www.elsevier.com/locate/surfcoat

Surface & Coatings Technolog

Analysis of localized damage in EB-PVD/(Ni, Pt)Al thermal

barrier coatings

Mei Wen a, Eric H. Jordan b,*, Maurice Gell a

aDepartment of Metallurgy and Materials Engineering, University of Connecticut, Storrs, CT 06269, United StatesbDepartment of Mechanical Engineering, University of Connecticut, Storrs, CT 06269, United States

Received 8 February 2005; accepted in revised form 27 May 2005

Available online 21 July 2005

Abstract

The structural integrity of the thermally grown oxide (TGO) in the EB-PVD/(Ni, Pt)Al thermal barrier coatings was examined using

photo-stimulated luminescence piezo-spectroscopy (PLPS). PLPS spectra exhibiting both a high stress component and a low stress (or stress-

free) component were observed during thermal cycling (i.e. bimodal luminescence). The fraction of bimodal spectra increases initially, then

decreases, and increases again when close to failure. It is shown that the bimodal luminescence originates from stress relaxation caused by

localized damage. It is proposed that the initial increase of bimodal luminescence is related to cracking caused by the volume change

associated with the phase transformation of TGO from u to a-Al2O3; whereas, the increase at the final stage is due to TGO cracking and

spallation. Area stress maps show the gradual accumulation of damage and indicate that PLPS is a useful tool for detection of the initiation

and progression of TBC spallation.

D 2005 Elsevier B.V. All rights reserved.

Keywords: Thermal barrier coatings; Photo-stimulated luminescence; Thermal cyclic oxidation; Localized damage

1. Introduction

Thermal barrier coatings (TBCs) are widely used in both

gas turbine engines and aircraft engines to protect metal

components from high operating temperatures in order to

improve durability and engine efficiency [1–4]. It is highly

desirable to have a non-destructive inspection (NDI)

technique to detect early damage of thermal barrier coatings

and assess the remaining life because coating spallation can

lead to premature component failure.

The photo-stimulated luminescence piezo-spectroscopy

(PLPS) technique is useful for non-destructively evaluating

the TBC condition and detecting early damage [5–11].

PLPS has demonstrated the ability for measuring the

residual stress in the TGO non-destructively by using the

frequency shift of R luminescence lines from the Cr3+ ions

0257-8972/$ - see front matter D 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.surfcoat.2005.05.044

* Corresponding author. Tel.: +1 860 486 2371; fax: +1 860 486 4745.

E-mail address: jordan@engr.uconn.edu (E.H. Jordan).

incorporated into the TGO [5–15]. The peak shift of the

collected spectra is a linear function of the value of the trace

of the stress tensor (proportional to the hydrostatic pressure)

in the sampled probe volume. Remaining life predictions

can be made based on the systematic change of stress level

with cycles at least for some TBC systems [16].

In addition to stress, the shape and broadening of the

luminescence spectrum may relate to damage. Peng and

Clarke [17] proposed that peak broadening is the result of

stress variation within the probe volume. Damage in the

TGO would be expected to affect the localized stress and

hence peak shape. To date such systematic shape changes

with cycles have not been found in experiments, except as

associated with bimodal peaks, as now discussed. One of

the extreme cases of stress variation is that of a spectrum

that contains two sets of R1–R2 peak pairs (also referred

to as bimodal), exhibiting high and low (or stress free)

stress components [9,15,17–19]. It has been suggested

that the low stress peak pairs are from the localized

damage region [9,15,17–19]. In such cases, the presence

y 200 (2006) 5193 – 5202

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–52025194

of a low stress (or stress free) component is an indication

of damage and it is potentially a means of quantifying

local damage in the TGO. Selcuk and Atkinson [15,19]

mapped the low stress regions for an EB-PVD/Pt diffusion

bond coat TBC and showed these regions were isolated at

first and tended to coalesce into larger regions toward the

end of life. They observed for some specimens that the

number and area fraction of low stress components

showed high initial level that gradually decreased before

increasing at the onset of spallation. Sohn et al. [9] also

found for an EB-PVD/MCrAlY system that the fraction of

low stress component increased initially, then gradually

decreased and then increased again when close to failure.

The increase close to failure can be related to TGO

spallation. However, the cause for the initial high values is

not clear.

In the present paper, we provide further experimental

evidence of the initial increase of bimodal spectra, followed

by a decrease, and finally an increase prior to failure for

another EB-PVD Pt modified NiAl bond coated TBC

system. Most significantly, direct evidence is presented

associating luminescence peaks having a stress free compo-

nent (bimodal spectra) with TGO cracks. The mechanisms

of TGO cracking in the different stages are studied as well.

