EVALUTION OF VARIOUS METHODS FOR MANUFACTURING ONE PIECE, SMALL TIP OPENING CENTRIFUGAL

COMPRESSOR IMPELLERS

Scot Laney Akiyoshi Ando

Senior Materials Engineer Welding Engineer
Elliott Group Elliott Group
Jeannette, PA, US Jeannette, PA, US

Derrick Bauer
Manager – Materials Engineering
Elliott Group
Jeannette, PA, US

Dr. Scot Laney is a Senior Materials ABSTRACT

Engineer in the Research and Development
team with Elliott Group, in Jeannette, Closed centrifugal compressor impellers have been
Pennsylvania. He joined Elliott Group in manufactured using several methods through the years. Due to
2007, and has been involved with materials limitations of the materials and machining processes, most of
related R&D projects, failure analysis, as these impellers have been manufactured in what is considered
well as production and aftermarket two piece or three piece methods. Despite the vast amount of
support. Dr. Laney received his BS (2001), experience with traditional construction methods, there is a
MS (2004), and PhD (2007) degrees from drive to move towards one piece construction, where there are
the University of Pittsburgh in Materials Science and no joints and, in theory, lower probability of preexisting
Engineering defects. Typically, the impellers that are being offered as one
piece are those with relatively large openings where 5-axis
milling machines can be utilized. This paper investigates
Akiyoshi Ando is Welding Engineer with several alternative methods to manufacturing small tip opening
Elliott Company, in Jeannette, impellers as a single piece. The methods discussed include
Pennsylvania. He has 12 years of Electrical Discharge Machining (EDM), investment casting,
experience in welding related with R&D Hot Isostatic Pressed Powder Metal (HIP’d PM), and Direct
projects and failure analysis. Mr. Ando Metal Laser Sintering (DMLS).
received his B.S. degree(2001) and M.S.
degree(2003) from Osaka University. INTRODUCTION
A customer of the authors’ company once stated “show us
your bleeding edge designs, but first show us an experience list
with at least 5 other running installations.” While at first
Derrick Bauer is the Manager of the reaction, the statement is rather humorous, it does speak
Materials Engineering Department at volumes about the turbomachinery industry. In general, the
Elliott Group, in Jeannette, Pennsylvania. industry is very conservative, but there is an element that exists
He joined Elliott Group in 2002, and has that is also always looking towards the future. It’s hard to place
been involved with materials related R&D blame for that when the amount of risk involved in operating
projects, failure analysis, production and turbomachinery and the potential rewards are considered.
aftermarket support, and remaining life Partly as a result of the conservatism, those who were present
assessments. Mr. Bauer received his BS during the infancy of the centrifugal compressor would have no
degree in 2002 and MS degree in 2011 from the University of problem recognizing a modern compressor. It’s a far cry, for
Pittsburgh. example, from comparing an early telephone and today’s
smartphones, since at a high level, not much has changed in the
compressor. If you look under the hood, it’s a different story.
OEMs have done a fair job through the years keeping abreast of
current technologies and have incorporated it into the design
and manufacturing of the equipment, despite the conservatism.
This paper looks briefly at the evolution of the manufacturing

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

methods traditionally used to make closed impellers before
discussing newer, alternative methods for making small
impellers as single pieces.

TRADITIONAL CONSTRUCTION METHODS

Three Piece
Three piece construction, so called due to the hub, blades,
and cover being separate pieces, is most traditional of the three
methods. The majority of the impellers made before the new
millennium were made in this fashion. This method has several
advantages. The multiple pieces and relatively simple designs
of the day allowed for easy forming of blades, machining, and
fabricating. Low alloy steels, such as 4100 and 4300 series
steels, were commonly used and the thin sections allowed them
to be through hardened. The thin sections also allow for a
variety of product forms to be used, including plate, sheet,
forgings, bar, etc. To fabricate the impeller, each individual
blade was hot formed or machined to shape, the hub and cover Figure 2. A three piece welded impeller being fit
pieces were machined, the blades were fit, and then the pieces
were joined. The most commonly used joining methods for became more common place. This essentially spelled the end
these impellers are riveting (Figure 1) and welding (Figure 2). of three piece construction, because these designs had
While making all of the pieces separate allows for the substantially less room for error to achieve a quality part. Due
advantages above, it also results in this method being very to this, OEMs moved towards the so called “two-piece”
manual and extremely dependent on the skill of the workers. construction and most impellers made today use this method.
For example, misshaped or poorly located blades could have As the name suggests, this type of construction uses separate
serious implications with respect to aero performance, modal hub and cover pieces with the blades integrally machined into
behavior, and structural integrity. Fortunately, the less one of the halves. The pieces are almost exclusively forgings.
sophisticated designs of the day were very robust and With the availability of 5 axis CNC milling, complex blade
operational requirements were less severe, which more than designs are able to be accurately machined to tight tolerances.
made up for the human factor and basically allowed this Being computer controlled the variability within and between
construction method to work well for decades. impellers is greatly reduced by removing much of the possible
deviations. The impeller halves are typically joined by fillet
Two Piece welding, slot welding, brazing, or a combination of these
Starting in the latter part of the 1990’s, more complicated methods. Much of joining is performed, at least in part,
three dimensional impeller designs were being introduced and manually. Even cases where robots or other automated

Figure 1. An example of a riveted impeller Figure 3. Two piece impeller brazed using foil and paste

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

methods are utilized, the impeller is typically manually tack appeal of one piece construction. This is not a new idea to the
welded together beforehand. Unfortunately, all of these joining industry, as cast impellers are not unheard of, particularly in the
processes require heat and can result in distortion, which can pump industry, but it is being revisited with modern
negate some of the gain from integrally machining the blades. manufacturing techniques that make it more cost effective.
Despite these faults, the two piece method has been very From the OEM perspective, the one piece impeller is about as
successful and allowed for great strides in performance and close as possible to the design, since the removal of all joints
reliability. In fact, Lüdtke (Lüdtke 2004) described 5 axis CNC virtually eliminates any manual processing and associated
milled and welded or brazed impellers as “… the highest defects. Given this, from the user perspective, if the
quality as far as precision, strength, material integrity, and compressor is operated correctly, the impeller should ideally
erosion and corrosion resistance are concerned”. run indefinitely. Being essentially a win –win situation, one
piece construction is gaining in popularity and is beginning to
be expected as a standard offering by some users and EPC’s
(Ross, S. personal communication, 2014).

