The Effect of Thermal Barrier Coating Surface Temperature on the Adhesion Behavior of CMAS Deposits

The ingestion of atmospheric contaminants and their deposition on hot-section hardware is one of the leading sources of distress for land-based and aircraft turbomachinery. As the demand for improved efficiency continues to rise, thermal barrier coatings (TBC) have enabled designers to push hot-section temperatures higher and higher [1]. The temperatures in the high-pressure turbines (HPTs) of modern turbofan engines are now so high that internal and film-cooling of blades, vanes, and shrouds are not sufficient, and the use of TBC is required for bulk metal temperatures to remain at safe levels. One of the most common ceramics used for TBC is Yttria-stabilized zirconia (YSZ). The deposits caused by atmospheric contaminants are generally formed from calcium magnesium aluminosilicate, usually referred to as CMAS [1–4]. At the high turbine inlet temperatures seen in modern commercial turbofans and land-based power generation turbomachinery, CMAS can melt and penetrate the surface of thermal barrier coatings [3,4]. The impact of cycling on TBC spallation in thermal gradient rig testing has been documented [5,6]. However, land-based turbomachinery used for power generation will often operate for months or years at a time without powering down; therefore, this type of equipment is not as susceptible to cyclic-induced effects. In addition, aircraft engines can experience high levels of particulate ingestion during a single cycle; thus, the impact of non-cyclic CMAS accumulation is of great interest. For example, Guffanti et al. indicate that particulate loading could be as high as 2000 mg/m³ in the ash cloud 150 km away from an active volcano [7].

Furthermore, the actual mechanisms responsible for deposit formation and rate of accumulation are less understood [2]. Current knowledge seems to indicate that only particles under 10 μm are able to reach, accumulate, and build in high-pressure turbines, as these are the size particles that can bypass particle separators and make it through the compressor [8].

Various studies have shown that the melting temperature of CMAS falls somewhere in the range between 1423 and 1523 K [2,3,5,6,9,10]. Using metallic targets with an impinging particle-laden jet, Laycock and Fletcher studied 0–10 μm coal ash deposition over a range of gas (1523–1673 K) and surface (1173–1423 K) temperatures [11]. They saw a monotonic increase in deposit capture efficiency with gas temperature, but a puzzling non-monotonic variation with surface temperature (first increasing and then decreasing below and above 1273 K, respectively). Accordingly, this same range of surface temperatures was targeted for the present evaluation of the characteristics of deposit formation of Arizona Road Dust (ARD) with a size distribution of 0–10 µm. Limited testing was also conducted with a smaller size distribution (0–5 μm).

The aim of this study is to understand the impact of TBC surface temperature on the adhesion and the rate of accumulation of CMAS deposits on a TBC coupon experiencing a representative thermal gradient. Interpretation of the results is aided by a companion computational fluid dynamics (CFD) model with particle tracking and deposition prediction.

The present study was conducted utilizing the High-Temperature Deposition Facility (HTDF) at The Ohio State University’s (OSU’s) Aerospace Research Center, see Fig. 1. This facility consists of a large burner encased in a ceramic combustion chamber that burns a propane and oxygen-enriched air premix. This hot gas is sent through a nozzle into an injection block where the upstream temperature is measured with a Type B thermocouple and particles are injected. This particle-laden hot gas then travels down the 19 mm inner diameter equilibrium tube and impinges on the target 2 diameters (38 mm) from the exit. The facility is heated over the course of hours and the jet exit temperature is allowed to reach a steady-state at the desired operating condition. Exit temperature is measured with a B-type thermocouple, and the target surface temperature is measured with an infrared (IR) camera.