(Ni, Pt)Al bond coats can form metastable alumina

phases, such as g,u-Al2O3, in early oxidation and then

transform to the stable a-Al2O3 phase. Besides stress

measurement and damage detection, PLPS can also be used

to determine the oxide type and its spatial distribution,

which makes it a valuable tool to study alumina phase

transformations [5,20,21].

2. Experimental procedure

The disk-shaped (2.54 cm in diameter and 0.32 cm in

thickness) TBC specimens investigated in this study were

supplied by an engine manufacturer. They consist of a 140

Am thick electron beam physical vapor deposited (EB-

PVD) 7 wt.% Y2O3 stabilized ZrO2 (YSZ) coating, a 50

Am thick grit blasted platinum-modified nickel-aluminide

bond coat [(Ni, Pt)Al] and a single crystal CMSX-4

superalloy substrate. The typical compositions of the

coating and the superalloy used in this study are shown

in Table 1. The specimens were bond coated on both sides

to allow study of the oxidation behavior of the bare bond

coat using the backside of specimens. The bond coats were

Table 1

Typical composition of various layers of the TBC system used in this study

Layer Material Composition (wt.%)

Substrate Single crystal

superalloy CMSX-4

9 Co, 6.5 Cr, 6.5 Ta, 5.6 Al,

6.0 W, 3.0 Re, 0.6 Mo,

0.1 Hf, 1.0 Ti, balance Ni

Bond coat Ni-rich Pt-Al 20 Pt, 21 Al, balance Ni

Ceramic TBC Yttria stabilized zirconia 7 Y2O3 in ZrO2

grit blasted prior to TBC deposition by the EB-PVD

process.

Cyclic oxidation tests were performed in air at 1151 -Cand 1121 -C in a CMi rapid temperature cycle furnace

(CM Inc., Bloomfield, NJ) [9]. The thermal cycles

consisted of a 10-min heat-up to peak temperature (1151

-C and 1121 -C), a 40-min hold at peak temperature (1151

-C and 1121 -C), followed by a 10-min forced air quench.

The temperature of 1121 -C was selected because it

corresponded to the sponsor’s goal temperature for the

future and 1151 -C was selected to correspond local hot

spots and to obtain result without having tests of excessive

duration. Specimens were removed at specified intervals to

perform PLPS measurements. A total of 12 specimens were

investigated and 8 of them were cycled to failure. Failure of

the specimen was defined by spallation of more than 50%

of the total area.

PLPS measurements were made using a RenishawiRamascopei 2000 (Renishaw, Glouchestershire, UK) in

conjunction with a Leicai DM/LM light microscope.

Detailed information on the experimental set-up can be

found elsewhere [9–11]. At specified intervals, a series of

spectra on either a coarse or a fine rectangular grid were

collected on each sample. The laser beam was focused on

the surface of YSZ and the spot was scanned over the

specimen by moving the computer controlled stage at the

specified step size. At each position, a luminescence

spectrum was acquired. The minimum step size possible is

1 Am. In coarse mapping, a total of 121 spectra were

acquired in a 12 mm�12 mm grid area, with a step size of

1.2 mm. In fine mapping, a total of 2601 spectra were

acquired in a 2 mm�2 mm grid area with a step size of 40

Am. For each specimen, the origin of the mapping was fixed

and matched so that the same general area was measured as

a function of thermal cycling. PLPS measurements can also

be made on polished cross sections, however in doing so

one must contend with the fact that cutting has changed the

stress state and local polishing damage must be ruled out as

an influence. Because of these issues such measurements

were not made in the present case. Such measurements

when done carefully have shown non-uniform stress

distribution in the direction of TGO thickness with higher

stress near the TGO/bond coat interface [22,23]. In the

present case the stresses reported were measured when the

laser penetrated through the YSZ top coat and are volume

averaged through the TGO thickness.

Spectra were analyzed by a constrained peak fitting

procedure. In this procedure, spectra are fitted using mixed

Gaussian-Lorentzian peaks, where the peak parameters

satisfy certain constraints. The constraints include the

relationship of R1–R2 peak spacing with peak position

and inequality constraints with respect to the R1–R2 peak

height ratio, the ratio of full width at half maximum height

of the R1 and R2 peak, and the Lorenzian/Gaussian

fractions. The best fit to the acquired spectra is assessed

by a merit function [15,19] and spectral characteristics can

0

5000

10000

15000

20000

25000

14300 14350 14400 14450 14500

Wavenumber (cm-1)

Wavenumber (cm-1)

Inte

nsi

ty (

a.u

.)