Figure 4. Two piece impeller prepped for fillet welding

Figure 6. One piece machined impeller

With the experience most OEMs have with 5 axis

machining of two piece impellers, it’s a natural progression to
utilize that process to make a one piece impeller. Similar to the
two piece method, forged material is used. Obviously in this
case, a single forging is used that is slightly larger than the
halves used in two piece construction to accommodate the
entire impeller. The articulation of the 5-axis machines allows
for most flow path profiles to be machined, and impellers of
any size are theoretically possible. The extent to which a
design can be machined depends on the accessibility and the
reach. Relative to accessibility, the tools have to be of a small
enough size to fit into the flow path and have line of sight to
machine the profile. Tool size is a major limitation with
Figure 5. Slot welded impeller in fixture respect to small impellers. Reach is an important consideration,
since the larger the impeller, the longer the supports must be to
hold the tools. As the flow path height decreases, the tool size
One Piece and the size of the supports must also decrease. With larger
When considering where to go to improve upon the two impellers with small tip openings, the long, thin supports are
piece method, it stands to reason that if two pieces is better than often not rigid enough to prevent deflection or tool bouncing,
three, then one piece must be better than two. It must be sound which can damage both the impeller and the tool. Small
logic, as that is the direction that the industry appears to be impellers with tight tip openings essentially have two strikes
taking. One piece construction, as the name suggests, consists against them, with typical tools being too large and extremely
of a single piece of metal with no joints. Given the stigma that thin supports required if the tools were available. Due to these
welds “are a necessary evil that should be considered defects factors, most one piece impellers currently encountered are
between two perfectly good pieces of metal” (Dowson, P. large, and relatively open designs. Milled one piece impellers,
personal communication, 2009), it’s not difficult to justify

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

as well as two piece impellers, are also very wasteful. In some
cases, upwards of 75% of the forging may be turned into chips.
This could, in theory, limit the possible materials to make the
impellers from. As an example, it would be a struggle to make
a titanium impeller cost effective given the high raw material
costs and the amount of waste resulting from manufacturing.

ALTERNATIVE METHODS
It is clear that before the industry can move solely to one
piece impellers, some gaps will need to be filled. When
making a plan of attack, the low hanging fruit of large size,
large opening impellers are very well covered by 5-axis
milling, although it remains to be seen if that is the overall best
method. It makes sense then to attack from that position of
strength to find where its limits lie. The authors company and
probably most OEMs, have done that and have a fairly good Figure 7. Model of the selected impeller for the project
feel for what 5 axis milling is capable of. The question then
becomes how one attacks the other end of the spectrum? PM, and additive manufacturing. In each case, vendors were
The remainder of this paper discusses four alternative contacted and given the same impeller design shown in Figure
manufacturing methods that the authors company is 7 to evaluate and quote. The impeller is roughly 300 mm
investigating as options capable of addressing the small diameter, has 3 dimensional blades, and a tip opening less than
impeller, small tip opening end of the spectrum. The specific 6 mm. The tolerances on the drawings were 5-axis milling
processes being evaluated are EDM, Investment Casting, HIP’d tolerances for two piece construction, although it was
understood that the tolerances are not likely possible for most
of the processes. 17-4 PH in the NACE MR0175 (2009)
compliant 1150oF double aged condition was selected as the
material of construction, although for reasons explained later,
the additive manufactured impeller was changed to Inconel
718. Due to the custom designed nature of impellers, rapid
prototyping and similar concepts were incorporated to reduce
costs and lead time associated with permanent tooling. The
impellers were subjected to a battery of tests to evaluate the
dimensional accuracy and structural integrity, as well as to
characterize the resulting material properties. Each process will
be discussed in detail separately. The results of the tests will
then be summarized and each process will be ranked with
considerations to cost and delivery, manufacturability, and
suitability for service.

EDM
Electrical Discharge Machining or EDM is the process of
removing material using a series of electric sparks. While the
observation of the damage from lightning strikes has been
observed since the beginning of human kind, it wasn’t until
1770 that the spark erosion phenomenon was characterized by
English scientist Joseph Priestly (Webzell 2001). It was not
until the 1940’s that Russian researcher Lazerenko developed a
power supply that paved the way for the phenomenon to be
harnessed for controlled material removal (Anonymous 1965).
It was initially relegated to the useful, but lowly task of being
used to remove broken drill bits from holes (Webzell 2001).
Since that time, EDM has developed to the point that it has
become the most popular of the unconventional machining
methods.

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 As stated above, EDM is an unconventional machining rate and is very prone to cracking. Typically parameters are set
process. Jameson (2001) provides an in depth discussion on the to minimize the remelted or recast layer and often surface
various aspects of EDM, some of which are summarized below. treatments are performed to remove the affected material.
Conventional machining is a mechanical process where a tool Control of the process keeps the “cut” width narrow and highly
physically touches the part and cuts material away. In EDM, accurate when the tool is attached to proper CNC equipment.
the tool is an electrode and it is separated from the part a certain EDM does require the material be electrically conductive, but
distance, known as the spark gap. Figure 8 illustrates the EDM beyond that, the process does not necessarily “care” what the
process. Dielectric fluid is flowed in the gap. The dielectric material is, which makes it useful for machining very hard
fluid is an electrical insulator under normal conditions. When materials that may not be possible to machine using
the voltage applied to the electrode exceeds a critical value, conventional methods.
ionization of the fluid occurs and it changes to an electrical There are two main types of EDM processes, namely wire
conductor. When this change occurs, a spark is discharged at EDM and sinker EDM. Wire EDM can be thought of being
the location of the shortest distance between the tool and the similar to a scroll or jig saw. Although it is capable of more, the
part. Sparks occur at about 2,000 to 500,000 times a second most simple and common application would be cutting shapes
and never at the same location. The temperature at either end out of sheet and plate. Basically the work piece must be thin
of the spark is extremely high and actually vaporizes a small enough that the wire can be passed through the thickness and
amount of material from both the must be started at an edge or in a through thickness hole.
Sinker EDM can be very complicated depending on
application, but in its most simple form is analogous to using a
drill to drill a series of holes into the work piece. By varying
the location and angle of the tool and the depth the tool is sunk
into the part, very intricate features can be machined. Given
the curvature and overall distance of the flow path in an
impeller, sinker EDM is the process required to machine an
electrode and the part. The vapor rapidly solidifies as a chip impeller.
and the flowing dielectric fluid washes it away. The local The process to manufacture the selected impeller via EDM
nature of the sparks and te rapid cooling of the dielectric fluid is not drastically different than that for machining the impeller
keep the overall part cool to using conventional methods. Two disk forgings of fully heat

Figure 8. Schematic of the sinker EDM process

the touch, but there are remelted areas that remain on the
surface. EDM is considered to be a thermal machining process.
In fact the only major issue with EDM is that it does leave
patches of remelted material on the surface. This material is
usually very hard and brittle due to the extremely rapid cooling

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 of four of the 15flow paths machined , two adjacent flow paths
separated by 180 degrees from the other two. One set of flow
paths was left as machined to evaluate the recast layer. As
mentioned above, the recast layer is commonly thought to be
very detrimental, although in recent years, machine shops have
been able to achieve very thin and discontinuous layers that
could possibly be left as is. It is also important to understand
what could be the worst case scenario if the removal process
fails. The other set was abrasive flow machined to remove the
recast layer. Abrasive flow machining is the process of forcing
an abrasive laden high viscosity fluid through the part. The
abrasive grinds the surface and can achieve a very fine, mirror
like finish. The other forging was machined as the full impeller
and was also abrasive flow machined. The full impeller (Figure
9) and sample (Figure 10) were returned and the remaining
processing and testing was performed in house.
The sample was cross sectioned in both the as EDM’d and
abrasive flow finished flow paths. The presence and thickness
of the recast layer was measured in each area. In the as EDM’d
surface, the recast layer is approximately 10 µm thick and is
very spotty. . Cross sections were taken and the length of each
recast island was measured and divided by the total length of
surface. The recast layer accounted for 9% of the flow path
surface of the cross section on the as EDM’d surface. Figure
11 shows the approximate locations of recast layer islands in
red. Interestingly, the recast layer is not visible near the eye
(inlet) of the impeller. Also shown in Figure 12 is a
Figure 9. Images showing the EDM’d full impeller representative cross section of the islands. The same process
was repeated for the abrasive flow machined surfaces. The
treated 17-4PH were sent to the supplier. The supplier process did not entirely remove the recast layer islands, as seen
performed a pre-turning operation to roughly shape the OD and in both Figures 13 and 14, but it did reduce the amount to 2%.
hub and cover contours. The flow paths were then machined The thickness of the islands also seems to have been reduced in
on a sinker EDM machine. After the flow paths were in place, many cases, but there are still instances of islands that are
the impellers were turned to final dimensions and ~10µm thick. If EDM is to be used for impellers, further work
dimensionally inspected. One of the forgings only had a total may be required to determine if the remaining recast islands are

