Full Stage Axial Compressor Performance Modeling Incorporating the Effects of Blade Damage Due to Particle Ingestion

In certain environments, solid particles such as sand, dust, ice, and volcanic ash can be ingested by a gas turbine engine in significant quantities. The higher inertia of some of these particles can cause them to deviate from the gas path streamlines and impact the fore-stages of axial fans and compressors [1]. The damage to the blades is caused by direct impingement of large particles as well as the impact and deposition of finer particles due to secondary flows through the blade passage [2]. In the case of aeroengines, particle ingestion into the core occurs during all phases of flight operation, however, it is most critical during takeoff and landing where particle/dust densities are typically higher. Damage modes associated with particle impact with compressor blades and endwalls include pitting, material loss, and curling of blade leading and trailing edges, increased effective tip clearances, and erosive blade surface roughness. The aerodynamic performance of the compressor blades is affected by these geometry changes. This, in turn, affects the efficiency and operating range of the compressor.

A number of research groups have studied these various damage modes in axial compressors. Balan and Tabakoff [3] studied the effect of sand erosion on axial compressor cascade aerodynamics and performance deterioration. They observed particularly severe erosion at the rotor leading edge and on the pressure surface. Tabakoff et al. [4] simulated and investigated the effects of single stage erosion on overall compressor performance. Sallee [5] reported significant in-service damage to various components, including the high pressure compressor for the JT9D engine. Also reported was an assessment of the effect of compressor operation time on rotor blade damage. Ghenaiet et al. [6] assessed the damage to an axial fan due to sand ingestion through a qualitative study of erosion patterns and quantitative measurements of degradation in aerodynamic characteristics. Batcho et al. [7] performed experiments on the Pratt and Whitney TF33 turbofan and J57 turbojet engines to assess particle ingestion damage on the performance of the axial compressors, and determined the increased susceptibility to surge at low power settings.

Several groups have studied compressor damage using computational fluid dynamics (CFD). Li and Sayma [8] assessed the effect of tip curl damage on the stall margin of a transonic axial compressor. They observed that the presence of one damaged blade prevented the stall cells from growing and improved stability. Fedechkin et al. [9] modeled different “nicks” and bending of the first stage rotor blade leading edge in a four-stage axial compressor. They determined limits for each of these damage modes for observable performance degradation at multiple operating conditions. Suzuki and Yamamoto [10] performed two-phase CFD analysis of a single stage axial compressor. The particle trajectories and impaction were used to model erosion depth explicitly used in a subsequent CFD analysis. They reported localized damage patterns and performance deterioration consistent with experimental results.

The research above focuses on ballistically impaired geometry. Other researchers have focused their studies on less severe damage modes including increased surface roughness and fouling/deposition. Suder et al. [11] performed laser anemometry experiments and some attendant CFD analysis on roughened and thickened compressor blading and observed significant performance degradation (>5% reduction in pressure ratio and adiabatic efficiency). Most of these losses were attributed to leading edge thickening/roughening and transition effects. Morini et al. [12] performed CFD simulations of NASA Stage 37 to assess the effect of compressor fouling by imposing different combinations of added thickness and surface roughness to the baseline geometry. They observed that the geometric modifications led to decreased mass flowrate, efficiency, and stagnation pressure ratio. Gbadebo et al. [13] performed experimental and CFD studies on an artificially roughened low-speed compressor stator. Total pressure traverses and surface flow visualization showed a significant roughness induced increase in the size of the hub endwall corner separation, this leading to attendant increased loss, blockage, and deviation. Syverud et al. [14,15] studied the impact of salt deposits on a jet engine compressor and again observed significant deterioration on stage performance. They also quantified the distribution of the salt fouling on the rotor and stator blades within the machine. Walton et al. [16] also quantified the significant leading edge surface roughness of ex-service compressor blades, their research focusing on the metrology of the roughness rather than its performance impact.

