The Effect of Gas and Surface Temperature on Cold-Side and Hot-Side Turbine Deposition

The subject of deposition in gas turbine hot sections has seen an explosion of interest in the last 30 years. Growth markets for commercial aviation (Asia, Africa, and the Middle East) are known to be regions with elevated micron-sized particulate, both naturally occurring and man-made. This increased traffic in regions of heightened airborne particulate levels has resulted in the loss of engine performance, increased maintenance intervals, and accelerated deterioration of engine components. At the same time, two key technological advances have combined to render modern gas turbines more susceptible to this increased particulate load. First, the pursuit of higher performance and efficiency has led to hot gas temperatures in modern engines that exceed the melting temperatures of nearly all ingested particulate. Second, higher fidelity computational tools have significantly shortened the design cycle of modern engines while at the same time allowing for more complex designs that are operating closer to their peak efficiency. In some cases, this has resulted in products that have a reduced ability to compensate for erosion, deposition, and wear.

For the operator, the concerns are significant in terms of operating cost and safety. Ingested airborne particulate reduces compressor efficiency through erosion, but arguably the more critical concern is found in the engine's hot section. Fine particulate can restrict or clog cooling passages such that critical parts exceed their red-line temperature. Molten particulate in the main gas path can build-up in the nozzle guide vane (NGV) throat, thus pushing the engine toward the surge line by restricting massflow through the gas generator core. In addition, combustor fuel nozzles can become irrecoverably clogged with deposits [1]. If engines could be designed to be more robust, this would surely be a welcome boon to operators. Of course, designing “dust-tolerant” engines requires a sufficiently mature understanding of the many physical processes associated with hot section deposition and degradation. The list of deposition processes and the necessary modeling tools include:

• Accurate steady and unsteady flow solutions for the continuous medium.

• Lagrangian tracking models for the discrete phase are suitable for use with non-spherical particles [2].

• Models for the turbulent dispersion of particles (discrete (DRW) and continuous random walk (CRW)) for Reynolds-averaged Navier–Stokes (RANS) or fully time-accurate simulations for large Eddy simulation (LES) [3].

• Modeling of mid-air particle collisions resulting in the agglomeration and/or dispersal of clumps.

• Impact/rebound models require accurate mechanical and thermal properties for micron-sized particles.

• Erosion of existing surface deposits by large particle impacts.

• Particle adhesion/cohesion.

• Particle break-up on impact

• Particle reentrainment by fluid shear.

“Effective” mechanical and thermal properties of non-homogeneous deposits formed by collections of cohesive (or molten) particles [2].

Progress with understanding each of these particle-related processes is hampered by significant composition dependencies for common airborne dusts. Rarely is the same dust composition ingested at the same engine operating point and with matched dust ingestion history. Yet, it has been shown that trace levels of critical elements/minerals (in the ingested dust or already deposited on the component surface) can dramatically change the deposition tendencies of heterogeneous dust mixtures [4]. Moreover, recent evidence suggests that airborne particles are not chemically inert as they pass through the engine; instead recombine into new substances as they deposit [5].

Having enumerated the complexity and significance of the subject at hand, the authors propose to explore one of the critical variables that influence deposition in gas turbine hot sections, namely temperature. This study comprises the effect of varying both fluid (T_f) and surface (T_s) temperature independently. A brief review of the significant body of work that has already been published on this topic is presented. Key findings, trends, and critical gaps in understanding are discussed. Furthermore, newly acquired data are provided for both internal cooling passage conditions (T_f < T_s) and hot gas impingement on cooled surface conditions (T_f > T_s). A companion computational fluid dynamics (CFD) study of both impinging jet geometries elucidates current gaps in deposition models relative to temperature sensitivity.

Hot gas path deposition studies can generally be divided into three categories: full-engine tests, engine component (cascade) testing, and heated impinging jet studies. Arguably the most heavily referenced full-engine deposition test campaign is that conducted by Dunn and coworkers at Calspan Corporation in the 1980s and 1990s [1]. Dunn reported results from six engines with a wide range of dusts from volcanic ash, to clay, to quartz. The average particle size was 37 µm, but measurements indicated mass mean diameters (MMD) under 10 µm by the time the particles reached the hot section (due to particle breakup in the compressor). A key finding of this test campaign relevant to our current topic is that a minimum temperature threshold must be reached to produce stubborn (glassified) deposits in the hot section. This threshold was nominally set at 1280 K, but was found to be dependent on particulate composition and surface temperature. In particular, calcium was singled out as a “glassifier” that is capable of producing a noxious melt in small concentrations, even if the melting temperatures of the majority of other dust constituents are much higher. A more recent example of full-engine dust testing was published by the US Army Research Lab using rotorcraft turboshaft engines [6]. Images provided by the authors clearly show extensive deposit buildup on the NGV from a “mission relevant” dose of AFRL03 (Air Force Research Laboratory) sand. Though specific operating conditions are not provided, the authors indicate that the turbine inlet temperature was at least 1570 K, a good 150 K above the expected AFRL03 sand melting point.

