Engineering Fracture Mechanics 77 (2010) 21362144

Contents lists available at ScienceDirect

Engineering Fracture Mechanics

journal homepage: www.elsevier.com/locate/engfracmech

Determination and analysis of crack growth resistance in plasma-sprayed thermal barrier coatings
Z.X. Chen a, L.H. Qian b, S.J. Zhu c,*
a

Research Institute for Applied Mechanics, Kyushu University, Kasuga-koen 6-1, Kasuga, Fukuoka 816-8580, Japan State Key Laboratory of Metastable Materials Science and Technology, Yanshan University, Qinhuangdao 066004, PR China c Department of Intelligent Mechanical Engineering, Fukuoka Institute of Technology, Fukuoka 811-0295, Japan
b

a r t i c l e

i n f o

a b s t r a c t
Ceramic thermal barrier coatings (TBCs) are increasingly applied to enhance the performance of advanced gas turbine engines. However, the delamination cracks initiated in these coatings limit their applications. In this research, a sandwiched four-point bend specimen is used to evaluate the crack growth resistance in plasma-sprayed TBCs. Well controlled, stable and measurable crack extension is obtained. A rising crack growth resistance curve is found. The steady state strain energy release rate is obtained to be about 170 J/m2. The delamination crack evolution behavior is in situ observed and simulated by the nite element analysis based on a crack bridging model. 2010 Elsevier Ltd. All rights reserved.

Article history: Received 16 December 2009 Received in revised form 5 February 2010 Accepted 8 February 2010 Available online 12 February 2010 Keywords: Thermal barrier coating Crack growth Interface Strain energy release rate Ceramic

1. Introduction Crack growth resistance is an important mechanical property of plasma-sprayed thermal barrier coatings (TBCs) which usually fail due to spallation of the coatings from the substrate. Although there have been a number of research activities to establish a standard methodology for evaluating interfacial fracture toughness in TBCs, all of them seem to not be satisfactory and the reported data on crack growth resistance in terms of interfacial strain energy release rate (G) or fracture toughness in plasma-sprayed TBCs vary largely. For example, the G values of air plasma sprayed (APS) Al2O3 coatings have been reported to be in the range from 12 to 40 J/m2 [1,2] and the values of APS ZrO2Y2O3 coatings vary from 25 to 200 J/m2 [25]. The used test methods can be one major reason for the scatter in the measured toughness data. It is known that the Vickers indentation technique [2,3], the wedge opening load technique [6] and the sandwiched three-point bend specimen [7] have been attempted, but each of them is not applicable for long crack propagation. The sandwiched double cantilever beam (DCB) specimen has been used for long crack propagation in TBCs [1]. However, this technique needs a compliance calibration curve for crack length prediction, which is non-linear and complicated [1]. Additionally, it is difcult to obtain a measurable, stable crack extension by using this specimen [1]. Recently, a sandwiched four-point bend specimen has been used, in which a complicated equation considering the specimen thickness has been adopted for calculating strain energy release rate, and nite element calculations of compliance have been employed for predicting crack lengths [8]. The questions are whether this method is applicable to TBCs and how to simplify it for application in engineering. Therefore, a more feasible and reasonable method needs to be developed in order to obtain reliable and reproducible interfacial fracture toughness in TBCs.

* Corresponding author. Tel.: +81 92 606 4265; fax: +81 92 606 0747. E-mail address: zhu@t.ac.jp (S.J. Zhu). 0013-7944/$ - see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.engfracmech.2010.02.007