For this test campaign, a target holder made of stainless steel and high-temperature alloys is fixed to a slider on a tandem traverse. The traverse allows the holder to be moved along the axis of the impingement jet, while the slider moves the holder transverse to this so that the holder can be removed from the hot flow. Coupons can then be cooled or removed and safely replaced. The coupon used in these experiments is a Hastelloy X ring that is solid on one side. A groove is cut into the ring so that it can be gripped by the coupon holder, and a small divot (<0.5 mm deep) is machined in the backside of the solid coupon surface for thermocouple locating. A bond coat and TBC layer were applied to the nominally 1.27 mm thick solid top surface of these coupons. The coatings were applied by Praxair Surface Technologies and consisted of plasma-sprayed nickel-based bond coat (0.2 mm thick) followed by a 7–8% Yttria-stabilized Zirconia with a vertically segmented microstructure (1 mm thick). Figure 2 shows the TBC coupon along with a schematic of the coupon in the holder. Figure 3 contains a cross section of the coupon.

The holder consists of a C276 block with plumbing into the backside and slots for two slides on the front face. The slides fit into the slots, and their semicircular lips grasp either side of the coupon’s groove. The slides are then clamped in place, and a cast ceramic shield is secured around the coupon and in front of the metal holder. The backside cooling is brought to the block via a flexible hose that allows easy movement of the holder between tests. This hose is connected to a 12.7 mm OD tube (9.4 mm ID) which passes through a compression fitting in a 25.4 mm NPT cross junction. The tube continues through a length of 25.4 mm NPT pipe that is fitted to the backside of the block so that the tube is coaxial with the coupon. The tube exhausts 38 mm from the backside of the coupon, where the flow impinges and is turned back around the tube before exiting through the top of the cross junction to ambient. The bottom leg of the cross junction holds another compression fitting which brings in a 1.56 mm grounded K-type thermocouple that continues to the coupon and fits into the divot, reading the backside metal temperature throughout the test.

This study involves a spread of tests conducted at constant hot jet temperature (1600 K ±15) and velocity (200 m/s ±10) with variable coolant flowrates impinging upon the backside in order to better understand the mechanisms controlling CMAS accumulation on TBC protected hardware.

During the test, a coupon is secured in the coupon holder with the backside thermocouple trapped in the divot. The coupon is then moved across the tandem traverse on the slider, into the far-field of the flow. Once the target is coaxial with the HTDF equilibrium tube, it is slowly moved forward along the traverse until the surface of the TBC is 38 mm from the exit and within the view of the stationary IR camera. The backside temperature is monitored as the coupon soaks in the flow until a steady-state backside temperature has been reached (ΔT < 0.5 K/min). Once the target has reached equilibrium, the coolant is ramped up to the desired flowrate. The backside temperature is again monitored for a steady-state. Once a balance has been reached, particles are added to the flow via a conveyor dropbox. The conveyor belt delivers particles to the flow at approximately 0.25 g/min for the duration of the test. For these tests, the belt ran for 4 min, resulting in a total of 1 g of ARD injected. During the test, an IR video is recorded to track the change in surface temperature and deposit evolution during the injection. Once the dust injection has been completed, the coupon is allowed to remain for 20 min in the hot jet. The holder is then traversed back along the axis of the jet until it is in the far-field where it is again brought out of the flow.

At the jet exit conditions of ∼200 m/s and 1600 K in this study, 1 g of ARD injected across four minutes resulted in an ARD concentration of 399 mg/m³, corrected to standard atmospheric density, and a particle volume fraction on the order 2e–8. Elghobashi showed that one-way coupling exists for volume fractions below 1e–6 [12], so this level of loading is not expected to have impacted the flow field in any appreciable manner. van Donkelaar et al. reported global average particulate concentrations ranging between 0 and 0.08 mg/m³ for sub 2.5 μm particles [13], while Bojdo and Filippone show data suggesting ambient levels approaching 10–100 mg/m³ in sand storms or aircraft generated dust-ups [14]. The overall particulate loading in this testing, therefore, represents a value above that typically seen in a dust-storm to simulate accelerated deposition.

An infrared camera provides the best non-contact method for monitoring TBC surface temperatures throughout the testing. Eldridge et al. demonstrated that a long-wavelength (approximately 12 µm) IR camera provides the best method to directly monitor TBC surface temperatures [15]. With a long-wavelength pyrometer, TBC surface temperatures can be directly read assuming an emissivity of 1. Unfortunately, in the absence of a long-wavelength camera, a short wavelength camera had to be substituted. The IR camera used was a 320 × 256 pixel Electrophysics Silver 420 shortwave (3.6–5.1 µm) camera. Some estimate is thus needed to convert infrared camera intensity readings to a surface temperature. To calibrate the infrared camera results, calibration images of a TBC coupon were taken as the HTDF was warmed up to the operating temperature of 1600 K. These images were taken without any forced impingement cooling on the coupon backside.