In this work, we study several ballistic compressor blade damage morphologies observed in the field. Specifically, we aim to study local aerodynamic and attendant performance implications of representative geometric models of three of these modes—ballistically bent/curved leading edges, spanwise cragged erosion of leading edges, and eroded leading/tailing edges at outer spanwise locations. We employ thermoplastic additive manufacturing (AM) to build a number of baseline undamaged stage 1 rotor blades, and then apply heat and tooling treatments to obtain the damaged geometries which are then optically scanned and incorporated into sublayer resolved Reynolds-averaged Navier–Stokes (RANS) analysis.

The paper is organized as follows: First, we summarize the technical approaches associated with geometry generation, CFD meshing, and CFD modeling. We present results for the well-studied NASA Rotor 37 configuration to ascribe credibility to the modeling we perform for the T700 stage. In axial compressors, damage to the first stage rotor significantly impacts the instantaneous and average flow field in the rotor and downstream passages. Accordingly, we perform a full stage unsteady analysis in order to avoid the significant washing out of local aerodynamic changes encountered in a steady analysis. We present an approach for appropriately accommodating operating point variation for a damaged compressor. Next, we present comparisons of the flow field and performance predictions between the undamaged and three damaged blades. A short summary of how results such as these can be used to predict overall engine performance is included.

The geometry of the first stage compressor rotor was obtained from GE and NAVAIR for the T700–401C helicopter engine [17]. Figure 1 shows several views of the stage 1 rotor blade geometry.

Since the damaged blades from the T700 experimental tests are forthcoming, an AM based approach was used to represent damage for the CFD studies. The blades were fabricated with a Creality3D Ender-3 V2 3D printer using polylactic acid plastic which is easily amenable to local mechanical and thermal deformation without compromising the shape and aerodynamic design of the blade. The damaged plastic blades were then 3D optically scanned to produce point clouds. The point clouds were then processed in solidworks [18] to produce 3D CAD models to be used for simulation.

Three damage modes were chosen for this study, designated here as, Cragged, Eroded, and Curled. These damage modes have been observed by many workers including Walton et al. [16], Aust and Pons [19], and Zuniga and Osvaldo [20]. Cragged and Eroded damages arise from the surface abrasion/wear of the blades over long exposure times. Curled tip damage arises due to impact from singular foreign large scale object ingestion. These and other damage modes are explored in Aust and Pons [19] and the definitions, descriptions, and photography in that reference were used to adapt geometries studied herein.

Figure 2 shows a compressor blade with representative cragged leading edge damage, and the corresponding 3D-printed blade used in this study. The blade leading edge has notches of depth 1 mm each at 20%, 35%, 46%, 65%, and 84% spans. The notches were made and then mechanically smoothed.

Figure 3 shows a blade with leading and trailing edge chord material loss, and the corresponding 3D-printed blade. At the leading edge, there is blade material loss from 75% span to 25% chord length at 100% span. At the trailing edge, there is blade material loss from 70% span to 30% chord length at 100% span.

Figure 4 shows a blade with a curled tip, and the corresponding 3D-printed blade. The blade tip is bent 45 deg toward the suction side along a line connecting the leading edge at 75% span to 10% percent chord at the tip. Further details of the 3D-printed damaged blades are available in the first author’s forthcoming Ph.D. thesis [21].

A computational domain with four rotor blades and six stator blades is considered for each case, and is shown in Fig. 5. This ratio corresponds very closely to the integer blade-vane-count in the T700 first stage. We use 4:6 rather than 2:3 to enable analysis of damage scenarios where all blades do not exhibit the same damage. In this work, we use three undamaged and 1 damaged blade for each four-blade annulus section.