(a)

R 1

R 2

R 1

R 2

0

5000

10000

15000

20000

14300 14350 14400 14450 14500

Inte

nsi

ty (

a.u

.)

(b)

R 2 (Stress-Free)

R 1 (Stress-Free)

R 2 (Stressed)

R 1 (Stressed)

R 2 (Stress-Free)

R 1 (Stress-Free)

R 2 (Stressed)

R 1 (Stressed)

Fig. 1. Typical photoluminescence spectra showing (a) one set of peak pairs

and (b) two sets of peak pairs.

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–5202 5195

be extracted from the fitting functions obtained. Detailed

information of some very similar procedures can be found in

Refs. [18,19]. The fraction of bimodal spectra (the spectra

exhibiting high and low (or stress free) stressed compo-

nents) is of particular interest, because the regions produc-

ing bimodal spectra may be related to localized damage. The

fraction of bimodal spectra were calculated according to the

equation:

Number fraction of bimodal¼Nbimodal= NunimodalþNbimodalð Þð1Þ

where: Nunimodal= the number of spectra that show only one

stress component; Nbimodal= the number of spectra that show

two stress components.

The oxidation of the bond coat was studied on the bond

coated only backside of one specimen. The sample was

polished before oxidation in order to remove the original

thin oxide layer. Then, it was subject to the same cyclic tests

at 1151 -C to form a new oxide layer. PLPS measurements

were carried out as mentioned before to study the fraction of

bimodal peak and the alumina phases and their spatial

distribution in oxide scales.

The microstructure of the sample surface was charac-

terized using a scanning electron microscope Philips

ESEM 2020. Quantitative analysis of crack density was

carried out using UNIX based image analysis software

microGOP 2000 (ContextVision, Linkoping, Sweden). The

SEM micrographs were converted into binary image in

which the black represents the cracks and the white

represents the region without cracks. The length for every

crack was measured for each image. The crack density was

calculated as the total length of cracks divided by the total

image area.

3. Experimental results

3.1. Typical luminescence spectra during cycling

PLPS measurements were carried out for as-coated TBC

specimens and for thermally cycled specimens at specified

intervals. Typically two types of spectra were observed in

the specimens. One type is the spectra showing only one

set of R1–R2 luminescence peaks and correspondingly

exhibiting one major stress level (i.e. unimodal lumines-

cence). The other type is the spectra showing two sets of

R1 and R2 luminescence peaks exhibiting two stress

components, with one of them being close to zero (i.e.

bimodal luminescence). Examples of each type of spectra

and the corresponding fitted peaks using the constrained

peak fitting procedure are shown in Fig. 1. After analyzing

the spectra, it was found that the stresses for the unimodal

spectra ranged from 1.2 to 2.5 GPa compression during

thermal cycling. Detailed information concerning the stress

evolution can be found in Ref. [24]. For the bimodal

spectra, the low stress component was usually close to zero

(below 0.5 GPa in compression), but occasionally reached

a maximum tension level of 0.8 GPa, whereas the high

stress component ranged from 1.6 to 3.0 GPa in

compression.

3.2. Evolution of fraction of bimodal spectra

The evolution of the fraction of bimodal spectra with

thermal cycling at 1151 -C for the TGO under the ceramic

top coat from fine mapping is shown in Fig. 2. The fraction

of bimodal spectra shows systematic change with thermal

cycling. The fraction increases initially up to 5 cycles,

gradually decreases and then increases again when close to

failure. The evolution from coarse mapping shows the same

trend as fine mapping.

3.3. Stress distribution

Stress mapping was carried out with thermal cycling at

both 1151 -C and 1121 -C. Stress distribution maps based

on the bimodal luminescence contribution were obtained.

Fig. 3 shows one example of stress distribution maps for a

specimen cycled at 1121 -C. The black regions represent

the regions exhibiting bimodal spectra with the appearance

of low stress components whereas the white regions

represent the regions showing unimodal spectra. The maps

show that the number and size of regions having bimodal

spectra increase with thermal cycling. The increase is more

0.0

1.0

2.0

3.0

4.0

0 50 100 150 200

Specimen No.

1 2 3

Thermal Cycles

Fra

ctio

n o

f B

imo

dal

Sp

ectr

a (%

)

1 2 3

Fig. 2. Fraction of bimodal spectra of TBC specimens as a function of

thermal cycles at 1151 -C.