A 1D heat transfer model was developed to provide a surface temperature estimate during IR calibration. A resistance schematic for the heat transfer model is shown in Fig. 3. Due to the complex geometry of the coupon holder, it is impractical to estimate a 1D thermal resistance for the last resistor shown in Fig. 3. Therefore, the backside thermocouple measurements were used as an initial boundary condition for the model, and the R_cond resistor in the schematic in Fig. 3 was not included in the initial calculations. To provide a starting surface temperature estimate, radiation was also neglected for the initial calculations. A value of 1 W/m K was used for the thermal conductivity of the TBC and 26 W/m K for the metal substrate, along with a TBC thickness of 1.01 mm and a combined substrate and bond coat thickness of 1.6 mm. The substrate thickness was measured from the cut-ups of several coupons and was slightly thicker than the nominal value.

The TBC surface radiates heat to three locations: the hot jet tube exit, the ceramic housing around the jet tube, and the ambient surroundings. Although the ceramic housing has a temperature gradient across it, the outer surface was measured at approximately 800 K; so, this value was used across the entire surface for a conservative calculation. The view factor from the TBC surface to the jet diameter was calculated to be 0.19, while the view factor to the ceramic housing was calculated to be 0.40, both using Bergman et al. [16] for two coaxial parallel disks. The ambient surroundings take the remainder of the view factor (0.41).

The initial (no radiation) TBC surface temperature estimate can be used with Eq. (4) to calculate the radiation heat flux from the TBC to the surroundings. Recognizing that the impinging jet provides all of the heat transfer to the TBC, the total heat flux is simply the summation of the three radiation terms and the prior conduction heat flux, as represented in Eq. (1). An updated surface temperature can then be calculated with the total heat flux and the impingement jet heat transfer coefficient. This radiation calculation can then be iterated until the updated surface temperature converges to within 0.1 K.

For Eq. (4), the emissivity of YSZ TBC was estimated to be 0.075 (at 1500 K) using data from Ref. [17]. The estimated radiative flux during calibration was approximately 4% of the total convective flux from the hot jet to the target.

The mean thermal conduction resistance shown in Fig. 4 was then used for each calibration point, and the model was evaluated using the complete thermal resistance network shown in Fig. 3, including the bracketed R_cond in Eq. (1) and the radiation effects. The measured backside temperature was no longer used as a boundary condition to the model, rather the ambient cool air was used. Figure 5 shows the delta between the impinging jet temperature and the model-estimated surface temperature versus impinging jet flowrate. Figure 5 shows no clear trend in the jet to surface temperature delta; therefore, the mean value of the data seen in Fig. 5 (185.9 K) was used as a constant offset when estimating TBC surface temperature during IR calibration.

The average IR digital count (referred to hereafter as the intensity) across the TBC surface was correlated to the estimated surface temperature. A logarithmic curve fit was applied to the combined data set so that when backside cooling was applied, and the TBC surface temperature could be estimated based on the measured mean intensity of the surface. Figure 6 shows this curve fit. The accuracy of the fit attests to the repeatability of the IR measurement and optical arrangement over the entire test campaign.

The sensitivity of this 1D model to variations in estimated parameters is shown in Table 1. The backside thermocouple is not an estimated parameter, but there is the potential for variations due to the depth of the divot that was drilled to locate the thermocouple. While there may be some level of uncertainty in these estimated parameters, the estimated surface temperatures all move in sync with any error in these parameters. In other words, the logarithmic fit in Fig. 6 would simply shift up or down. Accordingly, although there is some level of uncertainty in the absolute estimated surface temperatures, the relative differences between surface temperatures will remain the same.

