200 =

E

0

150

-.~

9

100

9

50 0

.~;,.

"...

.

~% "~

.

.:...." ,.

:~'.~)..'~,~.".~"...,:..'~

." i

i

i

i

I

10

i

i

I

[

,L

20

i

i

i

I

30

,

,

,

~_.1

i

40

i

i

i

50

Time (hours) Fig. 9. The frequency spectrum of AE signals from the experiment in which Ni 30Cr was oxidized for 20 hours and then naturally furnace cooled to room temperature (time increment was 6 minutes).

538

Z h a n g and Shores

80

'

'

'

'

I

'

'

'

I

'

'

'

'

I

'

'

'

'

I

'

'

'

,

1200

IO00oC

1000

soo 60

40

600

~

400 200 0

"~

cr

r162

20

i

.

.

,

,

I

10

,

,

,

20

,

I

,

,

,

9

Time

i

I

30 (hours)

i

i

40

i

50

Fig. 10. The p e a k - a m p l i t u d e s p e c t r u m o f A E signals f r o m the e x p e r i m e n t in w h i c h N i - 3 0 C r was oxidized for 20 h o u r s a n d t h e n n a t u r a l l y furnace c o o l e d to r o o m t e m p e r a t u r e ( t i m e i n c r e m e n t w a s 6 minutes).

Based on the observations of these previous studies, we infer that the events produced in the isothermal-oxidation stage and in the early stage of cooling came from through-thickness crack nucleation. From Figs. 8, 11, and 14 it can also be found that the duration of these events is very short, implying that the crack-propagation distances must be short. It should also be noted from Figs. 6 to 8 that after 2 hr oxidation, the peak amplitude,

3000

.... IO00~

~ ....

~ ....

2500

i ....

11200 200

i ....

~

1ooo 8oo

..

H

600 2000 ~

400

o L

1500

------.

""

"

.-';:'i. -..-'.

1000

200

~'g

0

"-~

9 :.r ...-,:" : ..:" ": ~.~,~,:~,'~.... . : 9.- ~"-: "....." 9 ..v

soo

"

"

"~

.

. . . .

:~

0

i "1 ~

0

"1

i

,

I

10

.

.

.

.

.

20 30 Time (hours)

*

%'-

: ,

,

40

50

Fig. 11. The e v e n t - d u r a t i o n s p e c t r u m o f A E signals f r o m the e x p e r i m e n t in w h i c h N i - 3 0 C r was oxidized for 20 h o u r s a n d t h e n n a t u r a l l y furnace c o o l e d to r o o m t e m p e r a t u r e ( t i m e i n c r e m e n t was 6 minutes).

Cracking and Spalling of CF203Scale 400

'

'

'

I

539 I

'

1200

I

10OO~ .

1000

~'m 350

B

800 300

600

2s0

400

t~

200 0

200 9 ~

9

v

150 &

lOO .

=

9- 9 9

50

.

.

20

0

"9 "'" " 9 " : " ~" 9 "-" :: %":'.4. :~-': , ' " . " 9 =, . , . : ~ ' ~ ' . ' . 9-..,.~...,~.?--,:.

.

40 60 Time ( h o u r s )

100

80

Fig. 12. The frequency spectrum of AE signals from the experiment in which Ni-30Cr was oxidized for 20 hours and then cooled to room temperature at constant rate (time increment was 6 minutes).

frequency, a n d event d u r a t i o n o f the A E signals collected d u r i n g coolin are smaller t h a n the same parameters for the 20-hr oxidation experimen This implies that when the scale is thinner, it is easier to be buckled tha to be cracked t h r o u g h the thickness, a n d b u c k l i n g c a n occur at a smalb flaw size. This is consistent with the prediction of Eqs. (1) a n d (2) a n d wil the S E M results (described later in Figs. 15 a n d 16).

~,

80

I

'

'

i

I

'

'

'

I

1200

I

1000~

800 60

"~

600

q~

=

b.]

100o

400 40

g3.

9

:.

9

...

. ..:.;=-'-..-

-

200 0

r

20 m_

i

0

i

i

[

20

i

i

,

]

40

,

,

I

60

,

J

80

~

,

,

100

Time (hours) Fig. 13. The peak-amplitude spectrum of AE signals from the experiment in which Ni-30Cr was oxidized for 20 hours and then cooled to room temperature at constant rate (time increment was 6 minutes),

540

Zhang

~" 3 0 0 0 = ~a "~ 2 5 0 0 =

52

' ' 1000~

I

'

'

'

I

I

i

1200 1000 800

e~

600 400

2000 1500

200 0

= 1000

.

