Proceedings of the Twenty-Ninth International Pump Users Symposium

October 1-3, 2013, Houston, Texas

INFLUENCE OF IMPELLER SUCTION SPECIFIC SPEED ON VIBRATION PERFORMANCE

David Cowan Thomas Liebner

Hydraulics Engineer Hydraulics Engineer
ITT Goulds Pumps ITT Goulds Pumps
Basingstoke, Hampshire, UK Seneca Falls, NY, USA

Simon Bradshaw
Director of API Product Development & Technology
ITT Goulds Pumps
Seneca Falls, NY, USA

David Cowan is a Hydraulics Engineer Simon Bradshaw is the Director of API

with ITT Goulds Pumps responsible for Product Development & Technology for
applied research and hydraulic design of ITT Goulds Pumps, in Seneca Falls NY.
engineered API process pumps. His His responsibilities include the design
responsibilities include the development and development of new products and
and analysis of new and existing hydraulic processes. Prior to joining ITT Goulds,
products through traditional and he worked for both Sulzer Pumps and
computational methods. He is also jointly responsible for Weir Pumps, where he held various positions of engineering
continuous development of the computational fluid dynamic and contractual responsibility. Additionally he has supported
analysis techniques. Prior to joining ITT Goulds, he worked as the Hydraulic Institute in the development of pump standards
a Hydraulic Engineer for ClydeUnion Pumps. and best practice guides.

Mr. Cowan has a B.Sc. in Aeronautical Engineering from the Mr. Bradshaw has a BEng (Hons) degree (Mechanical
University of Glasgow. Engineering) from Heriot Watt University. He is a registered
Chartered Engineer in the UK and a member of the Institute of
Thomas Liebner is a Hydraulics Engineering Designers.
Engineer with ITT Goulds Pumps
responsible for applied research and
hydraulic design of engineered API ABSTRACT
process pumps. His responsibilities The most commonly used hard limitation for pump suction
include – new product design, specific speed is 11,000 (US units). This hard limit grew out of
computational modeling, and hydraulic the recommendations from a 1982 reliability study by J.L.
analysis for performance prediction. Hallam (Hallam 1982). Concomitant testing of the vibration
performance of an OH2 4x6-11 pump was made with impellers
Dr. Liebner has a B.S. in Mechanical and Aerospace Eng. from designed for different suction specific speeds (Lobanoff and
SUNY at Buffalo. He completed his studies for his doctorate in Ross 1985). This study showed that all things being equal, a
Mechanical Engineering at Penn State University where he strong relationship existed between suction specific speed and
performed experimental and computational modeling of the pump vibration at off BEP operation.
particle adhesion and their liberation and entrainment in air jets.
Given the significant changes in impeller design methods
and computational tools in the subsequent three decades, this
paper seeks to investigate how these new methods/tools have
affected the relationship between suction specific speed and the
pump vibration.

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 1

 Experiments are performed using a series of impellers The landmark paper by Warren Fraser (Fraser 1981),
designed for different suction specific speeds using modern brought the consequences of relying on large impeller inlet
design techniques. These impellers are mounted in a subject diameters into focus. Pump users had already become
test pump which is also an OH2 4x6-11 in order to achieve increasingly concerned that while such designs minimized plant
equivalency with the prior testing. Vibration performance over 1st cost, it was at the price of reliability and overall life cycle
the pump operating range is recorded. The results are cost. However, there was no large scale study available of the
complemented with computational fluid dynamic (CFD) phenomenon in an actual pump population and hence the nature
analysis to further examine the performance of each impeller. of the trade-off between suction performance and reliability
was unclear.
BACKGROUND
The suction performance of a centrifugal pump is an This changed when Jerry Hallam (Hallam 1982) published
extremely important consideration for optimal pump the results of a large scale reliability study of 480 pumps over a
performance. Good suction performance allows for the use of 5 year period at the Amoco Texas City refinery. He found that
smaller piping, lower tank elevations, less excavation and a the reliability of a pump was meaningfully related to its suction
general optimization of plant design. These optimizations can specific speed (Nss). Specifically pumps with a Nss > 11,000
lead to significant 1st cost savings. (S > 213) failed twice as often compared to lower suction
specific speed pumps. Figure 2 shows the failure rate vs.
In the 1950's to 1980's the impeller design methods suction specific speed.
available to pump designers were more limited than they are 1
today. Impeller designs from that era were notable for their
achievement of good suction performance through the
deployment of large impeller inlet diameters (D1). It was not Failure frequency
understood until later that the enlarging of the impeller inlet
diameter caused impairment of the impeller performance at 0.5
flow rates lower than the best efficiency point (BEP). This
impairment exhibited itself as significantly increased vibration
and in some extreme cases an unstable NPSHr characteristic.

