Internal Factors Affecting the Surface Rumpling of a β-NiAlHf Coating

β-NiAl coatings on a superalloy substrate will inevitably result in severe rumpling at elevated temperatures; however, the associated rumpling mechanisms are not completely understood. The scale rumpling behavior of a β-NiAlHf coating deposited by electron beam physical vapor deposition (EB-PVD) on single crystal superalloy IC21 was investigated in this work. Some internal factors, including the mismatch in the coefficient of thermal expansion and the stress induced by the growth of oxide scale and the phase transformation, were taken into consideration. The thermal mismatch stress between the coating and substrate was the main internal factor responsible for rumpling behavior during thermal cyclic loads, while the phase degradation from β-NiAl to γ’-Ni3Al in the coating played a dominant role during static thermal loads.


Introduction
β-NiAl B2 intermetallic compounds have been extensively studied as bond coat materials in thermal barrier coating systems for the hot section parts of gas turbines because of their excellent properties, such as low density, high-melting point, and excellent isothermal oxidation resistance at elevated temperatures [1][2][3]. However, the β-NiAl coating is subjected to thermal cycles when turning the turbine on and off during the practical service, which could lead to a wavy surface due to its low strength at a high temperature. The uneven surface on the coating is often termed as "rumpling", which is indicative of the initial cracking of the protective oxide scale, which finally results in the premature spallation failure of the thermal barrier coating system [4,5].
Rumpling is frequently observed on aluminide coatings and NiAl coatings upon hightemperature oxidation. Given the reduced strength of the coatings at high temperatures, once the coating is subjected to a stress of sufficient magnitude to meet its yield strength, it deforms plastically at this point [6]. Accordingly, the magnitude of the stresses acting on the coating tends to play a decisive role in the coating deformation, namely rumpling behavior. To date, some progress has been made on factors such as thermal cyclic frequency and numbers, oxidation dwell time and temperature, coating thickness, and initial surface roughness [7][8][9][10][11]. Previous studies show that these factors indeed affect the rumpling behavior by varying degrees. However, in fact, factors such as oxidation dwell time and cyclic frequency contribute to surface rumpling by changing the duration time and frequency of thermal stress, instead of inducing stress. Hence, the aforementioned factors can be defined as external factors that are not directly responsible for surface rumpling behavior, and they cannot explain the mechanism of rumpling fundamentally. Obviously, it is necessary to determine where the stresses are generated in the coating systems. Analyzed by the coated system without ceramic top coat, three positions probably introduce stresses in the coating-substrate interface, the coating-oxide scale interface, and the coating body itself, which could induce thermal mismatch stress, oxidation stress, and stress induced by volume shrinkage associated with phase transformation, respectively.
In this work, different experiments were designed to investigate the effects of stresses from the three above-mentioned positions on the rumpling behavior. Bulk alloys as the reference specimens [12][13][14] for researching misfit stress were not considered in this work because bulk alloys are duplex in structure (scale-substrate) upon oxidation, while the coated specimens are triplex in structure (scale-coating-substrate). In this case, the coated specimen and the bulk specimens may exhibit totally different rumpling behaviors, even under the exact same thermal cyclic loads. Hence, NiAlHf cast alloy and IC21 single crystal superalloy were selected as the substrate materials, and a NiAlHf coating was deposited onto the two different substrate materials by EB-PVD to ensure the same structure of all samples (i.e., a small amount of Hf was added to the coating as a reactive element to improve the oxidation resistance and oxide scale adherence to avoid the peeling and spallation of the oxide scale [15,16]). Meanwhile, the coefficient of thermal expansion (CTE) was measured and calculated to interpret the coating deformation process quantitatively. In addition to the coating-substrate thermal mismatch stress, the mismatch stress between the oxide scale and NiAlHf coating was also investigated, which is apt to be neglected, with few investigations found in the previous literature. Moreover, most studies on coating phase transformation are conducted on NiPtAl coatings produced by electroplating following chemical vapor deposition or pack cementation [9,17,18]. In contrast, in this experiment, the NiAlHf coating with fine perpendicular grain boundaries was produced by EB-PVD, whose columnar structure may affect the mechanism of phase degradation from β to γ' and have an unexpected influence on surface rumpling to some extent. More detailed experiments focusing on coating rumpling were implemented, and the internal factors affecting rumpling behavior were systematically researched individually.