." , . ". ~ . . ; ~ : .:".'." .: ", . ..j,-~',": . . , , , . , . . - : ~ . 9

....

500 9 ""

0

and Shores

":"

....

~.".'-

20

9

9

,

'.,

':'-.

9, . . ~ - ~

.....

....,..,..

,,~..:...,.

-.

.

Time

...

"

",

9 .:.

. . . .

",r

40

r

,

I

60

,

~

,

-...

".'.:

J

80

~-:,.

9

.., r ""

,

100

(hours)

Fig. 14. T h e e v e n t - d u r a t i o n s p e c t r u m o f A E signals f r o m the e x p e r i m e n t in w h i c h N i - 3 0 C r w a s o x i d i z e d f o r 20 h o u r s a n d t h e n c o o l e d to r o o m t e m p e r a ture at constant rate (time increment was 6 minutes).

Fig. 15. S E M m i c r o g r a p h o f the f r a c t u r e c r o s s - s e c t i o n o f the C r 2 0 3 scale f o r m e d b y 2 - h o u r s o x i d a t i o n at 1000~

Cracking and SpaUing of

Cr203 Scale

541

Fig. 16. SEM micrograph of the fracture cross-section of the Cr203 scale formed by 20-hours oxidation at 1000~ When the oxide stress reaches a certain value, cracking, and spalling become intense upon further continuous cooling. This process will last over a range of temperature, indicating that there is no unique value of critical-fracture stress which applies to the entire scale. This phenomenon can be explained by a model proposed by H. E. Evans et al., 15'16 which postulates that there is a variation of properties of the oxide scale across the surface such that some areas will spall at low values of the specific strain energy for fracture, W7, and other areas will spall only at high values. Therefore, they suggested an intrinsic variation of W~s, such that a distribution of W7 will exist across the specimen surface. The critical strain energy Wf is achieved primarily by the stress induced by the thermal-expansion mismatch during cooling, i.e., ath. When the sample is cooled to a critical temperature, the low W~ tail of the distribution is entered, and cracking and spallation commences. As AT increases, more of the distribution will be covered, so that an increasingly larger fraction of the surface experiences a strain energy equal to or higher than the local WT. This model explains what happened during cooling, however it can not explain the phenomenon that cracking and spalling continues to occur after the temperature stabilized at room temperature for some time, as shown in Figs. 3-5. We suggest as an explanation of this phenomenon that when

542

Zhang and Shores

cooling stopped at room temperature, the subcritical cracks continue to propagate slowly toward the critical size under the residual stress. Once the crack size reaches the critical value for the local W?, spalling will be induced, Therefore, it is suggested that the key factor controlling the fracture of oxide scale be stress-intensity factor, K, a combination of stress and crack length. ~7 From Figs~ 6 to 14, it can be seen that some events collected after the sample reached room temperature have relatively high frequency and most events have very long event duration, indicating that the processes of crack nucleation, scale buckling, and spalling are all possible during this stage.

The Mechanism of Scale Cracking and Spalling Figures 15 and 16 are the SEM micrographs of the fractured cross-section of the Cr203 scales for exposure of 2 hr and 20 hr, respectively, at 1000~ in pure oxygen. The scale consist of two layers: an outer layer with coarse columnar grains and a thin inner layer with fine equiaxed grains. Many buckled regions were found on the remaining scale, and throughthickness cracks were only found on the top of buckled regions or along the perimeter, as shown in Fig. 17. Therefore it can be suggested that the

Fig. 17. SEM micrograph of a buckled area on the Cr203 scale formed by 20-hours oxidation at 1000~ The arrows indicate the cracks running along the buckled ridge.

Cracking and Spalling of Cr203 Scale

543

scale spalled predominantly via the Route II proposed by H. E. Evans et This mode of failure requires a relatively weak oxide/metal interface or pre-existing interfacial flaws. 6'7 Figures 18 and 19 show spalled areas from 2 hr oxidized sample and 20 hr oxidized sample. The clearly-exposed alloy grains suggest that the interface between the Cr203 scale and the N i - 3 0 C r alloy substrate must be relatively weak, and interfacial flaws may nucleate along the substrate grain boundaries. We also note the deep crevices delineating metal grain boundaries and the rounded corners of grains. These features suggest vapor transport of Cr across the gap separating metal and scale in a buckle could sustain scale growth. The spalling of the areas shown in Figs. 18 and 19 must have occurred at a lower temperature, since there is no new oxide formed on the spalled area. If the spalling had occurred at a relatively high temperature, further oxidation of the exposed alloy would have been observed. Figure 20 shows that some fine oxide nuclei had formed on the spalled area, suggesting that this area must have spalled at a higher temperature. When the interfacial crack meets a through-thickness shear crack, spalling will occur inside the scale, instead of occurring along the interface, as shown in Fig. 21. In this case, a small interfacial flaw can cause spalling under relatively low stress.

