Considerations for Using Additive Manufacturing Technology in Centrifugal Compressor Research

Additive manufacturing technologies have existed for decades, but in recent years, their benefits to the field of turbomachinery have been realized. Increasingly tight machining tolerances and intricate design geometry associated with fuel nozzles, blades, and vanes lead to components that are challenging to manufacture with traditional methods. Aerospace components, and gas turbine engines specifically, consist of many parts that are traditionally difficult to manufacture, because their complex geometry often depends on mathematics governed by fluid flow [1]. As a part becomes more difficult to manufacture, the cost increases significantly. Many members of the aviation and power generation industries incorporate additive manufactured (AM) technologies into their gas turbine designs.

One of the premier examples of AM in aviation engines is the fuel nozzle used in the leading edge aviation propulsion engine which was developed by CFM international. The nozzle design for the AM process transformed 20 individual parts into a single fuel nozzle, while eliminating the need for welding and brazing during assembly. This new design reduced weight by 25% and made the part five times more durable than the previous nozzle [2].

Blade and vane designs in gas turbines feature complex external and internal geometry. Turbine blades, specifically, utilize complex internal channels for heat transfer. Siemens has successfully tested AM turbine blades at full loading conditions in a 13 MW gas turbine for power generation. Using AM technology allowed engineers to design a new internal cooling geometry for the blades. This technology reduced the lead time for developing prototype hardware by 90% [3].

Rolls Royce utilized AM technology during the testing of the Trent XWB-97 engine to produce a fully AM front bearing housing with a diameter of 1.5 m. This part incorporated vanes that guided the flow into the engine's core, and these vanes featured internal geometry that could be used as part of an anti-icing system for use in cold flight conditions. These internal channels were only possible because of the AM technology used to manufacture the bearing housing [4].

NASA sought to design a fully nonmetallic gas turbine engine while utilizing AM technology to accomplish this goal. Ceramic and polymer composite AM materials were evaluated for various engine components. Polymer matrix composites were considered suitable material choices for an acoustic liner, the fan stator, and vane rows in the high pressure compressor. Ceramic matrix composites were considered for the core nozzle, blades, and vanes in the turbine, and the combustor liner. The polymer and ceramic AM components were lighter than the traditional engine components—leading to a weight reduction of ∼14.5%. The ceramics used in the turbine allowed for redesign of the cooling system which was projected to improve the design's specific fuel consumption by ∼2.6%. Together, the design improvements made possible by AM were projected to improve fuel burn by 4.9% [5]. AM technology has allowed gas turbine designers and manufacturers to produce complex geometry and improve designs through features that would not be possible otherwise.

Additive manufactured technology has also influenced turbomachinery research in academia. The larger design space allows researchers to leverage AM techniques to perform experiments that would otherwise be cost prohibitive. Ng and Coull investigated the losses associated with leading edge Kielheads on low-pressure turbine blades. This was accomplished by utilizing AM to print multiple Kielhead designs that were interchangeable on the turbine blade [6]. AM allows researchers to incorporate innovative measurement techniques into their experiments.

One of the more complex turbomachinery components to assemble is the vaned diffuser for the centrifugal compressor. Generally, a vaned diffuser is assembled as a two-piece assembly consisting of a forward and aft plate which are precisely aligned and brazed or bolted together through the vanes. A small difference in tolerance between the components will compound among the vanes and cause the mating to fail, in which case the diffuser must be remade. Because this entire process costs around $40,000 and requires at least 6–8 months to complete for a research diffuser, it limits the ability of a research program to respond to design changes quickly enough to impact engine development.

The flow entering the vaned diffuser is nonuniform and highly unsteady. The diffuser serves to maximize the pressure rise and efficiency of the compressor by reducing the velocity of the flow exiting the impeller; thus, it is an integral part of the centrifugal compressor design [7]. Also, the diffuser design often sets the stall margin of the compressor due to the small throat areas. Computational models struggle to accurately characterize stall margin, and the unsteady nature of the problem requires significant resources to model the instability correctly. However, introducing new design features without fully understanding the effect on the stability of the system is a significant risk to any engine development program. Therefore, it is necessary to develop a method to perform systematic experiments to understand new design features early in the engine development program. This method must be effective in optimizing compressor designs while not adding significant cost or time burdens to the engine development program.