Materials and Methods
The substrate material used in this study was NiAlHf ingots (nominal composition: 50 at.% Ni, 49.9 at.% Al, 0.1 at.% Hf) produced by arc-melting and single crystal superalloy IC21 (nominal composition: 1.51 at.% Cr, 7.33 at.% Mo, 12.61 at.% Al, 0.87 at.% Re, 0.95 at.% Ta, minor Y and balanced Ni). Disk-shaped coupons (14 mm in diameter and 2 mm thickness) were fabricated from the annealed NiAlHf ingots and superalloy bars, respectively. In addition, a target with a diameter of 68.5 mm was also cut from the same annealed NiAlHf ingots for the preparation of NiAlHf coatings. Prior to deposition, all the specimens were carefully ground with SiC abrasive paper up to 800 grit and ultrasonically cleaned in acetone and alcohol. An EB-PVD facility equipped with four electron beam guns and three crucibles was used to produce the NiAlHf coatings under a pressure level of 10 −3 Pa. During the processing of the coatings, two electron beam guns were used for heating, one of which was used for heating the samples so as to keep a substrate temperature of~900 • C, and the other used for heating and evaporating the target ingots. After coating deposition, the as-deposited samples were annealed at 1050 • C for 4 h in a vacuum for the purpose of homogenization. Then, the as-annealed samples were ground by 2000-grit SiC paper and polished by diamond paste to remove pre-existing undulations and preoxide films on the surface, imparting a uniform roughness of approximately 0.2 µm on all specimens. To facilitate in their description, the specimens with IC21 substrate and NiAlHf substrate developed in this study are abbreviated as NAIC and NANA, respectively.
Cyclic oxidation tests of an NAIC sample and an NANA sample for 50 one-hour cycles were carried out in a vertical tube furnace equipped with an automation system to move the samples in and out of the furnace. Each cycle consisted of 50 min heating at 1200 • C, and 10 min forced air cooling outside the furnace. The argon gas thermal cycle tests for 50 and 100 one-hour cycles were conducted in the same vertical tube furnace and followed the same cyclic regime. The difference from the cyclic oxidation tests is that the samples cycled in argon were put into crucibles and sealed under argon atmosphere in silica ampoules, including two NANA specimens that had been subjected to isothermal exposure at 1200 • C for 50 h. Isothermal oxidation tests of three NAIC samples were performed at 1200 • C for 50, 100, and 200 h, respectively, in a tubular furnace (GSL-1600X, HF-Kejing, Hefei, China). A flowchart of the tests and specifications can be seen in Figure 1. After thermal cycling and isothermal testing, samples for microstructure characterization were nickel-plated, embedded in epoxy, ground, and finely polished using standard metallographic techniques.
Coatings 2021, 11, x FOR PEER REVIEW 3 of 14 move the samples in and out of the furnace. Each cycle consisted of 50 min heating at 1200 °C, and 10 min forced air cooling outside the furnace. The argon gas thermal cycle tests for 50 and 100 one-hour cycles were conducted in the same vertical tube furnace and followed the same cyclic regime. The difference from the cyclic oxidation tests is that the samples cycled in argon were put into crucibles and sealed under argon atmosphere in silica ampoules, including two NANA specimens that had been subjected to isothermal exposure at 1200 °C for 50 h. Isothermal oxidation tests of three NAIC samples were performed at 1200 °C for 50, 100, and 200 h, respectively, in a tubular furnace (GSL-1600X, HF-Kejing, Hefei, China). A flowchart of the tests and specifications can be seen in Figure  1. After thermal cycling and isothermal testing, samples for microstructure characterization were nickel-plated, embedded in epoxy, ground, and finely polished using standard metallographic techniques. The morphologies of the coated samples were characterized by a field emission-scanning electron microscope (FE-SEM, Carl Zeiss, Jena, Germany) equipped with an energy dispersive X-ray spectrum (EDS). X-ray diffraction (XRD, D/max 2200, Rigaku Corporation, Tokyo, Japan) with Cu Kα radiation at 40 KV was applied to examine the major phase identification of the coatings in the 2θ range of 10°-90°. The CTE of the NiAlHf alloy and IC21 superalloy samples were determined by a Dilatometer (DIL802, TA Instruments, New Castle, DE, USA) with a dual push rod differential dilatometer. The coating surface roughness was characterized by a confocal laser scanning microscope (OLS4000, Olympus Corporation, Tokyo, Japan). The volume fraction of the γ'-Ni3Al phase in the coating was evaluated from the cross-section backscattered electron (BSE) images via the ImageJ analysis software. A total length of at least 200 μm with a step of 10 μm was measured in appointed figures.
Residual stresses in the oxide scale were measured by photostimulated luminescence spectroscopy. The Cr 3+ luminescence spectra were acquired using Raman spectroscopy (LabRAM HR Evolution, HORIBA Jobin Yvon, Longjumeau, France). For each specimen, more than 10 positions away from edges and the spalled area were measured. Then, residual stresses can be calculated by the equation [6,[19][20][21]: where σ (GPa) is the biaxial stress and Δν (cm −1 ) represents the frequency shift of the peak measured by Raman spectroscopy. Moreover, a stress-free Al2O3 spectrum as the The morphologies of the coated samples were characterized by a field emissionscanning electron microscope (FE-SEM, Carl Zeiss, Jena, Germany) equipped with an energy dispersive X-ray spectrum (EDS). X-ray diffraction (XRD, D/max 2200, Rigaku Corporation, Tokyo, Japan) with Cu Kα radiation at 40 KV was applied to examine the major phase identification of the coatings in the 2θ range of 10 • -90 • . The CTE of the NiAlHf alloy and IC21 superalloy samples were determined by a Dilatometer (DIL802, TA Instruments, New Castle, DE, USA) with a dual push rod differential dilatometer. The coating surface roughness was characterized by a confocal laser scanning microscope (OLS4000, Olympus Corporation, Tokyo, Japan). The volume fraction of the γ'-Ni 3 Al phase in the coating was evaluated from the cross-section backscattered electron (BSE) images via the ImageJ analysis software. A total length of at least 200 µm with a step of 10 µm was measured in appointed figures.
Residual stresses in the oxide scale were measured by photostimulated luminescence spectroscopy. The Cr 3+ luminescence spectra were acquired using Raman spectroscopy (LabRAM HR Evolution, HORIBA Jobin Yvon, Longjumeau, France). For each specimen, more than 10 positions away from edges and the spalled area were measured. Then, residual stresses can be calculated by the equation [6,[19][20][21]: where σ (GPa) is the biaxial stress and ∆ν (cm −1 ) represents the frequency shift of the peak measured by Raman spectroscopy. Moreover, a stress-free Al 2 O 3 spectrum as the reference spectrum was obtained by photostimulated luminescence spectroscopy from a stress-free single crystal sapphire.