The coupons in TS1125 and TS1455 had 1.46 mm thicker side-walls compared to the others, so the backside surface area where the impingement cooling was located had 25% less surface area. The coupon in TS1455 was insulated in the holder to reduce the amount of conduction heat loss through the back of the coupon into the holder. As expected, this coupon had the highest surface temperature (1455 K).

Figure 7 shows post-test images of the TBC coupons and their deposit structures after removal from the HTDF rig. Figure 8 shows a time progression of CMAS deposit interaction from infrared images taken with different levels of backside cooling. The top three rows show results with an uncooled backside, with surface temperatures in the 1380–1450 K range, but for two different size distributions of ARD. Neither visual nor IR evaluation show any noticeable difference in the deposition characteristics of the two different ARD size distributions at these conditions. This is as expected given that approximately 70% by mass of the 0–10 ARD dust falls under 5 μm. The subsequent rows show the deposit evolution when TBC surface temperature is lowered through the addition of forced convective cooling on the backside. The dashed circles in Fig. 8 show the location of the inner diameter of the HTDF pipe superimposed on the TBC surface. The 19 mm diameter jet was centered to within 3.1 mm radially of the 25 mm diameter coupon center for each test, with an average center radial offset of 2.6 mm.

A qualitative analysis of the deposit morphologies seen in Fig. 8 shows a clear difference in the deposit formation and growth characteristics. The higher surface temperature deposits tend to spread out further and faster along the TBC coupon surface, while the lower temperature deposits appear to accumulate more sporadically and far less evenly. Post-test visual examination of the deposit structures (Fig. 7) confirms the infrared evaluation. The cooler surface temperatures create rough and bumpy deposits that are less uniform, while the TBC coatings subjected to higher surface temperatures have smooth surfaces indicative of more molten conditions throughout the test. For the three uncooled cases, the deposit began to crack off as the coupon cooled off upon removal from the impinging jet. The temperature gradients and associated residual thermal stresses were so strong that the deposit did not simply fall off these coupons, but rather exhibited vigorous spallation off the surface during the post-test cooldown. For TS1297 and TS1233, the deposit stayed on the coupon even after cooling and removal from the test apparatus. For TS1192, the deposit simply felloff the TBC surface upon removal from the test apparatus. Physical examination of the deposits post-test showed that in the uncooled cases, where the surface temperature was the highest, the remaining deposit that did not flake off during cooling was firmly attached to the TBC surface. In the cooled cases, the deposit could easily be removed from the surface with a paintbrush.

The specimens were potted in a cold-mount epoxy and then sectioned for analysis under a scanning electron microscope (SEM). Figure 9 shows an SEM image of the coupon from TS1381, along with energy dispersive X-ray spectroscopy (EDS) results. SEM images were collected using a ThermoFisher Apreo Field Emission SEM at the Center for Electron Microscopy and Analysis (CEMAS) at The Ohio State University. To mitigate the effects of sample charging, the instrument was operating under low vacuum conditions of 50 Pa water vapor, and images were collected with the in-column backscatter detector. The beam was operated at 8 kV. EDS spectra were collected with an EDAX Octane Elect Plus 30 mm² Silicon Drift Detector and analyzed in the edax team software package. The EDS intensity values for the primary TBC constituents (yttrium and zirconium) and the CMAS constituents (calcium, magnesium, aluminum, silicon) show a clear delineation between the TBC and the CMAS deposits, with no indication of penetration of the CMAS elements into the TBC. Given that the average surface temperature in TS1381 was just below the reported CMAS melting temperature range (1420–1520 K), the lack of penetration of CMAS into the TBC is expected and aligns with the results shown by Mack et al., who showed that the conditions for TBC spallation at low cycles (five or fewer cycles) requires hotter surface temperatures (∼1500 K) and longer high-temperature dwell times (1–4 h) [5].