Advances in stereolithography (SL) additive manufacturing have reached a point where this technology can adequately be assessed for its applicability to centrifugal compressor aerodynamics research. A vaned diffuser produced using the stereolithography process reduces the lead time of a single diffuser to less than two weeks (compared to 6–8 months) and the cost to less than $10,000. Successful application of this technology in the research sphere permits exploration of sophisticated and timely compressor design programs where a large variety of diffuser features can be experimentally verified, such as stage matching, incidence, radius ratio, vane count, vane shapes, throat contouring, and 3D contouring. Furthermore, the layer-by-layer manufacturing allows additional instrumentation access in the vaned diffuser.

This investigation sought to use additive manufacturing to produce a vaned diffuser for compressor aerodynamics research. The primary goal was to use this technology to reduce the manufacturing time for a unique research diffuser. Regardless of the material choice, additive manufacturing would result in shorter lead times, since there is no longer a need for the laborious process of aligning and assembling the forward and aft plates of the diffuser. To further the usefulness of AM technology in compressor research, plastic materials were also investigated to find a method of reducing the cost of producing a research diffuser. AM technology allows for better understanding of the flow field since instrumentation can be embedded in the compressor geometry through complex channels that could only be formed by an AM process.

The AM diffuser was specifically designed to be implemented in the Purdue Centrifugal Stage for Aerodynamic Research (CSTAR) facility, Fig. 1. The CSTAR compressor consists of an impeller with 15 main blades and 15 splitter blades, and a diffuser with 35 wedge-style vanes. The diffuser also features locations where interchangeable, instrumented vane cartridges can be installed. A vaneless turn-to-axial channel follows the diffuser to guide the flow into the collector. The compressor geometry is representative of a centrifugal compressor that would form the rear stage of an axicentrifugal compressor series with a design rotational speed of 22,500 RPM. Station numbering and additional details about the baseline CSTAR facility can be found in Ref. [8].

For this investigation, the CSTAR facility was modified to allow the implementation of an AM diffuser. In the baseline CSTAR configuration, the diffuser is a structural member of the assembly. The facility required modifications to maintain allowable stresses within the AM diffuser throughout the range of compressor operating conditions. Two 15-5PH stainless steel components hold the AM vaned diffuser as illustrated in Fig. 2. The two components, called the diffuser inner holder and the diffuser outer holder, maintain the structural integrity of the compressor during assembly and operation. This design ensures the diffuser is only subjected to the unsteady aerodynamic vane loads present during an experiment.

These components were designed to allow for the thermal expansion of the diffuser at the designed operating conditions. To prevent leakage flow around the outside of the diffuser, O-rings were incorporated to seal the flow path. At design conditions, the diffuser expands and uses the O-rings to create a tight seal with the inner and outer holders. This was a necessary modification for implementing an AM diffuser in the CSTAR facility.

As additive manufacturing has continued to grow, many different methods and materials have been developed to accomplish similar goals. For a research diffuser, an AM process that maintains tight tolerances, results in a smooth surface finish, and maintains high strength at compressor operating temperatures was required.

Implementation of an AM diffuser in the CSTAR facility was divided into two phases—a subscale test and the full diffuser test. The subscale test consisted of using AM processes to develop instrumented vane cartridge inserts for the baseline CSTAR diffuser. AM cartridge inserts were first developed using direct metal laser sintering (DMLS) to match the legacy, traditionally machined, vane cartridges that are usually installed in the CSTAR diffuser. The primary purpose of this cartridge was to gain familiarity with sizing instrumentation channels within the diffuser vanes. The vane cartridge, Fig. 3, contains ten static pressure taps and three total temperature Kielheads. The minimum wall thickness near an instrumentation channel is 0.25 mm for the DMLS process [9]. After successful implementation of DMLS cartridges in the research compressor, efforts focused on utilizing a plastic material for a cartridge to save cost and lead time.