The “cascade” category of deposition testing is generally focused on replicating the physical geometry of the NGV flowpath at a relevant inlet Mach number (0.1–0.2) and hot gas path temperature (>1350 K). Notably, these component-level tests are generally conducted at approximately 2 atm absolute pressure, significantly lower than an operating engine. This is important in terms of matching particle trajectories since the higher density gas at elevated absolute pressure results in a lower effective Stokes number and less ballistic particle trajectories as documented in Ref. [7]. Over the last decade, the primary author in conjunction with earlier students published a dozen or more articles documenting various cascade deposition tests with and without cooling [8]. Tests were conducted using coal ash and Arizona road dust (ARD). Key takeaways from this cascade testing relevant to the focus of this study include the reaffirmation of an abrupt gas temperature threshold for deposition to occur. For the coal ashes, this could occur as low as 1250 K, but for the more commonly used ARD the threshold was established around 1350 K [9,10].

A particularly notable cascade test campaign was completed using a nozzle geometry with a single spanwise cooling slot at a 20% wetted distance on the vane pressure surface [11]. The NGV inlet temperature was 1353 K and the dust was 6 µm MMD sub-bituminous coal fly ash. The authors found that by delivering a coolant massflow equivalent to 1.3% of the vane passage massflow, deposition was reduced by a factor of three. Significantly, the reduction in deposition extended not just downstream but also well upstream of the slot. The 30 deg slanted slot passage reduced surface temperatures by greater than 100 K upstream of the coolant injection point, and nearly eliminated deposits upstream of the slot to 10% wetted distance.

Several decades earlier, Wenglarz and Fox conducted a similar component-level deposition study using an array of staggered NGV leading edge simulators in a hot gas path laced with coal ash (9–15 µm MMD) [12]. They found that reducing the hot gas path from 1370 K to 1250 K (at constant 1170 K surface temperature) produced a two order of magnitude drop in the deposition. Furthermore, at a constant 1370 K flow condition, cooling the NGV by 200 K (from 1370 K to 1170 K) created a factor of 2.5 reductions in deposit mass. Based on only two data points in each variable, the authors concluded that deposition sensitivity to gas temperature far exceeds that to surface temperature.

By far the most common deposition test used to simulate turbine hot gas path conditions is the simple impinging jet flow, with studies dating back nearly three decades [13–19]. Due to the multiplicity of test variables and experimental facilities, it is difficult to make direct comparisons between the various data sets, however, several universal trends emerge. A threshold temperature exists for deposition, after which capture efficiency is generally non-linear with gas temperature. For test sands and higher rank coal ash, this threshold is in the vicinity of 1250–1300 K while for volcanic ash the threshold can be more than 200 K lower [13]. This result is consistent with ash softening and melting temperature data assembled by Song et al. who observed that many volcanic ashes experience phase changes associated with softening and melting 200–300 K earlier than test sand (e.g., ARD) [20]. Most of the impinging jet studies were conducted with an uncooled (i.e., nominally “adiabatic”) target surface. The studies where constant T_f was maintained while dropping T_s through target backside cooling all observed that deposition could ultimately be eliminated with a surface that is adequately cooled [14,15,18,19].

Since the turn of the century, greater focus has been placed on cooling flow blockage due to deposition. Coolant temperatures are typically below the threshold softening/melting temperatures cited earlier, nonetheless strong dependencies of internal cooling passage deposition on temperature persist. This is puzzling since particulate that builds up in internal cooling passages is typically not sintered or glassified. So, what is the explanation for observed temperature dependencies in internal cooling passage deposition? In their comprehensive effusion hole blockage study, Varney et al. evaluated several possibilities including thermophoresis, van der Waals, and electrostatics [21]. Thermophoresis is the result of gas temperature gradients and is an extremely weak force that typically only plays a significant role with particles of order 0.1 µm (or smaller). Van der Waals forces are commonly considered to be the dominant inter-molecular force contributing to particle adhesion in a non-sintered deposit. Remarkably, recent work by Pinon et al. [22] indicates that increased particle temperature results in lower (not higher) van der Waals force. This brings us to electrostatics. Static charges build-up whenever two materials impact each other and exchange valence electrons. Particles passing through a gas turbine have many opportunities to rebound from surfaces and develop a net charge. These charged particles could then accumulate in stagnation or separated flow regions where fluid shear is very low. Crowe and Bons provided analysis indicating that van der Waals and electrostatic forces are nearly of the same order for particles in the 1–10 µm range [4]. Unlike van der Waals, there is a link between temperature and electrostatics since hot dry air is more conducive to building electrostatic forces. Thus, the hot dry environment in a gas turbine could be conducive to greater particle adhesion with temperature through the electrostatic force.

Internal cooling path deposition studies typically involve one of the following canonical cooling architectures: effusion (film cooling) hole array [4,21,23–26], impingement hole array [7,27], or double-walled impingement-effusion arrays [28–31]. The pressure ratio (PR) through the cooling test article is generally held constant while dust is fed through the circuit. As dust accumulates and blocks cooling passages, the massflow rate is adjusted to maintain constant PR. For greater fidelity, the test article can be supplied with an external thermal boundary condition that is elevated relative to the “hot” coolant, just as it would be in a gas turbine. Pressure ratios generally hover around 1.015–1.05, but have been studied as high as 1.7 [24]. Since coolant flow velocity scales with the square root of pressure ratio, higher pressure ratios generally yield fewer deposits. Coolant temperatures are typically in the range from 650 K to 1000 K. Internal deposition and cooling hole blockage at temperatures below particle softening are dominated by adhesive forces that act on the particle surface area. Because of this, adhesive forces only begin to rival inertial (ballistic) forces at small particle diameters and low flowrates. Thus, multiple researchers have found that internal deposition is primarily caused by particles smaller than 3 µm [23,26]. Larger particles can actually remove small particle deposition through the process of deposit erosion [26,30]. Several researchers [24,25,28] reported the opposite trend for particles larger than 30 µm. They observed that the trajectory of these large particles is more dictated by inertia, making it difficult for them to follow flow streamlines. Thus, they are more apt to remain in the cooling circuit and block small coolant holes.