Suction specific speed ranges (US units)

Figure 2: Failure frequency vs. suction specific speed.
D1
D1 Hallam concluded: "This study indicates that caution
should be exercised when purchasing hydrocarbon or small
Figure 1a: Effect of larger D1 on suction recirculation strength water pumps with a Nss greater than 11,000 unless operation is
closely controlled near BEP."

This conclusion was supported by the results of testing an

35
OH2 configuration 4x6-11 (100x150-280) pump in the book
Centrifugal Pumps:Design & Application (Lobanoff and Ross
25
1985). For this testing a series of eight impellers with differing
NPSHr

suction specific speeds were designed and tested at 3560 RPM.

15
The range of suction specific speeds varied from Nss = 7000 (S
= 135) to Nss = 20,000 (S = 387). For each impeller the flow
5
was varied until the pump vibration level exceeded the API 610
0 20 40 60 80 100 120 140
allowable level of 0.3 inches/sec (7.6 mm/s) peak. Those
Pump Flow % of BEP
limiting flow rates are shown for each impeller in Figure 3.
Figure 1b: Unstable NPSHr characteristic.

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 2

 INTRODUCTION
55
16 The purpose of this paper is to revisit the testing reported
by Lobanoff and Ross in 1985. The reasons for doing so are
14 primarily:
45
• changes to impeller design techniques, and
12 • improved design and construction standards.

NPSHr (m)
35 10
NPSHr (ft)

Impeller design techniques

Stable Operation
Impeller design techniques and tools have improved
Window 8
significantly in the last 30 years allowing impellers to attain a
25 required suction performance while minimizing the increase in
6
impeller inlet diameter. While not intended to be an exhaustive
4 list, some of the design options available to today’s designers
15 include:
2 • Small incidence blade angles coupled with small blade
20000 (387)
and approach flow angles (for better NPSH behavior at
5 0 part-load operation).
0 20 40 60 80 100 120 140 • Low blade loadings in the inlet region up to the
Pump Flow % of BEP impeller throat area. These help prevent the formation
Figure 3: Stable window according to Lobanoff & Ross. of low pressure zones where cavitation will begin.
• S shaped developments of the impeller camber line in
The testing showed that the impeller operating range with order to achieve the required impeller throat area while
acceptable vibration characteristics was strongly related to minimizing the eye diameter.
suction specific speed. • Backward swept blades to reduce the volume of any
cavitation that develops at the leading edge.
In the years following the publication of Hallam’s work the • Impeller leading edge carried well forward at the
Nss = 11,000 (S = 213) limit was widely adopted as a hard limit impeller hub in order to reduce the formation of
in the oil and gas industry to the extent that it is rare to see a cavitation at part load operation.
specification that does not invoke it in some form. It is • The deployment of better controlled leading edge
common to see the limit applied rigorously to the extent that profiles. These profiles effectively limit the leading
(for example) a pump with Nss = 10,950 (S = 212), is viewed edge pressure spikes and are less sensitive to part load
as acceptable while a pump with Nss = 11,050 (S = 214), is operation. For example prior research by the author's
viewed as unacceptable. company (Balasubramanian et al. 2011) has shown
that optimized impeller leading edge profiles improve
A number of authors have over the years studied and suction specific speed without requiring larger
reported that the influence of suction specific speed on pump impeller inlets.
reliability is diminished [(Stoffel and Jaeger 1996), • Utilizing computational analysis techniques the
(Hirschberger and James 2009), (Hergt et. al. 1996), (Gulich. impeller inlet design can be optimized for a given set
2001) and (Balasubramanian et al. 2011)]. Central to their of conditions, thus allowing greater control and
claim was the premise that modern impeller design techniques, understanding of the flow and pressure characteristics
ceteris paribus, allowed attainment of higher suction specific in the impeller passageway.
speeds without resorting solely to enlargement of the impeller
inlet diameter. However none of this work has altered the
widespread view that the original Nss = 11,000 (S = 213)
number is the main criteria that should be used in assessing a
pump's quality.

