Effect of substrate curvature on residual stresses and failure modes of an air plasma sprayed thermal barrier coating system

Available online at www.sciencedirect.com

Journal of the European Ceramic Society 33 (2013) 3345–3357

Effect of substrate curvature on residual stresses and failure modes of an air plasma sprayed thermal barrier coating system D. Liu a,b,∗ , M. Seraffon c,d , P.E.J. Flewitt a,e , N.J. Simms c , J.R. Nicholls c , D.S. Rickerby f a

Interface Analysis Centre, University of Bristol, Bristol BS2 8BS, UK Department of Mechanical Engineering, University of Bristol, Bristol BS8 1TR, UK c Cranfield University, Cranfield, Bedford MK43 0AL, UK d National Physical Laboratory, Hampton Road, Teddington TW11 0LW, UK e School of Physics, HH Wills Laboratory, University of Bristol, Bristol BS8 1TL, UK f Rolls-Royce plc, Derby DE24 8BJ, UK

b

Received 8 March 2013; received in revised form 30 April 2013; accepted 4 May 2013 Available online 15 June 2013

Abstract A set of aerofoil shaped air plasma sprayed thermal barrier coated (APS-TBC) specimens were adopted in this paper to investigate the stress distributions in the ceramic top coat (TC) and the thermally grown oxide (TGO), the mechanism of local crack generation and propagation at the TC/BC (bond coat) interface. The failure mode of the TBC system, the distribution of asperities at TC/BC interface, thickness of the TC and BC, and the TC microstructure were found to be influenced by substrate curvature. Residual stress was therefore measured across the thickness of the TC, along the undulating TGO and mapped at locations of asperities where failure tended to occur to interpret the initiation of local failure. The role of the TGO was investigated via its chemical bonding with the TC and the decohesion occurring at the TGO/BC interface. The crack propagation at the interface has been discussed with respect to the macro-failure of the TBC system. © 2013 Elsevier Ltd. All rights reserved. Keywords: Air plasma sprayed thermal barrier coating; Substrate curvature; Residual stresses; Interfacial failure modes

1. Introduction Blades used for land-based gas turbine power plants and aero engines operate at high gas temperatures above which the parent superalloy cannot withstand without additional internal cooling.1,2 To protect the blades and to further promote efficiency, turbine blades with ceramic thermal barrier coatings (TBCs) were introduced into service during the early 1970s.3 The benefits these low-thermal conductivity ceramic systems offer the base material of components in the engines are significant, including extended component lives, due to a reduced creep rate and lower rate of oxidation. TBCs normally consist of three layers: an outer ceramic top coating (TC) to insulate the superalloy substrate from high gas flow by sustaining an appreciable temperature difference of up to 300 ◦ C depending on TC

∗

Corresponding author at: Interface Analysis Centre, University of Bristol, Bristol BS8 1TL, UK. Tel.: +44 01173311174. E-mail address: [email protected] (D. Liu). 0955-2219/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jeurceramsoc.2013.05.018

thickness4,5 ; a metallic bond coat (BC) providing good adherence between the TC and substrate as well as oxidation and/or corrosion protection to the underlying superalloy. A thermally grown oxide (TGO), predominately Al2 O3 , develops between the TC and the BC during exposure to the service temperature.6,7 Within the BC, the Al present, either in solution or in the form of ␤-NiAl phase, creates and feeds this protective alumina layer. It constitutes the Al reservoir of the coating. Depletion of this reservoir results in compositional changes in the TGO that may form ␣-Al2 O3 and/or a mixture of chromia and spinels ((Cr,Al)2 O3 ) and (Ni(Cr,Al)2 O4 ).8 The growth of the TGO between the TC and BC can be divided into three stages: a transient stage, when all thermodynamically stable oxides can form (e.g. ␪-Al2 O3 , Cr2 O3 , NiO, etc.); a steady-state stage, when the high temperature stable, long-term phase (␣-Al2 O3 ) is established and grows; and finally a breakaway stage, when the growth of less protective oxides leading to the failure of the scale.9 At oxidation temperatures below 1000 ◦ C, the transient ␪-Al2 O3 phase is also thermodynamically stable and can co-exist with ␣-Al2 O3 in the TGO.10

3346

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

The failure of APS-TBC systems under oxidation is a complex mechanism and is still not fully understood.11 The process involves several general phenomena including: thermal expansion mismatch stresses; the growth of the TGO at an undulating BC/TC interface; the formation of spinel/mixed oxides due to Al-depletion in the BC; the sintering of the porous ceramic TC leading to a deterioration of strain tolerance and thermal resistivity; the degradation of the metal/ceramic interface toughness; cracking, crack coalescence, and delamination of the TBC at either interface or within an individual layer (TGO or TC).12 The residual stresses within the TBC arise from three main sources: (i) the deposition process,6,7,13 (ii) thermal stress mismatch during cooling14,15 and (iii) growth of the TGO. This is usually associated with the nucleation and growth of microcracks caused by stresses developing between the various layers or within the TGO.16,17 The degree of curvatures, concave or convex, that are characteristic of the geometry of turbine blades and vanes, influence the thermal/residual stress distribution within multilayer TBC system18,19 which greatly affects the energy release rates associated with interface delamination and crack propagation (circumferential or axial)18,20 during spallation. The roughness of TC/BC interface has been found to influence the nature and evolution of residual stresses in the ceramic as well as in the TGO.21 Protrusions of ZrO2 into the BC are preferred sites for damage initiation. The sintering effects at high temperature in such small volumes fixed by the geometry of the BC produce localised tensile stresses. At the same time, the stresses within the TGO layer contouring such protrusions change during oxidation. Undulations of TGO into the BC usually show higher compressive stresses compared to TGO undulations into the TC.21 In the TGO, besides the substrate curvature variation, the interface undulation and local geometrical features can also affect the stress distribution and relaxation under cyclic behaviour. Therefore the failure modes within this layer are complex.22,23 In this paper, a set of modified aerofoil-shaped specimens, with representative turbine blades’ curved features, were used to study the effect of the substrate curvature. The deposition process, the resulting TC microstructure, the TC, BC and TGO thicknesses, and the TC/TGO interface undulation were considered in this study as parameters that may influence the failure modes of the TBC system. Residual stress measurements were then undertaken on cross-sections along the thicknesses of TC and TGO after extended thermal exposure to correlate the residual stress evolution to the coating degradation in terms of crack initiation and propagation. Four types of microstructures at the TC/TGO interface were found to have the potential to start and eventually lead to interfacial failure. The residual stress maps at these locations are discussed with respect to mechanical debonding at the interface and are compared with existing models. 2. Materials and methods Modified aerofoil-shaped specimens were designed to recreate curvatures found on industrial gas turbine blades (Fig. 1).