Since the subject of this study is temperature effects on deposition, the experiments reported by Varney et al., Walsh et al., and Whitaker et al. are of particular interest [21,24,30,31]. All three evaluated cooling hole blockage while varying surface temperature (T_s) and/or coolant temperature (T_c) independently. They used the common ARD test sand, however, Walsh et al. employed 0–100 µm sand while the others used a 0–10 µm size distribution. Their test facilities also bore some similarities. Walsh et al. used an effusion hole pattern with 150 deg coolant flow turning and PR = 1.4 through 0.38 mm holes, while Varney et al. studied a similar 150 deg effusion panel with a much lower PR = 1.03 and a larger (0.51 mm) hole. Whitaker et al.'s cooling configuration was double-walled with an array of 0.64 mm impingement holes leading to a more dense array of 0.46 mm effusion holes. The pressure ratio through the double-walled configuration was 1.03, the same as Varney et al. All three facilities employed a large kiln to impose the desired external surface temperature (T_s). Results, in terms of massflow blockage per gram (BPG) of dust delivered from these three studies, are combined in Fig. 1. For ease of comparison, results obtained by varying T_c at constant T_s (open symbols) are plotted alongside results obtained varying T_s at constant T_c (filled symbols). The value of the temperature held constant in each data set is indicated in the plot legend. The only exception is for Whitaker et al. [30] where no attempt was made to control T_s for the variable T_c study (the kiln enclosure was not employed). In this case, the temperature of the impingement and effusion panel increased together with increasing T_c. Given the presence of hot coolant on both sides of the impingement panel, it is presumed that T_s was approximately equal to T_c throughout this particular series.

The data tell an interesting story when plotted together this way. It appears from the Whitaker et al. studies that blockage per gram increases linearly with temperature—T_c and T_s increasing together up to 866 K and then for T_s rising from 920 K to 1270 K at constant T_c = 866 K. Using the same dust but a different cooling configuration, Varney et al.'s variable T_c data (for constant T_s = 1113 K) follow this linear dependency of BPG(T_c) until just above 900 K whereupon the effusion hole blockage rises dramatically. Apparently, some material properties of the dust (softening or adhesion) breached a significant threshold at this condition. One is left to speculate that if the Whitaker et al. [30] study had continued above T_c = 866 K, it too may have departed from the linear trend. Though it is a significant scatter in the data, Varney et al.'s variable T_s results follow the general trend of Whitaker et al. [31] until approximately 1200 K, whereupon a non-linear increase in blockage rate was observed. From these data, it appears that deposit sensitivity to temperature in cooling passages is strongly coupled between T_s and T_c, depending on the absolute temperature relative to a material sticking/softening threshold. Prior to the threshold at 900 K, both parameters produce a gradual rise in deposition, whereas, above this threshold, T_c appears to have greater influence than T_s. The addition of data from Walsh et al. to Fig. 1 introduces two new variables—particle size and pressure ratio. The blockage is nearly an order of magnitude lower with the larger 0–100 µm dust and high-pressure ratio, even at temperatures above the 900 K threshold and with smaller effusion holes. This result is consistent with results from several studies as noted earlier [23,26,30]. Walsh et al.'s blockage data show a slightly greater dependency on T_s than T_c, though the data are limited. The authors commented that the softening temperature for their dust was between 1270 K and 1350 K, thus they suspected that the rise in BPG for T_s > 1300 K was due to material softening.

In addition to these three film cooling deposition studies, there have been several evaluations of particle-laden coolant jet impingement flows with flow temperature, both as single jets and in arrays [7,27,32]. In every case, a significant rise in capture efficiency was observed near 750 K when using ARD of various size distributions. Unfortunately, none of these studies were conducted with a fixed target surface temperature, so the individual sensitivity to coolant and plate temperature could not be ascertained.

Due to fluid property (µ, ρ, a) dependencies on temperature, particle Reynolds number decreases with increasing temperature at constant flow Mach number. As succinctly demonstrated by Lundgreen [33], the corresponding increase in particle drag coefficient reduces the “effective” particle Stokes number making particles less ballistic as they more readily follow flow streamlines. While it is true that accounting for this temperature sensitivity in the continuous phase (Eulerian) flow solution and subsequently in the discrete phase (Lagrangian) particle trajectory calculation affects particle impact angle and velocity, the net result is not significant enough to accurately capture the experimentally observed dependency of deposition growth with temperature [21,32]. Accordingly, modelers have resorted to other means to incorporate temperature dependence in their deposition (sticking) models, including Young's modulus (E) [16], particle viscosity (µ) [34,35], yield stress (σ_y) [36,37], deposit thermal conductivity [2], sticking probability [38], surface free energy (adhesion) [32], and yield and sticking velocities [39]. All of these models rely to some extent on composition dependent constants that must be derived empirically for each new dust. Particle size and shape, and substrate properties (including roughness) may also influence model validation.