The metal cartridge was printed using DMLS primarily to maintain material similarity between the legacy vane cartridges and the AM cartridges. However, it was desired to utilize a plastic material for new vane cartridges to reduce the manufacturing cost. Using a plastic material required evaluation of the material properties at compressor operating temperatures, since, in an AM material, they vary based on build direction and can be strong functions of temperature. Therefore, the plastic material was evaluated through extensive material testing and in the compressor environment through the use of a plastic vane cartridge in the brazed diffuser.

The plastic cartridges are printed using a stereolithography process. The stereolithography AM process is a layer-by-layer cross-linking of a thermoset polymer. Starting with a vat of monomer, additives, and photoinitiator, a UV laser provides an energy catalyst to produce a free-radical and initiate polymerization. The polymerization occurs only in the region traced by the laser, and therefore, a solid part emerges from the liquid vat [10]. A three-dimensional part is built up in this manner. This is a similar process to powder bed fusion techniques, such as DMLS, but instead of sintering or melting an aggregate, the laser polymerizes the liquid monomer. Stereolithography produces a smooth printed surface finish and is a cost-effective method for producing large parts with tight tolerances and fine feature resolution, making it a suitable choice for centrifugal compressor hardware. Because stereolithography is the oldest and most mature AM process, a large material bank and technical expertise exist for printing SL parts [11]. Some composite materials, such as DSM Somos^® PerFORM, are stiff and maintain strength at temperatures up to 450 K, which is appropriate for the diffuser of a single stage research centrifugal compressor. Due to the nature of the layer-by-layer manufacturing process, many materials are anisotropic when printed in an AM process. Although basic material properties are published by the producers of many AM materials, an independent investigation of the available stereolithography materials was required. It was necessary to determine whether a material was suitable for the pressure and thermal loads that exist in the CSTAR facility.

The DMLS vane cartridges were manufactured from 17-4 PH stainless steel because it matched the material of the legacy vane cartridges, and the material is readily available in powdered form. No material testing was performed on the DMLS print of 17-4 PH stainless steel.

Accura Bluestone and DSM Somos PerFORM were chosen for prototype production, because they were both SL materials quoted to maintain strength at temperatures over 420 K. Bluestone is an engineered nanocomposite plastic and PerFORM is an engineered thermoset. Both are materials designed for stereolithography printing and for developing wind tunnel prototypes for the aerospace and motorsports industries.

Vane cartridges were printed from both materials to perform visual inspections of instrumentation and rudimentary material tests. Instrumentation channels of 0.25 mm diameter appeared to fully resolve in the PerFORM cartridge [9]. The static pressure ports were fully formed and allowed compressed air to pass through the channel. However, the static pressure ports in the Bluestone cartridge did not resolve on the vane surfaces. The ports were incompletely formed and, therefore, did not pass compressed air. The Kielheads were also not fully formed. Additionally, knowledge of the two plastics' strength and thermal growth at high temperatures was required. Both cartridges were placed in an oven heated to 420 K. While in the oven, a load of approximately 70 N was placed on top of the vanes to determine if there would be significant dimensional changes under constant loading at a high temperature. After 6 h, the length of an average compressor experiment, both cartridges were removed from the oven. Significant bending moments were applied to each cartridge to test for deflection. The Bluestone cartridge fractured, as depicted in Fig. 4, when the moment was applied. This proved that the Bluestone stereolithography material would be too brittle to test in the research compressor. The PerFORM cartridge did not yield under the bending moment, and no significant dimensional changes were noted [9].

PerFORM has been used in other applications that require high printing resolution and performance at high temperatures. The Formula SAE team at the University of Connecticut used PerFORM for a fuel injection component on their race vehicle's engine [12]. PerFORM has also been used by Wehl and Partner Iberica to manufacture molds for a high-temperature injection molding process [13]. Based on the combination of the rudimentary inspection and test and previous applications of this material, PerFORM proved to be a viable material choice for the diffuser design.

The layer-by-layer manufacturing process used in AM technologies results in materials that are anisotropic. Therefore, a detailed set of material tests of the PerFORM material was required to determine properties that could be used for finite element method models. Properties and surface roughness of AM materials are dependent on the build direction of the part. Therefore, test coupons were printed in three different build directions, and each was tested to determine its properties. Figure 5 illustrates the three build directions of the samples, where the thin, black line represents one build layer. The build directions are described as follows:

• BD 1: Layers are built parallel to the build platform along the largest area face,

• BD 2: Layers are built parallel to the build platform along the second largest area face, and

• BD 3: Layers are built parallel to the build platform along the smallest area face.