Fig. 1. Picture of a modified aerofoil-shaped specimen. P1, P2, P3, P4 and P5 are convex areas whereas P6 and P7 are concave.

Deposited on a CMSX4 Ni-superalloy substrate by high-velocity oxy-fuel (HVOF) spraying was an AMDRY 995 CoNiCrAlY BC (Table 1), over-coated with an air plasma sprayed (APS) yttria-stabilised tetragonal zirconia (YSZ) TC. The manufacturing process caused the TC and BC thicknesses to vary with the curvature. The BC was measured to be between 50 and 135 ␮m and the TC between 160 and 300 ␮m. Long cycle thermal exposure tests were carried out on a set of eight specimens in resistance heated furnaces at 925 ◦ C in air for up to 10,000 h. After each cycle (250 h), the specimens were removed gradually from the hot furnace and cooled in air to room temperature. Selected specimens were removed for destructive examination after periods of 100, 2740, 4000, 7000 and 10,000 h. As shown in Fig. 1, the specimens were each ∼43 mm long and ∼22 mm wide with curvatures that changed around the periphery. Seven locations, P1-P7, the curvature of which are listed in Table 2, were chosen around the modified aerofoil-shaped samples for study (Fig. 1). The curvature was calculated as the inverse of the radius, thus, a flat surface would have a curvature equal to zero, with the sign indicating whether the surface was concave or convex. Both scanning electron microscopy (SEM) and energy dispersive X-ray spectroscopy (EDX) were used to observe cross-sections in order to characterise the microstructure of the specimens. Line-scans of TC residual stress and mapping of residual stresses along the undulation of the TGO were undertaken on the cross-sections at each position using non-destructive laser stimulated spectroscopy, both Raman spectroscopy (RS) and photo-stimulated luminescence piezospectroscopy (PLPS).24 The equipment is a Renishaw RamaScope spectrometer, model 2000, fitted with a laser source with a wavelength, λ, of 514 nm and an integrated microscope that allows the observation of the specimen surface. The residual stress measurement locations and traverses are shown in Fig. 2. Two Raman spectra for the TC are shown in Fig. 3. Peak 6 (640 cm−1 ), which has pronounced intensity in this case, was fitted by mixed Gaussian and Lorentzian method and chosen to calculated the stress by a conversion factor, 5.60 cm−1 GPa−1 , obtained previously using diamond anvil cell on this set of APSTBC.25 The other peaks located within the lower frequency range, as shown in Fig. 3, have significantly reduced intensity at many locations. This scenario limits the application of the peak centred at 465 cm−1 for calibration of APS-YSZ, as proposed by Limarga et al.26 based on bulk dense pure YSZ material. Hence,

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

3347

Table 1 Composition (in wt.%) of modified aerofoil-shaped specimen’s substrate and BC. Alloy

Ni

Cr

Co

Al

W

Ti

Mo

Ta

Re

Hf

CMSX-4 AMDRY 995

Base 32

6.9 21

9.6 Base

5.6 8

6.4

1

0.6

6.5

3

0.1

Y 0.5

Table 2 Substrate curvature of measured positions. Concave curvature defined as minus value. Position Curvature

(mm−1 )

P1

P2

P3

P4

P5

P6

P7

0.006

0.33

0.866

0.178

3.023

−0.062

−0.195

Fig. 2. Schematic of residual stress measurement along the thickness of TC using Raman spectroscopy and TGO stress along the undulating interface using PLPS.

in the present particular APS-TBC, the stress value, σ (GPa), has been calculated using the following relationship: σtri-axial =

v 5.60

(1)

where ν is the Raman shift of the peak (cm−1 ) from stress-free state. As proposed by Liu et al.25,27 it is considered to be tri-axial stress that is measured in the probed volume of the APS-TBC. A shift to higher frequency indicates compressive stress.

Fig. 3. Typical Raman spectra collected from TC characterising the six-peak pattern of tetragonal YSZ and the reduced intensity for peaks with lower frequency.

The residual stress in the TGO was measured according to the shift of PLPS based on characteristic luminescence R1 and R2 lines (∼693–694 nm) which originates from chromium ions (Cr3+ ) in Al2 O3 . A conversion factor of 5.07 GPa−1 cm−1 was adopted in this paper to convert the PLPS shifts to biaxial stress.24 3. Results 3.1. TC thickness and TGO microstructure change with curvature around the specimen The TC thickness was measured on cross-sections through the specimens as shown in Fig. 4. The red dotted line suggests that as the curvature became more convex, the thickness of the TC is reduced. No significant variation in TC thickness was found between the six specimens, therefore the set of coatings investigated have comparable stability. The high scatter in the data at locations P7 and P6 (concave locations) was caused by the high roughness of the TC in these regions (see Fig. 5); while at location P5 (convex location), it can be explained by the particular geometry of the trailing edge of the blade. Position P5, located at the trailing edge, had the largest curvature and the TC thickness measured was the thinnest. Position P2 near to the leading edge (also a convex location) was an exception which had a smaller curvature but thinner TC. There was no obvious