The present study contributes additional experimental data to both the “internal” and “external” deposition portfolios. Deposition experiments were conducted using a solo impinging jet with T_f < T_s (impinging “coolant” jet configuration) and T_f > T_s (“hot gas path” deposition configuration). In both configurations, experiments were conducted by varying T_f and T_s independently while holding the other temperature constant. All tests were conducted with ARD, with the fluid temperature for the “coolant” and “hot gas path” test series deliberately set below and above (respectively) the ARD dust softening limits (T_soft). A companion CFD study of both configurations is used to evaluate the effect of temperature on particle trajectories without incorporating any temperature sensitivity in particle or surface properties. In this way, the need for temperature-dependent sticking models is underscored.

Experiments were performed with a heated particle-laden jet impinging normal to a backside-heated plate of higher temperature (T_s > T_c) as shown in Fig. 2. For each 8 min experiment, nominally 1 g of aerosolized, 0–10 µm ARD dust was fed from a conveyor belt into a 3.66 m long, 6.35 mm inner diameter tube. A “declumper” device was employed to disperse clumps that formed during delivery by using a high-speed counter-flow jet of air in the drop-down tube from the dust delivery box. High-speed camera assessment verified this setup to be effective in dispersing 80–95% of clumps. Inside the tube, the two-phase flow was heated to 839 K, 894 K, or 950 K (depending on the experiment) and accelerated to an average velocity of 57 ± 0.3 m/s. After exiting the tube, the air-dust mixture impinged normally upon the center of a Hastelloy X plate, positioned 12.7 mm (two pipe diameters) downstream of the pipe exit. The plate was heated from behind with a methane-oxygen torch at the same time that it was cooled from the front by the impingement flow. Balanced between the heat fluxes of the torch and the flow, a stable surface temperature (1033 K, 1144 K, or 1255 K, depending on the experiment) on the front and center of the plate was verified prior to dust delivery. Each experiment was repeated three times. Capture efficiency, qualitative deposit morphology, and X-ray diffraction (XRD) analyses were performed on the deposits.

The capture efficiency is calculated by dividing the deposit mass on the plate by the dust loaded onto the conveyor belt, minus any dust that is residual on the belt or in the plumbing upstream of the impingement plate. This residual is collected by carefully brushing off the belt and blowing out the facility with high pressure air into a dense filter. The filter's change in mass then becomes the residual mass measurement. Typically, the residual amounted to 0.4 g out of the 1.4 g originally loaded onto the belt. The uncertainty associated with the residual recovery was estimated to be ±0.08 g. Since all other mass measurements were made with a scale having an accuracy of 0.1 mg, the residual uncertainty dominated the uncertainty in capture efficiency. For a typical η_c value of 25%, the uncertainty is ±2%. Uncertainty in coolant temperature and plate temperature is ±4 K and ±8 K respectively.

The manufacturer's reported chemical composition for ARD is shown in Table 1, according to the common practice in the mineralogy of reporting oxide content. Of course, naturally occurring dust from Arizona's Salt River Valley is not composed of pure oxides. Our own XRD analysis showed that our batch of ARD was approximately 40% quartz, 20% feldspar, 7% calcite, and 33% clay minerals, such as kaolinite and montmorillonite. These findings are corroborated by others [40].

Figure 3 provides capture efficiency results for both the constant T_c and T_s studies. A linear fit is added to the average data in each case to allow for trend comparison. Figure 4 includes images of typical deposits from each case. The dashed white circle overlaid on the images indicates the inner diameter of the delivery pipe that is located two diameters away. The images were taken after the deposits had cooled to room temperature, consequently, significant fracturing and spallation of the deposit had already taken place. Care was taken to collect all of these fractured pieces and include them in the capture efficiency data shown in Fig. 3.

As shown in Fig. 3, capture efficiency rises monotonically with temperature in both test series. The linear fit highlights a heightened sensitivity to flow temperature compared to surface temperature, roughly a factor of two in slope was noted (14%/100 K for 839 < T_c < 950 K versus 7%/100 K for 1033 < T_s < 1255 K). The images in Fig. 4 corroborate the trends noted in Fig. 3. The primary effect of increasing T_c is to produce a larger diameter central cone region of the deposit shown in Fig. 4(a). In each case, this central cone is surrounded by a region of isolated radial deposit streaks and ultimately a uniform dusting at the extremities of the deposit. The central cone has a unified structural character such that when the deposit cools, the boundary between the central cone and the region of isolated radial streaks is typically where the deposit fractures. There is however no noticeable alteration in the nature of the deposit, other than size. The same is not true for the variable T_s series. As T_s is increased from 1033 K to 1144 K, the ARD cone becomes taller but the diameter remains the same. The region of isolated radial streaks increases in width. Then at the highest plate temperature (1255 K), ARD becomes molten near the plate surface, and the deposits in the stagnation region spalled off upon cooling, revealing glassified deposits underneath. This underscores a significant difference between deposits that form in cooling passages as compared to deposits that form in the main hot gas path. Airborne particulate that deposits on a cooling passage surface acts as an insulator to subsequent heat transfer. Thus, even if the surface was initially below the dust melting temperature, the growth of deposit thickness will result in a rising metal wall temperature. This could eventually raise the temperature of deposits nearest the wall above the softening and melting temperatures of the dust. At the same time, the temperature of the deposit surface (where subsequent particles are impacting) is being reduced with the thickening of deposits. Thus, though η_c has a greater sensitivity to T_c than T_s, increasing T_s could ultimately be more damaging since deposits near the heated wall may become molten and irreversibly anchor themselves to the metal substrate.