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Figure 10. Images of the sample piece
likely to be an issue or revisit the processing to remove it 15. Once the deformation occurs, the material thereafter
entirely. behaves linearly at stresses below the maximum and assuming
The full impeller was treated as a typical production the maximum stress is in the elastic region, no further
impeller would be in the shop. The impeller was magnetic permanent deformation occurs at lower stresses
particle inspected. As would be expected on a single piece of (unloading/reloading line in Figure 15). Based on this, 17-4
forged material, there were no indications found. Being that PH impellers are spin tested the first time and allowed to
the forging was already heat treated, the properties did not deform or grow to increase the limit of proportionality. Then
change from the certification from the forging supplier it was they are machined to the final dimensions and spin tested again
not necessary to retest for material properties. It was to ensure no further growth occurred. The impeller survived
dimensionally inspected and found to be within the same the first spin and showed typical amounts of growth on the
tolerances as specific for a 5 axis milled two piece impeller. subsequent dimensional inspection. It was then finish
One slight deviation from typical processing was that the machined and spin tested again. Only one dimension was
impeller was spin tested at the maximum rated speed rather found to be out of specification after the second spin. The
than a speed based on a specific job requirement. Given that distance from the back of the hub to the end of the bore was
17-4PH exhibits significant continuous yielding behavior it is slightly out of tolerance. It is believed that this was a
actually spin tested twice. The reason for this is that the limit machining error prior to the second spin.
of proportionality or the elastic region where it behaves linearly It was mentioned above that one piece impellers should be
on a stress strain curve is very small in this type of material. the closest to achieving the ideal cases of the actual design. To
The material is essentially always yielding, resulting in
permanent deformation, as shown as the loading line in Figure

Figure 13. Schematic of the impeller flow path showing

Figure 11. Schematic of the impeller flow path showing instances of recast layer (red) after abrasive flow machining
instances of recast layer (red) in the as EDM’d passages

Figure 14. Cross sectional image showing a typical island of

Figure 12. Cross sectional image showing a typical island of recast material remaining after abrasive flow machining
recast material in the as EDM’d surface
perform a high level check on that theory, the impeller was rap

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

tested to evaluate the variation in blade frequencies. Due to the
high expected frequencies, a microphone with a range up to
70kHz was used. The standard deviation of the measured
frequencies was 0.2%. This shows very little deviation
between the blades.
The final check on the impeller is currently underway at
the time of this writing. The outer surfaces were measured
using a structured light process. The impeller will then be cut
apart using a well defined wire EDM process. The internal
surfaces will then be measured. All of the measurements will
then be reconstructed and compared against the model to
evaluate the overall and more importantly, the flow path
accuracy. Particular interest will be the transition area on the
cover, where it may be difficult to reach via EDM. This
process is planned to be repeated on the other impellers
discussed in this work.

Figure 16. Schematic showing an example of an area

inaccessible via EDM and the required modification

Being that EDM is a mature machining method, it is the

easiest of the processes evaluated to implement. Based on that,
the supplier was sent several designs to evaluate for
compatibility with the EDM process. Surprisingly, the
complicated, 3 dimensional designs which are difficult to
manufacture as two pieces were said to be easy for this process.
The problematic designs were the simple, 2 dimensional
designs with small flow path openings and nearly 90o transition
in the flow path. The reason for this can be explained by
considering the drill analogy for sinker EDM. Drill bits do not
bend, and as such, line of sight is required to machine the flow
path. As illustrated in first schematic in Figure 16, the
geometry results in an area on the cover side in the transition
that is not able to be reached. To be able to reach this area, the
bore at the eye of the impeller would need to be cut back, as
shown in the “modified” schematic in Figure 16. In these
instances, an evaluation would be required to determine if
enough contact surface remains to prevent movement of the
impeller on the shaft prior to making any changes.
Figure 15. Schematic illustrating the stress-strain behavior of
continuously yielding materials
Investment Casting
Investment or lost wax casting is by far the oldest of the
methods considered in this work. Archeologists have recovered
artifacts dating to ~3500BC that are believed to have been
investment cast (Hunt 1980). Some of these artifacts have
complex forms and intricate details that would be virtually
impossible to make any other way given the tools available
(Hunt 1980). The advantage of investment casting lies in the
way the mold was created. Initially the actual part that is

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

desired is made from wax. Wax is obviously easy to form with internet search for investment casting foundries will find a large
smooth surfaces and high detail using even primitive tools. The percentage of web sites for foundries with pictures showing
wax was then coated with a slurry containing clay. The examples of cast open and closed impellers. In fact; the
composite structure could then be treated essentially as pottery, authors’ company manufactured closed centrifugal compressor
being heated to melt and remove the wax and to fire the clay. impellers in the early 80’s using investment casting. These
The result was a strong mold. Molten metal, initially smelted impellers met all of the mechanical and dimensional
copper (Hunt 1980), was poured into the mold. When the metal requirements at the time. The per part cost was reasonable and
cooled, the mold was broken off, leaving a metal part with all the lead time was good. Unfortunately in this application, the
the detail of the wax pattern. greatest strength of investment casting is also the source of its
weakness. Conventional investment casting requires a die or
tool to make highly repeatable wax patterns in a very short
Through the years the basic concept has changed little, but period of time. There is a significant upfront capital cost for
the process has been optimized for mass production. Figure 17 the tooling and severe hit to the lead time. Although the
shows a schematic of a modern investment casting process. process can produce parts rapidly and relatively inexpensively
Rather than hand working the wax, permanent dies are typically once the tool is available, a large number of parts are required
used to injection mold the wax to shape. The wax pieces are to offset the initial commitment. The process was cost effective
attached to preform gating pieces to form pathways for the when impeller designs were reused regularly. As new design
molten metal to be fed to the casting. The gates also can be methodologies and computers became more prevalent, it
used to make trees of several patterns, allowing molten metal to became easier to make minor tweaks to optimize designs for
be fed to several molds simultaneously. Very fine specific applications. Each change would require either a new
slurries/stuccos are used in the investing process to build the tool or a permanent change to an existing one. Obviously the
initial layers of the mold to perfectly replicate the detail of the costs and lead times skyrocket in that situation and investment
wax pattern. Coarser slurries/stuccos are used to build the outer casting lost its competitive edge.
layers rapidly and with increased strength. Most foundries In recent years, the investment casting industry has been
have some sort of automated system for holding the trees and widening its net to capture specialized one off applications.
applying the ceramic. Alloys that are investment cast range This is being accomplished by incorporating additive
from simple copper to some of the most complex nickel based manufacturing (aka rapid prototyping or 3-D printing)
superalloys. Additional features have also been added. For techniques into the pattern making process. While wax based
example, chills and grain selectors are built into the molds to printers are available, they are not currently in widespread
give a certain grain orientation or single crystal materials that usage, nor do they currently have large build dimensions
are used in high temperature applications. It’s astonishing to (currently about the size of a sheet of paper in the horizontal
consider the critical technological applications, such as gas plane). As these machines are being developed, machines that
turbine blades and medical implants, which investment casting work in various polymers are filling the void. In fact many
is frequently used for given the ancient origins of the process. machines and materials used for the rapid prototyping in
As mentioned above, investment casting is not new to polymer materials are specifically designed for use as
rotating equipment or even more specifically to impellers. An investment casting patterns. The rapid prototyped plastic