It is noticeable (by its absence), that there has been no

similar follow-up large scale study of refinery pump reliability
in the past 30 years. This is a concern given the increased
emphasis on safety, life cycle cost and minimizing emissions.

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 3

Design and construction standards
Pump standards (e.g. API 610 11th edition), have 1.0E+04 API 610 App. K acceptance line
continued to evolve such that modern designs are more robust Test Pump
than designs existing in the 1980's.
1.0E+03 Older generation Pump

Specifically, the L3/d4 ratio has been reduced in order to

L3/d4 (in-1)
limit shaft deflection at the seal chamber to 0.002” (0.05mm)
under any operating condition. L3/d4 is calculated from the 1.0E+02
impeller overhang (L) divided by the shaft diameter at the
mechanical seal (d), see Figure 4. This mechanical constraint
was driven by the need to improve mechanical seal reliability 1.0E+01
and the use of L3/d4 as a cost factor weighting representing life-
cycle cost
1.0E+00
1.0E+01 1.0E+02 1.0E+03 1.0E+04 1.0E+05
QH/N (USGPM x ft / RPM)

Figure 5: Excerpt from API 610 11th edition Appendix K

API 610 7th edition (1989) also introduced the current

requirements for limiting the deflection of the pump under
specified nozzle loads including optional testing. API 610 9th
edition (2003) specifically prohibited the use of rear bearing
housing supports on OH2 pumps. This required an
improvement of the overall rigidity of the pump casing, bearing
frame and baseplate.

Figures 6 and 7 contrast the arrangement of a casing foot

typical of current designs with that of an older design.
Consequently the improved rigidity tends to improve overall
pump reliability and vibration performance.
Figure 4: L3/d4 for an overhung pump rotor

It is not unusual to see pumps designed to earlier versions

of API 610 having L3/d4 ratios that are 3x to 6x higher than the
industry average today. For example in a comparison between
the 4x6-11 (100x150-280) tested in this paper and a similar
pump from a model line designed to an earlier version of API
610, the older design had a L3/d4 of 213 in-1 (8.4 mm-1). This is
5x greater than the value of the pump tested for this paper
which has a L3/d4 of 42 in-1 (1.65 mm-1).

API 610 11th edition introduced non-binding criteria for

L /d in Appendix K of the standard. The criteria plots L3/d4
3 4

vs. a factor composed of the pump flow x head / speed. The

location of the test pump is plotted on the graph in Figure 5 as
compared with an older generation pump.

Figure 6: Pump foot fully compliant to API 610 11th edition.

