Development of an Additively Manufactured Stationary Diffusion System for a Research Aeroengine Centrifugal Compressor

In recent years, the development of turbomachinery for the gas turbine and fluid handling industry has increasingly explored the design space for the use of additive manufacturing to provide reductions in weight, increases in efficiency, and improvements in performance. In their 2019 review of 3D printing technologies, Novotny et al. [1] outline suitable 3D printing technologies for turbomachinery applications and specific purposes for such technologies, such as end-product customization, field repair and temporary replacement parts, and rapid prototyping. Major developments in the field of additive manufacturing for turbomachinery prior to 2020 can be found in Ref. [2].

Since Novotny et al.'s technology review, much of the additive manufacturing research for turbomachinery has concerned the use of metal 3D printing for implementation in thermally and mechanically challenging environments. Advancements include the use of powder bed fusion as a viable alternative to traditional manufacturing techniques [3] and the optimization of blade cooling using geometries only achievable using additive manufacturing [4,5].²

Metal additive manufacturing has not found widespread use in turbomachinery at lower operational temperatures because these methods tend to be expensive, slow, and of lower quality than traditionally machined turbomachinery equipment [6]. However, Andrearczyk et al. [7] have used polymer additive manufacturing to experimentally validate computational models for turbomachinery, and interest in the feasibility of additively manufactured (AM) polymer compressors for use in the medical and defense sectors exists [6,8].

One of the sectors listed by Novotny which has not been as heavily explored is the utility of additive manufacturing in rapid prototyping. Advances in materials and additive manufacturing technology have enabled the use of parts with representative performance to the final production diffusion system. The use of stereolithography (SLA) to manufacture a stationary diffusion system with novel, in situ measurement capabilities was first described by Meier et al. [2] The advantages of using additive manufacturing are twofold: the cost and lead time for the parts are significantly reduced, and in situ measurements are enabled which are otherwise impossible to implement in traditionally manufactured parts. Specific metrics for cost, complexity, and lead time for traditionally manufactured and additively manufactured diffusers are listed in Table 1, adapted from Ref. [9]. The cost of all parts is normalized by the cost of the corresponding metal fabricated part.

In the iterative design process, additive manufacturing has found utility for its low cost and lead time relative to traditional manufacturing methods. As the diffusion system design is refined over time, different iterations can be rapidly implemented, which probe the performance of different flowpath features. Furthermore, the in situ pressure measurements are useful to quantify individual component performance and can help to validate computational models in conjunction with overall performance metrics. These advantages help the designer to rapidly optimize their diffusion system design and validate their computational models at a lower cost than what is otherwise possible using traditional methods.

This work details the implementation of an additively manufactured stationary diffusion system on the Centrifugal Stage for Aerodynamic Research (CSTAR) at Purdue University, which is used to investigate the performance of a centrifugal compressor as the last stage in an aeroengine compressor. The additively manufactured parts have enabled the use of in situ steady and unsteady pressure measurements in the diffuser, building from the work of Meier et al. [2] and Adkins-Rieck [10]. The new generation facility has also implemented an additively manufactured turn-to-axial and deswirl system, which features both total and static pressure measurements along with total temperature measurements. Furthermore, data gathered from three design iterations have been used to validate design decisions, inform additional design iterations, and refine computational models. Lessons learned through these iterations have also provided valuable insight into further improvements to the manufacturing and design of additively manufactured parts for future design work.

The CSTAR Gen 2.5 compressor is a low-specific speed centrifugal stage designed for axi-centrifugal compressors. The facility is designed to study the aerodynamics of centrifugal impellers and stationary diffusers as a replacement for the final axial stages of an aeroengine compressor. This is motivated by the trend of smaller engine core sizes to achieve higher bypass ratios. As the passage size decreases toward the rear of the compressor, endwall losses become significant. Thus, the axi-centrifugal configuration is proposed to provide improved efficiency because of its ability to circumvent axial tip leakage flows and produce a significant improvement in total pressure ratio while maintaining a similar footprint to a comparable axial stage configuration.