3348

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

Fig. 4. TC thickness as a function of curvature on specimens oxidised at 925 ◦ C for 0, 100, 2740, 4000, 7000 and 10,000 h.

change of TC thickness at each location due to thermal exposure. The thickness variation in the TC around the specimens is expected to change the ability of the coating to accommodate local stresses and hence the failure modes observed. The BC surface of most locations around the modified aerofoil-shaped samples were relatively smooth, as shown in Fig. 5(a). However, roughness increased greatly on a specimen’s concave features. During manufacturing, the spray gun was programmed to travel around the sample and deposit an equal amount of molten powder in every location. As a molten particle hits the sample surface, it creates a splat. Around concave

Fig. 6. Amount of spinels/mixed oxides in TGO as a function of the curvature for samples exposed at 925 ◦ C.

features, a small part of the splat might spatter on the adjacent concave position, which can explain higher coating thicknesses as well as high roughness on concave features. At position P7, the TGO undulated more and was also highly uneven. The surface of the ceramic TC at this location became very rough as a consequence. In convex regions, the amplitude of the surface’s undulation was observed to reach a maximum of 30 ␮m, while it exceeded 50 ␮m (a quarter of the nominal TC thickness) for the rough concave regions (see Fig. 5(b)). The curvature of the sample also affected the growth of the TGO. Indeed, the amount of spinel/mixed oxides formed, instead of the protective ␣-Al2 O3 , at 925 ◦ C reduced significantly in areas of convex curvature (Fig. 6). The high fraction of spinel/mixed oxides at location P5, with a curvature of 2.86 mm−1 , may be explained by the thin section shape of the trailing edge. Aluminium depletion in this region occurs more rapidly, as the BC is thinner and the surface area to volume ratio is greater due to the sharpness of the convex curvature. The influence of the geometry of this modified aerofoil-shaped specimen on the concentration of spinels has been associated by Seraffon et al.28 with the increased undulation of the TGO in these concave locations. 3.2. Residual stress in the top coat

Fig. 5. SEM images of TC, TGO and spinels in specimens exposed at 925 ◦ C for 7000 h. (a) Location P1 with flat curvature and (b) location P7 with concave curvature.

The residual stress in the TC was measured at seven locations in each specimen from the as-coated condition to an exposure time of 7000 h in terms of line-scans from the TC/BC interface to the TC surface. As shown in Fig. 7(a), in the as-coated condition, the residual stresses were within a similar range across the TC thickness, with variations from ∼0.5 GPa to ∼1.5 GPa and a median value of 0.9 GPa. In the specimen oxidised for 100 h at 925 ◦ C, the median TC stresses generally deceased to ∼0.6 GPa (less tensile) but was compressive (∼−0.3 GPa) at some locations adjacent to the interface (Fig. 7(b)). Following further oxidation, up to 4000 h, the stress in the cross-section through the thickness of the TC becomes less tensile (median of 0.4 GPa), whilst the maximum compressive stress near the interface decreased to ∼−0.15 GPa, with some higher tensile stress regions being present near the interface (∼0.5 GPa). These tensile stress regions may contribute to cracking and decohesion along the interface, and the reduced compressive stress

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

3349

Fig. 8. The residual stress change with substrate curvature measured on ascoated and oxidised specimens.

Fig. 7. The residual stress across the thickness of TC (a) in as-coated condition, in specimen oxidised at 925 ◦ C for (b) 100 h and (c) 4000 h.

indicates reduced cohesion between the TC and TGO layers with extended thermal exposure. Comparing the three plots in Fig. 7, it is evident that the growth of TGO has an important role on the residual stress distribution across the cross-section; based on the oxidised specimens, 925 ◦ C for 100 h and for 4000 h, the residual stresses in the TC were constrained by the interface up to a distance of ∼20 ␮m into the TC from the BC/TC interface. Beyond this distance, in the specimen oxidised for 100 h for example, the TC stresses remained stable with fluctuations between ∼0.45 GPa and ∼0.75 GPa, which is about a third of the

magnitude of the variation observed for the as-coated condition. Not all the line-scans have been plotted for clarity, but the trend of residual stresses in Fig. 7 is considered to be representative of all. Data within each of the lines-scans was averaged (for the data at thicknesses above the interface constraint zone) and the standard deviation calculated to demonstrate the variation of residual stress around the specimen with substrate curvature (Fig. 8). In the as-coated condition, the TC deposited on the substrate’s flat area, P1, showed the least tensile stress in the TC. The change of curvature, whether concave or convex, both increased the stress values. Large changes in stress magnitude were observed for curvatures ranging from ∼−0.2 mm−1 to ∼0.35 mm−1 . Further changes from ∼0.35 mm−1 to ∼3 mm−1 did not lead to further variation in stress. The variation of stress as a function of curvature was similar for oxidised specimens. However, the magnitude of fluctuation in stress at each location reduced with increase in oxidation time, Fig. 8. In specimens oxidised for 7000 h for example, the variation reduced to ∼±0.05 GPa from ∼±0.2 GPa in the as-coated condition. This behaviour could also be observed in the data presented in Fig. 7, where the stress variation through the TC thickness decreases sharply with extent of thermal exposure. A large decrease in the magnitude of tensile stress within the TC was observed also with progressive oxidation. As shown in Fig. 8, a significant decrease occurred in the first 100 h of oxidation (from a mean value ∼1.05 GPa to ∼0.6 GPa), whereas it is necessary for oxidation up to 7000 h to produce a similar further reduction of stress, from a mean value of ∼0.6 GPa to ∼0.15 GPa. This is consistent with the fact that the much of sintering of the TC happened within the first several hundred hours of thermal aging,29 which contributes most significantly to the stress relaxation of the TC significantly. 3.3. Stress mapping associated with the four types of TC intrusions at the TC/TGO interface The microstructure and residual stress of the TC, close to the TC/TGO interface, was investigated to better understand