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 4

Figure 7: Pump foot design typical of a pump designed to pre- Figure 8: Cross-sectional assembly of the Test Pump
API 7th edition standards.
The pump was installed in a standard testing station in the
Hence it is timely to examine how these changes have large hot water tank (LHWT) test loop of company’s R&D
affected the attainable acceptable flow range as it relates to facility. The test setup complied with HI 14.6 test standards.
suction specific speed. Figure 9 shows the test pump as installed in the test loop. It is
important to note that all test loop setups are temporary
TEST PUMP SETUP constructions and the vibration levels measured on the pump
The test pump selected is a 4x6-11 (100x150-280) in a will necessarily be higher than those achieved in the final site
single stage overhung configuration with centerline mount installation. The absence of a large permanent foundation and
(OH2). It is fully compliant with 11th edition of API 610. In grout reduces the ability of the test setup to attenuate these
terms of overall construction it is unremarkable though vibrations effectively. Additionally, all of the fluid energy
consistent with the current best practice for a full compliant imparted by the pump needs to be dissipated within the test
API 610 OH2 design. Figure 8 shows a cross-sectional view of loop. This tends to cause vibrations that are fed back to the
the test pump. pump, and in extreme cases acoustic resonances can occur in
the typically short pipe runs.
The characteristics of the test pump are tabled below:
Hydraulic Institute recognizes this fact in their vibration
Parameter Value standard 9.6.4 which has higher allowable levels for factory
Running Speed 3560 RPM testing than for site testing. API 610 makes no such distinction
BEP Head 450 ft (137 m) and requires the same low levels be achieved in the factory test
BEP Flow 1670 USGPM (380m3/h) loop as in the final permanent site installation.
BEP power @ 1.0 SG 232 HP (173 kW)
Specific Speed Ns (nq) 1489 (28.8)
Design Pressure 750 psig (51.7 barg)
Materials of Construction API 610 code S6
Shaft dia. @ mechanical seal 2.362” (60mm)
L3/d4 ratio 42 in-1 (1.65 mm-1)
Table 1: Test Pump Specifications

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 5

 These vibration values would be used to determine the
allowable operating range of each impeller.

IMPELLER DESIGN
For the test rig, four single entry end-suction impellers
were designed. Details of the key geometry information are
tabled below. Constraints were placed on the maximum outlet
width dimension to ensure each impeller could fit within the
standard 4x6-11 case being utilized as well as ensuring similar
radial thrust values.

The impellers were designed with varying suction specific

speed (Nss) constraints, notably 8000 through 15000, with the
intent to maintain a standard generated head and best efficiency
Figure 9a: Pump installed in the test loop flow rate.

Maintaining a similar meridional geometry between

impellers is not possible due to the large increases in suction
specific speed. As such, the impeller eye diameters gradually
increase causing differences in the overall meridional shape.

There was some variation in discharge angle and discharge

width between the different designs. B2 and β2 are strongly
dependent, and were adjusted to achieve the appropriate
discharge area while accommodating the variation in inlet
geometry.

The inlet diameter for the highest Nss impeller was almost
20% larger than the lowest Nss design. An overlay of each of
the impeller meridional shapes can be seen in figure 10.

Figure 9b: Pump installation (bearing housing view)

For the purposes of the testing, the following allowable

vibration levels were used in accordance with API 610 11th
edition:

Parameter Vibration level

Overall unfiltered in the flow range 0.12 in/s (3.0 mm/s)
70% to 120% of BEP
Any discrete frequency in the flow 0.08 in/s (2.0 mm/s)
range 70% to 120% of BEP
Overall unfiltered in the flow range 0.156 in/s (4.0 mm/s)
MCSF to < 70% and > 120% of
BEP
Any discrete frequency in the flow 0.10 in/s (2.6 mm/s)
range MCSF to < 70% and > 120%
of BEP
Table 2: Vibration criteria for acceptable performance under Figure 10: Overlay of meridional geometries
API 610 11th edition.