Z.X. Chen et al. / Engineering Fracture Mechanics 77 (2010) 21362144

2137

This paper reports an in situ measurement of interfacial crack growth resistance in APS TBCs by using sandwiched fourpoint bend specimen. The technique reported in this paper, compared to that in Ref. [8], is more convenient for thin ceramic lms since the thickness of the thin lm is ignored and its Youngs modulus is not needed, the accurate value of which is difcult to be measured. Furthermore, crack lengths can be directly obtained from specimen side surface because crack growth is controllable, stable and measurable. Using this technique, the variation of strain energy release rate with crack length is measured and a rising R-curve is obtained. Finally, the delamination crack evolution behavior is in situ observed and simulated by the nite element analysis for understanding the fracture mechanism in the APS TBCs. 2. Experimental procedures A well-polished and cleaned SUS304 stainless steel plate of 5 mm in thickness was used as the substrate material. A NiCoCrAlY alloy (Co23Cr17Al13Y0.5 and balance Ni) was used as the bond coat layer. The bond coat layer was deposited on the substrate using APS method. A partially stabilized ZrO2 with 8 wt.% Y2O3 was used as the top coat layer, and the top coat layer was deposited on the bond coat layer by APS method. The average thickness of either the bond coat or the top coat was about 150 lm. A sandwiched four-point bend (4 PB) specimen (Fig. 1) was used for the experiment. The thickness of the substrate of the as-deposited specimen was reduced to about 3 mm. Then, a stiffener, which was made of the same material and the same thickness as the substrate, was xed to the surface of the top coat layer using a strong epoxy base adhesive. The thickness of the adhesive layer was about 50 lm. Hence, the coating layers were sandwiched between the substrate and the stiffener. The width and length of the sandwiched 4 PB specimen were 7 and 50 mm, respectively. After preparing the sandwiched specimens, the side surfaces of the specimens were carefully polished up to 1 lm diamond paste nish. A notch was carefully cut through the stiffener and the adhesive layer, and into the top coat layer with a diamond saw such that the notch tip was laid within the top coat layer. Pre-cracks were introduced in the notched specimens by fatigue on a servo-hydraulic test machine in an ambient air, by using three-point bend xture with a span of 40 mm. The compression-compression load ratio was 0.1 and the frequency was 20 Hz for the fatigue pre-cracking. The pre-cracks initiated from the notch tips, then propagated and kinked symmetrically towards both sides of the notch, and nally entered into the plane close and parallel to the interface between the bond coat and top coat layers, as is shown in Fig. 2. After pre-cracking, the specimens were monotonically loaded under four-point bend at a constant loading rate of 2 N/s. The inner and outer spans of the four-point bend xture were 20 and 40 mm, respectively. The testing was performed in an ambient air using the same servo-hydraulic test machine as for fatigue pre-cracking. The load and the crosshead displacement were continuously recorded using a digital recording system at a sampling rate of 100 ms. The crack length was directly measured from the polished surface during loading by means of a video camera and a traveling microscope. The crack length was dened as the sum of the initial crack length and the crack extension. The initial crack length was determined as the distance from the notch root to the fatigue pre-crack tip, and the crack extension was dened as the distance from the fatigue pre-crack tip to the crack front generated by the monotonic loading. In the test, the specimen was unloaded

Fig. 1. Geometry of sandwiched four-point bend specimen.

Fig. 2. Typical crack path in the sandwiched TBCs specimen after the bend test (TC: top coat; BC: bond coat).

2138

Z.X. Chen et al. / Engineering Fracture Mechanics 77 (2010) 21362144

after a crack extension of about 0.2 mm, and then reloaded to get another crack extension of about 0.2 mm. The loading unloading process was repeated until the cracks on both sides were propagated to the inner loading lines. A linear loading behavior was observed during the whole loading stage of the rst cycle. No crack extension was detected after the rst cycle. During each of the following loadingunloading cycles, the initial linear loading behavior was followed by a non-linear behavior, which occurred near the maximum load on each curve and was accompanied by a crack extension. With increasing the number of loadingunloading cycles the applied maximum load was incremented, the crack length increased and the slope of the linear portion in loading stage decreased. After complete unloading in each cycle, a permanent residual displacement appeared. The residual displacement was correlated with crack extension after each loadingunloading cycle.Since each sandwiched four-point bend specimen was constructed by bonding a thin coating layer into two halves of substrate material, the whole bulk of the specimen was homogeneous except for the thin coating layer in the structure with a pre-existing crack lying close to the interface. According to Suo and Hutchinson [9], strain energy release rate determined for homogeneous specimen without interlayer can be used to characterize the interface crack in the presence of an interlayer by ignoring the layer when its thickness is small compared with the crack length and other in-plane sizes of the specimen. Accordingly, in some works [10,11], the expressions for strain energy release rate for homogeneous DCB and compact tension (CT) specimens were used for sandwiched DCB and CT specimens to measure the interfacial fracture toughness of glass/Cu and Al2O3/Al interfaces. Similarly, the expression for strain energy release rate of a four-point bend specimen without a sandwiched layer [12], G1, was adopted in this investigation to characterize the interfacial strain energy release rate in sandwiched 4 PB plasma-sprayed TBCs specimens:

G1 21P2 L2 1 t2 =16Eb h

2 3

provided that the propagating interfacial cracks are located within the inner loading lines of the specimen, where P is the total load, L is the spacing between the inner and outer loading lines, b and h are the width and half height of the specimen, and E and t are the Youngs modulus and Poissons ratio of the substrate material. In the present experiment, the nominal specimen sizes necessary for the calculation of strain energy release rate in Eq. (1) were L = 10 mm, b = 7 mm, and h = 3 mm. The Youngs modulus and Poissons ratio of the substrate material were taken to be 200 GPa and 0.3, respectively. 3. Finite element modeling For the analysis of strain energy release rate of the 4 PB specimen, a nite element (FE) model has been constructed and solved using the software package ABAQUS 6.5. The symmetry about the midsection of the bend specimen depicted in Fig. 1 allows the requisite solutions to be obtained by considering one half of the specimen (Fig. 3). Based on the actual situation, the interlayer in Fig. 1 was subdivided into three layers in the present FE model. They are adhesive layer, top coat (TC) layer and bond coat (BC) layer. The mechanical properties of these layers as input data are listed in Table 1. The bend specimen is discretized using two-dimensional eight-noded isoparametric elements. The size of elements varies along the thickness of the bend specimen with a very ne mesh used near the crack tip. The crack tip is modeled with a ring of collapsed quadrilateral elements to capture the singularity of stress and strain. The roller support at the inner loading line is constrained to x the displacement in the vertical direction, whereas the symmetry is enforced by inhibiting horizontal displacements along the midsection. The crack surfaces are traction-free. The setting scheme of crack propagation path during bend loading can hence be examined. 4. Results and discussion The cross sectional microstructures of the as-deposited coatings showed that a lot of pores and curved inter-splat microcracks distributed uniformly in both the top coat and bond coat layers, as is shown in Fig. 4a. In the top coat layer, the curved microcracks were longer and sharper in the horizontal direction than in the vertical direction, whereas there was not such a difference in the bond coat layer. Closer examination indicated that the top coat was composed of lamellae, typically several

Fig. 3. Finite element model for the analysis of strain energy release rate of sandwiched four-point bend specimen.

Z.X. Chen et al. / Engineering Fracture Mechanics 77 (2010) 21362144 Table 1 Mechanical properties of adhesive, TC and BC layers [1315]. Properties Youngs modulus (GPa) Poissons ratio Yield strength (MPa) Adhesive 1.89 0.33 50 TC 37 0.18 BC

2139

200 0.3 1000

Fig. 4. Cross sectional microstructures of plasma-sprayed thermal barrier coatings: (a) at low magnication; (b) at high magnication.