Figures 1 and 2 show a cutaway of the stage. The impeller operates at a design corrected speed of approximately 21,000 rpm and consists of 15 main blades and 15 splitter blades. In addition, 20 variable inlet guide vanes (VIGVs) are upstream of the impeller, which can be adjusted ±9 deg in increments of 3 deg, to provide swirl to the impeller as would exist if the impeller was swallowing flow from upstream axial stages in the axi-centrifugal compressor arrangement. Downstream of the VIGVs, the hub of the transition duct feeds into the impeller and can be either rotating or stationary. The stage's stationary diffusion system, which consists of the diffuser, turn-to-axial bend, and deswirl, is additively manufactured using Somos PerFORM, a commercially available ceramic-like stereolithography resin. This is the same material which was selected among a field of candidate materials by Meier et al. The baseline diffuser is a conventional wedge diffuser consisting of 37 diffuser vanes and parallel endwalls. The baseline return channel consists of a turn-to-axial bend and a vaned deswirl containing 99 vanes. Downstream of the stationary diffusion system, the flow enters the collector and then exhausts vertically into the facility's exhaust ducting.

The CSTAR Gen 2.5 configuration was designed as a modification to a previous generation of the facility (“Gen 2”). In the prior iteration, the diffusion system was a structural component of the assembly, as opposed to Gen 2.5, where the diffusion system does not support the structure. Thus, hardware changes were required to accommodate the new additively manufactured components into the assembly. An exploded view that shows the Gen 2.5 assembly and the changes from Gen 2 is depicted in Fig. 3. Three new structural components were designed to house the AM parts: the aft mount, aft retainer, and forward mount. The aft mount connects to the metering plate of the compressor (not shown in Fig. 3) and serves as an attachment site for the aft retainer, which both locates and provides rear support to the deswirl system and turn-to-axial bend. The AM diffuser then rests on the aft mount and on the outer diameter of the impeller backface. Finally, the forward retainer fits over the AM diffusion system and attaches to the aft retainer while also providing the attachment points and structural support for the shroud and exhaust collector. The new metal components isolate the AM components from static structural loads from the compressor configuration and ensure only aerodynamic and thermal loading during operation. Because the 17-4 pH stainless steel parts used for the new metal support system have a different coefficient of thermal expansion (CTE) than the AM parts, the metal support structure was designed to allow for the thermal expansion of the AM parts during rig operation at full speed.

In the Gen 2.5 configuration, the diffusion system is split into three components: diffuser vane segments, turn-to-axial segments, and deswirl vane segments. The parts are segmented in a similar manner to that described by Meier et al. [2], with the goal of meeting maximum print volume sizes and ensuring better printability. Figure 4 shows a cross section of the assembly and renderings of the diffusion system components. The diffuser is split into 10 circumferential segments, while the turn-to-axial segments and deswirl segments are split into 11 segments each. Because of the odd number of diffuser vanes, the diffusion annulus is composed of three- and four-vane segments.

The AM parts are manufactured at “cold” geometries which do not necessarily match the “hot” geometry at running speeds. To facilitate mating of the segmented diffuser and deswirl parts, they are designed with overlapping shiplap joints. As the parts heat up due to the compressor operation, the thermal expansion of the AM parts closes the gaps that are designed into the parts at cold conditions—in other words, the parts “grow” into each other as they are heated by the compressor—which creates an airtight seal across the shiplap joints that prevent leakage out of the assembly.

Despite this, the parts are designed and installed with small gaps due to the inevitable thermal expansion of the parts at part speed. Thus, there exist many potential flow leakage paths through the compressor. Therefore, a collection of O-rings is used to seal between the AM components, shown in Fig. 5. At cold conditions, the AM components do not adequately seal against the metal components. However, at part speed, the thermal expansion causes both the shiplap joints and O-rings to seal and prevent flow leakage from the compressor. All O-rings that interface with the AM components use low-durometer silicone rubber to prevent excess stress on the AM components.

The parts have anti-rotation features which not only fix the parts in place but also help to locate the parts on the compressor rig during buildup. The anti-rotation features vary for each component. The diffuser and deswirl segments both have cylindrical slots, which mate with cylindrical pins on the aft mount and aft retainer, respectively. This fixes the circumferential location of both the diffuser and deswirl segments. The turn-to-axial components are designed with extruded tabs which mate to matching slots on the deswirl vanes, which locate and fix the turn-to-axial bend parts. The anti-rotation features are shown in Fig. 6.