The mesher in star-ccm+ (v2021.1) [22] was used to create a sublayer resolved unstructured polyhedral mesh with prism layers at the blade surfaces and endwalls. A structured grid built in Pointwise [23] was used to extend the outlet domain one axial chord downstream of the stator trailing edge plane. This is done to reduce overall mesh count and to facilitate accurate spanwise averages at the stage exit. The tip clearance gap was explicitly resolved using 31 prism layers. Views of the mesh used are shown in Figs. 6 and 7.

The RANS simulations were performed in star-ccm+ (v2021.1) [22]. All cases were run time accurate, using the available coupled solver, with second-order space and time discretization. Perfect gas air was the working fluid. The Menter-k–ω SST turbulence model [24] was used. An exactly conservative sliding interface was defined between the rotor and stator meshes. The model was run in the absolute frame-of-reference, with rotor mesh motion. A physical time-step of 1.37 × 10⁻⁶ s was used, corresponding to 50 time-steps per rotor passing period. Each simulation was run for 4000 rotor passing periods to achieve near statistical stationarity for the four rotor blades (1 damaged) : six stator vane configurations studied.

Rotationally periodic boundaries were used. At the inlet, radial profiles of the stagnation pressure and stagnation temperature were specified based on information provided by GE. Inlet turbulence intensity and length scale were specified to accommodate expected test conditions. At the outlet, hub static pressure was specified, and a simplified radial equilibrium pressure distribution imposed. The blade and endwall surfaces were defined as adiabatic non-slip walls.

A grid independence study was performed on a single undamaged rotor blade passage to determine an appropriate mesh for the CFD analyses. A 3.50 × 10⁶ cell mesh and a finer 7.03 × 10⁶ cell mesh were run. Predicted stagnation pressure ratio, stagnation temperature ratio, and adiabatic efficiency varied by less than 0.1% between the two meshes. Accordingly, we have used the meshing parameters of the 3.50 × 10⁶ cell mesh for all damaged and undamaged rotors blade and stator passages in this paper. This led to full stage cell counts of approximately 3.4 × 10⁷ cells for the 4:6 rotor:stator stage simulations presented below.

Damaged blading implies off-design compressor operation. In order to accommodate real-world off-design operational conditions (i.e., pilot control), several considerations were made regarding the target conditions for the simulations with the damaged rotor blades. The propulsion plant output power needs to be maintained for adequate aircraft performance. So, in the T700 turboshaft application, the mass flowrate to, and the shaft speed of, the power turbine is maintained. Hence, for the stage with damaged blades, the target mass flowrate needs to match that with the undamaged rotor operating on-design. Therefore, compressor shaft speed must be increased. In addition, the stagnation enthalpy rises across the entire six-stage compressor (Δh_0,1−6) should also match that of the undamaged compressor.

Additionally, as per meanline velocity triangles for these rotors, increasing the compressor shaft speed Ω leads to a change in the tangential component of the absolute velocity (v_θ) at constant mass flowrate (v_z remains the same) as shown in Fig. 8.

In summary, for the damaged stage simulations, the rotation speed and the outlet hub static pressure are adjusted, during iteration, such that the converged mass flowrate matches that of the on-design mass flowrate, and the stagnation enthalpy rise for the damaged stage satisfies Eq. (4).

Experimental data from T700 field tests are forthcoming, therefore the CFD approach was validated against the well-studied NASA Rotor 37 transonic axial compressor. Rotor 37 was designed and tested by Reid and Moore [25], and retested by Suder [26]. The same meshing and simulation approaches presented above were used to analyze the rotor (here, single blade row, steady-state, relative frame-of-reference simulation) at a given experimental operating point near peak efficiency, i.e., at 98% of the experimental choked mass flowrate. Here, results are compared to the experimental measurements of Suder [26] and the CFD results of Ameri [27] and Bruna and Turner [28]. Table 1 shows a comparison of the overall performance parameters. Stagnation pressure ratio (SPR), stagnation temperature ratio (STR), and adiabatic efficiency predictions are in very close agreement with the experimental data and other CFD results. Figure 10 shows predicted contours of instantaneous relative Mach number at 70% span, illustrating the fairly complicated shock system, attendant boundary layer interaction, and wake field in this flow.