micrometers thick, as is shown in Fig. 4b. The measurement of Raman spectra indicated that there were tetragonal phase and a small amount of monolithic phase of ZrO2 in the top coat layer. During bend loading, the crack evolution and propagation path were observed on one polished surface of the specimen using an optical microscope. It was found that the onset of the crack growth occurred from the initial fatigue pre-cracks. The delamination cracks were always extended within the top coat layer close and roughly parallel to the interface between the bond coat and top coat layers, as is shown in Fig. 5a. The crack propagation path was slightly wavy, which was caused by the curved splat structure. The entire crack usually propagated within 1060 lm away from the interface. Occasionally, the crack approached to a plane very near the interface between the bond coat and top coat layers, but it did not enter the interface, as is shown in zone B in Fig. 5a. This crack trajectory is consistent with the failure mode most frequently observed in plasma-sprayed TBCs [16,17]. This implies that the sandwiched four-point bend technique is a feasible way to reproduce the in-service delamination, which can thus be used to evaluate strain energy release rate for the delamination crack propagation in TBCs. Closer examination indicated that the delamination cracks propagated predominantly along inter-splat boundaries. Multiple, parallel short cracks were found in the crack tip region. Between the overlapping parallel cracks there were the unbroken ligaments, which bridged the parallel cracks, as is shown in Fig. 5b and c. The crack bridging existed even in the wake zone far behind the main crack tip although it was more sparsely separated (Fig. 5a). During the crack growth, some previously formed ligaments began to fracture under the increased bend loading, as is shown in Fig. 6a. With the fracture of bridging ligament, linking of the parallel cracks became visible. A typical cracking surface of the bend specimen is shown in Fig. 6b. From this gure, two kinds of cracking surfaces can be easily observed. They are relatively smooth and curved inter-splat cracking surface and rougher intra-splat cracking surface. The smooth and curved morphology of the

2140

Z.X. Chen et al. / Engineering Fracture Mechanics 77 (2010) 21362144

Fig. 5. Cross sectional crack morphology showing (a) crack propagation prole far behind the main crack tip and (b, c) unbroken ligaments between two cracks enlarged from zone A and zone B in (a).

former reects the inter-splat boundaries. This indicates that the cracking occurred along the inter-splat boundaries, which is consistent with the observed cross sectional crack path (Fig. 5). The linking of inter-splat boundary cracks became more evident with the increasing of crack length and the opening of cracking surfaces. The morphology of rough intra-splat cracking surface, which exposed the internal ne structures of the spraying splats, suggested that inter-splat boundary cracks on different planes linked to each other by intra-splat cracking of bridging ligaments, as is shown in Fig. 6a. The observed crack growth behavior (Figs. 5 and 6) indicate that the bridging ligaments were prevalent in the TBCs system under the initial bend loading, and then some of them would fracture with further increase of bend loading. The maximum load in each loadingunloading cycle was used to calculate the strain energy release rate at the corresponding crack length. A typical curve of the measured strain energy release rate as a function of crack extension is shown in Fig. 7. An obvious rising R-curve is visible. Multiple specimens for the measurement demonstrated the same behavior. The R-curve, starting at a very low value, increased rapidly over a crack growth distance of about 6 mm, and then gradually rose to a steady state value, G1 , about 170 J/m2, at a crack length of about 8 mm. The steady state value of G1 obtained SS SS in this work was close to the strain energy release rate measured in Refs. [5,6] (150200 J/m2), but twice the fracture work of delamination crack in Ref. [7] (7181 J/m2), and much higher than those measured in Refs. [2,3] (2535 J/m2) for APS ZrO2 Y2O3 coatings. The difference in measured strain energy release rate could be caused by two factors: TBCs microstructure and test method. Although the tested materials were all composed of the standard composition (78 wt.% YSZ), the coating process and spray parameters might be different, which caused the variation in TBCs microstructure. As fractographic observations indicated in Ref. [18] and Fig. 6b, there were many smooth surfaces and fresh rough surfaces on the fracture surface. The smooth surfaces were the faces of micropores, or inter-splat boundaries. The rough surfaces were formed as a result of discrete translamellar microcracks between neighboring micropores or inter-splat boundaries. Macro-fracture occurred when sufcient micropores or inter-splat boundaries have been connected by trans-lamellar micropores to separate that region from the remainder of the coating. Therefore, the micropores or weak splat boundaries acted as the crack initiation site and contributed to the major reductions in fracture resistance of TBCs. Thus the size, shape, orientation and size distribution of pores, and porosity level should be responsible for the difference in the reported fracture toughness. It should also be argued that different test methods might generate different results. A notched specimen (2.6 4.6 3.5 mm3) with a notch depth of 1.8 mm was used in wedge opening load technique in Ref. [6], and a pre-cracked

Z.X. Chen et al. / Engineering Fracture Mechanics 77 (2010) 21362144

2141

Fig. 6. Cross sectional crack morphology (a) and cracking surface (b) showing the fracture of bridging ligament.