al. 5,6

Fig. 18. SEM micrograph of a spalled area on the Cr203 scale formed by 2-hours oxidation at 1000~

544

Zhang and Shores

Fig. 19. SEM micrograph of a spalled area from the Cr203 scale formed by 20 hr-oxidation at 1000~

Fig. 20. SEM micrograph showing a sheared fracture surface of the CF203 scale formed by 20-hr oxidation at 1000~ and the new oxide formed on spalled area as indicated by arrows.

Cracking and Spalling of Cr203 Scale

545

Fig. 21. SEM micrograph showing a spallation inside the scale.

There are two possibilities for the buckled regions to spall. One is that tensile stress will be developed at the perimeter of the buckled region which will fracture the buckled scale and produce spallation. 6 Usually this process will produce a fracture surface nearly perpendicular to the metal/scale interface, as shown in Figs. 15, 16, 18, and 19. Another possibility is that when the rupture strength of the interface is locally low, the interfacial crack at the perimeter of the buckled region will propagate to a critical size at which spallation will o c c u r . 6'7 A. G. Evans et al. 7 suggested that at this critical crack size, the K H at the crack tip will become negative and the interfacial crack will tend to deflect toward the surface, resulting in spalling. The critical-crack size is predicted by Eq. (2). It can be expected that this process produces a shear-fracture surface which makes an acute angle with the interface, as indicated by the arrow in Fig. 20. It has been suggested I that vacancies can condense along the oxide/ substrate interface during oxidation to form interfacial voids. The coalescence of these voids can form interfacial flaws, as schematically shown in Fig. 22. Interfacial flaws can also be formed when scale loses contact with the substrate locally, is During cooling, shear stresses will be generated along the interface. If two small flaws are located close to each other, the shear stress may drive these two flaws to grow into one larger flaw.

546

Zhang and Shores

Fig. 22. Schematicshowingof the process of interfacial-flawformation.

Although the transport of Cr from the alloy interior to the surface requires the flux of vacancies away from the metal/oxide interface, voids could be formed in metal grain boundaries, and espe-.ially at grain triple points. Therefore the interfacial flaws will most likely nucleate along the substrate grain boundaries, as shown in Figs. 15 and 16. The Mechanism of the Intrinsic Distribution of Critical-Fracture Stresses

From the discussion of Section 2 we can suggest that the interface between the Cr203 scale and Ni-30Cr substrate is relatively weak. Therefore buckling is expected to occur under compressive stresses. Equation (1) predicted the critical half length of the interracial flaws required to produce buckling. Equation (1) also shows that at a given value of scale thickness, larger interracial flaws will buckle at lower stress and smaller flaws will buckle at higher stress. Therefore if ao is spatially uniform, the intrinsic distribution of the fracture stresses is mainly derived from a distribution of the sizes of the interracial flaws. The intrinsic distribution of the fracture stresses can be indirectly measured by AE, therefore we can use Eq. (2).to calculate the distribution of the sizes of spalled areas based on AE data, and Eq. (1) to calculate the distribution of sizes of the interfacial flaws which produced the buckling required by spallations. As mentioned earlier, in-situ strain measurement by X-ray diffractional indicates that the growth stress generated on the Cr203 scale formed on Ni-25Cr alloy at 950~ is about - 550 MPa, if Young's modulus is taken as 273 GPa. 1~ Although no similar strain measurements have yet been carried out for Cr203 on Ni-30Cr under our experimental conditions, we would expect the growth stresses to be similar. Therefore, in the following calculation, we will take the growth stress agw = - 5 5 0 MPa, Young's modulus Eox=273 GPa, (~A--%x)= 1.58 • 10 -5 (K-a), 1~ and o =0.3. The creep of substrate during cooling and the effect of temperature on Eox and (~A- %x) have been ignored.