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Figure 17. Schematic of the traditional investment casting process

materials are able to achieve a high amount of detail and are essentially machining tolerances, which competing
surface smoothness. Some processes even impregnate the processes, such as EDM are capable of. The narrow flow
surface of the plastic with wax to further improve the surface passages in the impeller are also a source of difficulty.
finish such that they can rival fully wax patterns. As an Passages in the size range of those in this impeller and smaller
example, Figure 18 shows a pattern of an impeller printed using are problematic to invest properly. The ceramic slurry tends to
the QuickCast process. The process uses a photoreactive bridge and can close off part of the passage before adequate
polymer resin and prints bulk portions of the pattern as a thickness of ceramic is achieved. The thin, weak areas of the
honeycomb structure to reduce weight. The honeycomb also mold can rupture during casting. This obviously ruins the part,
allows the pattern to collapse in on itself when heated to burn it but also presents a safety risk to the foundry workers. The fact
out. These patterns are able to be coated with the ceramic that an impeller is designed to move fluids through it is
slurry, so there are little to no changes required for the advantageous for this method. Despite the risks, a few
investing process. The major difference is in the pattern foundries believed that with care in the investing process, even
removal. The plastics are generally burned out rather than to the point of manually dipping the impeller, the flow paths
melted and poured. It is therefore important to consider thermal could be properly filled. The relatively large masses of the hub
expansion of the plastic, so it doesn’t break the low expansion and cover are connected by the thin blades also complicates the
mold materials. Other considerations are venting of the matter. Depending on the pattern and gating designs, one of
gaseous products generated by the burn out and the amount and these large masses may be required to be fed molten metal
removal of any residual ash. After the pattern is removed, the through the blades. If care is not taken, the thin blades may
remainder of the processing is unchanged. solidify before the mold is filled. Since this is the most cost
effective method to investment cast this impeller, an impeller
was made using this method.
While it was not an entirely smooth process, the impeller
was successfully cast into metal using the invested flow path
method. The mold was highly scrutinized prior to casting and
found to be very sound, particularly in the flow paths.
Unfortunately, there was a mishap during casting and the
bottom of the mold broke before the metal solidified. This is a
known risk with the investment casting process. Given that the
failure was not related to the investing of the flow path, a new
pattern was quickly made and the process restarted. Once
again, the mold was more than acceptable and this time casting
process went smoothly. Due to the setback, there is a limited
data set to present on this impeller. Figure 19 shows the hub
and cover sides of impeller after the gating was removed and
Figure 18. Photo of a QuickCast pattern of a double flow pump the part was HIP’d. Stock was added to the model to aid in
impeller casting and to provide machining stock to ensure the impeller
can be trued up during final machining. Figure 20 shows a
Given that investment casting was successfully used for closer view of the eye of the impeller. The flow path appears
impellers in the past and rapid prototyping appears to overcome even and smooth. The majority of the blades appear very
the reasons why it fell from favor, investment casting was smooth and well formed, similar to the blade on the right hand
explored as a possible method to make the test impeller. side of Figure 20. The other two blades in this view appear to
Several foundries were contacted before selecting one to work have indications at the leading edge, but it is unclear at this
with. The common consensus from all the foundries was that point as to the cause and severity. Being an in process casting,
while impellers are commonly made, this particular impeller these may even be repaired during an upgrade cycle by the
would be pushing the limits of the process. Impellers typical supplier. An upgrade cycle includes inspection, removal of any
investment cast that have the same number of blades and of a indications found, and weld repairing the excavation. Figure 21
similar shape are almost exclusively open impellers. Most shows the OD of the impeller. Once again, the flow path is
closed impellers that are cast are pump impellers, which tend to very smooth in appearance.
have fewer, heavier blades and more open flow paths. This Since the casting is still being processed, only rough
impeller is closed with 15 thin blades and very tight tolerance dimensions have been able to be taken at this point. Overall,
requirements. Investment foundries often quote +/-0.005 to +/- the dimensions are on the large side, but they are relatively
0.010 in/in as a standard on dimensional tolerances. Premium consistent around the impeller. For example, the mean flow
quality castings frequently achieve better than that with path opening is ~0.027” oversized, but they only range +/-
traditional wax patterns, but the use of rapid prototype patterns 0.005” from the mean. The supplier has stated that relatively
adds to the variability. The desired tolerances on this impeller conservative parameters were used to determine the pattern

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

dimensions required to achieve the final metal dimensions. machines large enough to make a core for this impeller are not
Given the knowledge of how this impeller behaved will allow readily available. To evaluate the feasibility of using a ceramic
them to be less conservative on future impellers of similar core, a two piece core was made. One source of possible
geometry. problems with using a printed ceramic core is that ceramics
There are expected to be limits to how small of any differ from metals and polymers in that they are not melted in
opening can be to allow proper investing. Other methods were the printing process. They are loosely held together by a binder
also considered with this possible issue in mind. To get around and then sintered, which is simple terms is heating to cause the
particles to diffusion bond. The sintering process results in
shrinkage and can introduce distortion. Prior to investing using
the core (Figure 22) it was dimensionally inspected to
determine if it was accurate enough to be worth processing
further. The tolerances were fairly large for the core, but they
were close enough to have value to pursue further. To make
the mold, hub and cover patterns (Figure 23) would be made
using the same plastic process used above. The pieces are then
assembled and the assembly would be invested. The first
impeller being made with this process suffered the same fate as
the first invested flow path impeller. Since the core takes more
time to print than the plastic patterns, this replacement impeller
has not been cast at the time of this writing.

Figure 20. Eye of investment cast impeller

Figure 19. Hub and cover sides of the investment cast impeller
after gate removal

the problems with investing small passages, solid ceramic cores

in the shape of the passage are often incorporated into the
pattern and designed to remain in the mold after burn out. In
the case of an impeller, the core would be the negative of the
flow path (Figure 22). For mass produced parts, the cores are
made with permanent tooling similar to the wax patterns, which
is obviously not desirable for this application. Once again,
rapid prototyping, this time of ceramic materials, was
investigated. Similar to wax, processing ceramics in this
manner is not nearly as mature as plastics or even metals and
Figure 21. OD and flow path of investment cast impeller