The CSTAR Gen 2.5 compressor is heavily equipped with an array of steady pressure and temperature measurements to evaluate the stage performance. Total pressure and temperature measurements are acquired with rakes that span from hub to shroud with either three or four elements. In addition, static pressure and surface temperature measurements are scattered throughout the flow path in areas of interest. Overall, the compressor stage utilizes 239 pressure measurements and 88 temperature measurements. Figure 7 is a schematic of the instrumentation orientation throughout the flowpath.

The AM diffuser is heavily instrumented with pressure measurements on the hub, shroud, suction side, and pressure side of the diffuser vanes. The flexibility of the SLA process allows for internal channels to be printed within the diffuser segments at various points along the flow path. They are routed within the parts from the measurement location to a connection point external to the flow path. There, the parts are fitted with stainless steel tubulations using a two-part epoxy, depicted in Fig. 8. These instrumentation locations would be previously unachievable from a traditional manufacturing standpoint. Furthermore, the diffuser segments have different instrumentation quantities and locations, highlighting the flexibility of the AM components.

The diffuser comprises six different diffuser segments that vary the number of vanes and pressure measurement locations. The diffuser segments have three types of pressure measurements: total pressure rakes, shroud static pressures, and vane static pressures. Figure 9 shows a schematic of the pressure measurement locations in the diffuser collapsed into one passage. The total pressure is measured at the inlet and outlet of the diffuser passage via rakes which are printed into the structure of the diffuser parts. The inlet rakes are at three evenly spaced circumferential locations and contain three Kiel heads each. The inlet rakes are printed directly onto the leading edges (LEs) of the diffuser vane to provide minimal disturbance to the diffuser inlet flow region. The outlet total pressure rakes are spaced at four pitch wise locations in the passage, starting at the trailing edge of the vane. There are eight total pressure rakes that contain four Kiel heads each. To prevent flow path obstruction, only one exit rake is present per diffuser passage.

Static pressure ports are located from the inlet to the exit of the diffuser segment on the shroud side of the flow path. In the inlet region, static pressure is measured at the inlet of the semi-vaneless space at six circumferential locations and the diffuser throat at three circumferential locations. Downstream, static pressure is measured at the diffuser trailing edge and exit in three diffuser segments. Static pressure is measured along the diffuser vane surface at multiple locations at 50% span. The suction-side vane has two locations in the semi-vaneless space, one at the diffuser throat and one in the diffuser passage. On the pressure side of the diffuser vane, there is one tap at the diffuser throat and two downstream in the passage. In total, six diffuser vanes are instrumented with such taps, resulting in three sets of pressure-side and suction-side measurements.

The deswirl segments feature similar pressure measurement locations to the additive diffuser and are shown in Fig. 10. Static pressure measurements are acquired at the inlet and exit of the deswirl vanes. However, the deswirl features hub-side, as well as shroud-side, static pressure measurements. The deswirl also has static pressure measurements along the vane surface at 50% span for a single vane. Total pressure rakes are printed integrally along the span of four deswirl vane leading edges, each containing three Kiel heads. Furthermore, eight rakes at four pitch wise locations measure the total pressure at the exit of the deswirl. Each exit total pressure rake has three elements across its span. Finally, four rakes outfitted with three T-type thermocouple wires each measure the exit total temperature of the machine. For both exit total pressure and total temperature rakes, the four span wise rakes are distributed among different segments to reduce disturbances to the flow field.

Internal channels are routed to egress sites on the aft face of the deswirl segments equipped with stainless steel tubulations. These tubulations are used to attach nylon pressure tubes to the physical parts, which are then routed through the front of the compressor to a data acquisition station. Thermocouple wires are fed into the printed total temperature rakes and fixed in place with a two-part epoxy. Figure 11 shows the tubulations and the thermocouple routing for the total temperature rakes.

The performance of several different stationary diffusion systems has been measured in this facility to facilitate design improvements. To show that the thermal growth of the printed hardware is sufficient to seal all potential leakage flowpaths, data at the lowest speed tested, 70% speed, were compared for different ambient air (the compressor working fluid) temperatures. During a particular test campaign, data were acquired at an ambient temperature of 23 °F (−5 °C) and 55 °F (13 °C). Figure 12 shows the comparison for the speed lines. The data have been normalized by the values at the aerodynamic design point at 100% corrected rotational speed. The difference in total pressure ratio (TPR) at the highest loading point was within 0.9%, and the difference in choked corrected mass flowrate was within 0.04%. The changes in isentropic efficiency were also comparably small, showing high repeatability, even across different ambient temperatures at the speeds where thermal growth would be smallest and the potential for leakage through the joints between segments would be highest [9].