3350

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

Fig. 9. SEM images of TC/BC interface from (a) location P2, (b) location P4, (c) location P5 and (d) location P3, in sample oxidised at 925 ◦ C for 100 h. Dashed lines represent the outline of APS-YSZ powder particles.

the correlation between the failure modes at the interface and residual stress state of the TC. SEM images of cross-sections through the TC/TGO interface at each location around the specimens were compared. Since these are 2-D images, it was important to integrate many images to understand if the 3-D structure could be classified. As a result, four types of TC intrusions into the BC were observed, whose microstructure depended upon the splat particle configuration (Fig. 9). Type 1, in Fig. 9(a), occurred on deep and broad undulations filled by a single splat particle. This type of intrusion was created during manufacture when a single molten powder particle was deposited on the bare BC. After exposure, the TGO contoured that intrusion. Type 2 intrusions, in Fig. 9(b), were also filled by a single splat particle but unlike type 1, the intrusion was more convoluted and smaller. Stresses involved during the TGO growth might deform these asperities during oxidation causing a part of the splat to be trapped by the growing oxide. Type 3, in Fig. 9(c), occurred when the intrusion was deep and broad with splat particles following the shape of the BC surface. Unlike type 1, the particle did not retain a round shape but flattened and elongated, suggesting it was fully molten on impact. Type 4 intrusions, in Fig. 9(d), the most common, were shallow and narrow undulations filled with one or more small splat particles. The upper boundary of the splat was usually aligned with the edges of the intrusion. This type of asperity probably formed when a high velocity particle lodged into an initially smooth BC during TC deposition, causing deformation of the BC. Residual stress mapping has been undertaken for the four types of intrusions. In type 1 intrusion, as shown in Fig. 10(a),

Fig. 10. (a) Type 1 intrusions and (b) the stress mapping in TC and TGO. Numbering of the locations of measurements is from left to right from 1 to 12.

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

residual stress measurements have been taken both in the TC and TGO along the interface with numbering from 1 to 12. The compressive TC and TGO stresses both decreased in magnitude towards zero from measurement points 1–6, Fig. 10(b). At the concave trough, the TGO stress tended to become slightly more compressive, whereas the TC stress in this region moved from compressive towards zero or even tensile. After the trough, the residual stress difference between TGO and TC were reduced to ∼0.2 GPa, together with a shift in the stress levels in both the TGO and TC towards less compressive and more tensile stress, respectively. The decohesion between the TC and TGO occurred off-centre of the concave peak into the TC, where the stress in TC became tensile. For type 2 intrusions, the TC and TGO interface were more tortuous and a strong mechanical bond between the two layers were created by physical diffusion of ions from the Al2 O3 into the YSZ and vice versa, Fig. 11(a). Higher stress/strain accumulated at these links during thermal exposure, causing a larger mismatch and eventual failure. In Fig. 11(b), the failure of a type 2 intrusion occurred by separation of the intrusion from the outer TC (the TC further from the interface) and cracks were present at the narrow neck. The numbering from 1 to 16 of the residual stress measurements along the undulating interface follows the path of the arrows, Fig. 11(b). Inside the intrusion, measurement numbers 6–13, the TGO stress fluctuated between a tensile stress range (∼0.1 GPa) and compressive stress range (∼−1.3 GPa), which is the location where TC decohesion occurred. After the intrusion, measurement numbers 14–16, a higher stress difference was measured between the TGO and TC (∼1.6 GPa) and delamination was also observed. Therefore, the fluctuation in stresses in the TC or the large stress differences between the TC and TGO are indicative of the potential for interfacial debonding or micro-cracking. A type 3 intrusion had a similar failure mode to type 2 (Fig. 12(a)) that the intrusion splat separated from the outer TC. Stress measurements were undertaken along the undulation of the interface numbered from 1 to 10. Decohesion happened at the trough of the undulation. At this location, the stress in the TGO was compressive and sustained a large difference with that of the TC, which was in a tensile state (Fig. 12(b)). After the intrusion, measurement numbers 9–10 in Fig. 12(b), the residual stresses in the TC became compressive, and the stress difference between TC and TGO stress was modified as a consequence. A type 4 intrusion consisted of a group of splats of smaller size lying in the concave asperity (Fig. 13). Discontinuities between these small splats had the potential to open as cracks. The residual stress was measured along the undulating interface and was numbered from 1 to 7. At the trough, two additional points (a and b in Fig. 13(a)) in the TC were measured to investigate the residual stress variation with increasing distance from the TC/TGO interface. As shown in Fig. 13(b), the TGO residual stress exhibited a large variation at locations where TGO had small intrusions extending into the TC. The measurements taken at location a and b showed that the tensile stress increased within the TC with increasing distance from the TC/TGO interface.

3351

Fig. 11. (a) Discontinuity in TC, and TC/TGO bonding for a type 2 intrusions; (b) cracks at the neck area of the intrusion, and the locations of stress measurements marked from 1 to 16 for both TGO and the TC inside the intrusion, 1–4 for the outer TC; and (c) stress distribution.

To summarise, although the four types of intrusions vary in microstructure and modes of failure, the residual stress variations within the TC and TGO showed similar features: (i) the difference between TC and TGO residual stresses is larger on one side of the trough and much smaller on the other side. Take type I intrusion as an example, Fig. 10, point 1–5 sustains a ∼1.2 GPa TC/TGO difference, whereas point 10–12 ∼0.3 GPa difference; (ii) There are large fluctuations of the TGO and TGO residual stresses at the site of decohesion as described above for all four types of intrusions. Thus, the intrusions are found to modify the residual stress differences between TC/TGO in terms of local decohesion or micro-cracking.