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 making tungsten filaments for light bulbs (Angelo 2008). The
process really hit its stride in the 1955 with the invention of Hot
Isostatic Pressing or HIP by researchers at Battelle (ASME
1985). Prior to this, pressure had to be applied in discrete
directions. This led to inhomogeneity in terms of powder
consolidation and some anisotropy in the material properties,
which limited the complexity of the parts. HIP is performed in
a pressure vessel with high pressure inert gases applying
pressure uniformly over the surface. This addressed the
downfalls of previous methods to apply the necessary pressure.
Today powder metallurgy is a large industry and competes in
some of the same high tech markets as investment casting.
Prior to the inception of the projects that are the subject of
this paper, the authors company had worked on a powder metal
impeller. Same as the work discussed here, the material
selected was 17-4PH, but the impeller was larger with a more
complex flow path contour. The first part of the process to
make the impeller was to design the carbon steel can that would
be used to contain the powder. The bottom piece of the can
(grey in Figure 24) was essentially dish shaped to accommodate
the tail end of the bore, the back wall of the impeller and some
Figure 22. Ceramic core of the reverse of the impeller flow of the flow path height. To reduce the amount of powder
path required, a 17-4PH forging was machined to serve as the bore
and part of the back wall of the impeller. This is shown as the
brown piece in Figure 24. It was expected that this part would
fuse with the powder to make a single part. Similar to what
was discussed for the ceramic core for investment casting, the
reverse of the flow path was machine out of carbon steel and
placed in the can (pink in Figure 24). The top of the can
consisted of two pieces. The first was essentially a ring that
made housed the remainder of the flow path height at the tip
and the horizontal part of the outer cover surface (dark grey).
The second part made up the vertical part of the cover and had
the proper accommodations for purging and filling the can with
powder (light grey). Figure 25 shows a cross section of the
can. The blue in this figure indicate open areas to be filled with
powder. After the can was properly filled with powder, it was
evacuated, sealed, and finally HIP’d to sinter the powder.
When the HIP cycle was completed, as much of the carbon
Figure 23. Example of an impeller pattern using a ceramic core steel can as possible was machined away. The part was then
submerged in an acid tank to leach out the carbon steel flow
Hot Isostatic Pressed Powder Metal path insert and any remaining carbon steel on the outer
For the purposes of this paper, powder metallurgy can be surfaces. At this point, the impeller was processes as any other
defined as the process of making solid components from 17-4PH impeller. The final impeller is shown in Figure 26.
powder precursors by the application of heat and pressure.
Powder metallurgy can trace its origins back nearly as far as
investment casting. There is evidence in India and from the
Incas of direct reduction of iron ore with charcoal to produce
sponge iron powders. These powders were subsequently
formed into solids by forging. An example of this is the
Ashoka Pillar of Dehli in India (Angelo 2008). Strangely, with
great works being created using the process, it is believed to
have been forgotten from that time until the 1800’s when it was
recorded to be used for producing platinum (Ramakrishnan
1983). The first major industrial use was in the early 1900’s

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 Figure 25. Cross section of the can

In the time from that project to the current project, the

vendor had gained more experience with impeller type parts
and believed they had worked out the calculation problems.
They found the impeller for this project less problematic in
terms of the complexity. Unfortunately, the dimensional
Figure 24. Model showing the can, bore piece, and insert for tolerances they quoted were very large, particularly given the
the powder process small size of the impeller. The reason for this is once again
related to the complex nature of the calculations. The typical
The testing program was similar to what was done for the process for developing a powder metal solution would be to do
current project. The impeller passed NDT inspections and the several iterations to optimize the calculations. This would then
mechanical properties were also acceptable. The impeller did result in a rapid and highly repeatable process. Due to the fact
not pass the dimensional inspection. All of the dimensions that impellers are mostly custom designed for each job, it was
measured on the outer surfaces were acceptable. Measurement not felt that this process is currently feasible for impellers and
of the flow path contour showed that the vertical to horizontal was not pursued further.
transition region of the cover was incorrect and with the outer
surface machined properly, the cover was thin in that area. The Additive Manufacturing
impeller was not able to be safely spin tested as a result. The Additive manufacturing is a generic term that encompasses
root cause of the deviation was determined to be an error in a several processes that are used to create parts by adding
calculation by the vendor, which is not unusual in a first article material, which can be metal, plastic, or ceramic, rather than
in a new process. In this process, volume shrinkage due to removing it as would be done in traditional machining (which
sintering and thermal expansion are both occurring and are is by extension a subtractive process). Most of these processes
competing processes. The can design, and more specifically work by adding layers of material to build up a part, not unlike
the insert design must properly account for both. In this case, building with LEGO® blocks. In fact, from that perspective,
the problem region was a transition where both processes are one could argue that the basis for additive manufacturing has
occurring significantly in multiple directions, which makes it a been known long ago in human history and great works like the
complex problem to work out.

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 With respect to additive manufacturing of metals, welders
would argue that they have been doing it forever. In fact, in
1926, Baker patented a process of building three dimensional
shapes with layers of molten metal using an electric arc as the
heat source (Ding 2011). Fast forwarding to today, and there
are several A/M processes available using metal. The basis of
the vast majority of these systems is a weld process applying
layers of metal, not unlike Bakers concept, although most are
typically using powders and sophisticated positioning
equipment for higher speed and precision.
There are two general types of metal additive
manufacturing systems (Herderick, 2011). The first type can be
thought of as free form welding, where layers of weld metal are
deposited on a substrate and built up in space. The other types
of systems are called powder bed systems, which is analogous
to a stereo lithography type process. There are advantages and
disadvantages to using these systems for an impeller
application.

The free form welding systems are conceptually relatively

simple. They basically consist of a welding head attached to
positioning equipment (often a robot arm) that is integrated
with additional positioning equipment for the part. There is a
litany of advantages to these systems compared to the powder
bed systems. The most significant advantage of the free form
welding systems is that they are less limited in the build size
than the currently available powder bed systems (Herderick
2011). The size is really only limited by the size/range of the
positioning equipment being used or, if something like electron
beam welding is being used, the size of the vacuum chamber or
other containment available. These types of systems are very
fast, with several weld processes that could be considered that
can deposit material in the range of 10 kg/hr or faster (Ding
2011). There are a multitude of weld consumables available in
virtually any weldable material, particularly for machines that
use wire. This opens the doors with respect to material
possibilities, because more materials are available, multiple
materials could be used, and it allows for a variation of
chemistry/properties throughout the part to optimize the final
product. The positioning equipment also allows for some
features, such as overhangs, be made without supports, which is
a problem that will be discussed with the powder bed systems.
Despite the numerous advantages, there appears to be fewer
commercial options available and significantly less published
work on these systems (Herderick 2011). Part of the reason for
this is that in general, unless the part is extremely complex, it is
often less expensive to work with castings or in some cases,
forgings. These machines also shine due to the build size and
Figure 26. Photos of the final HIP’d PM impeller build rate. Unfortunately, these attributes are not normally
conducive to high precision. In turn, the lower level of
precision relative to the powder bed machines results in surface
Egyptian Pyramids are examples. Others do not have such a finishes that are not likely able to be used in the as printed state
grand view of origins and trace the idea of using a layered for most applications (Herderick 2011). Subsequent machining
approach to making complex shapes to the 1880’s and the is required. Although the machining can be minimal relative to
processes invented to make topography maps (Bourell 2009). a forging or even a casting, it cuts into the time advantage and

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

adds cost. This is obviously not acceptable for an impeller developed for near full density in small part sizes. It is not a far