Cracking and Spalling of Cr203 Scale

547

In our experiment, the number of AE events generated inside a temperature interval can be measured as a function of AT as n =f(AT)

(5)

where n is the number of AE events collected in a given AT interval. In the regime of intense AE activity, n vs. f i A T ) is approximately a normal distribution. If we take as a simple model that the spalled area is square and each spalling will produce Z AE events, Z-1 from through-thickness cracks along the perimeter of the buckled region and 1 from the buckling, then the number of the interfacial flaws to be spalled, N, can also be expressed as a function of AT, as f(AT) N - - Z

(6)

where Z would be 5 for a square spallation area. Z might typically range from 5 to 7 or 8. Combining Eq. (2) and Eq. (4) and rearranging: AT =

{

} 1o)

E~ - agw 14.44 (2as)2 Eox~AZaox)

(7)

It is noted that the first term inside the brackets is significantly larger than agw, so the outcome of these calculations is not much affected by the value of agw. Since the function f ( A T ) is experimentally measurable, we can substitute Eq. (7)' into Eq. (6) to obtain the number of spalled areas with size 2as, N, as a function of 2as: N = F(2as)

(8)

where F(2a~) = f ({14.44 E~ (2as) 2

I (1 - o ) O'gw) Eox(0~ A -- 0~OX)J

Using Eq. (8), we can convert the measured distribution of AE events in the temperature domain to the distribution of the sizes of spalled areas. By combining Eq. (1) and Eq. (2) we can get: 1.1 ac = ~ a s

(9)

Using Eqs. (8) and (9) we can convert the size distribution of spalled areas into the size distribution of interfacial flaws. Figures 23-25 show the calculated size distributions of spalled areas for the 2-hr oxidation experiment and the 20-hr oxidation experiments using Eq. (8). The bold lines in these three figures are the normal-distribution curves fitted to the calculated data. The parameters for the distributions

548

Zhang and Shores

81

-

-

C a l c u l a t e d distribution

Normal distribution

based o n A E data

standard deviation = 2 p m

7

mean = 50.2 pm ~-

6

=

4 @

.m

2

= Z

44

46

48

50

52

54

56

58

S i z e o f s p a i l e d a r e a s (txm) Fig. 23. The calculated size distribution of the spalled areas from the sample which was oxidized for 2 hours.

are listed in Table I. It can be seen that the curves calculated by Eq. (8) fit the normal distribution satisfactorily except that the sizes of the spalled areas are truncated at a lower limit. Because the flaw size is directly related to the spall size, the sizes of the interfacial flaws produced in the isothermal-oxidation stage are inferred to have a normal distribution. This is the

16 -

-

Calculated distribution based o n A E data

~

Normaldistribution standard deviation = 9 p m m e a n = 130 ~tm

t_

12

~

8

@ L.

.o

4

=

z

0

~

90

~

100

110

120

130

140

150

160

170

S i z e o f s p a l l e d a r e a s (ktm) Fig. 24. The calculated size distribution of the spalled areas from the sample oxidized for 20 hours and then cooled to room temperature at constant cooling rate.

Cracking and Spalling of Cr203 Scale 25

~J t_

m

- -

549

Calculateddistribution based on AE data

Normal distribution standard deviation = 13 gm mean = 130 pm

20

15

t,. O

10

t_

5

Z 0

90

100

110

120

130

140

150

160

170

Size of spalled areas (gm) Fig. 25. The calculated size distribution of the spalled areas from the sample oxidized for 20 hours and then furnace cooled to room temperature.

reason why there is a distribution of critical-fracture stresses across the scale. The normal distribution of the spall size of the constant-cooling-rate experiment is slightly different from the one of the furnace-cooling experiment. This small difference may be due to the effect of the substrate creep. Since the sample in the constant-cooling-rate experiment spent a longer time at higher temperatures, a part of the oxide stresses might be relaxed by the substrate creep. Therefore, the calculated oxide stress ao from Eq. (4) will be higher than the actual stress value. Using Eq. (9) and the AE data one can also construct a distribution of interfacial flaw sizes corresponding to the onset of buckling. One such Table I. Parameters of the Size Distributions of Spalled Areas

Calculated mean dia. (#m) (AE data) Calculated std. dev. ~ m ) (AE data) Measured mean dia. (pm) (SEM) Measured std. dev. Gum) (SEM)