To pursue this temperature dependency further, powder XRD was performed on pulverized ARD deposit samples. With the sole exception of the T_s = 1255 K case, the relative proportion of quartz to feldspar and clay minerals generally stayed consistent for all samples. At the highest surface temperature, the quartz peak in the XRD was reduced in intensity by 25% compared to feldspar and clay peaks. Other indications of a rise in amorphous mineral phases suggest widespread molecular assimilation (melting) and glassification. Some minerals lose their identity altogether compared to the raw dust. Notably, the small initial mass fraction of calcite (7%) completely disappeared even at the lowest T_c and T_s, signaling its decomposition into CaO and CO₂.

The complementary “hot impinging jet” study (with T_f > T_s) was conducted utilizing the high temperature deposition facility (HTDF) at the Ohio State University (OSU). As shown in Fig. 5, the facility consists of a combustion chamber, an injection block, an equilibration tube (EQ), and a target. High temperature flow is generated by burning propane and oxygen-enriched air from a stainless-steel burner in a large refractory combustion chamber. The hot products (nominally 14 g/s) then flow through a converging nozzle into the injection block where temperature at the center of the flow is measured with a type-B thermocouple. Just downstream of the thermocouple, a secondary flow with air from ambient temperature enters the main gas path with a mass flowrate of 1.3 g/s through a 3.2 mm diameter pipe. The secondary flow tube is at a 60 deg angle to the main flow direction and is used to inject aerosolized dust once the desired operating conditions are reached. The particle-laden flow then travels down a 0.61 m long and 1.9 cm inner diameter alumina EQ tube. For all the results presented herein, the mean jet exit velocity is 215 ± 15 m/s. Additional tests (not included) with mean velocities from 150 m/s to 300 m/s did not indicate a consistently significant dependence on velocity. The alumina pipe is weighed before and after the test to calculate the deposit mass on the pipe and a new pipe is used for every test. The particle-laden flow then impinges on the target plate located at two pipe inner diameters from the pipe exit at a 90 deg angle. A type-B thermocouple is placed at the center of the EQ tube exit to measure the exit temperature and then removed just prior to particle injection. The deposited mass on the plate is determined by subtracting the plate mass after and before the test.

By changing the mass flowrates of propane, air, and oxygen, the flow temperature can be adjusted to desired operating conditions. The temperature at the center of the pipe exit is around 200–300 K lower than the upstream temperature due to the secondary injected flow and heat loss in the refractory enclosure. Though not measured explicitly, the temperature of the dust injection fluid was estimated using a conjugate CFD model to be 550 K when it enters the main flowpath. Once the desired conditions are reached and a steady-state is confirmed, dust is injected with a uniform delivery rate; either 1 g over a period of 4 min or 5 g over a period of 5 min. The delivery rate is controlled with a conveyor belt assembly and declumper as with the impingement cooling deposition rig (ICDR) in Fig. 2. Three different size distributions of ARD were used including 0–10 µm, 0–5 µm, and 5–10 µm. The dust was provided by Powder Technology Inc. and the size distributions for the dust batches are shown in Fig. 6.

Two types of targets are used in this study. The first is a cast ceramic square plate that measures 51 mm on each side and is 9.5 mm thick. Given the low thermal conductivity of the ceramic, this could be considered a nominally adiabatic boundary condition. The second type is a 25.4 mm diameter, 1.27 mm thick Hastelloy coupon coated with thermal barrier coated (TBC). The TBC coating was applied by Praxair Surface Technologies and consisted of plasma sprayed nickel based 0.2 mm thick bond coat followed by a 7–8% yttria-stabilized Zirconia with a vertically segmented microstructure (1 mm thick). The TBC coupon is placed in a holder with a ceramic shield surrounding the front face. A type-K thermocouple is fitted into a divot at the backside of the TBC coupon to measure the backside metal temperature where cooling air impinges. The TBC surface temperature just prior to dust injection is measured using an infrared camera that is calibrated using a 1D heat transfer model as described in Clark et al. [19].

Once the dust injection is complete, the temperature is slowly ramped down over several hours to not disturb the accumulated deposit. The target capture efficiency is computed as a ratio of the change in target mass to the mass of dust loaded on the conveyor belt minus the sum of the residue dust mass left in the control box and injection pipe collected after the test and the deposit mass adhering to the pipe wall. For the 25 mm diameter TBC coupons, any deposit mass found on the ceramic shield is not included in the target η_c measurement. This smaller deposit area and the fact that the TBC coupon tests delivered only 1 g versus 5 g of dust for the ceramic target tests result in smaller values of η_c for the TBC series. As with the ICDR tests, the primary source of uncertainty in η_c comes from the residual dust collection. Because the TBC tests only deliver 1 g of dust, the uncertainty in η_c is larger (±2% for η_c = 25%) than for the ceramic target tests (±1% for η_c = 25%). The uncertainty in equilibrium tube exit and TBC surface temperature is estimated at ±10 K and the calculated exit velocity uncertainty is ±3%.