Figures 11 and 12 show the spanwise distribution of mass weighted STR and enthalpy weighted SPR at the outlet. Again, good agreement is observed between the present results and the experiment and previous CFD analyses. Discrepancies between the CFD models and experiments are likely associated with the outer wall thermal boundary condition [28], hub leakage [29], and turbulence modeling [30]. Further discussion of the Rotor 37 analysis is beyond the scope of this paper. These results justify that the meshing and modeling methods used are suitable for the parametric study of the T700 stage carried out below.

Results for the baseline undamaged T700 first stage are presented first. Like Rotor 37, the first stage rotor is transonic and a complicated compressible wave field arises, as illustrated by the instantaneous relative Mach number contours at 90% span shown in Fig. 13. This and all subsequent instantaneous visualizations are obtained at the same physical time-step (i.e., relative rotor-stator positions). The bow shock propagates upstream on the suction side and intersects the adjacent suction surface aft of midchord as a lambda shock which decelerates the boundary layer there. Expansion waves/acceleration are observed on the pressure side. Downstream of the shock-boundary-layer interaction a weak normal shock appears and at this near casing location separates the pressure side boundary layer. The shock system in the blade passage clearly gives rise to significant boundary layer and wake thickening.

Figure 14 shows a front view of instantaneous relative Mach number contours at (x/c)_hub = 0.27 downstream of the hub leading edge. The spanwise variation of the shock location is evident. A low momentum region at the casing is induced by the impinging shock there. The undamaged rotor has a tip clearance of 0.4% of span, and the tip clearance flow is visualized in Fig. 15, using relative frame streamlines integrated through the time-averaged flow field. Double leakage above the adjacent blade is observed as the tip clearance streamlines migrate across the blade passage. Figure 16 shows a top view of the same streamlines, with a 0.98 span surface contoured by time-averaged relative velocity. In both views, we observe the interaction of the tip vortex with the passage shock, which is seen to redirect and decelerate the vortex. The redirected vortex structure develops as it migrates across the blade passage and impinges at the pressure side of the adjacent blade aft of midchord. These observations are consistent with the Rotor 37 studies of Yamada et al. [31].

We present qualitative and quantitative comparisons of the damaged configurations with the baseline Undamaged case. The Cragged configuration shown in Fig. 2, is considered first. Figure 17 shows instantaneous relative Mach number contours at 90% span. In this figure, and all subsequent blade-to-blade representations, the damaged blade is labeled. Comparing the two contour plots, modest differences are observed, and this is somewhat expected since damage to the rotor blade is confined to the leading edge along the span. Slight differences in shock and boundary layer fields are observed, particularly in the passages on either side of the damaged blade. Downstream of the bow shock on the suction side, we observe an increased acceleration of the flow compared to the undamaged case. Although relative Mach numbers are supersonic, the axial Mach number does not exceed one, so upstream influence of the Cragged passage flow arises, leading to off-design physics in all four of the blade passages. The wake downstream of the cragged blade is thinner and dissipates more quickly at this location.

Figure 18 shows instantaneous relative Mach number contours at 90% span for the Eroded damage case. Here, we can clearly observe the impact of chord loss at the leading and trailing edges, and the resultant modifications to the flow field. The leading edge and its attendant bow shock are further downstream, and a richer shock-expansion field arises. We see a stronger near-normal shock in the suction side passage adjacent to the damaged blade. There is significant difference in the flow physics in all the blade passages. Also, the shock emanating from the blade’s pressure surface, is significantly stronger than in the Undamaged case, and impinges on the wake of the adjacent blade, further downstream than in the Undamaged case. The wake behind the Eroded blade is small and quickly dissipates at this outer span location.