Infrared videos of the ARD injection periods were analyzed frame by frame to better understand the rate of deposit formation. The pixel location of the TBC coupon in the frames of each video was identified, and then the infrared digital count (or intensity) was interrogated for the duration of the injection. The average intensity across the range of pixels corresponding to the coupon was calculated for each frame. In addition, a frame by frame intensity delta relative to a pre-injection “baseline” frame was calculated for each pixel identified as belonging to the coupon surface. When deposits form on the TBC, they are immediately evident due to a greater emissivity and higher surface temperature when compared to the surrounding TBC. Once the intensity delta of a pixel passed a certain threshold from the pre-injection baseline, that pixel was flagged as having accumulated CMAS, and the percentage of total pixels covered was tracked frame by frame through the injection. To determine an appropriate value for the deposit intensity threshold, the standard deviation of counts for the TBC pixels in each pre-injection baseline frame was calculated for each test injection. The standard deviations were averaged together across all tests, and the threshold was set at three times the average standard deviation of intensity. This value was approximately 800 counts. Setting the deposit intensity threshold to three times the average standard deviation ensures that slight fluctuations in the surface are not erroneously identified as CMAS deposit locations.

Figure 10 shows the fraction of CMAS coverage versus normalized injection time. For these tests, since 1 total gram of ARD was injected, the normalized injection time is synonymous with grams of ARD delivered. Figure 10 shows a clear trend in the rate of accumulation of CMAS relative to surface temperature. As surface temperature decreases, so do both the rate of CMAS accumulation and the total level of CMAS coverage. These data show a different trend than what Fletcher and Laycock reported for metallic surfaces, where capture efficiency decreased with an increasing surface temperature above 1273 K when the gas temperature was held constant at 1673 K [11]. Given the significant differences between these two studies, this result is not unexpected. TS1125, which was just inside the observed sticking threshold, has the slowest accumulation rate, as expected. The lower coolant flowrates (higher surface temperatures) accumulate CMAS more quickly. The insulated coupon, which had the highest surface temperature, shows the fastest rate of accumulation. All of the uncooled coupons (TS1455, TS1406, and TS1381) along with the coupon from TS1297 show similar accumulation rates, while accumulation slows down markedly in TS1233. This indicates there is a significant change in deposition behavior in the surface temperature range of ∼1230–1300 K.

Figure 11 shows the average intensity of pixels flagged as having accumulated CMAS. The values for the first several frames jump around significantly simply because the number of pixels with CMAS is quite low. The average intensity of the CMAS is a good proxy for deposit thickness. The deposit maintains a thermal gradient between the TBC surface and the impinging hot air; therefore, an increase in deposit intensity indicates the deposit is growing thicker. Generally, the average deposit intensity increases over time, indicating growing deposit thickness. The deposits in the higher surface temperature tests (TS1455, TS1406, and TS1381) began to saturate the IR camera, which is why these intensity values appear to plateau. Not all of the pixels on the deposit surface are saturated, however, which is why those three tests plateau at different intensities.

It should be noted that no deposit surface temperature inferences can be made using the intensity values seen in Fig. 11. The correlations for intensity to temperature seen in Fig. 6 only apply for the TBC surface, since the thermal properties of the deposit (k, t, and ɛ) are unknown.

Inspection of the trends for TS1233 and TS1192 in Figs. 10 and 11 shows an apparent offset in deposit formation relative to the other tests. This was not a delay in the actual injection of ARD, but rather an indication of the slower nature of deposit formation. The overall average intensity of the entire TBC coupon region begins to rise immediately, but no deposits are identified until the intensity passes the deposit threshold. The super-cooled coupon, TS1125, does not have this same delay because there was a small region of residual deposit on this coupon from operation at slightly higher coolant flowrates while identifying the sticking threshold.

Figure 12 shows an image of the TS1455 coupon (surface temperature ∼1455 K) at the end of the injection period, versus the TS1233 coupon (surface temperature ∼1233 K). These two images provide insight into the nature of deposit formation and morphology and their variation with surface temperature. When the TS1455 coupon was immersed in the hot jet, it was difficult to distinguish the presence of isolated deposits on the surface due to the uniform appearance of the deposit. This is indicative of the deposit on the surface being near the melting temperature and therefore able to build evenly along the surface. In contrast, the deposit formed at the lower surface temperature in TS1233 was variegated and non-uniform. The deposit did not form or grow evenly along the surface. Rather, impinging molten particles were more likely to stick to existing deposits than the TBC surface itself.