Figure 7 presents the ceramic target HTDF data series alongside legacy data from Taltavull et al. [13] and Ai and Fletcher [16]. These new data were obtained using the ceramic target with 5 g of 0–10, 0–5, or 5–10 µm ARD delivered in 5 min. Based on the new data, the threshold “softening” temperature of ARD appears to fall between 1300 K and 1350 K, consistent with data mentioned earlier [10,37]. The inclusion of two additional data sets underscores the dependency on dust chemistry. The η_c data from Ai and Fletcher with sub-bituminous coal ash particles exhibit a very similar initiation temperature for deposits. Other differences in the facilities (lower velocity of 170 m/s, 45 deg impingement angle, and metal versus ceramic target) all influence the lower levels of η_c thereafter. The Icelandic volcanic ash used by Taltavull et al. is rich in Si and Fe with an estimated 80% amorphous content [13]. The authors measured the glass transition temperature to occur around 950 K with a crystalline melt temperature 400 K higher. The low glass transition threshold of the predominantly amorphous content is no doubt responsible for the significantly higher η_c values at much lower temperatures compared to coal ash and ARD. This is despite the larger particle size (0–100 µm), lower velocity (100 m/s), and metal target. The roughly 300 K offset of the volcanic ash data relative to ARD is in line with the conclusions reported by Song et al. [20] as noted earlier.

A visual inspection of the 0–10ARD deposits on the ceramic target clearly shows signs of molten minerals flowing away from the point of jet impingement to create a raised “crown” of large solidified deposits around the impingement zone. This formation becomes more pronounced as the jet impingement temperature increases as shown in Fig. 8. Deposit evolution during the 5 min injection period shows evidence of molten particle impingement, solid and liquid adhesion, and resolidification. Comparing the hot jet deposit (Fig. 8) to the cold jet deposit of the same dust composition and size distribution (Fig. 4) the effect of temperature and velocity is evident. Even at the lowest jet temperature (Fig. 8(a)) the hot jet deposit has no elevated central impingement cone of smooth deposit. Elongated radial structures dominate the morphology out well beyond the equilibration pipe diameter. At higher temperatures, the 215 m/s impingement velocity creates a high shear situation where the impinging jet is at its hottest. This results in molten deposits that flow to the edges of the impingement zone before solidifying to form the deposit “crown”. Admittedly, the target surface (ceramic versus Hast-X), and wall thermal boundary condition (T_f > T_s versus T_f < T_s) are also different between the ICDR and HTDF facilities, but the dominant parameters are arguably the particle-laden gas temperature relative to the particle softening (sticking) temperature and the high shearing velocity.

To provide greater insight into the relative influence of gas versus surface temperature on hot impinging jet deposition, 25.4 mm diameter TBC coupons were tested at a lower dust loading and slower delivery rate (1 g of dust delivered in 4 min) than the ceramic targets. The jet exit velocity was again maintained in the range of 215 ± 15 m/s with 90 deg impingement. Tests were primarily conducted with 0–10 µm ARD, with select cases repeated using 0–5 and 5–10 µm ARD. Backside impingement cooling allowed the frontside surface temperature to be controlled to a desired set point prior to dust delivery. Once deposition commenced, the coolant flowrate was maintained through the four-minute test.

Figure 9 includes the results from both the variable T_f study at constant T_s and the variable T_s study at constant T_f. The “constant” T_s and T_f values listed in the legend were target values and the actual value could vary from the target by ±15 K depending on the test. Due to the density of data in this presentation, results from the same test series are connected with a solid line for clarity only. Presenting all the data together in this manner confirms what can be discerned visually, namely that the sensitivity of η_c to flow temperature is larger than the sensitivity of η_c to surface temperature (approximately double, 10%/100 K versus 5%/100 K). Remarkably, this is the same conclusion deduced from the cold jet study in Fig. 3 and the legacy blockage per gram data in Fig. 1. Deposition appears to be more sensitive to changes in flow temperature than the surface temperature on both the hot side and cold side in the gas turbine hot section.

Laycock and Fletcher also studied particle deposition with a hot impingement jet using 0–10 µm sub-bituminous coal ash at impingement velocities comparable to the present study (205 m/s) [18]. Their impingement angle was 45 deg and the target substrate was uncoated Inconel. The authors conducted a similar test campaign, varying T_s and T_f independently, and found η_c to have comparable sensitivity to both temperatures. Capture efficiency varied from 6% to 12% over 1520 < T_f < 1670 K (4%/100 K) at constant T_s = 1270 K and from 6.5% to 16% over 1170 < T_s < 1390 K (4.3%/100 K) at constant T_f = 1670 K. As shown in Fig. 7, the coal capture efficiency values are again roughly half those of ARD for comparable temperature ranges. Thus, we again find a significant deposition dependency on dust composition (substrate and impingement angle differences notwithstanding).