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–52025196

pronounced when the sample is close to failure. It is noted

that the fraction of bimodal spectra extracted from these

maps does not decrease after initial increase which is

different from the evolution of the fraction of bimodal

spectra at 1151 -C shown before. The possible reason is

that we did not catch the correct time in the early stage to

make measurements and missed the data points showing

the initial increase and then decrease. However, we did

(c) After 470 cycles

μm

μm

μmμm

(a) After 27 cycles

1000

500

-500

-1000-1000 1000-500 0 500

0

1000

500

-500

-1000-1000 1000-500 0 500

0

1

-

-1

Fig. 3. TGO stress mapping distribution of TBC specimens after (

observe that the number and size of regions having

bimodal spectra increased when sample was close to

failure for tests at both 1151 -C and 1121 -C.

3.4. Combined PLPS and SEM study of formation of

alumina scales

The backside of specimen which was bond coated but

not ceramic coated was polished to remove the original thin

oxide layer and then heated at 1151 -C to form a new

alumina scale.

Fig. 4 shows an optical image of the alumina

morphology after 10 min oxidation and the corresponding

PLPS spectra. Both unimodal spectra (one set of peak

pairs) (Fig. 4b) and bimodal spectra (two set of peak pairs)

(Fig. 4c and d) of a-Al2O3 were observed. The stress

value for the low stress component of bimodal spectra was

close to zero. The regions producing bimodal spectra

corresponded consistently to location of microcracks (Fig.

4a, points 2 and 3) whereas all regions without visible

cracks gave unimodal spectra (Fig. 4a, point 1). The

fraction of bimodal spectra for the oxide layer without a

ceramic top coat as a function of cyclic oxidation time is

shown in Fig. 5. It increases sharply up to 10 min (point

(b) After 240 cycles

Intact

Damaged (Bi-modal)

μm

μm

000

500

500

000-1000 1000-500 0 500

0

a) 27 cycles, (b) 240 cycles and (c) 470 cycles at 1121 -C.

20 ∝∝ m

3

2

1

(a)

20 ∝ m

3

2

1

20 μm

3

2

1

(a)

014200 14300 14400 14500 14600

Wavenumber (cm-1)

Inte

nsi

ty (

a.u

.)

(b) Point 1

30000

40000

14400

Inte

nsi

ty (

a.u

.)

(b) Point 1

0

10000

20000

30000

40000

14200 14300 14400 14500 14600

Wavenumber (cm-1)

Wavenumber (cm-1)

Inte

nsi

ty (

a.u

.)

(c) Point 2

Inte

nsi

ty (

a.u

.)

(c) Point 2

0

10000

20000

30000

14200 14300 14400 14500 14600

(d) Point 3

Inte

nsi

ty (

a.u

.)

(d) Point 3

20000

10000

Fig. 4. Optical image (a) and corresponding luminescence spectra of

alumina scale formed on different location of bare bond coat after oxidation

for 10 min at 1151 -C showing: (b) unimodal, (c) bimodal, and (d) bimodal.

0

10

20

30

40

50

0 50 100 150

Oxidation Time (Hour)

(a)

(b)

(c)

(d)

Fra

ctio

n o

f B

imo

dal

Sp

ectr

a (%

)

Fig. 5. Evolution of fraction of bimodal spectra on bare bond coated sample

after oxidation at 1151 -C for (a) 10 min, (b) 1 h, (c) 100 h, and (d) 150 h.

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–5202 5197

a), gradually decreases (points a to c) and then increases

(points c to d) again after 100 h until spallation. The

fraction of bimodal spectra for the oxide layer without a

ceramic top coat shows the same trend with cyclic

oxidation as that for TGO under the ceramic top coat

with the initial increase, a subsequent decrease and a final

increase before failure.

Fig. 6a–d shows the microstructure of aluminum oxide

surface after different exposure times. The alumina layer,

formed on the surface of bond coat after oxidation for 10

min at 1151 -C, has a large number of microcracks. With

increasing oxidation time, the microcracks start to heal and

result in a decrease in the number of microcracks. In some

places, microcracks form outward protrusions (see arrows in

Fig. 6b). The density of microcracks reaches a minimum at

100 h and the grain structure becomes evident. With

increasing oxidation time, the alumina scale spalls, accom-

panied by the appearance of large cracks (Fig. 6d). The

crack density as a function of time is shown in Fig. 7. The

density of cracks as a function of oxidation time increases

initially, decreases and then increases again after 100 h. The

change of the fraction of bimodal spectra as function of

oxidation time is consistent with the change of the density of

cracks (compare Figs. 5 and 7). The fraction of bimodal

spectra and the density of cracks exhibit a qualitative

relationship and change in a similar manner with thermal

cycling.