For the uncooled cases, the initial TBC surface temperatures were just under the range of reported CMAS melting temperatures. As the CMAS particles deposit, they cool down and harden before penetration into the TBC can take place. As the deposit builds, the deposit surface temperature goes up, as evidenced by the gradual rise in intensity seen in Fig. 11, and may even exceed the CMAS melting temperature. However, the underlying layers of CMAS stay below the melting temperature due to the thermal gradient that builds across the deposit. This thermal gradient creates a stratified deposit that is somewhat visible in Fig. 7 for TS1455 and TS1406. The stratified nature of deposits has been shown previously in cascade testing of turbine nozzle guide vanes by Lundgreen et al. [18].

Figure 13 shows a summary of the jet, surface, and backside temperatures versus heat flux ratio for the calibration points and the test points. Note that all of the calibration points are at a zero heat flux ratio; so the points were spread out for visual clarity. The shaded region in Fig. 13 shows the spread of surface temperatures for which the IR calibration is valid. The deposition threshold line shows the observed sticking threshold temperature of 1104 K. The coupon from TS1125 was held at a temperature just above this threshold. A hotter jet temperature should require a cooler surface to prevent deposition; therefore, the observed threshold is specific to the tested jet temperature of 1600 K and jet velocity of 200 m/s.

The motivation of this companion CFD study is to explore the effects of decreased surface temperature on particle deposition to aid in the interpretation of the experimental results already presented. To that end, a CFD model was created that is representative of the HTDF experimental facility. Particles were tracked through a steady Reynolds-averaged Navier–Stokes (RANS) flow solution using Lagrangian multi-phase particle tracking (LMPT). Particle-wall impacts were handled using the OSU deposition model [19] including a new molten model presented in Plewacki et al. [10] that reduces particle yield stress at particle temperatures above 1362 K. Particles with lower yield stress at high temperatures are more likely to deposit upon impact in their softened or molten state.

Two 3D domains were created in star-ccm+ in order to explore this phenomenon. The first domain models the equilibrium tube of the HTDF, just downstream of the injection point to 0.2 diameters from the exit. The second domain takes the output from the first as its inlet conditions and models the last 0.2 diameters of the pipe and the impingement zone. Both domains can be seen in Fig. 14 where L₁ = 584.2 mm, L₂ = 305 mm, and D = 19.05 mm.

A polyhedral mesh with a base size of 1 mm was used to discretize the domain. Fifteen prism layers were used to adequately model the flow near the boundary, and the value of y+ is less than 0.2 throughout the domain. The fluid under consideration is air, and the ideal gas law was used to calculate the density of the fluid. The dynamic viscosity and thermal conductivity were calculated using Sutherland’s law. A temperature-dependent polynomial was used to calculate specific heat in the domain. The k − ω SST model was used to model the effects of turbulence. The Lagrangian particle tracking feature in star-ccm+ was used to model the trajectory of the particles. The interaction between the particle and the fluid is considered to be one-way coupled based on the criteria suggested by Elghobashi [12]. All the grid and flow parameters apply to both the domains shown in Fig. 14.

This two-domain setup allows one flow solution to be obtained for a given upstream jet condition in Domain 1 without changing any boundary conditions near the target. Because the flow and particles greater than two pipe diameters upstream of the target are not affected by the level of cooling on the surface, the exit flow properties and particle information from Domain 1 can be used as the exact input condition for every simulation in Domain 2 by using a table injector. To study the effect of backside cooling on the impingement plate, the surface temperature of the plate was held constant at a chosen value in the range of the experimental values from Table 2. Using two domains like this, saves significant computational time.

A mass flow inlet condition is used at the beginning of Domain 1 resulting in a spatially uniform velocity and temperature profile upstream. The total temperature at this point is calculated from the mass flow average of the injection line and primary hot gas path. Constant heat flux is specified on the pipe walls to match the average exit temperature measured experimentally. The mass distribution of particles injected in the simulation was matched to the experimental dust distribution provided by the supplier Powder Technology, Inc. as shown in Fig. 15.