The limited set of measurements taken with other than 0–10 µm ARD provides some insight into the role of particle size in high temperature deposition. By selecting 0–5 µm and 5–10 µm ARD distributions, it was expected that one would yield higher η_c than 0–10 µm ARD while the other would have a lower η_c, with an average close to the 0–10 µm ARD result. Surprisingly, both for the ceramic target case (Fig. 7) and the TBC coupon case (Fig. 9), both sand distributions yield lower capture efficiencies than the 0–10 µm ARD result. Could this be evidence of a synergy that occurs when both large and small sizes are present that enhances the net deposition rate above what occurs with just small or just large particles? Using a CFD simulation, Clark et al. showed that particles larger than 4 µm experience a negligible change in temperature in the thermal boundary layer of the impinging HTDF flow [19]. Thus, the smaller size distribution should experience more cooling before impact as well as having a lower impact efficiency due to the smaller Stokes number and greater tendency to follow the flow streamlines. As such, a lower η_c is understandable for the 0–5 µm ARD dust. On the other hand, the 5–10 µm ARD distribution is more ballistic and less affected by cooling in the thermal boundary layer. Thus, it might be expected that these larger particles would raise the η_c above the value obtained for 0–10 µm ARD. And yet it is the 5–10 µm distribution that deposits the least.

Two possible explanations for this conundrum are as follows. First, referring to the size distributions in Fig. 6, it is clear that the three size distributions have peaks at 2.5, 4.5, and 6.5 µm respectively for 0–5, 0–10, and 5–10 µm ARD. Thus, it could be that particles near the 4.5 µm mean of the 0–10 µm ARD distribution have the greatest likelihood to impact (due to a ballistic Stokes number) and stick (due to higher particle temperature) at these flow conditions. Another possibility is that the particle distribution exiting the equilibration tube is not the same size distribution that is originally injected upstream. Perhaps different particle sizes preferentially deposit on the walls of the equilibration tube leaving an altered distribution to impact the target. Since the particle size distribution exiting the equilibration tube was not measured, this postulate cannot be evaluated at this time.

To gain insight into the effect of variable temperature on particle impingement and deposition, computational models were constructed for both jet impingement configurations. In each case, the flow model was divided into two regimes for computational accuracy and efficiency. The equilibrium pipe flow was first solved using a RANS solver (Fluent^TM) with an Reynolds stress model (RSM) turbulence model on a structured grid. This allowed turbulent dispersion to be modeled using a CRW model developed by Lo et al. [3]. The fully developed pipe flow direct numerical simulation (DNS) data of Dreeben and Pope [41] were used to correct the near-wall turbulence statistics of the RSM turbulence model as suggested by Dehbi [42]. For each diameter in the 0–10 µm ARD distribution (Fig. 6), 15,000 particles were injected from a uniformly distributed grid at the upstream pipe inflow (injected at mean flow velocity and temperature). The pipe length modeled was 20 diameters for both configurations. The discrete phase tracking feature in fluent was used to track particle trajectories from inlet to pipe exit. The non-spherical drag law of Haider and Levenspiel [43] with a shape factor of 0.65 was used to calculate the trajectories. Saffman lift force and gravity were also accounted for. Particles rebounded specularly from the pipe wall and no sticking was permitted. A convective wall thermal boundary condition accounted for heat loss and allowed the radial temperature distribution at the tube exit to match experimental measurements.

As particles exit the pipe, the fully developed pipe flow turbulence data [41] are no longer applicable. Therefore, a separate flow domain was created for the impingement region from the pipe exit to the target surface. Exit flow and particle properties from the pipe flow model were used as inlet conditions for the impingement region flow model. The SST k–ω turbulence model was employed to obtain the mean flow solution and turbulent dispersion was switched to the DRW model in Fluent. The CRW turbulent dispersion model is necessary for the long pipe flow to obtain accurate particle distributions at the pipe exit since turbophoresis plays a dominant role there. By contrast, using DRW in a turbulent pipe flow erroneously yields particle distributions that are skewed to the outer radius of the pipe as shown by Lo et al. [3] and others. However, in the short (two pipe diameter) gap between the pipe exit and the target, turbulent diffusion is not significant, so the simpler DRW model is adequate. The pipe inlet flow temperature and mass flowrate are specified in the pipe flow model while the target surface temperature was matched to the experimental conditions in the impingement zone model. The ideal gas assumption was employed throughout. For particles impacting the target surface, the OSU deposition model was employed to determine the rebound trajectory or particle capture (deposition). For the HTDF, a total of 4.6M cells were used while for the ICDR configuration 7M cells were employed. These grid densities are comparable to those arrived at previously through a rigorous grid independence study documented in Refs. [19] and [2] for the HTDF and ICDR configuration respectively.

The goal of this computational study was to quantify the effect of variable flow and plate temperature on particle trajectories, surface impacts, and deposition without including any explicit temperature dependency for particle properties or adhesion. This was done to demonstrate how critical such temperature sensitivities are to modeling deposition at temperatures relevant to gas turbines. As such, the OSU deposition model was employed with a constant surface free energy (adhesion parameter), particle yield stress, and Young's modulus.

Figure 10 contains CFD model predictions of target capture efficiency versus particle diameter for 0–10 µm ARD. Results for the “hot” and “cold” jet impingement configurations are provided. In each case, the target surface temperature (T_s) and flow temperature (T_f) were selected to cover the range of experimental conditions reported in Figs. 3 and 9. Results for both the OSU deposition model and the “all stick” condition (sticking efficiency = 100%) are included since this provides a minimum and maximum bound to the range of possible capture efficiencies from the OSU model. The stark conclusion from these results is the absence of any significant sensitivity to temperature. Indeed, when the data are integrated with their appropriate mass fraction weighting (as per Fig. 6) to yield an aggregate capture efficiency for each case, the results show a negligible effect of temperature (Table 2). The subtle variation is due to slightly altered particle trajectories due to the lower particle Reynolds number with increasing temperature.