Coating Characterization
The average chemical compositions of the as-annealed coatings are summarized in Table 1 based on the analysis of the energy dispersive spectrum (EDS). Compared with a designed coating composition, some deviations of the element contents happened after deposition, on account of losing aluminum during the vapor deposition. Fortunately, the phase composition and crystal structure of the coating on IC21 and NiAlHf was determined to be single-phase β-NiAl with high intensity peaks at the (110), (210), and (211) directions ( Figure 2a) by the examination of XRD, which is consistent with the experiment design. reference spectrum was obtained by photostimulated luminescence spectroscopy from a stress-free single crystal sapphire.

Coating Characterization
The average chemical compositions of the as-annealed coatings are summarized in Table 1 based on the analysis of the energy dispersive spectrum (EDS). Compared with a designed coating composition, some deviations of the element contents happened after deposition, on account of losing aluminum during the vapor deposition. Fortunately, the phase composition and crystal structure of the coating on IC21 and NiAlHf was determined to be single-phase β-NiAl with high intensity peaks at the (110), (210), and (211) directions (Figure 2a) by the examination of XRD, which is consistent with the experiment design.  The cross-section morphologies of the specimens were characterized by SEM using a backscatter detector. As shown in Figure 2b, the coating on the IC21 superalloy consists of a β-NiAl outer layer about 82 ± 3 μm and an interdiffusion zone (IDZ). The IDZ layer with a thickness of 8 μm contains a number of white granular precipitates, which are rich in refractory elements such as Mo, Ta and Cr that diffuse from the substrate during the deposition process and homogenization treatment. In Figure 2c, no IDZ or precipitates are observed, since both the coating and substrate alloy have a chemical composition of NiAlHf without refractory elements. As is well established, the surface roughness profiles of a coating upon deposition are affected by the initial surface state. Additionally, as Figure 2b,c illustrate, the surface is so rough that it likely disturbed the observation of the rumpling behavior after oxidation. To rule out the potential interference above, all the annealed specimens were grounded to a 2000-grit SiC and polished following standard metallographic approaches before tests to obtain flat and uniform surfaces (Figure 2d,e). Figure 3 shows the cross-section morphologies of NAIC and NANA after 50 onehour cycles in the air atmosphere at 1200 °C. As Figure 3a shows, the IDZ in the NAIC sample has reached a thickness of 37 μm. It is indicated that the interdiffusion behavior occurs between the NiAlHf coating and the IC21 substrate during cyclic oxidation. In Figure 3b, some white linear precipitates appeared at the coating/substrate interface. The EDS The cross-section morphologies of the specimens were characterized by SEM using a backscatter detector. As shown in Figure 2b, the coating on the IC21 superalloy consists of a β-NiAl outer layer about 82 ± 3 µm and an interdiffusion zone (IDZ). The IDZ layer with a thickness of 8 µm contains a number of white granular precipitates, which are rich in refractory elements such as Mo, Ta and Cr that diffuse from the substrate during the deposition process and homogenization treatment. In Figure 2c, no IDZ or precipitates are observed, since both the coating and substrate alloy have a chemical composition of NiAlHf without refractory elements. As is well established, the surface roughness profiles of a coating upon deposition are affected by the initial surface state. Additionally, as Figure 2b,c illustrate, the surface is so rough that it likely disturbed the observation of the rumpling behavior after oxidation. To rule out the potential interference above, all the annealed specimens were grounded to a 2000-grit SiC and polished following standard metallographic approaches before tests to obtain flat and uniform surfaces (Figure 2d,e). Figure 3 shows the cross-section morphologies of NAIC and NANA after 50 one-hour cycles in the air atmosphere at 1200 • C. As Figure 3a shows, the IDZ in the NAIC sample has reached a thickness of 37 µm. It is indicated that the interdiffusion behavior occurs between the NiAlHf coating and the IC21 substrate during cyclic oxidation. In Figure 3b, some white linear precipitates appeared at the coating/substrate interface. The EDS results show that they are Hf-rich phases ( Table 2). For the higher magnification of the NAIC and the NANA samples shown in Figure 3c,d, the oxide scale is exclusively Al 2 O 3 without spinel because the β-phase is Al-rich, while Hf as a reactive element in the coating can slow the oxidation rate and provide a strong interfacial bonding at the scale-coating interface. It can be noted  Figure 3c that the surface of NAIC appears rumpled with a maximum amplitude of 15 µm. In contrast, the NANA surface is flat and straight in Figure 3d. The different surface states in Figure 3c,d indicate that the rumpling behavior after cyclic oxidation is indeed associated with the substrates. The two aspects are considered to explain the results, as follows.