Four cases were run to determine the effectiveness of the cool boundary layer adjacent to the target surface in lowering the particle temperatures before impact. One case was run with an adiabatic wall to establish a baseline comparison. The other cases were run with constant target surface temperatures of 1450 K, 1300 K, and 1150 K. The boundary layer seen in the adiabatic wall case and the cooled target cases are interrogated in Fig. 16 to determine the near-wall thermal boundary layer. These temperatures are taken from the CFD domain along the centerline of the target, as shown in Fig. 14.

In the boundary layer temperature profiles shown in Fig. 16, there is a steep gradient near the wall for the cooled cases, over a distance of only 0.4 mm. Because the thermal boundary layer is so thin, the particles are only affected by the lower temperature region of the flow for a short amount of time. It is expected that only small particles (<4 µm) with low thermal Stokes numbers will be influenced by this cool region, and larger particles will retain a temperature closer to their inlet temperature when they impact the surface.

Figure 17 shows the average temperature of the particles upon impact for each diameter that was injected. To match the experimental setup, only impacts within the 25.4 mm diameter circle on the target plate are evaluated. This plot indicates that for particles above 4 μm, the thermal boundary layer has a negligible effect on the particle temperature, as both the mean and the spread do not change significantly from the adiabatic wall to the cooled wall cases. It should be noted that the large particles in the CFD simulations are not at the jet temperature when they impact the surface. This may be due to the high velocity of the flow, leaving the particles with less time to come up to temperature before exiting the pipe or due to their radial location closer to the cool walls of the pipe. Below 4 μm, as wall temperature decreases relative to the jet temperature, the particle temperature upon impact also decreases while the spread of temperatures seen at impact increases. Average particle impact temperature may exceed the mass flow-averaged (MFA) temperature when that particle size is concentrated near the center of the pipe where gas temperature is maximum. The lower limit of the impact temperature of the 1 μm particles is very close to the wall temperature, indicating that some of these particles are strongly affected by the wall thermal boundary layer, despite the small boundary layer thickness.

Because the particle yield strength is dependent on particle temperature, the molten model predicts no change in the deposition in the cooled wall cases for particle diameters greater than 4 μm. This represents nearly 50% of the cumulative mass distribution in Fig. 15. For particle diameters less than 4 μm, deposition is reduced, but not significantly. When integrated over all diameters, the result is only a modest decrease in capture efficiency with decreasing wall temperature, as shown in Fig. 18.

Impact efficiency is approximately constant (97.4% ± 0.05) between the four cases, indicating that the cool wall does not significantly affect the bulk flow patterns or the particle trajectories. In Fig. 18, there is about a 0.4–0.9% absolute decrease (1.4–3.6% relative decrease) in the capture efficiency as the wall temperature is lowered by 150 K. Because the impact efficiencies are the same, the decrease in the capture efficiency indicates that the lower wall temperature causes the model to predict less deposition as expected. This is in agreement with the trends seen in the experimental data. It should be noted, however, that while the trend of decreasing deposition with decreasing surface temperature is the same as the observed experimental trend, the magnitude of the effects is not equal.

The experimental surface coverage data from Fig. 10 are plotted at three successive times (0.2 g, 0.6 g, and 1 g injected), also in Fig. 18. It is noted that early in the deposition process (0.2 g injected), the wall temperature has a dramatic effect on the deposition rate (as measured by CMAS coverage area). Test coupons with temperatures below 1250 K experience an order of magnitude less deposition than the hotter coupons. As the deposit begins to cover the surface, however, even the cooler walled cases eventually experience deposit coverage fractions within 20% of the hot wall cases. The thermal boundary layer modeled by the CFD is most closely analogous to the experimental condition where the target surface is relatively clean, i.e., the very beginning of each test. At this point, the CFD predicts only a 2.5% relative decrease in capture efficiency (from 1450 K to 1150 K), whereas the experimental deposit area coverage drops by more than 90% over the same surface temperature interval (at 0.2 g injected). This disparity between the experimental and computational deposition rates can likely be attributed to the molten model formulation. The computational model determines sticking based on the particle temperature and velocity alone, which does not change appreciably for the larger particles that make up half of the mass of particulate. If the sticking was instead based on some combination of the particle and wall temperature at impact, the model would match more closely the order of magnitude difference in early deposition rates. Note that in Fig. 18, the left axis is at a magnified scale to show the decrease in predicted capture efficiency with temperature, as this trend would not be clear otherwise when presented with the surface coverage fractions.