3352

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

Fig. 14. Schematic of the stress state of splats after deposition.

4. Discussion 4.1. The accommodation of stress and crack propagation at the interface

Fig. 12. (a) Separation between intrusion and outer TC and decohesion occurred for a type 3 intrusion and (b) the stress mapping in marked locations.

Fig. 13. (a) Type 4 intrusion with the cracking along the TGO/BC interface, and the locations of stress measurement, marked from 1 to 7 together with point a and b; (b) the residual stress distribution in TC and TGO at the marked locations in (a).

Measurement of the residual stress along the TC/TGO interface at the four types of intrusions showed that the local roughness and structure of the ceramic influences greatly the stress state of the system for a thermally sprayed TC. The stress difference between the TC and TGO layer is an important driver for crack initiation and propagation. Moreover, as mentioned in the results section, the TGO stress can constrain the TC stress at the near-interface region, for distance of up to ∼20 ␮m. If the TGO and TC were well adhered, the TC stress should change with the TGO at a similar rate. The concave trough in type 1 intrusions showed that the change of TC stress and TGO stress are not in-phase, thereby indicating the decohesion of TC from the TGO, Fig. 10(a). However, type 1 intrusions do not allow cracks to propagate through the TC because the asperities are too large for the crack to penetrate as no short splat boundaries were present to provide a crack path. In this case, the crack propagated into the TGO or just followed the TC/TGO interface. For type 2 intrusions, the stress in the TC changed sign because of the ‘neck’ shaped convoluted structure. Cracks were measured at three locations: (i) the concave bottom of the neck, (ii) the entrance of the neck and (iii) the discontinuity between this splat and the outer TC. The cracks originated in the TGO continued to propagate along the splat boundary and then across the fine strip of ceramic. There is one obvious similarity between type 2 and type 1 intrusions and that is the stress difference between the TC and TGO are larger on one side of the intrusion compared with the other. This is a result the intrusion morphology allowing the stress levels to re-adjust along the undulating interface by introducing decohesion or cracking into the system. For type 3 intrusions, upon deposition, the residual stress in the first layer of splats was tensile as the substrate constrained the splat during cool down and shrinking7 as shown schematically in Fig. 14. Delamination in this type of intrusion happened at concave locations where the TGO stress was large and compressive. Sometimes, the cracks were able to propagate along the splat boundary; however, if the splat boundary was too far from the TC/TGO interface, the crack was observed to continue propagating within the TGO. The cracks arrested when the chemical bonding between TC and TGO was sufficiently good, as observed at many other type 3 intrusion locations (Fig. 15). The stress gradient distributed between the small splats contained in the intrusion may have the potential to open existing defects and discontinuities to form incipient cracks. There were

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

3353

suggesting that the mixed zone was more a physical phenomenon than chemical diffusion controlled. This means that the thermodynamically stable ␪-Al2 O3 phase, present in the TGO at this temperature, grew between the micro-cracks in the TC. This bonding can partially explain the constraint that the TGO can apply to the TC stress observed at the interface. The stress would start to accumulate at the interface until the stored energy reached some threshold value before being released in the form of micro-cracks. The strength of this bonding also affected the propagation of cracks (Fig. 15), and promoted discontinuities in the TC and TC/TGO interface, which in turn open as cracks34 as mentioned in the study of type 4 intrusions. Tuan et al.35 studied the mechanical properties of Al2 O3 –ZrO2 composites and reported that the strength of Al2 O3 was reduced by adding t-ZrO2 or m-ZrO2 . They found that increasing the ZrO2 content led to a reduced densification of the Al2 O3 , and a decrease of the elastic modulus.35 Larger scale energy release behaviour at the interface resulted in the nucleation of cracks and decohesion of the TC from TGO or BC. The TGO growing at the TC/BC interface plays an important role on the stress distribution within the multi-layered (BC + TGO + TC) coating system by generating stress gradients through the thickness of TC. On the other hand, these stress gradients may modify the microstructures of the TC during thermal cycling, by opening or closing such features as microcracks, splat boundaries or defects, in turn reducing the local stresses. Fig. 15. (a) The bonding of the TC and TGO arrests or alters the crack path, and (b) stress difference between region 1 and 2 promote the crack propagation within a near TGO interface splat.

4.2. Substrate curvature and the failure modes at intrusions

two forms of cracking observed in this type of intrusion: (i) microcracks within the TC which could propagate at the splat boundary, often at the edge of the asperity to cause decohesion, and (ii) the relaxation of the accumulated stress can occur by the separation of two adjacent TC splats. For this geometry of interfacial intrusion, another less frequently observed failure mode was the large crack following the TGO/BC interface, Fig. 13(a). From the SEM examinations of cross-sections, bonding formed by the TGO growing into the TC was observed at the TC/TGO interfaces. This physical bond, known as mixed zone29–31 was first observed between EBPVD-TCs and MCrAlY-BCs. It was found to be caused by the BC heat treatment before deposition of the ceramic.30–32 According to Murphy et al.30 , the formation of the Al2 O3 + ZrO2 mixed zone is promoted by the presence of transient Al2 O3 (␥ or ␪), and this was confirmed by Levi et al.31 , as well as Jarayam et al.33 , who stated that ZrO2 is much more soluble in metastable Al2 O3 than in stable ␣-Al2 O3 . Indeed, an initial layer of ␥-Al2 O3 could dissolve and contain up to 17 mol% YSZ; and ␪-Al2 O3 up to 40 mol%. Levi et al.31 proposed that Zr and Y are able to slow down the conversion of ␥ to ␣ and thus delay the transition between TGO growth mechanisms by Al outward diffusion and O inward diffusion.30 Exposures in these studies were usually at temperatures above 1000 ◦ C, making the solid state diffusion of elements easier between these two layers. However, in this paper the modified aerofoil-shaped samples were exposed at 925 ◦ C where diffusion rates would be some 4–5 times slower,