u-Al2O3 can be detected by the PLPS method due to the

fact that u-Al2O3 has characteristic luminescence peaks

easily distinguished from those of a-Al2O3 [20,21]. Quite

fortuitously, two small regions in which a-Al2O3 and u-

Al2O3 coexisted were observed to confirm the transforma-

tion after the initial oxidation stage. Fig. 8a shows an optical

image of one region in which a-Al2O3 and u-Al2O3

coexisted after oxidation of 10 min at 1151 -C. The small

island of u-Al2O3 was surrounded by a-Al2O3 matrix. The

existence of a-Al2O3 (R2=14420 cm�1) and u-Al2O3

(R2=14613 cm�1) was identified by the different frequen-

cies of the characteristic peaks in PLPS (Fig. 8b). Similar

results were also seen after oxidation of 2 min. With

increasing oxidation time, the relative amount of a-alumina

increased by the transformation of u to a alumina. Fig. 8c is

an oxide-type distribution area map indicating the distribu-

tion of alumina (alpha and non-alpha). The figure shows the

(a) 10 min (b) 1 hr

(c) 100 hr (d) 150 hr

5 ∝∝ m

5 ∝ m 5 ∝ m

(a) 10 min (b) 1 hr

(c) 100 hr (d) 150 hr

5 ∝ m5 μm

5 ∝ m5 μm 5 ∝ m5 μm

5 μm

Fig. 6. Oxide morphology on bare bond coated sample after oxidation at 1151 -C for (a) 10 min, (b) 1 h, (c) 100 h, and (d) 150 h.

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–52025198

initial presence of an island of u alumina surrounded by a-

alumina. The island is reduced in size, by the transformation

of u to a alumina, after 1-h exposure, and then disappears

after 5 h.

The aluminum oxide layer formed on the bare bond coat

surface started to become wavy or rumpled after oxidation

for 5 h. The amplitude of rumpling increased with

increasing oxidation time (Fig. 9). Measurements of the

local residual stress in the oxide scale were made using

PLPS. Since the oxide rumpling wavelength (the distance

between the ridges and the valleys) is sufficiently large, the

0.00

0.05

0.10

0.15

0.20

0.25

0 50 100 150

(a)

(b)

(c)Cra

ck D

ensi

ty (

1/μm

)

(d)

Oxidation Time (Hour)

Fig. 7. Evolution of crack density on bare bond coated sample after

oxidation at 1151 -C for (a) 10 min, (b) 1 h, (c) 100 h, and (d) 150 h.

local stress in the vicinity of the ridges and the valleys can

be resolved with a probe (about 2 Am) whose size is

smaller than rumpling wavelength. Fig. 10a and b show a

typical morphology of rumpled a-Al2O3 scale after

oxidation for 100 h and the frequency shift of this oxide

layer from a series of PLPS measurements, respectively. It

is clear that the frequency shift is smaller at the ridges than

at the valleys. Although the exact stress value cannot be

obtained due to the wavy surface, the corresponding

hydrostatic compressive stress is smaller at the ridges than

at the valleys. This is as expected from finite element

calculations [25–27] and is consistent with the expectation

that wrinkled scale produces tensile stress across coating-

metal interface at the ridges and compressive stress in the

valley. The oxide stress measured by PLPS is the overall

stress in the small probe volume of oxide. The in-plane

stress for the oxide is in compression at both the ridges and

valleys; whereas, the out-of-plane stress is tensile at the

ridges and compressive at the valleys. Thus, the overall

hydrostatic compressive stress for the oxide scale at the

ridges should be smaller than that at the valleys due to the

normal stress contribution. The experimental results which

were obtained from PLPS (Fig. 10) show the frequency

shift is indeed smaller at the ridges than at the valleys. We

also note that no nickel oxide on the surface layer of bond

coat was observed in the experiments, consistent with other

results for this type of bond coat [21]. It may also be

m

θ θ , α α - Al

2O

3

α α - Al2O

3

10 μm

(a)

0

5000

10000

15000

20000

25000

30000

35000

14140 14240 14340 14440 14540 14640

-Al 2O 3

-Al 2O 310 min

1hr

5hr

Wavenumber (cm -1)

Inte

nsi

ty (

a.u

.)

0

5000

10000

15000

20000

25000

30000

35000

14140 14240 14340 14440 14540 14640

α -Al2O3

θ -Al2O310 min

1hr

5hr

Wavenumber (cm -1)

Inte

nsi

ty (

a.u

.)

(b)

m

30

m μm μm

10 min 1 hr 5 hr

0

30 30

30 30 3000 m mμm

30

μm m m

10 min 1 hr 5 hr

α α + θ α

0

30 30

30 30 3000 μm μm

(c)

Fig. 8. Optical image (a), PLPS spectra (b) and oxide type distribution maps (c) of alumina scale on bare bond coated sample showing the distribution of a and

u alumina after oxidation at different times at 1151 -C.