Fig. 7. Measured strain energy release rate as a function of crack extension.

specimen with a height of only 1.3 mm was used in three-point bending approach in Ref. [7], where the crack was extended along the height direction. Clearly, cracks in such tests could not grow a sufcient distance to obtain potential R-curve behavior in TBCs, while such a distance is accessible in this work. The rising R-curve measured in the present research was attributed to the formation of bridging zone. In the wake zone behind the main crack tip there existed many bridging zones where there were unbroken bridging ligaments. The bridging of

2142

Z.X. Chen et al. / Engineering Fracture Mechanics 77 (2010) 21362144

Fig. 8. Crack bridging model for nite element analysis.

Table 2 Width evolution and number of bridging ligaments (lm in unit). BL1 20 30 60 120 Fracture Fracture BL2 20 30 60 120 Fracture BL3 BL4 BL5 BL6

20 30 60 120

20 30 60

20 30

20

Fig. 9. Simulated strain energy release rate as a function of crack extension.

cracks in the wake zone prohibited crack opening and contributed to the toughening. The length of each bridging zone increased with crack growth, and the bridging-caused toughening contribution was enhanced accordingly. As a result, a rising R-curve formed. With a further increase in crack length, some unbroken ligaments began to fracture. Gradually, the bridging zone approached to a saturation state, and the size of this zone attained the saturation. Thus, the toughness reached its steady state value. The evolution process of the bridging zone under bend loading can be modeled using the nite element method. As is shown schematically in Fig. 8, a bridging ligament (region between the dotted lines in Fig. 8) was set between two cracks. The width (L) and number of bridging ligaments (BL) were changed during crack growth simulation. As is shown in Table 2, it

Z.X. Chen et al. / Engineering Fracture Mechanics 77 (2010) 21362144

2143

Fig. 10. The deformation pattern of bend specimen with bridging ligaments.