Figure 19 shows instantaneous relative Mach number contours at 90% span for the Curled damage case. The bow shock is much stronger since the leading edge curl acts as a much larger leading edge. A reverse flow region arises immediately downstream of the curl feature. The flow field within the damaged blade pressure side passage is more complex, with multiple shock expansion interaction features arising. On the suction side, much more expansion is observed. The wakes of all four blade passages are significantly impacted.

In Figs. 20 and 21, instantaneous relative stagnation pressure contours are plotted at (x/c)_hub = 0.66 downstream of the hub leading edge for the damaged and undamaged cases. In the undamaged blade passages, a high loss region is observed at the casing-suction surface corner associated with the shock-tip clearance vortex physics discussed above (Fig. 15). A number of additional observations apply. First, a low p_0,rel region arises on the suction surface near mid-span for all passages/cases. This high loss region is due to boundary layer interaction with the adjacent blade’s bow shock. The size and shape of this loss region are significantly impacted by all three of the damaged blade geometries. The curled tip and eroded blades in particular give rise to significantly larger regions of high loss adjacent to the blade surfaces in passages on either side of the damage.

Figures 22 and 23 show the tip clearance flow for the Eroded and Curled cases visualized with the same viewing angle and seeding locations as Fig. 15. Comparing the three streamline plots, they are generally similar, however some differences can be observed. In the Eroded case, the tip vortex remains more closely confined to the blade, not extending all the way to the pressure side of the adjacent blade. Unlike the Undamaged case, double leakage over the adjacent blade is not observed. In conjunction with Fig. 20, we see evidence of the tip vortex developing along the shroud creating a broad region of low momentum fluid. In the Curled case, we observe a significantly larger tip vortex, with migration across the blade passage. In conjunction with Fig. 21, we see the attendant regions of low momentum fluid near the shroud, closer to the pressure side of the adjacent blade. In this case, we observe minimal double leakage over the adjacent blade. The streamlines for the Cragged case are not included here, since the near-tip geometry is similar to the undamaged blade, and the streamlines are thereby quite similar.

Stage performance parameters are summarized in Table 2. These are defined based on computational domain outlet and inlet boundary values. These parameters have been normalized with the results for the Undamaged case. Specifically, enthalpy averaged SPR, mass weighted STR, and adiabatic efficiency (based on mass weighted enthalpies) are tabulated. As discussed in Sec. 2.4, specifying appropriate operating conditions for the damaged runs involves specifying a higher shaft speed and lower rotor 1 stagnation enthalpy rise, Δh_0,1, and these are listed as well. All three damaged rotors exhibit significant reductions in the stagnation pressure and temperatures ratios, and adiabatic efficiency. Consistent with the observations above, the cragged rotor exhibits modest performance deterioration, followed by more significant reductions for the curled blade, and yet more severe impact for the eroded damage.

Figure 24 shows the spanwise distribution of normalized adiabatic efficiency at an axial plane 0.34 stator axial chord downstream of the stator trailing edge plane. All profile data have been normalized using the mid-span values of the Undamaged case. For all cases, the normalized adiabatic efficiency decreases across the span, from approximately 1.05 near the hub to 0.9 at 80% span, then dropping precipitously to near 0.6 at the casing, this decrease arising due to increasing M_rel and shock strength with radius, tip clearance flow, and relative casing rotation.

Figures 25 and 26 show the spanwise distributions of mass-weighted STR and enthalpy-weighted SPR, also at an axial plane 0.34 stator axial chord downstream of the stator trailing edge plane. Figure 27 shows the spanwise profile of absolute flow angle at the rotor-stator interface. Flow angles are also referenced to the mid-span Undamaged case. A number of observations are forthcoming. As shown in Fig. 25, all four cases exhibit decreased stage work and rotor turning with span across the inner 80% span, a general characteristic of the stage design. All three of the damaged blades give rise to significant reductions in STR and attendant rotor under-turning across most of the span. This redistributive unloading is observed for all stages despite the damage itself being restricted to outer spans for the curled and eroded blades. As shown in Fig. 26, the stagnation pressure rise decreases correspondingly with the increasing severity of the damage morphology.