25 ∝∝m 25 ∝m

(a) 5 hr (b) 150 hr

25 ∝m25 μm 25 ∝m25 μm

(a) 5 hr (b) 150 hr

Fig. 9. Top surface of bare bond coated sample showing rumpling after oxidation for (a) 5 h and (b) 150 h at 1151 -C.

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–5202 5199

Ridge

V

50 μm

Ridge

Valley

(a)

0

5

10

15

20

25

0 5 10 15

ValleysRidges

Measurement Number

0

5

10

15

20

25

0 5 10

(b)

Ridges

Fre

qu

ency

Sh

ift

(cm

-1)

Fig. 10. Optical image (a) and luminescence results (b) of a series of

measurements of the rumpled alumina scale after oxidation for 100 h at

1151 -C.

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–52025200

rationalized by the thermodynamic stability of a-alumina

in the Ni–Al–O system [4].

1 ∝∝m1 μm

Fig. 11. Top view of alumina scale morphology after oxidation for 10 min at

1151 -C showing whiskers.

4. Discussion

4.1. Relationship between bimodal spectra and localized

damage and its implication in NDI use

In the present work, the bimodal spectra have been

clearly connected with TGO cracking (localized damage).

First, there was a perfect one to one correspondence

between the measured bimodal spectra and observed cracks

in the bare TGO formed on the bond coat. That is, bimodal

spectra were observed only from regions of the TGO in the

immediate vicinity of cracks and never detected where

cracks were not seen. Secondly the crack density and

fraction of bimodal spectra change consistently with thermal

cycling. The association of cracks with bimodal spectra is

illustrated in Fig. 4. When the probed regions in spectro-

scopy contain both intact and damaged region, the probed

regions will have stress variations and produce two sets of

peak pairs with the high stress component corresponding to

intact regions; whereas, the low stress component corre-

sponding to damaged regions. It would be reasonable to

assume that the likelihood of getting bimodal spectra

increases with increasing localized damage. In fact, there

were consistent trends in the fraction of bimodal spectra and

the crack density as a function of oxidation time. The

fraction of bimodal spectra increased when the density of

cracks increased (compare Figs. 5 and 7). The correspond-

ence of the trends in the crack density and the fractional

presence of bimodal spectra is consistent with the idea that

the bimodal spectra is caused by the small regions of stress

free TGO associated with cracking.

Tolpygo and Clarke [21] studied the theta–alpha trans-

formation in alumina scales on platinum-modified nickel

aluminides and found that bimodal spectra can originate

from the alumina scales containing whisker morphology.

TGO consisting of continuous and whiskers morphology is

constrained and unconstrained, respectively, under in-plane

compression with the underlying metal. The luminescence

from regions containing both constrained and unconstrained

TGO would give rise to bimodal spectra [21]. In the present

work bare alumina scales were also observed containing

whiskers morphology (Fig. 11) and the whiskers disap-

peared after oxidation for 25 h. In contrast with Tolpygo’s

study, however, bimodal spectra were not observed from the

scales containing whiskers but only from the regions of the

TGO in the immediate vicinity of observed cracks. The

possible reason is that the volume of TGO scale with

whiskers are too small compared to the continuous and

constrained TGO layer in the present case making the low

stress peaks from the whisker region difficult to detect.

Because the association of bimodal spectra and cracking

was directly observed on the bare TGO layer, the fraction of

bimodal spectra measured through the ceramic layer gives

an indication of extent of damage. In the case of measure-

ments made through the ceramic TBC layer, the diameter of

the spatial region from which spectra are collected in a

single measurement is estimated to be around 70 Am due to

scattering by the columnar structure of the EB-PVD TBC

[28]. As a result, the presence of bimodal spectra through

the TBC has a statistical aspect representing the degree of

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–5202 5201

cracking of an area of TGO. The change in the fraction of

bimodal spectra seen through the TBC parallels the change

observed in the occurrence of bimodal spectra in the bare

oxide and also in crack density observed on the bare oxide.

In addition, the stress distribution mapping based on

bimodal spectra contribution can provide a visual record

of the number and size of the damage sites. The stress

distribution maps (Fig. 3) show clearly that damage

accumulation occurs with thermal cycling. Both the number

and the size of continuous damage sites increase with

thermal cycling. Therefore PLPS is a potential method for

determining the number and size of damage that could be

used to assess remaining life.