It was observed that towards the end of the deposition cycle, the net CMAS area coverage is much less dependent on initial TBC temperature. It was shown in the experimental results for lower surface temperatures that deposition occurs preferentially where a deposit already exists on the surface. This is due to the increased surface temperature in locations where deposits have already begun to form. As the thickness of a deposit increases, the deposit surface is further insulated from the coolant and thus becomes hotter than the underlying TBC. This increased surface temperature in turn causes the inbound particles to impact the hotter deposit surface at higher temperatures, thus resulting in higher deposition rates comparable to the hot TBC wall temperature cases.

These effects can be properly modeled using mesh morphing with conjugate heat transfer between the target, deposit, and impinging flow. Morphing the surface as particles stick to form deposits would allow the surface accumulation to be properly represented geometrically in the simulation. Conjugate heat transfer would then account for the insulating effect of the deposit. As deposition increases, and the surface is morphed further from the cooled wall, the surface temperature of the deposit would exceed that of the TBC surface. With an increase in the surface temperature, particles would more readily stick to the area where deposits already exist in the simulation. If the molten model included a particle yield strength dependency on some combination of particle and surface temperature, the trends shown in Fig. 10 could perhaps be replicated.

The role of TBC surface temperature on the rate and morphology of CMAS deposition on engine-relevant TBC-coated surfaces was investigated experimentally and computationally. Modern commercial turbofan high-pressure turbine operating conditions were simulated by subjecting test coupons to a 1600 K, 200 m/s impinging jet laden with 0–10 μm ARD particles, while the TBC surface temperature was regulated through the use of variable backside cooling. A computational model of these experiments was created, and the ability of the combined OSU and molten deposition models to predict deposition was compared to the experimental results.

The testing of CMAS deposition onto TBC coupons at various surface temperatures provides valuable information to guide the development of deposition models suitable for high temperatures. A clear dependence on surface temperature is observed in both the rate of accumulation and morphology of CMAS deposits. As the surface temperature nears the CMAS melting temperature, the impinging molten particles deposit evenly on the surface and appear to remain in a near-molten state, thus forming a uniform deposit structure. These deposits were also firmly attached to the TBC surface, although some portion flaked off when the coupon was cooled post-test. Conversely, at lower surface temperatures, the molten particles appear to be quenched and solidify immediately upon deposition and are far more likely to stick to existing deposits than on the TBC. These deposits could be easily removed from the TBC surface post-test. A sticking temperature threshold was observed at approximately 1100 K for these operating conditions.

Finally, it was shown that modeling the cooled wall in CFD allowed for the combined OSU and molten deposition models to predict somewhat lower particle capture. The direct correlation seen between wall temperature and sticking is confirmed; however, the CFD predictions of reduced sticking do not match the magnitude seen in experiments. This discrepancy highlights the need for higher fidelity deposition modeling to be used in these cases. It is postulated that a better representation of the true experimental phenomena through mesh morphing and conjugate heat transfer would allow the computational results to approach the experimental findings.

The authors would like to thank Steve Boona and Henk Coljin at CEMAS at The Ohio State University for their expertise and aid in preparing and evaluating the TBC samples. Thanks are also due to the Ohio Supercomputer Center for their computational resources. This work was partially sponsored by the Office of Naval Research with Dr. Steven Martens as the program manager. The views expressed in the article are those of the authors and do not reflect the official policy or position of the U.S. Government.

There are no conflicts of interest.

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request. The authors attest that all data for this study are included in the paper. Data provided by a third party are listed in Acknowledgment.

D =

pipe diameter

R =

thermal resistance

Z/D =

wall distance over pipe diameter

ɛ =

radiative emissivity

η_c =

capture efficiency

η_i =

impact efficiency

σ =

Stefan–Boltzmann constant