Figure 27. Schematic of the DMLS process (courtesy EOS)

design where machine tools can not fit into the flow path after stretch to imagine that this process would be hampered by the
the part is printed. To address this, there are new hybrid same complications as the powder metal process.
systems becoming available that incorporate 5-axis machining Given the current deficiencies of the other systems, the
into the system. Essentially these machines build a few layers DMLS process was chosen for this project. The timing was
with the additive process and then machine the surface and fine very fortunate. First, this project was in the early stages when
features as required before adding more material. Given the Allison, et,al, (2014) reported on making closed impellers using
potential for impeller applications, plans are in place to include DMLS as well, which helped this project move from lab
the hybrid processes in future evaluations. curiosity to a process with a legitimate chance of success in the
eyes of many at the authors’ company. Secondly, machines
Within powder bed systems, there are two types, those with large enough build volumes to accommodate this impeller
that use welding processes, such as laser or electron beam, and were only introduced to the market in mid 2014 and were just
the so called binder jetting systems. Figure 27 shows the basic starting to become available. The vendor selected had not yet
concept for these systems. The welding based systems, more received this larger machine, so in the meantime, a segment of
commonly known as Direct Metal Laser Sintering (DMLS), the impeller containing two flow paths was made on existing
Selective Laser Sintering (SLS), or Electron Beam Melting smaller machine to determine feasibility and if acceptable, the
(EBM), apply a thin layer of metal powder over a base plate. A full impeller was built when the larger machine was
high power laser or electron beam scans over the areas to be operational.
fused and melts the powder only in those locations (Herderick As with the other techniques discussed thus far, this
2011). The machine lowers the base plate down and the impeller is not straight forward to manufacture using DMLS.
machine applies another layer of powder on top and the process The major issue with this design and closed impellers in
continues until the part is complete. The unfused powder is general, is that as relatively flat rotating parts, it is preferred to
blown out and collected for reuse, leaving behind the printed be built horizontally. This minimizes the build height, which is
part. The binder jetting systems are the closest to literally a major contributor to the build time and in turn, the cost.
being a 3D printing process, as they use modified print heads to Being dynamically loaded, it is important that the properties be
apply a binder resin to a powder bed similar to the laser powder as uniform a possible about the axis of rotation, which also
bed system (ExOne 2014). This essentially glues the powder favors a horizontal build. As a consequence of this, the faces of
together rather than directly melting it. The “green” part is the hub and cover are nearly parallel to the base as the impeller
dried and the binder is burned out during furnace sintering. is built. The process is currently only able to build surfaces like
Currently, most parts made by this process are very porous and this for extremely short distances at an angle less than 25o and
must be back infiltrated by a lower melting point material such it must be off of the build platform or a previous layer
as a bronze. There are however, some materials that have been (Brancher, 2015). The powder bed by itself does not provide

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

adequate support to build disconnected layers, as would be the Figure 29. Schematic showing the extent of supports required
case for example, for the start of the cover if built with the in the flow path
cover on top. Being essentially a weld, the residual stresses in A second consideration with the DMLS method is that of
the layer of solidified metal would tend cause it to curl up and material selection. Most suppliers use powders provided by the
be destroyed by the recoater blade when it applies the next manufacturer of the machine, as powders with the proper
layer of powder (Brancher 2015). To get around these issues, characteristics are just now becoming available elsewhere
any surfaces less than 25o from the base require support (Higgs 2015). This limits the number of materials available to
structures be built into the part to allow the horizontal pick from. The options that could be used for impellers are 17-
structures to be built. This is typical of this type of machine, 4 PH, 15-5PH, Inconel 625, Inconel 718, and Ti-6Al-4V (EOS
even those that print in plastic. Figure 28 is a printed dental 2014). All of these have some drawbacks.
apparatus showing the as printed part with supports and the At first glance, 17-4 and 15-5 PH, are obviously ideal
final part with the supports removed. While supports on the choices since they are already are widely accepted in the
outer surfaces of the impeller can be machined off fairly easily, industry. It is interesting that they are considered separate
there is a significant distance in the flow path where the hub materials, as their chemistries largely overlap. The reason for
and cover are nearly parallel and requires supports. The yellow the difference is in the way the powders are made. The 15-5
region in Figure 29 shows the extent of the supports required in PH powder is atomized using argon gas. Parts printed with this
the flow path of the subject impeller to build the cover. The powder are stated to be able to meet all the properties of
supports typically are built with perforations, as shown in wrought 15-5PH. The problem is that it is prone to cracking in
Figure 28, to make them as easy as possible to remove, but they highly restrained parts, as an impeller would be. Suppliers can
are still metal and require tools to break off. Given the small and will build using this powder, but for a complicated part, it
tip opening, reaching all of the supports and removing any may be difficult to establish a robust process with a high
residual material may be difficult, if not impossible. probability of success, which will most likely increase costs.
The 17-4PH powder is nitrogen atomized and the powder picks
up a significant amount of nitrogen in the process (Murr 2012).
Nitrogen is an austenite stabilizer and the parts made using this
powder have a greatly increased amount of retained austenite as
a result. This affects the properties of the material. In
particular, the yield strength can drop as low as 80 ksi from the
typical 100 ksi plus yield strengths expected from the NACE
approved double 1150oF aging treatment. As an example of
this, Allison (2014) reported on an attempt to make a 17-4PH
impeller using similar equipment. They stated that impeller
cracked on liquid quenching. This alloy does not typically
require liquid quenching, so it is assumed that they were
Figure 28. As printed (left) and finished dental aparatus attempting to overcome the effects of excess nitrogen. There
(courtesy EOS) are mixed findings as to whether the properties can be
recovered. Some researchers (Murr 2012) show that using an
argon purge during the DMLS process can cause the material to
respond correctly to heat treatment. The suppliers experience
shows that this is not likely (Brancher 2015). Starr, et.al
(2013). report that parts can regain properties through heat
treatment, but once again, this does not appear to be the case in
all instances.
With the stainless steel materials not being straight
forward, the more exotic nickel and titanium alloys were
evaluated. Inconel 625 would be able to be applied to the
widest range of applications of any of the materials, including
the stainless steels. The issue is that the minimum yield
strength is approximately half that of 17-4PH. Inconel 718 and
Ti-6Al-4V have essentially the opposite problem in that they
are too high of strength, and despite being acceptable for
NACE applications, they have yields strengths that violate the
120 ksi maximum yield strength required by paragraph 4.5.1.11
of API 617 (2014) for hydrogen service. Note that API 617
does not distinguish between material types in this requirement.