Fig. 16 is a schematic representation of the four intrusions types. These four recurrent TC asperity types were found to affect the crack propagation at the TC/TGO interface (Fig. 17). The red arrows in Fig. 16 represent the observed crack paths and are in agreement with the residual stress measurements undertaken on similar microstructures. Type 1 did not result in cracks propagating through the TC, only within the TGO or at the TGO/TC interface. The large asperity, the absence of short splat boundaries, together with the less likely observation of microcracks made the propagation of a crack difficult for this intrusion geometry and in that case, the crack would be expected to travel within the TGO. It is evident that the type 2 intrusions allowed a lot of stress to concentrate in a small area of the TC. Microcracks were observed to develop in the convoluted space of the asperity and cracks propagated into the TGO and continued their path at the splat boundary and across the fine strip of ceramic formed as part of this intrusion geometry. The large undulation encountered in type 3 did not stop cracks from propagating at splat boundaries. However, if that splat boundary was too far away from the TGO/TC interface, then cracking occurred in the TGO. In type 4, again a lot of stresses were shown to form within the asperity causing the formation of micro-cracks. A propagating crack from the TGO would travel along the splat boundary at the edge of the asperity, with the linkage of micro-cracks within the asperity also observed. Finally, in the absence of any undulation, the cracks would propagated mainly at the TGO/BC interface or in the middle of the TGO.

3354

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

Fig. 16. Schematic representation of YSZ splats found in TC protrusions into the BC. Red arrows and lines are potential cracks paths. (For interpretation of the references to colour in text, the reader is referred to the web version of this article.)

Work done by Karadge et al.36 can help explaining how cracks propagating at either the TC/TGO or the TGO/BC interface manage to link when encountering highly convoluted features of the oxide layer (Fig. 17(a) and (c)). These workers studied the microstructure of the oxides growing at the BC/TC interface in EBPVD- and APS-TBC systems and found that cracks nucleated either at the TC/TGO or at the TGO/BC

interfaces, but were able to traverse inter-granularly across the TGO thickness around severely rumpled geometries (in the case of EBPVD-TBCs). In addition, they showed that the crystallographic texture of the TGO in the APS-TBC system led to large inter-granular stresses that could explain micro-cracking observed within the oxide layer.36 In a previous paper on these modified aerofoil-shaped specimens, it has been shown that failure was influenced by geometry and happened first in convex areas.28 Features common to plasma sprayed coatings, such as horizontal splat boundaries, micro-cracks within splats and pores could be observed in all the TCs. Convex areas examined showed the ceramic TCs comprised, even in the early stages of exposure, a significant amount of short splat boundaries that linked together during exposure, along with the sintering of pores and pre-existing micro-cracks, (Fig. 18(a) and (b)). Stress measurement in the TC, presented in the results section, has revealed that stress varies with the microstructure of the TC. Different features, splat boundaries, defects, micro-scale cracks can accommodate different amounts of stress. Stress gradients generated over these features caused the TC defects to close or open, see Fig. 7. Not only was the variation of stress reduced by the physical evolution of the TC, but also the entire stress level was observe to decrease with thermal exposure (Fig. 8). 4.3. Failure modes of the entire coating

Fig. 17. SEM images of crack propagating within the TC around (a) type 2, (b) type 3, and (c) type 4 intrusions.

Echsler et al.37 observed a reduction of porosity with exposure time and found that it was accompanied by an increase in the elastic modulus of the TC. An augmentation of the elastic modulus can result in a ceramic less tolerant to strain, which would make it more subject to cracking. In this study, after 7000 h at 925 ◦ C, cracks were observed to start to propagate at the TGO/BC surface of the specimens in convex regions. Some vertical branching cracks, linking splat boundaries of different sprayed layers (caused by several passes made by the deposition gun) were observed (Fig. 18(c)). These are caused by a primary crack propagating at the TGO/BC interface, coming to an undulation and being deflected to the TC. The widening of these vertical cracks, and their growth from the BC to the surface, can be associated with cracking in the TGO. This would eventually lead to delamination and spallation of the TC. This phenomenon was only observed for convex regions. For the case of the trailing edge, location P5, the curvature was extremely convex. The spraying layers could be counted as they were separated by thick splat boundaries due to the powder particles cooling down between each passage of the spraying gun. It

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

3355

Fig. 19. SEM image of TC in location P5 in sample oxidised at 925 ◦ C for 4000 h.

Fig. 18. SEM image of TC in location P3 in sample oxidised at 925 ◦ C for (a) 100 h, (b) 4000 h and (c) 7000 h. White arrows (c) point at vertical cracking.

was found that these boundaries did not follow the geometry of the blade as can be seen in this trailing edge region (Fig. 19(a) and (b)); the final layer did not go around the trailing edge but ended up breaking the top surface leaving splat boundaries linked with the surface. This behaviour was not observed at either flat or concave locations. Large splat boundaries/cracks opening to the surface are more likely to promote the spallation of the TC during exposure. This manufacturing defect could be observed in other convex areas of the specimens (P3 and P5), but was more noticeable as the curvature gets more convex. Crack initiation and propagation mechanisms are complex as they depend on many parameters including the stress state of the three different layers of the system, their microstructure and the local geometry. Several of these factors are systematically influenced by oxidation temperature and time. Suggested mechanisms14,38,39 agreed that the stresses caused by the growth of the TGO will promote the initiation of cracks in the TC