Effect of the Substrate during Cyclic Oxidation
Coatings 2021, 11, x FOR PEER REVIEW 5 of 14 results show that they are Hf-rich phases ( Table 2). For the higher magnification of the NAIC and the NANA samples shown in Figure 3c,d, the oxide scale is exclusively Al2O3 without spinel because the β-phase is Al-rich, while Hf as a reactive element in the coating can slow the oxidation rate and provide a strong interfacial bonding at the scale-coating interface. It can be noted from Figure 3c that the surface of NAIC appears rumpled with a maximum amplitude of 15 μm. In contrast, the NANA surface is flat and straight in Figure 3d. The different surface states in Figure 3c,d indicate that the rumpling behavior after cyclic oxidation is indeed associated with the substrates. The two aspects are considered to explain the results, as follows. On the one hand, considering the differences in the chemical composition of the β-NiAl coating and IC21 superalloy substrate, a large amount of aluminum in the coating diffuses into the substrate, as well as nickel diffusing into the coating. As Figure 3c shows, part of the β phase has degraded to γ' phase, which might be a cause for inducing rumpling behavior since the phase degradation is accompanied by volume reduction. The composition of the boxed zones B and C determined by quantitative EDS analysis is shown in Table 2, which confirms that the aluminum content of the coating on NAIC decreases to 33.37 at.%. Since the aluminum in NAIC is consumed by the generation of the alumina scale and diffuses into substrate alloy. As for the coating on NANA, the aluminum content maintains a relatively stable level of about 38.18 at.%, because Al can be replenished from the NiAlHf substrate in time due to the concentration gradient so as to stabilize the β phase [22,23], even after a long-term oxidation. In view of this, a specialized experiment for investigating the β-NiAl to γ'-Ni3Al phase transformation in the NAIC coating was designed, with the results discussed in Section 3.4. Table 2. Average chemical composition of the marked zone in Figure 3 (in at.%).  On the one hand, considering the differences in the chemical composition of the β-NiAl coating and IC21 superalloy substrate, a large amount of aluminum in the coating diffuses into the substrate, as well as nickel diffusing into the coating. As Figure 3c shows, part of the β phase has degraded to γ' phase, which might be a cause for inducing rumpling behavior since the phase degradation is accompanied by volume reduction. The composition of the boxed zones B and C determined by quantitative EDS analysis is shown in Table 2, which confirms that the aluminum content of the coating on NAIC decreases to 33.37 at.%. Since the aluminum in NAIC is consumed by the generation of the alumina scale and diffuses into substrate alloy. As for the coating on NANA, the aluminum content maintains a relatively stable level of about 38.18 at.%, because Al can be replenished from the NiAlHf substrate in time due to the concentration gradient so as to stabilize the β phase [22,23], even after a long-term oxidation. In view of this, a specialized experiment for investigating the β-NiAl to γ'-Ni 3 Al phase transformation in the NAIC coating was designed, with the results discussed in Section 3.4. Table 2. Average chemical composition of the marked zone in Figure 3 (in at.%). of the thermal expansion mismatch stress [4,24]. It is essential to measure the thermal expansion coefficients and calculate the stress quantitatively to further investigate the coating-substrate thermal expansion mismatch. For an elastic thin-film coating on a superalloy substrate, the equibiaxial in plane stress is quantified as follows [25]:

Element
where a C and a M are the CTE of the coating and matrix, respectively, E C represents the coating elastic modulus, ∆T is the temperature difference and ν C denotes the Poisson's ratio in the coating. According to Equation (2), the plus or minus, namely the direction of the thermal stress σ C , is determined by the difference in the CTE between the IC21 superalloy and NiAlHf alloy (a IC21 − a NiAlHf ). Figure 4 shows (a IC21 − a NiAlHf ) as a function of temperature from room temperature to 1200 • C. It can be seen from Figure 4 that three temperatures at 102, 415, and 1067 • C corresponding to the value of (a IC21 − a NiAlHf ) are zero. Thus, the stress σ C is tensile at the initial heating stage, then becoming compressive between 415 • C and 1067 • C and finally returning to tensile stress when the temperature exceeds 1067 • C. For the heating process of NAIC from 1000 to 1200 • C, the value of (a IC21 − a NiAlHf ) shown in Figure 4 varies from −3.02 × 10 −5 to 8.34 × 10 −5 K −1 . Thus, for the coating with an elastic of 155 GPa and a Poisson's ratio of 0.35 [26,27], the mismatch stress σ C calculated by Equation (2) is from about −68.4 to 228.6 MPa during the temperature variation. It should be mentioned that the yield stress of the coating measured at 970 • C is about 25 MPa [28], which is lower at a higher temperature. Undoubtedly, both the compressive mismatch stress and tensile mismatch stress exceeded the yield stress of the coating during the heating process from 1000 to 1200 • C, at which point the plastic deformation of the coating occurs, while the NANA samples whose coating and substrate share the same material (NiAlHf) have the same thermal expansion coefficient. Thus, both the value of (a M − a C ) and σ C is 0, indicating that no thermal mismatch stresses exist between the coating and the substrate. Therefore, it is reasonable to speculate that no rumpling behavior occurs after cyclic oxidation. The analysis mentioned above sheds light on the relationship between surface rumpling and the thermal expansion coefficient, which is highly consistent with the experiment results in Figure 3.
pansion coefficients and calculate the stress quantitatively to further inves ing-substrate thermal expansion mismatch. For an elastic thin-film coating o substrate, the equibiaxial in plane stress is quantified as follows [25]: and are the CTE of the coating and matrix, respectively, coating elastic modulus, ΔT is the temperature difference and ν denote ratio in the coating. According to Equation (2), the plus or minus, namely t the thermal stress σ , is determined by the difference in the CTE between t alloy and NiAlHf alloy ( − ). Figure 4 shows ( − ) a temperature from room temperature to 1200 °C. It can be seen from Figu temperatures at 102, 415, and 1067 °C corresponding to the value of ( zero. Thus, the stress is tensile at the initial heating stage, then becomin between 415 °C and 1067 °C and finally returning to tensile stress when th exceeds 1067 °C. For the heating process of NAIC from 1000 to 1200 °C ( − ) shown in Figure 4 varies from −3.02 × 10 −5 to 8.34 × 10 −5 K − coating with an elastic of 155 GPa and a Poisson's ratio of 0.35 [26,27], the m σ calculated by Equation (2) is from about −68.4 to 228.6 MPa during th variation. It should be mentioned that the yield stress of the coating meas is about 25 MPa [28], which is lower at a higher temperature. Undoubtedly pressive mismatch stress and tensile mismatch stress exceeded the yield str ing during the heating process from 1000 to 1200 °C, at which point the mation of the coating occurs, while the NANA samples whose coating and the same material (NiAlHf) have the same thermal expansion coefficient. value of ( − ) and σ is 0, indicating that no thermal mismatch stresse the coating and the substrate. Therefore, it is reasonable to speculate tha behavior occurs after cyclic oxidation. The analysis mentioned above she relationship between surface rumpling and the thermal expansion coeffi highly consistent with the experiment results in Figure 3.