Test coupons of PerFORM were tested in accordance with the ASTM D790 Flexural Modulus testing standard. This test is a bending test which measures the deflection of a beam under a continuously increasing load [14]. For this test, the load was applied at the center of the sample on the largest surface area face. No high-temperature data were provided by the manufacturer, and therefore, it was necessary to perform the bend test for each build direction over a range of temperatures. The tested temperatures were representative of compressor operating temperatures in the CSTAR facility. The results of this test are listed in Table 1. The material is weakest when the load is applied to BD3, since the load is applied parallel to each build layer. The lowest flexural moduli and ultimate stresses obtained in the bend test for BD3 were used for the design of the AM diffuser, since they represented the most conservative material property values.

Since a diffuser vane is subjected to a cyclic force during a compressor experiment, it was necessary to evaluate PerFORM's performance for high cycle fatigue. The ASTM flexural modulus test provided the data for a steady loading case, but a dynamic mechanical analysis (DMA) was necessary to determine the properties for a high frequency loading. In addition to demonstrating the changes in flexural modulus under cyclical loading, DMA can be used to demonstrate the phase lag between the applied stress and applied strain, which can be used to explore fatigue in polymers. The ability of a material to store energy is indicated by the phase angle (δ), which is the difference in phase between the applied stress and the measured resulting strain. For DMA results, the tangent of the phase angle (tan δ) is determined and is equal to the ratio of the loss modulus to the storage modulus of the material. It was desired to find a material with a high stiffness, which coincides with a low tan δ [9]. Performing DMA tests at a total number of cycles equal to the number of cycles a vane experiences during a compressor test was cost prohibitive. Instead, four material samples were tested at a variety of frequency and temperature combinations: one sample was tested at 296 K for 18 × 10⁶ cycles, one at 380 K for 12 × 10⁶ cycles, one at 441 K for 12 × 10⁶ cycles, and a final sample at 441 K for 18 × 10⁶ cycles. The data from the material testing indicate that the PerFORM material stiffens when subjected to a cyclical load, which is consistent with behavior for degrading polymers. When polymers degrade, the sliding of polymer chains is inhibited which leads to higher stiffness and lower ultimate strength [9]. Figure 6 depicts the behavior of PerFORM when subjected to DMA at temperatures ranging from 296 K to 441 K. The changes in flexural modulus, Figs. 6(a), and tan δ, Fig. 6(b), are shown as functions of the loading cycles. The flexural modulus of PerFORM progressively stiffens as a function of the quantity of loading cycles. The low tan δ for the tested samples indicates the material remains stiff under cyclical loading. The tan δ was only evaluated for the 380 K and 441 K test cases, because the material behavior at elevated temperatures was the desired outcome of the test. The material tests of PerFORM did not indicate significant material degradation when subjected to heat or cyclical loads.

In addition to structural properties, the surface roughness of the flow path components is a quantity that is dependent on build direction. The Bluestone and PerFORM cartridges' average surface roughness (R_a) were measured and compared with machined 17-4 SS, which is the material of the baseline diffuser. The components were placed in three different orientations, and the measurement was made three times for each orientation. The average of those three measurements is reported in Table 2.

The surface roughness for AM materials is a strong function of the build direction, as shown in Table 2. The average roughness for the machined stainless steel is approximately constant in all orientations. However, the average surface roughness of PerFORM is similar to the roughness of the machined stainless steel in two of the measurement orientations. Therefore, the build direction of the diffuser should be oriented to obtain minimal surface roughness on the flow surfaces.

Multiple components were designed to accommodate a plastic flow path while still maintaining structural integrity of the compressor, rapid assembly, and instrumentation access. Two 15-5PH stainless steel components hold the 3D-printed plastic vaned flow path. These two pieces maintain the structural integrity of the compressor during assembly and operation. The diffuser has been specifically designed for printability, ease of assembly, and to balance the thermal stress due to the difference in thermal expansion coefficient between the plastic and stainless steel.