and/or the TGO, and that local undulating morphology of the interface would create out-of plane and shear stresses leading to micro-crack linkage and propagation. It is also agreed that cracks preferentially propagate at the TC/TGO, TGO/BC interface, through the TGO and in the TC depending on the local roughness.38,40 Although the importance of roughness at interfaces on the mechanisms of failure have been recognised, little research has been undertaken on the effect of the discrete splat geometry and distribution within the TC near the TC/TGO interface, and the effect on residual stress state of the APS-TBC system. In this study this has been addressed. The particular layout of the YSZ splats and the TGO was found to have a great influence on crack propagation. This, in turn, is dependent on the macroscopic geometry of the samples representative of turbine blades. Indeed, in concave curvatures, the roughness on the TC/BC interface was more important when the amplitude of the TGO undulations between the two layers was high. Larger amplitudes were more likely to promote type 1 and type 3 ceramic microstructures and configurations which make cracks initiation and propagation more difficult (Fig. 16). On the other hand, the TC/BC interfaces at convex curvature were more likely to promote type 2 and type 4 TC asperities, because of the smaller amplitude of the TGO undulations were dictated by a smaller roughness. Those two types of undulation were found to promote cracking into the TC, the creation of an Al2 O3 –ZrO2 mixed zone and the development of high residual stress gradients between both sides of the interface. The probability of crack propagation inside the ceramic, when the curvature of the coating was convex, was therefore enhanced. Cracks encountering type 2 and

3356

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357

type 4 intrusions did not always continue propagating into the TC, but it increased the risk of creating a critical crack into the ceramic. The higher probability for cracks to generate critical damage close to the TC/TGO interface at convex features of the modified aerofoil-shaped sample and the poorer physical state of the whole TC, caused by the spray deposition in these areas (thinner coating not following the geometry of the blade), made this interface a preferred location for critical cracking which would dictate the lifetime of the entire system. 5. Conclusion The stress measurement together with the microstructural examination of the aerofoil shaped CMSX4/HVOF ‘Amdry 995’/APS-TBC specimens showed that substrate curvature has an influence on the failure modes hence the integrity of the TBC system. • The substrate curvature affects the splats deposition behaviour and therefore the microstructure of various locations. • The residual stress in the TC changes with the curvature but with an opposing trend. • Residual stress was measured across the thickness of the TC, along the length of the TGO around intrusions. It was found to change with various microstructural features and a local debonding is responsible to the redistribution of residual stresses both in TC and TGO. • Spinel formation and interface roughness affects the distribution of four types of intrusions. • The general failure of the entire TC layer can be modified by the substrate curvature because of the varied failure modes for the four types of intrusions. Undulation of the TC/BC interface and different TC asperity microstructure influences the propagation of cracks. Cracks into the TC promote the critical failure of the TBC system by initiation life limiting vertical macro-cracks within the ceramic. This failure is enhanced for TCs with convex curvatures. TCs with concave curvature did not promote the propagation of vertical cracks through the ceramic. • The higher probability for cracks to generate critical damage close to the TC/TGO interface at convex features of the modified aerofoil-shaped sample and the poorer physical state of the whole TC caused by the spray deposition in these areas make it a preferential location for critical cracking which would dictate the lifetime of the entire system. Acknowledgements We would like to acknowledge the support of The Energy Programme, which is a Research Councils UK cross council initiative led by EPSRC and contributed to by ESRC, NERC, BBSRC and STFC, and specifically the Supergen initiative (Grants GR/S86334/01 and EP/F029748) and the following companies; Alstom Power Ltd., Doosan Power, E.ON, National Physical Laboratory, Praxair Surface Technologies Ltd., QinetiQ, Rolls-Royce plc, RWE npower, Siemens

Industrial Turbomachinery Ltd. and Tata Steel, for their valuable contributions to the project.

References 1. Miller RA. Oxidation based model for thermal barrier coating life. J Am Ceram Soc 1984;67:517–21. 2. Fairbanks JW, Hecht RJ. The durability and performance of coatings in gas turbine and diesel engines. Mater Sci Eng 1987;88:321–30. 3. Meier SM, Gupta DK, Sheffler KD. Ceramic thermal barrier coatings for commercial gas turbine engines. JOM J Miner Met Mater Soc 1991;43: 50–3. 4. Miller RA. Current status of thermal barrier coatings: an overview. Surf Coat Technol 1987;30:1–11. 5. Birks N, Meier GH, Pettit FS. Introduction to the high-temperature oxidation of metals. 2nd ed. New York: Cambridge University Press; 2006. 6. Rickerby DS, Winstone MR. Coatings for gas turbines. Mater Manuf Process 1992;7:495–526. 7. Elsing R, Knotek O, Balting U. Calculation of residual thermal stress in plasma-sprayed coatings. Surf Coat Technol 1990;43/44:416–25. 8. Shillington EAG, Clarke DR. Spalling failure of a thermal barrier coating associated with aluminum depletion in the bond-coat. Acta Mater 1999;47:1297–305. 9. Haynes AJ, Rigney DE, Ferber MK, Porter WD. Oxidation and degradation of a plasma-sprayed thermal barrier coating system. Surf Coat Technol 1996;8:6–87. Part 1: 102-8. 10. Brumm MW, Grabke HJ. The oxidation behaviour of NiAl-I. Phase transformations in the alumina scale during oxidation of NiAl and NiAl–Cr alloys. Corros Sci 1992;33:1677–90. 11. Miller RA, Lowell CE. Failure mechanisms of thermal barrier coatings exposed to elevated temperatures. Thin Solid Films 1982;95:265–73. 12. Schlichting KW, Padture NP, Jordan EH, Gell M. Failure modes in plasmasprayed thermal barrier coatings. Mater Sci Eng A Struct 2003;342:120–30. 13. Teixeira V, Andritschky M, Fischer W, Buchkremer HP, Stöver D. Effects of deposition temperature and thermal cycling on residual stress state in zirconia-based thermal barrier coatings. Surf Coat Technol 1999;120/121:103–11. 14. Evans AG, Mumm DR, Hutchinson JW, Meier GH, Pettit FS. Mechanisms controlling the durability of thermal barrier coatings. Prog Mater Sci 2001;46:505–53. 15. Martena M, Botto D, Fino P, Sabbadini S, Gola MM, Badini C. Modelling of TBC system failure: stress distribution as a function of TGO thickness and thermal expansion mismatch. Eng Fail Anal 2006;13:409–26. 16. Mumm DR, Evans AG, Spitsberg IT. Characterization of a cyclic displacement instability for a thermally grown oxide in a thermal barrier system. Acta Mater 2001;49:2329–40. 17. Busso EP, Wright L, Evans HE, McCartney LN, Saunders SRJ, Osgerby S, et al. A physics-based life prediction methodology for thermal barrier coating systems. Acta Mater 2007;55:1491–503. 18. Hutchinson JW. Delamination of compressed films on curved substrates. J Mech Phys Solids 2001;49:1847–64. 19. Mao WG, Jiang JP, Zhou YC, Lu C. Effects of substrate curvature radius, deposition temperature and coating thickness on the residual stress field of cylindrical thermal barrier coatings. Surf Coat Technol 2011;205:3093–102. 20. Faulhaber S, Mercer C, Moon MW, Hutchinson JW, Evans AG. Buckling delamination in compressed multilayers on curved substrates with accompanying ridge cracks. J Mech Phys Solids 2006;54:1004–28. 21. Singh JP, Nair BG, Renusch DP, Sutaria MP, Grimsditch MH. Damage evolution and stress analysis in zirconia thermal barrier coatings during cyclic and isothermal oxidation. J Am Ceram Soc 2001;84:2385–93. 22. Khor KA, Gu YW. Effects of residual stress on the performance of plasma sprayed functionally graded ZrO2 /NiCoCrAlY coatings. Mater Sci Eng A 2000;277:64–76. 23. Chen WR, Wu X, Marple BR, Lima RS, Patnaik, Pre-oxidation PC. TGO growth behaviour of an air-plasma-sprayed thermal barrier coating. Surf Coat Technol 2008;202:3787–96.