4.2. Mechanisms of crack formation during thermal cycling

The early increase in bimodal spectra and associated

observed cracking in this work is attributed to the effect of

volume change associated with the transformation of u-

Al2O3 to a-Al2O3 [21]. Numerous studies of nickel

aluminides [21,29–32] have shown that metastable alumina

phases such as u-Al2O3 may form and then transform into

stable phase a-Al2O3 in the very early stage of oxidation

which results in a volumetric shrinkage due to the phase

change. The present experimental observations are consis-

tent with the idea that transformation of u-Al2O3 to a-Al2O3

is responsible for the early increase in crack density and

associated increase in bimodal spectra. First, the trans-

formation happened early in the oxidation, as confirmed by

the occasionally observed regions in which a-Al2O3 and u-

Al2O3 coexisted (Fig. 8). Second the appearance of

maximum peak crack density observed for the bare oxide

was close to and before the time at which u-Al2O3

disappeared based on the PLPS spectra. Finally, the u-

Al2O3 island was observed surrounded by a-Al2O3 matrix

as would lead to isolated cracks as observed here in Fig. 8.

The subsequent decrease in the occurrence of bimodal

spectra is associated with crack healing. This crack healing

is directly observed as decreasing crack density seen on the

bare oxide and indicated by the decrease in the fraction of

bimodal spectra for both bare and coated oxide.

By direct observation the final increase in the fraction of

bimodal spectra is associated with TGO cracking which

occurs nearly exclusively near the highest parts of protrud-

ing regions of the TGO as shown in Figs. 6d and 9b. It is

worth noting that these cracks are quite open and of the type

expected if there existed in-plane tension in the TGO. These

cracks are therefore different from delamination cracks that

result from out of plane tensile stresses caused by in plane

compressive stresses in regions with high positive curvature

found in other investigations [25–27,33]. Thermal expan-

sion mismatch effects in the TGO will result in a high state

of compression which was measured to be between 1.6 and

3.0 GPa in the present case. Accordingly the observed

cracks are probably not due to thermal expansion mismatch

related in-plane compressive stresses. The proposed cause

of these local tensile stresses relieved by cracking is tension

produced by shape change due to a combination of rumpling

and volume changes associated with diffusive transport that

includes not only aluminum loss from oxide formation but

also Ni influx from the substrate [34]. Final resolution of

this issue requires additional evidence. In summary, the

increase in the fraction of bimodal spectra near the end of

cyclic life is caused by increased TGO cracking which is

almost exclusively in regions of positive curvature. The

local tensile stress associated with rumpling and transport

induced shape change seems to be the best candidate

explanation of cracking at this time. Arguments that valleys

have higher frequency shift (i.e. higher compressive stress)

than ridges is based on an assumption of uniform in-plane

compression and resulting out of plane stresses required by

force balance using membrane theory [25]. The cracking

and shape change observed, and the implied tensile stresses,

suggest that the stresses in the TGO are also greatly

influenced by shape change. Accordingly it is not surprising

that there is a large stress difference for different measure-

ments in Fig. 10b, as the influence of shape change history

could easily vary stress greatly from place to place but does

not quite overwhelm the expected pattern based on

membrane theory.

5. Conclusions

The structural integrity of the thermally grown oxide

(TGO) in the EB-PVD/(Ni, Pt)Al thermal barrier coating

was examined using photo-stimulated luminescence piezo-

spectroscopy (PLPS). It is shown that bimodal luminescence

originates from stress relaxation caused by local cracking.

There is a perfect one to one correspondence of cracked

regions with bimodal spectra and uncracked regions with

unimodal spectra. The fraction of bimodal luminescence

spectra as a function of oxidation time increases initially,

decreases, and then increases again when close to failure.

The corresponding crack density follows the same trend.

Experimental observations are consistent with the idea that

the initial increase in bimodal spectra is caused by cracking

related to the volume shrinkage associated with the phase

transformation of TGO from u to a-Al2O3. The final

increase is associated with cracking near ridges at the

TGO surface. Both area stress maps, based on the bimodal

luminescence and average fraction of bimodal spectra with

cycles, have the potential for non-destructive prediction of

impending failure.

Acknowledgements

This research was funded by the Department of Energy

under the UTSR 02-01-SR097, administrated by the South

Carolina Institute for Energy Studies, Clemson University,

with Dr. Richard Wenglarz as the program monitor.

M. Wen et al. / Surface & Coatings Technology 200 (2006) 5193–52025202

References

[1] R.A. Miller, Surf. Coat. Technol. 30 (1987) 1.

[2] N.P. Padture, M. Gell, E.H. Jordan, Science 296 (2002) 280.