The presence of in situ pressure measurements enables the use of previously unobtainable measurements, pitch wise total pressure measurements at the diffuser and deswirl exit planes, which help to quantify differences in performance among different diffusion system designs. This is enabled using rakes at four separate pitch wise positions around the circumference of the annulus, which allow for a reconstruction of the passage flow field at 16 distinct points at the diffuser and deswirl exit planes. Figure 13 shows contours of the dimensionless diffuser exit plane total pressure loss coefficient $K$ for three different points at 100% corrected speed, defined as the total pressure (⁠$P0$⁠) loss from the diffuser leading edge (station 3) to the diffuser trailing edge (station 5) divided by the dynamic pressure at the diffuser throat (station 4):

Three different points were evaluated: choke, the best efficiency point, and the last stable operating point before compressor surge was observed. The total pressure loss coefficient is determined using total pressure data at the diffuser leading edge and trailing edge, which is only possible because of the integral pressure tubes built into the leading edge diffuser vanes and the total pressure rakes located at the same meridional location at the diffuser trailing edge. The reconstruction of the total pressure field at the diffuser exit plane shows that in addition to higher losses toward the vane suction surface into choke, losses are also stronger near the hub relative to those at the casing. As the compressor is throttled to its best efficiency point, losses at the hub and the suction surface both attenuate, and approaching surg surge, losses are concentrated to the vane pressure surface as the incidence on the diffuser vane increases.

After several iterations of manufacturing diffuser, deswirl, and turn-to-axial bend parts, many manufacturability concerns have arisen which have contributed to the failure of some important instrumentation features of the additively manufactured parts. These failures have ranged from minor, resulting in the loss of individual pressure measurements, to major, resulting in the scrapping of an entire additively manufactured part. These defects are direct results of technological limitations, whether within the stereolithography machine itself or, in certain cases, material limitations of the photopolymer resins.

Because manufacturing defects and part scrapping became an issue during implementation of some builds, an alternative materials study was conducted to vet other possible materials which may be able to provide similar or improved printability compared to Somos PerFORM while retaining its desirable material properties. Surveying commercially available stereolithography resins at the time of publication, desirable material properties generally correlate well with resin viscosity, which is itself a parameter which should be minimized to improve the chances of reducing channel blockages.

Of note is the trend in currently available resins in which as the penetration depth decreases, the viscosity correspondingly increases. This is undesirable for the manufacturability of the resins, as highly viscous uncured resin is difficult, if not impossible, to effectively remove from narrow passages. Future material improvements, then, would include materials which are able to provide smooth surface finish through reductions in penetration depth, providing the smoothest possible surface finish, while simultaneously reducing their viscosity, which is beneficial for postprocessing internal pressure tubes which can be blocked with uncured resin.

The measure of the ability of a polymer to resist deformation at elevated temperatures is measured by its heat deflection temperature (HDT), which is the temperature at which the solidified polymer deflects under a specified load. For materials considered for use in this application, higher heat deflection temperatures are beneficial because they enable the parts to withstand the hot gas temperatures downstream of the impeller without deforming the parts in undesirable ways. Figure 16 shows the variation in heat deflection temperature and viscosity for different candidate resins. Resins labeled as circles have similar or lower CTE than Somos PerFORM, while squares indicate resins with higher CTE. Black triangles indicate materials for which there is no published CTE.

Although many of the resins listed are marketed as “high temperature,” many of their heat deflection temperatures fall below the impeller outlet temperature on hot days of approximately 300 °F. Thus, there is a lack of materials which can withstand running temperatures that exist in the exhaust of this impeller.

Of the materials which are available, the heat deflection temperature correlates positively with increased viscosity. Notably, Accura SL 5530 has been proposed as a candidate resin for this application [10] because of its high heat deflection temperature and low resin viscosity, but it was ultimately rejected after test prints due to its high surface roughness and high coefficient of thermal expansion.