The AM diffuser was printed using a Viper SL machine in high-resolution 0.05 mm layers from the PerFORM stereolithography resin [9]. As the printed layer thickness decreases, the feature resolution of the part increases. The Viper SL offered a small layer thickness and, therefore, provided a fine resolution for printing the diffuser. Due to the machine size restrictions and the part build direction requirements, the AM Diffuser was printed in seven sections as illustrated in Fig. 7. The build orientation in Fig. 8 was selected to balance feature resolution requirements while also eliminating difficult support removal. As a part is built through an AM process, a support structure is required to prevent the part from collapsing especially parts that incorporate cavities or large passages. Therefore, if the diffuser was printed vertically from the hub surface to the shroud surface, it would require extensive support material to prevent the shroud from collapsing between the vanes. These supports would have been difficult to remove from the small vane passages. Also, the flow surfaces where the supports had been attached would not have been smooth, which would adversely affect the diffuser's performance. Printing the diffuser in the vertical orientation shown minimized the need for most of the support material. The SL machine had the highest resolution in the horizontal plane. Therefore, the most critical components of the flow path needed to be printed in that plane. The vertical build direction given in Fig. 8 minimizes the angle between the vanes and the vertical plane. The center vane is printed perpendicular to the build platform, and the remaining four vanes are printed at a maximum angle of 20 deg from the vertical plane. The dashed lines in Fig. 8 demarcate the relative angle between the build platform and the respective diffuser vane. Since the vanes are printed at a moderate angle, there is less chance of the vane deforming during the printing process—a concern that would be present if the vanes were printed at a larger angle from the vertical plane.

Segmenting the diffuser into seven sections was accomplished using a novel lap joint incorporated into the part design. This type of joint, Fig. 9, was necessary to constrain the seven sections and to facilitate insertion of each segment into the diffuser holder assembly.

The sections are fit together in an interlocking annulus to form the full diffuser ring (Fig. 7). The seven segments are inserted into the diffuser holding hardware to form the full AM diffuser assembly. Instrumentation channels from the diffuser vanes are routed to the exterior of the compressor through several egress points in the shroud. The diffuser sections and egress points are modular to allow for different diffuser and instrumentation configurations.

Small gaps between the metal holders and the plastic diffuser were sized to balance the contact necessary to hold the diffuser while minimizing the thermal stress. A thermal analysis was performed to appropriately design these components and ensure that flow-facing steps and gaps were reduced at the design conditions. The diffuser dimensions were offset so it would expand into the correct dimensions at the design temperature. The coefficient of linear thermal expansion for PerFORM is 2–13 times higher than that of stainless steel. This results in much more thermal growth of the AM diffuser than the brazed diffuser, but since the temperature gradients are small, the total dimensional change is still small. Finite-element-analysis (FEA) techniques were performed to generate the thermal offsets for the cold-to-hot transformation.

Two-dimensional axisymmetric and 3D cyclic symmetric analyses were utilized to design the metal and plastic components. These analyses were coupled structurally and thermally with boundary conditions extracted from computational fluid dynamics (CFD) and experimental data. The analyses were performed over a wide range of inlet temperatures and compressor loading conditions to ensure the design of the AM diffuser and supporting hardware was acceptable at a series of design and off-design conditions. Further details of the finite element models can be found in Ref. [9].

Although stereolithography allows the tightest tolerances of the AM processes, dimensional variation is still a challenge when producing parts. Variation can be due to remnants of support structure, uneven cooling of the material that leads to warping, or due to the layer-by-layer manufacturing process. This causes the implementation of AM to be an iterative process. Sections of the AM diffuser were inspected by laser scans and by a coordinate measuring machine (CMM) to compare the printed geometry with the modeled geometry. These inspections were possible for the AM diffuser because of its small, modular design, which allows it to be easily oriented for the laser scanner or CMM probe. Similar measurements may not be possible on a classically manufactured diffuser, because the full annulus would limit accessibility of the probe. The probe would not be able to enter a vane passage from the throat to take measurements, because it would interfere with the adjacent vane passage. Laser scans also would not be able to access the throat regions of vane passages in a brazed diffuser, because of these same spatial constraints.