The damaged stage performance profiles show generally consistent spanwise trends with the undamaged stage, since the qualitative physics of the damaged passage flow field remain largely the same (boundary/operating conditions, wave fields, secondary flows), and only one blade of every four is damaged in this study. However, there is less work done and increased losses across the entire span for all three damage modes. Damage severity can again be rank ordered from Cragged, which shows only modest deterioration of the four parameters considered, across the span, followed by Curled and Eroded, which exhibit highly compromised performance deterioration.

In Sec. 2.4, a method was presented for setting appropriate operating and boundary conditions for a damaged compressor CFD analysis, i.e., as a pilot would operate the engine under these circumstances. Maintaining full power with significantly reduced stage 1 compressor efficiency of course detrimentally impacts overall engine specific fuel consumption (SFC). Estimating this powerplant SFC reduction is of importance to aircraft operators in that it impacts mission duration, fuel costs, and off-platform refurbishment planning.

In Fig. 29, predicted values of SFC, normalized by the Undamaged model value, are plotted for the four configurations. As the nature of the damage becomes more severe, the normalized SFC increases as expected, to a maximum of just over a 1.4% increase for the Eroded damaged stage. This is attended by an overall reduction in CPR (also normalized) of 6.8% as plotted in Fig. 30.

CFD studies were performed to model the effects of ballistic airfoil damage on axial compressor blading. Physically representative geometric models of three damage modes were constructed and analyzed using a RANS approach which was validated here against Rotor 37. A scheme to implement boundary and operating conditions consistent with damaged compressor operation was presented. Full stage unsteady calculations for the three damaged and one undamaged configuration were studied in terms of compressible wave field and secondary/tip flows, spanwise performance parameter distributions, and efficiency. A simple scheme to estimate the effect of rotor damage on engine SFC was presented. The results enabled the rank ordering of the various damage modes in terms of expected impact on compressor and engine performance. The present study invoked three representative damage morphology geometries, and a particular damaged blade count (every fourth blade). The validated best-estimate CFD methods pursued here are likely to provide good engineering estimates for similar damage morphologies.

This material is based upon work supported by the Office of Naval Research under grant N00014-19-1-2232, with Dr. Steven Martens as Technical Monitor.

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.

r =

radius

$m˙$ =

mass flowrate

h₀ =

specific stagnation enthalpy

p_0,k =

stagnation pressure at inlet of rotor stage k

v_z =

absolute velocity in axial direction

v_θ =

absolute velocity in tangential direction

T_0,k =

stagnation temperature at stage k inlet

$m˙c$ =

mass flow rate at choke

crag =

cragged rotor 1 stage

curl =

curled rotor 1 stage

erod =

eroded rotor 1 stage

und =

undamaged rotor 1 stage

Δh_0,k =

stagnation enthalpy rise for stage k

Δh_0,1,tgt =

target Δh_0,k for stage 1

Δv_θ,k =

flow turning for rotor stage k

CPR =

compressor pressure ratio

SFC =

specific fuel consumption

SPR =

stagnation pressure ratio

STR =

stagnation temperature ratio

α =

absolute flow angle

η_tt =

adiabatic efficiency

Ω =

compressor shaft speed

a =

station 2a at outlet of stage 1

b =

station 2b at outlet of stage 5

f =

final/off-design condition

i =

initial/on-design condition

mag =

magnitude

mid =

mid-span value

rel =

relative frame-of-reference

SC =

centrifugal compressor stage

S1 =

axial compressor stage 1

S2 − 5 =

axial compressor stages 2–5

und =

undamaged rotor 1 stage

2 =

station 2 at compressor inlet

3 =

station 3 at compressor outlet