was assumed that there were six bridging ligaments emerging successively during the crack propagation. Every new emerged bridging ligament lied behind the main crack tip about 1 mm. With the continuously emerging of new bridging ligaments during the crack propagation, the width of previously formed ligaments increased progressively until they fractured. In addition, a nite element model without crack bridging was also built for comparison. For both the considered nite element models, the strain energy release rate was calculated once the crack length (the distance from the midsection to the main crack tip) reached 2 mm. Hereafter, the strain energy release rate was calculated again following each crack extension of 1 mm. The simulation results are shown in Fig. 9. It can be seen from Fig. 9 that, for the model with crack bridging, the predicted results have a similar trend to that of the experimental results although the absolute values of them are slightly different. When the simulated crack length was less than 5 mm, the strain energy release rate increased continuously. Such an increase in the simulated value corresponded to the continuous increase in the number of bridging ligaments along the crack propagation path, which shielded the crack tip from the far-eld loading. As is shown in Fig. 10, the bridging ligaments bore the bending load partly. When the simulated crack length was larger than 5 mm, the strain energy release rate became stabilized due to the balance between the emerging of new bridging ligaments and the fracture of previously formed bridging ligaments. In contrast, for the model without crack bridging, the strain energy release rate did not obviously increase with crack extension (Fig. 9). As is demonstrated and analyzed above, the rising crack growth resistance curve of plasma-sprayed thermal barrier coatings strongly depends on the crack bridging behavior, and the bridging behavior relies on the microstructure of these coatings. Thus, control of the coating microstructure plays an important role in improving the crack growth resistance in the thermal barrier coatings system. 5. Conclusions The interfacial crack growth resistance in APS ZrO28 wt.% Y2O3 TBCs was studied using the sandwiched four-point bend specimen. Delamination cracks always propagated in the top coat layer but tended to approach to the interface between the bond coat and top coat layers. The strain energy release rate increased with crack extension, showing an obvious rising Rcurve behavior. The rising R-curve in the APS TBCs was caused by the crack bridging in the wake zone, which shielded the crack tip from the far-eld loading. Acknowledgements S.J. Zhu thanks for KAKENHI (18560094) and a researcher grant (2005) in Fukuoka Institute of Technology, Japan. References
[1] Ostojic P, McPherson R. Determining the critical strain energy release rate of plasma-sprayed coatings using a double-cantilever-beam technique. J Am Ceram Soc 1988;71:8919. [2] Rabiei A, Evans AG. Failure mechanisms associated with the thermally grown oxide in plasma-sprayed thermal barrier coatings. Acta Mater 2000;48:396376. [3] Bartlett AH, Maschio RD. Failure mechanisms of a zirconia-8 wt.% yttria thermal barrier coating. J Am Ceram Soc 1995;78:101824. [4] Guo DZ, Wang LJ. Measurement of the critical strain energy release rate of plasma-sprayed coatings. Surf Coat Technol 1992;56:1925. [5] Heintze GN, McPherson R. Fracture toughness of plasma-sprayed zirconia coatings. Surf Coat Technol 1988;34:1523. [6] Wessel E, Steinbrech RW. Crack propagation in plasma-sprayed thermal barrier coatings. Key Eng Mater 2002;223:5560. [7] Thurn G, Schneider GA, Bahr HA, Aldinger F. Toughness anisotropy and damage behavior of plasma sprayed ZrO2 thermal barrier coatings. Surf Coat Technol 2000;123:14758. [8] Honger I, Oechsner M, Bahr HA, Swain MV. Modied four-point bending specimen for determining the interface fracture energy for thin, brittle layers. Int J Fract 1998;92:21320. [9] Suo Z, Hutchinson JW. Sandwich test specimens for measuring interface crack toughness. Mater Sci Eng A 1989;107:13543.

2144

Z.X. Chen et al. / Engineering Fracture Mechanics 77 (2010) 21362144

[10] Cannon RM, Dalgleish BJ, Dauskardt RH, Oh TS, Ritchie RO. Cyclic fatigue-crack propagation along ceramic/metal interfaces. Acta Metall Mater 1991;39:214556. [11] McNaney JM, Cannon RM, Ritchie RO. Fracture and fatigue-crack growth along aluminumalumina interfaces. Acta Mater 1996;44:471328. [12] Charalambides PG, Lund J, Evans AG, McMeeking RM. A test specimen for determining the fracture resistance of bimaterial interfaces. J Appl Mech 1989;56:7782. [13] Karlsson AM, Evans AG. A numerical model for the cyclic instability of thermally grown oxides in thermal barrier systems. Acta Mater 2001;49:1793804. [14] Busso EP, Lin J, Sakurai S, Nakayama M. A mechanistic study of oxidation-induced degradation in a plasma-sprayed thermal barrier coating system. Part I: model formulation. Acta Mater 2001;49:151528. [15] You M, Yan ZM, Zheng XL, Yu HZ, Li Z. A numerical and experimental study of adhesively bonded aluminium single lap joints with an inner chamfer on the adherends. Int J Adhes Adhes 2007;28:716. [16] Miller RA. Thermal barrier coatings for aircraft engines: history and directions. J Thermal Spray Technol 1997;6:3542. [17] Bose S, DeMasi-Marcin J. Thermal barrier coating experience in gas turbine engines at Pratt & Whitney. J Thermal Spray Technol 1997;6:99104. [18] Callus PJ, Berndt CC. Relationships between the mode II fracture toughness and microstructure of thermal spray coatings. Surf Coat Technol 1999;114:11428.