A laser scan creates a three-dimensional surface of the inspected part, which is then compared to the modeled geometry. Figure 10 demonstrates this comparison of the scanned surface to the modeled geometry. A tolerance of 0.05 mm was applied for the comparison of the scanned surface and the printed segment. The manufacturing tolerance for the brazed CSTAR diffuser was 0.10 mm for similar dimensions. As shaded in Fig. 10, most surfaces' dimensions were within the 0.05 mm tolerance. This represents the majority of the diffuser segment with the exception of the outer radius. The conformance to the 0.05 mm tolerance indicates that the AM process can maintain a tighter tolerance than traditional manufacturing methods for some surfaces. The outer radius varied by as much as 0.76 mm, which is likely due to the part “bowing” during the printing process. Since it was printed in a vertical orientation (Fig. 8), the relatively large, vertical force due to the weight of the part could cause the structure to bend slightly during the printing process and, therefore, cause the discrepancy in outer radius. This difference in outer radius was not significant enough to negatively impact the assembly of the seven diffuser segments. However, it is an indicator of the importance of optimizing the build direction of a printed component.

Laser scans are useful for highlighting general variations of a part's geometry. However, CMM measurements provide a much more precise inspection that can be used to check critical dimensions of a part. CMM measurements were used to measure the throats of a diffuser segment, since it is a critical component of the diffuser. These measurements provide precise coordinates of the features of interest. The coordinates are used to calculate the width and span of the throats. These measurements were necessary to compare throat uniformity to the brazed diffuser.

The diffuser throat areas of the baseline, brazed diffuser were inspected using a precision machined block and feeler gauges. This inspection was performed to determine the variability between vane passages for the two manufacturing methods—stereolithography and brazing. The baseline throat area measurements were compared to the CMM measurements of the AM diffuser. Table 3 lists the range of measurements for the throat width, span at the throat, and the calculated throat areas. These values represent the difference of the maximum and minimum values of the measured vane passages.

These measurements demonstrate that the reduction in cost and lead time by the stereolithography process does not sacrifice uniformity in the critical dimensions of the diffuser. The AM diffuser maintains the same range of uniformity in diffuser throat width as the brazed diffuser. The range of the measured span of the AM diffuser is even smaller than the range for the brazed diffuser. In the build direction presented in Fig. 8, the profile of the span is printed in the horizontal plane, where the SL machine has the highest resolution. In the braze process, the span is set by fixing gauge blocks to the hub of the diffuser and then placing the shroud plate of the diffuser onto the assembly. The entire assembly is then furnace brazed. This process leaves room for error in maintaining constant span across the entire diffuser. The properties of an SL machine can be leveraged to ensure uniformity of critical dimensions, like the span, of a diffuser.

The AM diffuser allows for significant instrumentation access in the hub, shroud, vane suction surface (SS), and vane pressure surface (PS). This AM diffuser was printed in seven sections, and four of the sections contained instrumentation in this configuration. Because the large number of static pressure taps would not fit in a single passage, they are distributed throughout the diffuser in multiple sections, and the locations are collapsed into a single passage in Fig. 11. For this AM diffuser, 20 pressure measurement channels were incorporated into the diffuser segment design. The maximum number of channels that can be incorporated into one segment depends on the thickness of the vane. A thicker diffuser vane allows more space for routing the channels; therefore, more instrumentation could be incorporated into a diffuser with thicker vanes. The total pressure and total temperature Kielheads were printed directly into the vane leading edges and the instrumentation routed through the body of the vane.

The use of an AM process to manufacture the diffuser allows the designer to include instrumentation routing within the design of the test article. Traditional manufacturing processes would not allow for instrumentation in these areas of sensitive geometry. Figure 12 demonstrates the routing of instrumentation channels in one of the diffuser segments. The channels presented in Fig. 12 are routed from the suction surface of the vane, through the vane itself and out of the hub structure of the diffuser. For total temperature measurements, T-type thermocouples were inserted into channels connected to the Kielheads and held into place with an epoxy resin. All static and total pressure ports are connected to pressure tubing using pressure tubulations inserted into the ports on the forward-facing surface of the diffuser shroud structure.