Effect of the Stress in the Oxide Scale
As can be seen in Figure 3d, even the surface on the NANA sample implies that the oxidation process appears to have little effect on rumpling behavior. However, when comparing the surface residual stress of NAIC and NANA after 50 one-hour cycles in Figure 5, it should be mentioned that the rumpled sample NAIC exhibits residual stress of about 2.05 GPa, while the NANA sample with an even oxide scale was measured at 3.91 GPa. The residual stress value of the NANA is nearly twice as high as that of the NAIC, which is confusing as the NAIC is considered to require higher residual stress because it suffered extra thermal mismatch stress compared to the NANA sample. However, in fact, the deformation of the coating contributes to releasing the stress in the NAIC sample, while the undeformed surface on the NANA sample restrains the stress release. Presumably, once the confined stress in the NANA scales is released under certain conditions, there is a high likelihood of film or coating deformation. Therefore, it is of urgent necessity to establish whether the stress in the oxide scale has an effect on the rumpling behavior.
As can be seen in Figure 3d, even the surface on the NANA oxidation process appears to have little effect on rumpling behavio paring the surface residual stress of NAIC and NANA after 50 on 5, it should be mentioned that the rumpled sample NAIC exhibits 2.05 GPa, while the NANA sample with an even oxide scale was The residual stress value of the NANA is nearly twice as high as t is confusing as the NAIC is considered to require higher residual s extra thermal mismatch stress compared to the NANA sample. H formation of the coating contributes to releasing the stress in the N undeformed surface on the NANA sample restrains the stress re the confined stress in the NANA scales is released under certain co likelihood of film or coating deformation. Therefore, it is of urge whether the stress in the oxide scale has an effect on the rumpling Previous studies show that two kinds of stress-oxide grow coating mismatch stress-could exist in the oxide scale [29]. In ord stresses separately, two groups of experiments were conducted introduced in the subsequent sections. Figure 6a shows the cross-section of the NAIC specimen afte in the Ar atmosphere at 1200 °C. It is obvious that remarkable su be clearly observed without the oxide film, which implies that th by the cyclic oxidation is determined by the cyclic process instead o In addition, some white granular phases can be found on the coat The EDS results confirm that the white precipitates are mainly Mo ical compositions of 39.66 at.% Ni, 17.78 at.% Al, 0.05 at.% Hf, Previous studies show that two kinds of stress-oxide growth stress and the filmcoating mismatch stress-could exist in the oxide scale [29]. In order to investigate the two stresses separately, two groups of experiments were conducted and detailed results are introduced in the subsequent sections. Figure 6a shows the cross-section of the NAIC specimen after the cycle examination in the Ar atmosphere at 1200 • C. It is obvious that remarkable surface distortion can still be clearly observed without the oxide film, which implies that the rumpling introduced by the cyclic oxidation is determined by the cyclic process instead of the oxidation process. In addition, some white granular phases can be found on the coating surface in Figure 6a. The EDS results confirm that the white precipitates are mainly Mo-rich phases with chemical compositions of 39.66 at.% Ni, 17.78 at.% Al, 0.05 at.% Hf, 42.51 at.% Mo. Mo-rich phases tend to react with oxygen to form a volatile phase when suffering oxidation, whereafter micropores and fissures are induced in the film. It is worth mentioning that these microdefects have little impact on coating deformation. The cross-section morphology after 50 h of isothermal oxidation is shown in Fi 6b. It is noteworthy that secondary reaction zone (SRZ) consists of γ/γ' matrix and ne like TCP phases can be found beneath the IDZ. The formation of SRZ is determine the Al activity and the residual stress state of the coating, which results in a signif reduction in the mechanical properties of the superalloy substrate [30]. Most of the o scale is straight and flat in Figure 6b, apart from a few local buckles as the inserted nified image Figure 6c shows. The profile of the coating remains relatively straight the profile of the oxide scale is in a convex state so that the film is totally separated the coating. Considering that the growth stress of oxide scale and phase transform are the two potential factors influencing the surface depression under the conditio isothermal oxidation, while no γ' phase can be observed in Figure 6b, then the buck oxide scale was probably induced by the scale growth stress, which is insufficient t form the coating surface but will rather induce the oxide scale buckle. This is consi with the high residual stress results of the NANA sample in Figure 5.