D. Liu et al. / Journal of the European Ceramic Society 33 (2013) 3345–3357 24. He J, Clarke DR. Determination of the piezospectroscopic coefficients for chromium-doped sapphire. J Am Ceram Soc 1995;78:1347–53. 25. Liu D, Lord O, Flewitt PEJ. Calibration of raman spectroscopy in the stress measurement of air plasma sprayed yttria stabilized zirconia. Appl Spectrosc 2012;66:1204–9. 26. Limarga AM, Vassen R, Clarke DR. Stress distributions in plasma-sprayed thermal barrier coatings under thermal cycling in a temperature gradient. J Appl Mech Trans ASME 2011;78:3–11. 27. Liu D, Lord O, Stevens O, Flewitt PEJ. The role of beam dispersion within YSZ in the application of raman and photo-stimulated luminescence piezo-spectroscopy in multi-layered coatings. Acta Mater 2013;61: 12–21. 28. Seraffon M, Simms N, Sumner J, Nicholls J. Performance of thermal barrier coatings in industrial gas turbine conditions. Mater High Temp 2011; 78:7. 29. Thompson JA, Clyne TW. The effect of heat treatment on the stiffness of zirconia top coats in plasma-sprayed TBCs. Acta Mater 2001;49: 1565–75. 30. Murphy KS, More KL, Lance MJ. As-deposited mixed zone in thermally grown oxide beneath a thermal barrier coating. Surf Coat Technol 2001;146/147:152–61. 31. Levi CG, Sommer E, Terry SG, Catanoiu A, Rühle M. Alumina grown during deposition of thermal barrier coatings on NiCrAlY. J Am Ceram Soc 2003;86:676–85. 32. Leyens C, Schulz U, Fritscher K. Oxidation and lifetime of PYSZ and CeSZ coated Ni-base substrates with MCrAlY bond layers. Mater High Temp 2003;20:475–9.

3357

33. Jayaram V, Levi CG, Whitney T, Mehrabian R. Characterization of Al2 O3 –ZrO2 powders produced by electrohydrodynamic atomization. Mater Sci Eng A 1990;124:65–81. 34. Chen WR, Wu X, Marple BR, Patnaik PC. The growth and influence of thermally grown oxide in a thermal barrier coating. Surf Coat Technol 2006;201:1074–9. 35. Tuan WH, Chen RZ, Wang TC, Cheng CH, Kuo PS. Mechanical properties of Al2 O3 /ZrO2 composites. J Eur Ceram Soc 2002;22:2827–33. 36. Karadge M, Zhao X, Preuss M, Xiao P. Microtexture of the thermally grown alumina in commercial thermal barrier coatings. Scripta Mater 2006;54:639–44. 37. Echsler H, Renusch D, Schütze M. Mechanical behaviour of as sprayed and sintered air plasma sprayed partially stabilised zirconia. Mater Sci Technol 2004;20:869–76. 38. Naumenko D, Shemet V, Singheiser L, Quadakkers W. Failure mechanisms of thermal barrier coatings on MCrAlY-type bondcoats associated with the formation of the thermally grown oxide. J Mater Sci 2009;44: 1687–703. 39. Echsler H, Shemet V, Schütze M, Singheiser L, Quadakkers W. Cracking in and around the thermally grown oxide in thermal barrier coatings: a comparison of isothermal and cyclic oxidation. J Mater Sci 2006;41: 1047–58. 40. Trunova O, Beck T, Herzog R, Steinbrech RW, Singheiser L. Damage mechanisms and lifetime behavior of plasma sprayed thermal barrier coating systems for gas turbines: part I. Experiments. Surf Coat Technol 2008;202:5027–32.