Instrumentation channels can be routed freely throughout the structure of the diffuser because of the design freedom offered by the stereolithography process. However, care must be taken to ensure that minimum wall thicknesses around the channels is maintained. This minimum wall thickness is dictated by the resolution of the particular SL machine used to produce the part. If the minimum wall thickness is not maintained, the structure can collapse, causing the channel to form improperly. For this case, the minimum wall thickness around the channels is 0.25 mm [9]. When wall thicknesses are maintained, channels can be routed creatively to maximize the instrumentation in one vane, as illustrated in Fig. 12.

The instrumentation ports depicted in Fig. 12 measure the static pressures along the suction surface of the vane. Static pressure ports were also printed on the pressure surface of a separate vane within the diffuser, as depicted in Fig. 11. These data provide vane loading information, as displayed in Fig. 13 with the ordinate axis normalized by the chord length (c). This dataset was acquired at 100% corrected speed, during a test when the compressor's TPR and corrected inlet mass flow were within 2% of those quantities for the brazed diffuser [9]. The points between the experimental measurements are cubic spline interpolated. Overall, the shape matches the CFD prediction well, except between 20 and 40% of the suction surface. The diffuser throat is located in this region and the flow is highly unsteady. Gooding et al. [15] concluded that steady CFD computations do not accurately predict the flow in this region, specifically on the suction surface. The trailing edge pressure deviates by less than 1% on both the PS and SS which suggests the steady CFD appropriately predicts the static pressure rise in the diffuser. Yoshinaga et al. [16] also proved that in fair pressure recovery diffusers, the compressible analytic solution closely matches experimental pressure distribution diagrams. For higher pressure recovery diffusers, the experimental pressure recovery and CFD may not match as closely. The shape of the loading diagram becomes closer to the prediction as loading increases, which may indicate the CFD design point is not matched correctly. Furthermore, the reverse loading at the low loading condition could indicate a negative incidence. The reverse loading decreases as the mass flow through the diffuser is increased, up to a point near the lower operating limit of the compressor.

Experimental vane loading diagrams of this nature are only possible, in a diffuser of this size, through utilization of AM technology. Locating static pressure taps along the surface of a vane would not be physically possibly via traditional manufacturing methods. The flexibility of AM processes allows designers to route instrumentation through the structure of a diffuser and, therefore, reach portions of geometry not previously explored. It is a powerful tool for design, as it elucidates the diffusion at several loading conditions.

The Purdue CSTAR facility was modified to accept a fully 3D-printed vaned diffuser flow path. The flow path is printed from PerFORM plastic using a stereolithography process capable of producing features and surface finishes comparable to a milled diffuser. The stereolithography additive manufacturing method allows for enhanced instrumentation access to the hub, shroud, pressure surface, and suction surface and embedded leading edge Kielheads. The AM diffuser was inspected with a 3D laser scan and CMM, and the results indicated that the stereolithography process produced a diffuser that was as dimensionally uniform as a brazed diffuser. Throughout the experiments, the printed diffuser flow path did not yield during normal operation or exhibit any signs of high cycle fatigue failure. The diffuser vane LE did not degrade, and flow visualization did not expose any signs of leakage. The diffuser performed as expected over a wide range of compressor operating conditions. The modularity and increased access for instrumentation allow for thorough understanding of the extremely sensitive diffuser flow and will result in higher performance vaned diffuser designs. The improved lead time, reduced cost, and increased instrumentation capabilities will make it possible to perform systematic investigations into the complex flow of vaned diffusers for centrifugal compressors.

The authors would like to thank Rolls Royce Corporation for sponsoring this research. Additionally, Mr. Grant Malicoat's assistance with assembly is appreciated.

• Rolls Royce Corporation (Funder ID: 10.13039/501100000767).

AM =

additive manufacturing

c =

chord length

CFD =

computational fluid dynamics

CMM =

coordinate measuring machine

CSTAR =

centrifugal stage for aerodynamic research

DMA =

dynamic mechanical analysis

DMLS =

direct metal laser sintering

LE =

leading edge

P_S =

static pressure

P_0,3 =

total pressure at diffuser inlet

PS =

pressure surface

R_A =

average surface roughness

SL =

stereolithography

SS =

suction surface

TE =

trailing edge

TPR =

total pressure ratio