Effect of Mismatch Stress between Scale and Coating
Based on the analysis mentioned in Section 3.2, the mismatch stress between the ing and substrate contributes to the surface rumpling under thermal cycle loads. Th is of great necessity to investigate the mismatch stress between oxide scale and coa which is a potential factor responsible for surface rumpling during thermal cyclic.
To validate the presumption above, an experiment focusing on oxide scale and ing was designed. NANA samples were preoxidized (isothermal oxidation at 1200 ° 50 h) to generate an oxide scale about 2 μm thick on the coating, and thereafter ex enced 50 one-hour cycles between 1200 °C and room temperature in an argon atmosp In this experiment, the influence of the coating-substrate thermal expansion mismat negligible because of the same thermal expansion coefficient of the coating and subs Furthermore, the argon protection environment during the thermal cycle could exc the interference from oxide scale growth stress. Additionally, it can be inferred from ure 5 that some residual stresses exist in the oxide scale after preoxidation, but as pe results discussed above, these residual stresses have a negligible influence on surface pling. Therefore, the residual stresses in the oxide scale are not considered in this ex ment. The cross-section morphology after 50 h of isothermal oxidation is shown in Figure 6b. It is noteworthy that secondary reaction zone (SRZ) consists of γ/γ' matrix and needlelike TCP phases can be found beneath the IDZ. The formation of SRZ is determined by the Al activity and the residual stress state of the coating, which results in a significant reduction in the mechanical properties of the superalloy substrate [30]. Most of the oxide scale is straight and flat in Figure 6b, apart from a few local buckles as the inserted magnified image Figure 6c shows. The profile of the coating remains relatively straight, but the profile of the oxide scale is in a convex state so that the film is totally separated from the coating. Considering that the growth stress of oxide scale and phase transformation are the two potential factors influencing the surface depression under the condition of isothermal oxidation, while no γ' phase can be observed in Figure 6b, then the buckle of oxide scale was probably induced by the scale growth stress, which is insufficient to deform the coating surface but will rather induce the oxide scale buckle. This is consistent with the high residual stress results of the NANA sample in Figure 5.

Effect of Mismatch Stress between Scale and Coating
Based on the analysis mentioned in Section 3.2, the mismatch stress between the coating and substrate contributes to the surface rumpling under thermal cycle loads. Thus, it is of great necessity to investigate the mismatch stress between oxide scale and coating, which is a potential factor responsible for surface rumpling during thermal cyclic.
To validate the presumption above, an experiment focusing on oxide scale and coating was designed. NANA samples were preoxidized (isothermal oxidation at 1200 • C for 50 h) to generate an oxide scale about 2 µm thick on the coating, and thereafter experienced 50 one-hour cycles between 1200 • C and room temperature in an argon atmosphere. In this experiment, the influence of the coating-substrate thermal expansion mismatch is negligible because of the same thermal expansion coefficient of the coating and substrate. Furthermore, the argon protection environment during the thermal cycle could exclude the interference from oxide scale growth stress. Additionally, it can be inferred from Figure 5 that some residual stresses exist in the oxide scale after preoxidation, but as per the results discussed above, these residual stresses have a negligible influence on surface rumpling. Therefore, the residual stresses in the oxide scale are not considered in this experiment. Figure 7a shows the surface morphology of the NANA sample upon 50 one-hour thermal cycles at 1200 • C. The flat zone labeled as A1 and rumpled zone labeled as B1 were proved to be oxide scale by EDS results in Table 3. Figure 7b shows the cross-section morphology of the NANA sample. By comparing Figure 7a,b, the oxide scale shows a duplex structure with outer layer A1 and inner layer B1. As Figure 7b shows, layer A1 is relatively flat and straight, while layer B1 exhibits quite obvious undulations, showing a group of sinusoidal periodic undulations of oxide scale. The surface and cross-section morphologies indicate that layer B1 has strong interfacial bonding with the coating, which guarantees the combined deformation as the coating deforms. Meanwhile, the unevenness of layer B1 reduces the contact area with layer A1, thus contributing to the spallation of layer A1. oatings 2021, 11, x FOR PEER REVIEW guarantees the combined deformation as the coating deforms. Meanwhile, t of layer B1 reduces the contact area with layer A1, thus contributing to th layer A1. One more test with 100 one-hour thermal cycles was performed unde perimental conditions, with the cross-section morphologies (Figure 7d) exh lar rumpling pattern, as Figure 7b shows. Comparing Figure 7b,d, it is wor more scale spallation instead of more severe surface rumpling occurred afte 50 one-hour thermal cycles. This result indicates that the thermal cyclic loa oxide samples contribute to the scale spallation, but are not responsible f undulations. Base on the surface morphology in Figure 7c, three different A2, B2 and C2 can be seen on the coating surface, corresponding to the scale, the middle-wrinkled scale, and the innermost layer with plenty of tively. Energy spectrum analyses confirmed that the main elements of zone Al (Table 3), which indicates that zone C2 is exclusively a NiAlHf coating. T of the surface morphology and cross-section morphology shows that the coating surface are uppermost for these undulations and wrinkles (Figure 7  One more test with 100 one-hour thermal cycles was performed under the same experimental conditions, with the cross-section morphologies (Figure 7d) exhibiting a similar rumpling pattern, as Figure 7b shows. Comparing Figure 7b,d, it is worth noting that more scale spallation instead of more severe surface rumpling occurred after an additional 50 one-hour thermal cycles. This result indicates that the thermal cyclic loads on the preoxide samples contribute to the scale spallation, but are not responsible for the surface undulations. Base on the surface morphology in Figure 7c, three different layers labeled A2, B2 and C2 can be seen on the coating surface, corresponding to the outermost flat scale, the middle-wrinkled scale, and the innermost layer with plenty of voids, respectively. Energy spectrum analyses confirmed that the main elements of zone C2 are Ni and Al (Table 3), which indicates that zone C2 is exclusively a NiAlHf coating. The correlation of the surface morphology and cross-section morphology shows that the voids on the coating surface are uppermost for these undulations and wrinkles (Figure 7c,d). Table 3. Average chemical composition of the marked zone in Figure 7 (in at.%). For the purpose of determining whether the formation of voids was at the stage of preoxidation or not, the surface morphology of a preoxidation sample was employed. Most of the oxide film of the preoxide sample examined by SEM was continuous and intact, but as presented in Figure 7e, there was a small quantity of spallation on the coating surface. The coating is completely naked on the surface with a number of voids, and even the NiAlHf grain boundaries can be clearly observed in these voids. This is attributed to the consumption of a large amount of Al ions during the growth of alumina scale and different diffusion speed between Al and Ni (known as the Kirkendall effect), leaving vacancies at the coating-oxide scale interface. Finally, such vacancies are condensed and grown into voids during the oxidation process [6,16,31]. The sinusoidal periodic undulations stemmed from voids having a maximum amplitude of about~5 µm when measured by ImageJ, which are one order of magnitude lower than that caused by mismatch stress between the coating and substrate (about~15 µm). However, the peeling oxide film and naked coating caused by these voids similarly imply that they deserve further attention.