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Many of the impellers in this size range are used in applications be easily changed if desired. The eye of the impeller is shown
such as hydrogen recycle compressors, so this is a significant in Figure 35. The center blade is the only blade that was
concern. Titanium also has the issue of increased material cost printed fully according to the model. This blade appears to be
and special handling since the material is highly reactive. very well formed, as it appears straight, with a good leading
Titanium powders are notorious for causing fires due to the edge radius and fillet. The surface roughness of the flow path
intense heat generated by reaction with oxygen. at the eye is exceptional. Overall, the visual appearance of the
While all of the other projects used 17-4PH, with the impeller segment was very good and although some finishing
possible questions as to where material in this condition fit in work is required, it far exceeded expectations.
terms of issues like NACE acceptance, it was decided to us
Inconel 718. This alloy is a work horse alloy for the gas
turbine industry and due to that is one of the most well
characterized and frequently printed materials. It was felt this
would provide the overall best chance to properly investigate
the abilities of DMLS to build an impeller to the desired
tolerances. The API 6ACRA (2015) heat treatment was used
for NACE compliance. Bhavsar, etal (2001) report that this
heat treatment also appears to significantly improve the
resistance to hydrogen embrittlement. This would have to be
evaluated in more detail, but shows potential. It is also
possible, if the DMLS process is shown to yield acceptable
results, that processing parameters can be developed for other
materials that may require fewer compromises.
Due to the fact that the supplier had not had their new
larger machine delivered yet, it was decided to make a two flow
path segment of the impeller on an existing smaller machine
and then follow up with the full impeller if the process Figure 30. Cover side of the DMLS impeller segment
appeared to show potential. Figures 30 and 31 show the top
and bottom of the segment as it was received from the supplier.
As mentioned above, the part was printed in Inconel 718. After
printing, it was HIP’d to consolidate and homogenize the
structure. At that point, it was heat treated per API 6A718.
Figure 32 shows the OD of the segment still attached to the
build platform. This figure illustrates the number of supports
required to build the hub off of the platform and the supports
required in the flow path to allow the cover to be built. The
hub supports were removed when the segment was parted off
from the platform prior to heat treat. After the heat treatment,
the flow path supports had to be removed. Figure 33 shows the
flow path after the supports had been removed. The hub side of
the flow path is fairly clean and smooth. While all the supports
were able to be removed, some areas on the cover side were
substantially more difficult to reach and some residual
roughness remained after processing with standard procedures. Figure 31. Hub side of the DMLS impeller segment
Subsequent finishing operations, such as abrasive flow
machining, will likely be required to minimize or remove this
roughness. Figure 34 shows outer surfaces of the hub and
cover, respectively. The hub surface was relatively smooth, but
did have a cross hatch pattern of rougher areas. The way the
machine prints large sections can cause a cross hatch pattern
(Beuth 2015) or it could be due to the supports. Regardless, the
rough areas are fairly shallow and could be easily cleaned up
during subsequent operations if desired. The cover surface had
steps visible; giving it the appearance as groove on a record.
This is caused by the low angle of the cover combined with the
layer wise build process. Once again, this surface profile could Figure 32. As printed OD of the DMLS impeller segment

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 fit to the model. Figure 36 shows color scale for the deviation
plots that follow. Note the corresponding dimensions are in
inches. An overall view model is shown in Figure 37 for the
cover side and Figure 38 for the hub side. Being that the part is
only a segment, the large red and uncolored areas represent
missing material. The other thing to note is that the solid blue
areas in the seal areas are due to the fact that the seal teeth were
not machined into the segment. Figures 39 and 40 show closer
views of the cover and hub side, respectively, at the OD. There
is a thin red line at the OD that once again indicates missing
Figure 33. OD of the DMLS impeller after supports were material. In this case, the diameter was undersized about
removed 0.015” (0.381 mm) on a side. Just inboard of the red line is
another deviation visible, appearing as orange in Figure 39 and
blue in Figure 40. The combination of these two areas
indicates that the cover has pulled down about 0.010” toward
the center of the flow path. This is likely due to distortion from
residual stresses and is a common occurrence in welded
impellers. There is a high probability that this can be corrected
for future parts with similar geometry. The final deviation
observed in the OD views is the blue areas near the edges of the
Figure 34. Surface roughness of the hub (left) and cover (right) segment. These areas indicate that the part is bending up at the
edges, once again due to distortion as a result of residual
stresses. It is possible that this distortion may not be present in
a full impeller. Moving to the ID, Figures 41 shows the eye
area of the segment and Figure 42 shows the scan 180o away.
The axial face of the cover appears to very accurately located,
but the cover appears to be thick and is reducing the flow path
by about 0.006”. The middle blade shows little to no deviation,
nor does the hub. The bore is a different story and shows poor
concentricity. This is evidenced by the fact that the bore
matches the model well in Figure 42, but is about 0.010”
undersized in the Figure 41. This is also most likely due to the
partial segment rather than full impeller that would help
balance the distortion. Overall, these results are very
encouraging, particularly given that this is a first attempt at an
impeller of this type. Most of the deviations are related to only
printing a segment or are thought to be correctable. Assuming
printing the full segment corrects much of the deviation, adding
some stock to the part should be able to accommodate any
remaining distortion.

Figure 35. Eye of the DMLS impeller

While looking like an impeller is a good first step, the

more vital test is how accurate is the part to the model. The
supplier reported that the segment was essentially printed to the
final machining model, i.e. there was no stock on the part. This
is not ideal for a rotating part with tight tolerances on the
runout, concentricity, and balance, but it does afford the
opportunity to evaluate the accuracy of DMLS as a net shape
method. A portable CMM arm was used to dimensionally
inspect the segment to the extent possible and the data was best Figure 36. Color scale for Figures 38-42

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 Figure 40. Deviation plot showing the hub and OD

Figure 37. Cover side deviation plot

Figure 41. Deviation plot showing the eye of the impeller and
the bore

Figure 38. Hub side deviation plot

Figure 42. Deviation plot showing the bore 180o from the
printed flow paths
Figure 39. Deviation plot showing the cover and OD
The final checks on the DMLS impeller segment were to

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

assess the properties and to evaluate the microstructure and
integrity of the material. Table 1 shows the results of tensile
tests performed on samples that were printed an processed with
the segment. Samples were printed in both the horizontal and
vertical directions to determine if there are any significant
differences between the two directions. The values were
compared to the requirements of API 6ACRA for UNS N07718
– 120K. As can be observed in the table, the properties meet all
of the mechanical properties. Furthermore, the hardness was
33.5-35 HRC, which is acceptable per NACE MR0175-2009
and to API 6ACRA. API 6ACRA required impact testing at -
75oF (-59.4oC). The sample size was not large enough for a full
size Charpy test bar, so the ft-lb requirement was not able to be
determined, but the lateral expansion should not change
drastically with bar size. Per the specification, the lateral
expansion should be 0.015”(.38mm). For the purposes of the
authors company, tests were performed at room temperature Figure 43. Cross section of the horizontally built DMLS
and -150oF(-101oC). For both temperatures and both build sample (waterless Kalling’s etchant)

Table 1. Comparison of mechanical properties in both build

directions

directions, the lowest lateral expansion measured was

0.021”(.508mm). The samples were then cross sectioned to
determine the presence of any porosity and to evaluate the
microstructure. This is part of the API 6ACRA requirements as
well. The structure is expected to be free of Laves phases,
acicular phases, and continuous phases at grain boundaries.
Beyond those requirements, being HIP’d material for a high
stress application, the amount of porosity is expected to be very
low or nonexistent. Figure 43 shows an image of the Figure 44. Cross section of the vertically build DMLS sample
horizontal sample and Figure 44 shows the vertical sample. As (waterless Kalling’s etchant)
can be observed, there are no deleterious phases and no
continuous phases. There are no indications of porosity. With
respect to the material, all these tests were found to be well
within the expected limits.