[3] R.T. Jones, in: K.H. Stern (Ed.), Metallurgical and Ceramic Protective

Coatings, Chapman and Hall, London, 1996, p. 194.

[4] A.G. Evans, D.R. Mumm, J.W. Hutchinson, G.H. Meier, F.S. Pettit,

Prog. Mater. Sci. 46 (2001) 505.

[5] M. Gell, S. Sridharan, M. Wen, E. Jordan, Int. J. Appl. Ceram.

Technol. 1 (2004) 316.

[6] R.J. Christensen, D.M. Lipkin, D.R. Clarke, K.S. Murphy, Appl. Phys.

Lett. 69 (1996) 3754.

[7] D.R. Clarke, R.J. Christensen, V. Tolpygo, Surf. Coat. Technol. 94/95

(1997) 89.

[8] M. Gell, K. Vaidyanathan, B. Barber, J. Cheng, E. Jordan, Metall.

Mater. Trans., A Phys. Metall. Mater. Sci. 30 (1999) 427.

[9] Y.H. Sohn, K. Vaidyanathan, M. Ronski, E.H. Jordan, M. Gell, Surf.

Coat. Technol. 146–147 (2001) 102.

[10] L. Xie, Y.H. Sohn, E.H. Jordan, M. Gell, Surf. Coat. Technol. 176

(2003) 57.

[11] S. Sridharan, L. Xie, E.H. Jordan, M. Gell, Surf. Coat. Technol. 179

(2004) 286.

[12] Q. Ma, D.R. Clarke, J. Am. Ceram. Soc. 77 (1994) 298.

[13] Q. Ma, D.R. Clarke, J. Am. Ceram. Soc. 76 (1993) 1433.

[14] D.M. Lipkin, D.R. Clarke, Oxid. Met. 45 (1996) 267.

[15] A. Selcuk, A. Atkinson, Acta Mater. 51 (2003) 535.

[16] E.H. Jordan, Y.H. Sohn, W. Xie, M. Gell, L. Xie, F. Tu, K.R. Pattipati,

P. Willet, Autotestcon, IEEE, Valley Forge, PA, 2001, p. 593.

[17] X. Peng, D.R. Clarke, J. Am. Ceram. Soc. 83 (2000) 1165.

[18] J.A. Nychka, D.R. Clarke, Surf. Coat. Technol. 146–147 (2001)

110.

[19] A. Selcuk, A. Atkinson, Mater. Sci. Eng., A Struct. Mater.: Prop.

Microstruct. Process. 335 (2002) 147.

[20] Q. Wen, D.M. Lipkin, D.R. Clarke, J. Am. Ceram. Soc. 81 (1998)

3345.

[21] V.K. Tolpygo, D.R. Clarke, Mater. High Temp. 17 (2000) 59.

[22] T. Tomimatsu, S. Zhu, Y. Kagawa, Acta Mater. 51 (2003) 2397.

[23] T. Tomimatsu, S.J. Zhu, Y. Kagawa, Scr. Mater. 50 (2004) 137.

[24] M. Wen, E. Jordan, M. Gell, Mater. Sci. Eng., A Struct. Mater.: Prop.

Microstruct. Process 398 (2005) 99.

[25] X.Y. Gong, D.R. Clarke, Oxid. Met. 50 (1998) 355.

[26] A.G. Evans, M.Y. He, J.W. Hutchinson, Mater. Sci. Eng., A Struct.

Mater.: Prop. Microstruct. Process. 245 (1998) 168.

[27] J. Cheng, E. Jordan, M. Gell, Acta Mater. 46 (1998) 5839.

[28] J.A. Nychka, D.R. Clarke, S. Sridharan, E. Jordan, M. Gell, M.J.

Lance, C.J. Chunnilall, I.M. Smith, S.R.J. Saunders, R. Pillan, V.

Sergo, A. Selcuk, A. Atkinson, K.S. Murphy, Surf. Coat. Technol.

163–164 (2003) 87.

[29] D.M. Lipkin, D.R. Clarke, M. Hollatz, M. Bobeth, W. Pompe, Corros.

Sci. 39 (1997) 243.

[30] G.C. Rybicki, J.L. Smialek, Oxid. Met. 31 (1989) 275.

[31] M.W. Brumm, H.J. Grabke, Corros. Sci. 33 (1992) 1677.

[32] J. Doychak, M. Ruhle, Oxid. Met. 31 (1989) 431.

[33] K. Vaidyanathan, E. Jordan, M. Gell, Acta Mater. 52 (2004) 1107.

[34] M.W. Chen, M.L. Glynn, R.T. Ott, T.C. Hufnagel, K.J. Hemker, Acta

Mater. 51 (2003) 4279.