Effect of Phase Transformation
As mentioned in Section 3.2, different chemical composition between the coating and substrate contributes to interdiffusion, including refractory elements upward into the coating and the aluminum element downward into the substrate. Obviously, the outdiffusion of refractory elements is not responsible for the surface undulation, because the content of these elements is too low to induce remarkable volume change of the coating. The in diffusion of aluminum, however, could result in a phase degradation from β-NiAl to γ'-Ni3Al, which is accompanied by a reduced volume of between 8% and 38% [32]. Figure 8 shows the phase evolution and distribution of the NAIC samples after different oxidation times. For the sample after 50 h isothermal oxidation in air, it is clearly observed in Figure 8a that neither prominent phase transformation nor rumpling occurs in the coating. The coating evolution for 100 h oxidation is shown in Figure 8b. There are already a number of β-NiAl transformed into γ'-Ni 3 Al, and of particular note in Figure 8b is that the γ' phase concentrates into a continuous path from the substrate towards the oxide scale. It is considered that the columnar structure of the EB-PVD deposit coating provides the perpendicular grain boundaries; furthermore, the diffusion of aluminum ions into the coating is primarily grain-boundary diffusion that promotes the precipitation of γ' phase at the grain boundaries. Meanwhile, surface depression appears right above the γ' phase position. This result is in strong agreement with the conjecture that phase transformation could affect the rumpling behavior. Figure 8c shows that a much larger amount of β phase has transformed into γ' phase after 200 h oxidation, while the parts closest to the substrate have turned into γ' phase completely. Moreover, the path of the γ' phase in Figure 8c becomes wider and the surface depression becomes deeper than that in Figure 8b. It appears that the evolution of the surface rumpling of the coating is due to a function of the oxidation time-namely, the amount of transformed phase. However, as presented in Figure 8d, some other regions of the same specimen undergoing 200 h isothermal oxidation accompanied with a large amount of phase transformation exhibit straight oxide scale. This seems to be in contradiction with the experiment result that rumpling behavior depends on the amount of the transformed phase. More details and analysis about phase transformation should thus be discussed meticulously.

Discussion
In the present work, the aim was to determine the effect of the internal fa coating and scale rumpling behavior under thermal cycle and isothermal conditio air/Ar atmosphere. It can be concluded from the experimental results that the mism thermal expansion coefficient between the coating and substrate, and the phase d tion from β to γ' are the determining factors in inducing rumpling in cyclic oxida isothermal oxidation, respectively. Compared with previous studies, the result study supply quantified evidence for interpreting the rumpling mechanism with understanding. Figure 4 illustrates how the thermal mismatch stress between the coating a strate has an impact on the coating deformation by an instance of a heating proces is in agreement with the finding of Clarke et al. [9] that cyclic oxidation with a sm perature decline, such as from 1150 to 1000 °C, is sufficient to induce rumpling. factor of note is that the temperature is a precondition of rumpling. Since a highature condition reduces the coating yield strength to a value lower than the therm match stress, then the plastic deformation of the coating can be induced. Meanwhi factors like temperature variation, length of holding time, and the number or fr of thermal cycles investigated in previous studies [10,11] affect the rumpling beh substantially influencing the thermal mismatch stress. When the coating and s possessed the same expansion coefficients as the NANA sample, the rumpling v and these external factors had no impact. Thus, it can be concluded that (1) the lo strength of coating due to high temperature, and (2) the thermal mismatch stress the coating and substrate are the primary causes of rumpling behavior under the cycle.
In the condition of isothermal oxidation, few thermal mismatch stresses exi coating, but the rumpling can still be observed, as seen in Figure 8b,c. This is bel be related to the phase degradation from β to γ' due to the aluminum depletion the oxidation and interdiffusion process, which gives rise to a reduction in the volume. However, a considerable amount of phase transformation with a straight