2-hr oxidation furnace cooling

20-hr oxidation furnace cooling

20-hr oxidation constant-rate cooling

50.2

130

130

2

13

9

22.5

45

--

7.5

15

--

550

Z h a n g and Shores

graph is shown in Fig. 26 for a 20-hr oxidation followed by furnace cooling. However, such an approach is likely to underestimate the number of interfacial flaws for two reasons. First, the AE data are truncated at a predicted flaw size of 72 #m, which is the smallest flaw able to cause a buckle, and which in turn led to cracking and spalling by the stress produced by 20 hours isothermal oxidation and by cooling from 1000~ to room temperature. The left tail of the normal distribution curve in Fig. 26 tends to approximately zero at about 60 # m for this experiment. However, this does not necessarily mean that the smallest interfacial-flaw size is 60 #m; certainly smaller flaws would be expected. But since flaws with size smaller than 72 p m do not cause AE activity under these conditions, we have no experimental information about smaller flaws, and we have tentatively represented them as following a normal distribution. Second, the model has ignored any interactions between nearby flaws. When a spallation occurs, the local stress in the adjacent region is reduced by the permanent strain accompanying the spallation. Thus, flaws in regions near a spallation will experience a lower stress than the nominal value calculated by Eq. (4), and they will cause spallation only when the local stress has risen to the critical value, or when the flaw itself has propagated to the critical size. Thus there may exist flaws larger than as in regions where the local stress is less than tro. Figures 27 and 28 are SEM-measured size distributions of spalled areas of a 2-hours oxidized sample and a 20-hours oxidized sample. More

25 -

~1

"d d,_

-

distribution standard deviation = 7.5 I.tm

Calculated distribution

Normal

based on AE data 2 0

mean = 75 p.m

15

.9

,gl

Z 50

60

70

80

90

100

Interfacial flaw size (ktm)

Fig. 26. The calculated size distribution of the interfacial flaws which caused buckling of the scale for the sample oxidized for 20 hours and then furnace cooled to room temperature.

Cracking and Spalling of

Scale

Cr203

551

200 9

v

SEM measured data

~

Normal distribution standard deviation = 7.5 l.tm mean = 22.5 txm

150

100

50

0 0

10

20

30

40

e~e'' 50

60

S i z e o f s p a l l e d a r e a s ([tin) Fig. 27. The SEM-measured size distribution of the spalled areas from the sample oxidized for 2 hours.

than 500 spalled areas were measured on each sample. These data fit a normal distribution reasonably well. The average size of spalled areas is about 22.5/tm for the 2-hours oxidized sample and is about 45 pm for the 20-hours oxidized sample (Table I). Both mean diameters are smaller than the calculated values using Eq. (8). This difference probably arises because of defects in the scale, whereas the derivation of Eqs. (1) and (2) modeled

300 250 200 "

150

100 so

0 0

20

40

60

80

100

S i z e o f s p a l l e d a r e a s ([tm) Fig. 28. The SEM-measured size distribution of the spalled areas from the sample oxidized for 20 hours.

552

Zhang and Shores

the delaminated region of the scale as a clamped smooth, defect-flee circular plate. 8'19The real oxide scale has defects such as grain boundaries and many small sharp V-notches on the scale surface, as shown by all the SEM micrographs. These defects will cause local stress concentrations which make the scale easier to be buckled and spalled. Therefore, an interracial flaw with a size smaller than predicted by Eq. (1) can cause buckling. Similarly, the stress concentrations around the perimeter of the buckled region, caused by the oxide grain boundaries and the small sharp V-notches, will make the buckled region spall before the interfacial crack under the buckled region propagates to the size predicted by Eq. (2). Figures 27 and 28 also show that the SEM-measured average size of spaUed areas is reduced when the scale becomes thinner. The amount of the reduction in the average size of spalled areas is close to the value predicted by Eq. (2). Equations (1) and (2) show that under certain stress, the thinner scale will buckle and spall at a smaller size. This can be verified from Figs. 6-8 and 18. It is also possible that in the early stage of a long-time isothermal ~xidation, when the scale is still thin, buckling will occur on small interfacial flaw if the growth stress is large enough. SUMMARY After 2 hours oxidation at 1000~ in pure oxygen flow, a Cr203 scale about 2-#m thick was formed on the Ni-30Cr alloy. The scale grew to 5 #m thick after 20 hours oxidation and consisted of two layers: an outer layer with coarse columnar grains and a thin inner layer with fine equiaxed grains. AE data show that through-thickness cracks (and possibly also interfacial cracks) nucleated during the isothermal-oxidation stage and the early stage or cooling. These crack nuclei can propagate upon further cooling and enable scale spallation. Intense fracturing of the scale started after a certain amount of cooling, indicating that a minimum stress is required to start the process. The scale cracks and spalls over a temperature range, implying the existence of an intrinsic distribution of critical fracture stresses across the entire scale. AE data characterized by high-frequency, high-peak amplitude and short-event duration are taken as evidence of the nucleation of cracks. On the other hand, AE signals having low frequency, low-peak amplitude and much-higher-event duration are taken as evidence of buckling. Some evidence suggests that the event duration is proportional to the area of the crack. SEM observations of the surface o f oxidized samples indicates that most spallation events occurred via the Route II proposed by H. E. Evans et al., 5"6 i.e., the scale buckled first under the compressive stress, then the buckled region fractured along the perimeter.to cause spalling. This indi-