For direct comparison, Table 2 also includes the range of experimentally measured η_c found in Figs. 3 and 9. The temperature sensitivity in the experimental data is an order of magnitude larger than is registered by the model. Of course, at this point, various particle and substrate material properties could be sensitized to particle and/or surface temperature in order to better match the trends in the experimental data using property models or empirical trendlines, but such is not the purpose of this study. The point is that accurate temperature sensitive models are absolutely essential to advance the state of the art in deposition prediction for gas turbines. It is hoped that the data shared in this manuscript will aid modelers in creating and validating such models in the future. As a parting observation, it is noted that the OSU model results in Fig. 10 and Table 2 (achieved without any temperature dependency) were supposed to be the lowest possible η_c achievable. Increasing temperature would only increase η_c from that level, through increased adhesive forces, more compliant material properties, or other physical processes. Yet for both configurations, the lowest experimental η_c data point (mean η_c minus range) is lower than the OSU model prediction. This is primarily a result of the lack of mesh adaptation to account for changes in the flow (and particle trajectories) as the deposit grows into the fluid domain. As previously demonstrated by Bowen et al. [32] and others [2,44], adjusting the flow to account for a conical impingement jet deposit can reduce the predicted capture efficiency by a factor of two or more due to reduced impact efficiency and shallower impact angles.

The effect of temperature on the deposition of ARD test dust in gas turbine hot sections has been studied using two experimental facilities. The impingement cooling deposition rig replicates the thermal conditions on the cold side of an impingement cooling array such as what might exist inside a turbine vane or blade or in a combustor liner. Studies were conducted varying coolant and surface temperatures independently. When the flow temperature was varied from 839 K to 950 K at constant surface temperature of 1144 K, dust capture efficiency increased by roughly 14%/100 K. The morphology of the deposit structure was self-similar over this range, increasing monotonically in size. When the target plate surface temperature was varied from 1033 K to 1255 K at constant impinging “coolant” jet temperature of 866 K, deposits increased at a lower rate of approximately 7%/100 K with surface temperature. At the highest surface temperature (1255 K), the insulating layer created by the deposit thickness resulted in metal/deposit interface temperatures in excess of the deposit melting temperature. Glassification of the deposit ensued and cracking/spallation were seen upon cool down.

The high temperature deposition facility was employed for a second study documenting hot-side deposition from a similar particle-laden pipe flow. Deposits on an adiabatic (ceramic) target increased non-linearly with impinging jet temperature, corroborating findings from legacy deposition data in the literature. Experiments using cooled TBC coupons permitted the study of flow versus surface temperature sensitivity. Over the range of temperatures studied, the flow temperature sensitivity was again roughly double that of the surface temperature sensitivity. The TBC coupon tests were conducted with ARD, but with three different size distributions (0–5, 0–10, and 5–10 µm). This allowed for particle size sensitivity to be evaluated. For both target types (ceramic and cooled TBC), the 0–10 µm ARD test dust produced the highest deposition rate.

A companion CFD study of both impingement jet configurations was used to evaluate the effect of temperature on particle trajectories without incorporating any temperature sensitivity in particle or surface properties. Model predictions show a negligible effect on the flow and particle trajectories. As such, capture efficiency predictions remain approximately constant over the wide range of surface and flow temperatures studied experimentally. This result underscores the need for accurate temperature-dependent sticking models to predict deposition at conditions relevant to gas turbines. Also, since cold side deposits occur in regions where particle temperatures typically remain below the particle softening temperature, whereas hot side deposits are most often above T_soft, the authors postulate that appropriately modeling deposition temperature sensitivity will require two different models. The former should sensitize particle adhesion/cohesion forces to temperature while the latter should sensitize particle mechanical properties (yield stress, Young's modulus, viscosity, etc.) to temperature.

The authors would like to acknowledge the critical assistance provided by Blair Peterson, without whom the HTDF test series could not have been completed. The donation of TBC coupons from Dr. Vaishak Viswanathan of Praxair Surface Technologies is also gratefully acknowledged. This work is supported by the Office of Naval Research with program manager Dr. Steven Martens. The views expressed are those of the authors and are not necessarily endorsed by the US Navy. Computational resources for this study were provided by the Ohio Supercomputer Center.

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.

AFRL =

air force research laboratory dust

ARD =

Arizona road dust

BPG =

blockage per gram (%/g)

CRW =

continuous random walk

DRW =

discrete random walk

EQ =

equilibrium tube

HTDF =

high temperature deposition facility

MMD =

mass mean diameter (µm)

NGV =

nozzle guide vane

PR =

pressure ratio

TBC =

thermal barrier coating

XRD =

X-ray diffraction

a =

speed of sound (m/s)

m =

mass of dust (g)

E =

Young's modulus (GPa)

T =

temperature (K)

η_c =

capture efficiency, m_dep/m_del

η_i =

impact efficiency, m_imp/m_del

η_s =

sticking efficiency, m_dep/m_imp

µ =

viscosity (kg/ms)

ρ =

density (kg/m³)

σ_y =

yield stress (MPa)

c =

coolant or capture

f =

fluid

i =

impact

p =

particle

s =

surface or sticking

del =

delivered to the test article

dep =

deposits on the surface

imp =

impacts the surface

soft =

softening