Discussion
In the present work, the aim was to determine the effect of the internal factors in coating and scale rumpling behavior under thermal cycle and isothermal conditions in the air/Ar atmosphere. It can be concluded from the experimental results that the mismatched thermal expansion coefficient between the coating and substrate, and the phase degradation from β to γ' are the determining factors in inducing rumpling in cyclic oxidation and isothermal oxidation, respectively. Compared with previous studies, the results in this study supply quantified evidence for interpreting the rumpling mechanism with a deeper understanding. Figure 4 illustrates how the thermal mismatch stress between the coating and substrate has an impact on the coating deformation by an instance of a heating process, which is in agreement with the finding of Clarke et al. [9] that cyclic oxidation with a small temperature decline, such as from 1150 to 1000 • C, is sufficient to induce rumpling. Another factor of note is that the temperature is a precondition of rumpling. Since a high-temperature condition reduces the coating yield strength to a value lower than the thermal mismatch stress, then the plastic deformation of the coating can be induced. Meanwhile, other factors like temperature variation, length of holding time, and the number or frequency of thermal cycles investigated in previous studies [10,11] affect the rumpling behavior by substantially influencing the thermal mismatch stress. When the coating and substrate possessed the same expansion coefficients as the NANA sample, the rumpling vanished and these external factors had no impact. Thus, it can be concluded that (1) the low yield strength of coating due to high temperature, and (2) the thermal mismatch stress between the coating and substrate are the primary causes of rumpling behavior under the thermal cycle.
In the condition of isothermal oxidation, few thermal mismatch stresses exist in the coating, but the rumpling can still be observed, as seen in Figure 8b,c. This is believed to be related to the phase degradation from β to γ' due to the aluminum depletion during the oxidation and interdiffusion process, which gives rise to a reduction in the coating volume. However, a considerable amount of phase transformation with a straight and flat surface in Figure 8d appears to refute the assumption above. In order to clarify the relationship between phase transformation and rumpling, the fraction of γ' phase in Figure 8 was measured horizontally by ImageJ and then schematized in a color waterfall plot graphic ( Figure 9). A dashed line is used for the sample after 50 h oxidation in Figure 9a, because the NiAlHf coating is still dominant with the β phase, as Figure 8a shows, while the other three solid lines in Figure 9 correspond to the fraction of γ' phase changing with the distance in the horizontal direction. Figure 9d shows the sample with a large amount of phase transformation after 200 h oxidation, but scarcely any rumpling behavior emerged on its surface, which implies that the amount of γ' phase is not responsible for rumpling. Meanwhile, Figure 9b exhibits a visualized conclusion that centralized γ' phase transformation corresponds to rumpling sites in Figure 9b. Similarly, the more centralized β to γ' phase transformation in Figure 9c is more favorable for the local reduction in volume, and thereby the more severe rumpling in Figure 8c. The analysis above reveals that the distribution mode of γ' phase in the coating, instead of the amount of phase degradation, determines the surface depression status in isothermal oxidation.
Ultimately, commonly voids on local coating caused by the Kirkendall effect in the oxidation stage may be regarded as rumpling due to the wave profile shown in Figure  7b,d. In comparison with those rumpling induced by thermal mismatch stress or phase degradation, the surface distortion originating from Kirkendall voids neither deforms the coating nor has a considerable amplitude, which has less influence on the surface morphologies. Moreover, the experiment results manifest that the growth stress of the oxide scale and the thermal mismatch stress between the scale and coating are insufficient to impact the rumpling behavior. Therefore, it can be concluded that oxidation may contribute to surface rumpling in some aspects, but it is not the internal factor that determines coating rumpling.

Conclusions
Various internal factors affecting the surface rumpling behavior of NiAlHf coating were investigated at 1200 °C, and several conclusions can be drawn as follows: 1. During the thermal cycling test, the reduced yield strength of the β-NiAl coating due to high temperature and the thermal mismatch stresses between the β-NiAl coating and IC21 superalloy substrate are the major internal factor to induce surface rumpling.
2. The growth stress of oxide scale and thermal mismatch stress between the oxide scale and the β-NiAl coating is insufficient to deform the coating, while the Kirkendall A dashed line is used for the sample after 50 h oxidation in Figure 9a, because the NiAlHf coating is still dominant with the β phase, as Figure 8a shows, while the other three solid lines in Figure 9 correspond to the fraction of γ' phase changing with the distance in the horizontal direction. Figure 9d shows the sample with a large amount of phase transformation after 200 h oxidation, but scarcely any rumpling behavior emerged on its surface, which implies that the amount of γ' phase is not responsible for rumpling. Meanwhile, Figure 9b exhibits a visualized conclusion that centralized γ' phase transformation corresponds to rumpling sites in Figure 9b. Similarly, the more centralized β to γ' phase transformation in Figure 9c is more favorable for the local reduction in volume, and thereby the more severe rumpling in Figure 8c. The analysis above reveals that the distribution mode of γ' phase in the coating, instead of the amount of phase degradation, determines the surface depression status in isothermal oxidation.
Ultimately, commonly voids on local coating caused by the Kirkendall effect in the oxidation stage may be regarded as rumpling due to the wave profile shown in Figure  7b,d. In comparison with those rumpling induced by thermal mismatch stress or phase degradation, the surface distortion originating from Kirkendall voids neither deforms the coating nor has a considerable amplitude, which has less influence on the surface morphologies. Moreover, the experiment results manifest that the growth stress of the oxide scale and the thermal mismatch stress between the scale and coating are insufficient to impact the rumpling behavior. Therefore, it can be concluded that oxidation may contribute to surface rumpling in some aspects, but it is not the internal factor that determines coating rumpling.

Conclusions
Various internal factors affecting the surface rumpling behavior of NiAlHf coating were investigated at 1200 • C, and several conclusions can be drawn as follows: 1.
During the thermal cycling test, the reduced yield strength of the β-NiAl coating due to high temperature and the thermal mismatch stresses between the β-NiAl coating and IC21 superalloy substrate are the major internal factor to induce surface rumpling. 2.
The growth stress of oxide scale and thermal mismatch stress between the oxide scale and the β-NiAl coating is insufficient to deform the coating, while the Kirkendall voids formed in the oxidation stage are noteworthy, bringing a group of sinusoidal periodic undulations onto the surface and degrading the oxide scale adherence.

3.
Concentrated phase transformation is the major internal factor inducing rumpling during isothermal oxidation. Continuous and concentrated β→γ' transformation brings distinct volume reduction, thereafter contributing to the rumpling of the coating.