Cracking and SpaUing of Cr203 Scale

553

cares that the interface between C r 2 0 3 scale a n d N i - 3 0 C r alloy substrate is relatively weak. If we assume that most spallation occurred via the R o u t e II, the critical-interfacial-flaw size can be calculated as a f u n c t i o n of the oxide stress using the A. G. Evans v model (Eqs. 1 a n d 2). A p p l y i n g Eq. (2) to the A E data, a size d i s t r i b u t i o n of spalled areas can be obtained. We can convert the size d i s t r i b u t i o n of spalled areas into the size d i s t r i b u t i o n of pre-existing interfacial flaws required by the buckling process by c o m b i n i n g Eqs. (1) a n d (2). The results show that the size distributions of spalled areas are n o r m a l . The n o r m a l distribution of interfacial-flaw sizes is the origin of the d i s t r i b u t i o n of the critical-fracture stress. We have measured the d i s t r i b u t i o n of spall diameters from SEM images. The average measured diameter is less t h a n the average diameter predicted from A E data, b u t close e n o u g h to lend s u p p o r t to the present simple model.

ACKNOWLEDGMENT This work was supported by a grant from the U.S. D e p a r t m e n t of Energy u n d e r contract D E - F G 0 2 - 8 8 E R 4 5 3 3 7 .

REFERENCES 1. J. Stringer, Corros. Sci. 10, 513 543 (1970). 2. D. J. Baxter and K. Natesan, Rev. High Temp. Mater. 5, 149 250 (1983). 3. P. Hancock and R. C. Hurst, in Advances in Corrosion Science and Technology, M. G. Fontana and R. W. Staehle, eds. (Plenum Press, New York, 1974), pp. 1-84. 4. J. K. Tien and J. M. Davidson, in Stress Effect and the Oxidation of Metals, J. V. Cathcart, ed. (The Metallurgical Society of AIME, 1974), pp. 201-219. 5. H. E. Evans and R. C. Lobb, Corros. Sci. 24, 209 222 (1989). 6. H. E. Evans, Mater. Sci. Eng. A120, 139 146 (1989). 7. A. G. Evans and R. M. Cannon, Mater. Sci. Forum 43, 243 268 (1989). 8. A. G. Evans and J. W. Hutchinson, Int. J. Solids Struct. 20(5), 455-466 (1984). 9. A. G. Evans, G. B. Crumley, and R. E. Demaray, Oxid. Met. 20(5/6), 193 216 (1983). 10. J. J. Barnes, J. G. Goedjen, and D. A. Shores, Oxid. Met. 32(5/6), 449 469 (1989). I 1. D. Zhu and D. A. Shores, Unpublished work. 12. J. H. Stout, J. G. Goedjen, Q. Guo, and D. A. Shores, in Proceedings of the Symposium on X-Ray Methods in Corrosion and Interfacial Electrochemistry, A. Davenport and J. G. Gordon II, eds. (The Electrochemical Society, 1992), pp. 101 114. 13. B. B. Jha,~ B. Raj, and A. S. Khanna, Oxid. Met. 26, 263-273 (1986). 14. W. Christi, A. Rahmel, and M. Schutze, Mater. Sci. Eng. 87, 289-293 (1987). 15. H. E. Evans and R. C. Lobb, "Eurocorr 87" (Dechema, Frankfurt, 1987). 16. H. E. Evans, Mater. Sci. Teehnol. 4, 415 420 (1988). 17. Y. Zhang and D. A. Shores, in Proceedings of the Symposium on Oxide Films on Metals and Alloys. B. R. MacDougall, R. S. Alwitt, and T. A. Ramanarayanan, eds. (The Electrochemical Society, Toronto, Canada, 1992), pp. 250 264. 18. F. H. Stott, Mater. Sci. Technol. 4, 431 438 (1988). 19. S. Timoshenko and J. M. Gere, Theory of Elastic Stability, 2nd Ed (McGraw Hill, New York, 1961).