One challenge when using additively manufactured parts is the high thermal expansion of plastics relative to their metal counterparts. Although the thermal expansion plays a crucial role in sealing the individual segments from air leakage, excessive thermal expansion would require significantly under sizing the parts at their cold geometries or inducing high compressive stresses at operating temperatures. Thus, candidate materials must feature a similar or lower coefficient of thermal expansion as the currently used resin, Somos PerFORM. Figure 17 shows the correlation between resin viscosity and coefficient of thermal expansion for available high-temperature resins. Circles denote resins with higher HDT than Somos PerFORM, while squares denote those with lower HDT.

With the exception of the Accura Phoenix resin, the trend for high-temperature resins is that as viscosity increases, the coefficient of thermal expansion decreases. Thus, suitable materials for this application are necessarily restricted to those with high viscosity. From the materials survey, desirable properties for surface finish, heat resistance, and thermal expansion correlate with higher resin viscosity, which becomes a problem for postprocessing small internal passages.

In general, the materials that show the best properties are so-called “ceramic-like” materials, including Accura PEAK, Accura Bluestone, and Somos PerFORM. These same materials also feature higher viscosity than other high-temperature resins.

As material technology improves, additional candidate materials should be considered for their ability to deliver low penetration depth, high heat deflection temperature, and low CTE, while minimizing resin viscosity.

Because of the material limitation of Somos PerFORM, the geometrical requirements of the additively manufactured parts have been adapted to ensure that printability standards are being met by the manufacturer, including minimum wall thicknesses of 0.015 in. (0.381 mm) and minimum channel diameters of 0.040 in. (1.016 mm). These features primarily affect the geometry of leading edge pressure rakes and static vane surface taps where the leading edge thickness is lowest.

The Purdue CSTAR facility was updated to be more amenable to 3D printed stationary diffusion system components. The addition of the turn-to-axial and deswirl additive components expanded the facility's capability to rapidly incorporate engine representative diffusion system designs. Like the previous additive manufacturing configuration, internal channels are seamlessly incorporated within the AM components, strategically positioned along the flow path. As a result, the AM components demonstrate remarkable versatility in terms of the quantities and locations of instrumentation. Specifically, the diffuser comprises six distinct segment designs, each featuring variations in the number of vanes and pressure measurement points. Furthermore, the deswirl system comprises seven component designs, each with individual instrumentation channels and locations. Overall, the AM diffusion system enables a total of 135 internal instrumentation channels, effectively capturing vital pressure and temperature data to evaluate the performance of different stationary diffusion system designs. The added capability of instrumentation, along with the immense reductions in cost and lead time, allow for diffusion system designs to be experimentally evaluated at unprecedented speeds and at lower cost. Lessons learned from the iterative design and build process have informed future work and improvements to be made to additively manufactured systems for use in engine flowpath testing.

While developments in materials technology have enabled the use of photopolymer resins for aerodynamic testing of stationary diffusion systems, there remains much to be desired in terms of favorable material properties. A survey of commercially available stereolithography resins has revealed the tradeoff in high-temperature resins between resin viscosity with properties such as CTE, HDT, and penetration depth. Further improvements in material technologies which would improve manufacturability of parts and fidelity of experimental measurements would include resins with lower viscosity while preserving or improving the listed properties. The limitations imposed by currently available resins define the envelope within which instrumentation features are designed.

The authors would like to thank Rolls-Royce Corporation for sponsoring this research. Additionally, the authors are appreciative of the assembly assistance and advice provided by Mr. Dean Dilley, without whom this research would not be possible.

There are no conflicts of interest.

The authors attest that all data for this study are included in the paper.

h =

blade height

$x$ =

percent chord

$E$ =

incident energy

$K$ =

total pressure loss coefficient

$Cp$ =

static pressure recovery coefficient

$Dp$ =

penetration depth

$Ec$ =

critical energy of polymerization

$Ps$ =

static pressure

$P0$ =

stagnation pressure

$τ$ =

tip clearance height

AM =

additive manufacturing

CTE =

coefficient of thermal expansion

CPS =

centipoise

CSTAR =

Centrifugal Stage for Aerodynamic Research

LE =

leading edge

PBF =

power bed fusion

PS =

pressure surface

SLA =

stereolithography

SS =

suction surface

TE =

trailing edge

VIGV =

variable inlet guide vanes