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

 Figure 45. Comparisons to a brazed impeller

RESULTS AND DISCUSSTION There are some surprising results for a similar analysis of
the lead times for each method. Once again, these were based
The original intent of this paper was to present identical off of quotes from the suppliers. All of the methods were as
data sets for each manufacturing method described. Being that fast or faster than the standard time for the brazed impeller
each of these is in a different state of completion, it is difficult (green bars in Figure 45). The DMLS impeller has a major
to make direct or absolute comparisons as planned. At this advantage in this area and was half the time of the standard.
point in time, some initial estimates of the costs, lead times, and The segment that was made was reported to take just under two
tolerances achieved can be discussed. days to print and a full impeller would not take significantly
When each supplier provided a quote for the longer. Most of the time and cost of this method is tied to the
developmental work for this project, they were also asked to build height and not the extent or complexity required in the
provide an estimate of the cost if the impeller were for horizontal plane. EDM was the second shortest time and was
production. The in house processing was assumed to be the 75% of the time of the brazed impeller. Once again, both of
same for each method, regardless of incoming condition. The these could be faster if the equipment was in house, as most of
blue bars in Figure 45 are the comparison of the cost for each the lead time is tied up in scheduling rather than actual work.
method relative to a 17-4PH brazed impeller. In this While investment casting does also have a the scheduling issue,
comparison, the investment cast impeller using the 3-D printed there is a significant time in processing these impellers. The
pattern and invested flow path (SLA IC) was the least costly pattern print time is similar to the build time of the DMLS
and was actually less expensive than the brazed impeller. impeller, the investing process can take a week, heat treating
Using a ceramic core in the mold (Core IC) increases the cost can take a week, and upgrade cycles could take significant
such that it is about 50% higher cost than the brazed impeller. time. The ceramic core also adds significant time, so the
The EDM and DMLS are approximately 30% and 45% more invested flow path option is expected to be faster, but the
costly, respectively. These could be considered over estimates, supplier quoted the same lead times for both.
because these impellers would be coming in with much less It was mentioned above that the tolerances the suppliers
finish machining required. The DMLS impeller is also a more were given were 5 –axis milling tolerances. It’s not shocking
expensive material. The equipment for these two processes are that EDM showed the best performance in meeting those
also within reason to purchase and develop in house, which also requirements. It is a mature, entirely subtractive process that is
should reduce the costs. computer controlled. It easily met all of the dimensions that are

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

typically checked in a production impeller. Based on the
quoted tolerances and preliminary results, the DMLS impeller API Standard 6ACRA, 2015, Age-hardened Nickel-based
was slightly better than the investment cast impellers. While Alloys for Oil and Gas Drilling and Production Equipment,
the suppliers may be able to tighten the ranges up on future 1st edition, American Petroleum Institute, Washington, DC.
impellers, it is highly unlikely they will be able to match the
tolerance requested. Aerodynamics engineers at the authors ASME, 1985, Evolution of HIP: Commemorating the First Hot
company are working on a sensitivity study to determine how and Cold Isostatic Processing Vessels, American Society
wide the tolerances can be before any significant changes to the of Mechanical Engineers.
performance are expected. Early indications are that the
acceptable tolerances for design will be within the range of Beuth, J., 2015, Process Mapping for Additive Manufacturing,
what is achievable with these processes. Additive Manufacturing for Small Manufacturers,
Canonsburg, Pa, Lecture
CONCLUSIONS
Bhavsar, R.B., Collins, A, and Silverman, S., 2001, Use of
Overall, all of the processes being evaluated in this work Alloy 718 and 725 in Oil and Gas Industry, Superalloys
are capable of producing impellers in this size range. It is truly 718, 626, 706 and Various Derivatives, TMS, pp 47-55.
a matter of perspective if one was to rank them in terms of
suitability. If the situation was different and the same design of Brancher, C., 2015, Private Communication.
impeller was mass produced, HIP’d powder metal and
investment casting suppliers could perfect the required tooling Bourell, D.L., Beaman, J.J, Leu, M.C., Rosen, D.L., 2009, A
and would be difficult to match with additive manufacturing or Brief History of Additive Manufacturing and the 2009
EDM for cost and possibly lead time. Conversely, each of Roadmap for Additive Manufacturing: Looking Back and
these is in a different state of readiness with respect to Looking Ahead, RapidTech 2009: US-TURKEY
implementation in a custom, one off type application. From Workshop on Rapid Technologies
this perspective, EDM is clearly the most advanced and ready
for application and in fact, the authors company is in the Ding, J., Colegrove, P., Mehnen, J., Ganguly, S., et.al., 2011,
process of submitting a quotation for a compressor that includes Thermo-mechanical analysis of Wire and Arc Additive
impellers of this type that would most likely be made using Layer Manufacturing process on large multi-layer parts,
EDM. The work on the other processes hasn’t progressed far Computational Materials Science, 50(12), pp 3315-3322.
enough for hard conclusions, but investment casting and
additive manufacturing appear to be very viable. Additive, in EOS, 2014, Materials Brochure
particular, is exciting, because it is fairly new and the surface
has barely been scratched with respect to what it might be Ex One, M-Flex Product Brochure, 2014.
capable of.
Herderick, E, 2011, Additive Manufacturing of Metals: A
Review, Materials Science and Technology (MS&T) 2011,
Columbus, OH
REFERENCES
Higgs, F, 2015, Mechanics of Additive Manufacturing Powder
Allison, T.C., Rimpel, A.M., Pelton, R., Moore, J.J., Wilkes, Spreading, Additive Manufacturing for Small
J.C., Wygant, K., 2014, Manufacturing and Testing Manufacturers, Canonsburg, Pa, Lecture
Experience With Direct Metal Laser Sintering for Closed
Centrifugal Compressor Impeller, 43rd Turbomachinery Hunt, L.B., 1980, The Long History of Lost Wax Casting:
and 30th Pump Users Symposia Over Five Thousand Years of Art and Craftsmanship, Gold
Bulletin, , Volume 13, pp 63-79.
Angelo, P.C., 2008, Powder Metallurgy: Science, Technology,
and Applications, Prentice-Hall of India Pvt.Ltd Jameson, E.C., 2001, Electrical Discharge Machining, Society
of Manufacturing Engineers, pp 1-21.
Anonymous, 1965, History and Development in the techniques and
practice of spark Erosion machining, Spar Catron Limited, Lüdtke, K.H., 2004, Process Centrifugal Compressors: Basics,
Gloucester, UK, pp.6 Function, Operation, Design, Springer-Verlag Berlin
Heidelberg, pp. 76-77.
API Standard 617, 2014, Axial and Centrifugal Compressors
and Expander-compressors, 8th Edition, American
Petroleum Institute, Washington D.C.

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station

Murr, L.E., Martinez, E., Hernandez, J., Collins, S., Krista N.,
Amato, K.N., Gaytan, S.M., and Shindo, P.W., 2012,
Microstructures and Properties of 17-4 PH Stainless Steel
Fabricated by Selective Laser Melting, Journal of Materials
Research and Technology, 1(3), pp167-177

NACE Standard MR0175/ISO 15156, 2009, Petroleum and

natural gas industries - Materials for use in H2S-containing
Environments in oil and gas production - Part 3: Cracking-
resistant CRAs and other alloys, 2nd edition, NACE
International, Houston, TX.

Starr, T., Rafi, K., Stucker, B., and Scherzer, C.M., 2012,
Controlling Phase Composition in Selective Laser Melted
Stainless Steels, Proceedings of 2012 Solid Freeform
Fabrication Symposium, the University of Texas, Austin.

Ramakrishnan, P, 1983, History of Powder Metallurgy, Indian

Journal of History of Science, 18(1), pp 109-114.

Webzell, S., 2001, The First Steps into EDM in Machinery,

Findlay Publishing LTD, Kent, pp 41.

ACKNOWLEDGEMENTS

The authors would like to that Elliott Group for allowing

this work to be published. We would also like to thank the
numerous suppliers that provided valuable insight into the
processes and realistic evaluations of the suitability of the
application for their process.

Copyright© 2016 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station