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 6

 Design Design Design Design
1 2 3 4
8000 11,000 13,000 15,000
Nominal Nss (S) (155) (213) (252) (290)
D2 Impeller outlet
11 11 11 11
diameter (in)
B2 Impeller outlet
1 0.9 0.85 0.95
width (in)
β2 Impeller vane
24 26.3 29 27.5
angle @ outlet (deg)
D1 Impeller inlet
4.9 5.3 5.5 5.8
eye diameter (in)
β1t Impeller vane
29 13.2 14.7 11.7
angle @ inlet (deg)
D1 / D2
Impeller inlet / 0.44 0.48 0.5 0.53
impeller outlet dia.
Table 3: Basic dimensions for the four impeller designs. Figure 11a: Nss= 8000 nominal impeller

As discussed previously, in research by the authors

company (Balasubramanian et al. 2011), it was demonstrated
that cavitation is better controlled and higher Nss values
achieved by employment of optimized leading edge profiles. As
such, a parabolic leading edge profile was adopted for each of
these designs, but the benefit of the leading edge profile was
not considered in the impeller design calculations (and impeller
design system utilized for these designs), as the exact
improvement that could be realized was uncertain.

To reduce variability between the impellers, a constant

wear ring diameter has been used. Wear ring clearances were in
conformance to API 610 11th edition Table 6.

In standardizing the wear ring geometry the consequential

volumetric loss is constant across the four impellers. This Figure 11b: Nss= 11000 nominal impeller
ensures a standard fluid damping effect. Wear ring length has
been held constant across the impellers to normalize the
favorable centering “Lomakin” effect. While API 610 does not
allow this effect to be considered when calculating the shaft
deflection, it does provide some additional stiffness and
damping and hence it was necessary to keep it constant for all
impeller designs.

The impellers were manufactured directly from the 3D

model using rapid investment casting techniques (pattern less
manufacture) and the cast impellers using SLA rapid
prototyping process. Pictures of the resulting impellers are
shown in Figures 11a to 11d

Figure 11c: Nss= 13000 nominal impeller

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 7

 size is chosen as 0.015x the maximum length of the passage.
This allowed for an average of 5 cells across the width of the
passage. The mesh size for the four models varied between
450,000 and 600,000 nodes.

A grid refinement study was performed for one of the

design cases to ensure that the mesh was properly constructed
and would produce results of sufficient accuracy. Three
meshes of increasing refinement were utilized. The results of
this sensitivity study are described in table 4.

Figure 11d: Nss= 15000 nominal impeller

COMPUTATIONAL FLUID DYNAMICS (CFD)

To verify the hydraulic designs, a computational study,
conducted within the framework of the ANSYS-CFX solver,
[ANSYS CFX-14.5, 2012], was undertaken. The initial
motivation for the computational analysis was to ensure that
each design achieved its target Nss at the best efficiency point
(BEP) while maintaining comparable performance.
Additionally, the CFD results can provide insight into the
development of cavitation on the leading edge of the blade and
into the onset of recirculation within the impeller. The onset of
suction side recirculation as the flow rate through the impeller
is reduced should signal an increase in vibration characteristics.

For simplicity, a single blade-centered passage with a

steady-state flow condition was utilized for this analysis. This
has certain limitations as it neglects the effect of the casing and Figure 12. Sample mesh used during computational study.
any unsteady characteristics including blade pass and system
response. However, it makes the size of the mesh and the time As described in the book Centrifugal Pumps by Johann
to convergence manageable such that multiple flow conditions Güilich, approximating a grid independent solution (Hnu), the
could be analyzed. discretization errors (eh) and the order (p) of the solution can be
calculated utilizing solutions of grid sizes that differ by a factor
MESH STRUCTURE of 2. The equations are listed below.
An unstructured mesh with tetrahedral mesh elements was
1 −
generated using the Simmetrix grid generation software ≈
2 −
[Simmetrix MeshSim, 2012]. A boundary layer mesh with
hexahedral mesh elements was placed on wall surfaces. A
minimum ∆y was established such that the average y+ value on −
=
the vane surface was between 10-20. The k-ω model with the 2 −1
shear stress transport (SST) adaptation is utilized to model the
turbulence and near-wall structures. For this turbulence model,
−
a y+ of less than 30 has produced repeatable results while ≈ +
2 −1
sufficiently capturing the near-wall characteristics. The global

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 8

 Mesh Size Predicted Head Predicted 3% NPSHr Multiple runs were conducted for each of the impellers.
Nodes Ho/ Hnu NPSHr/NPSHrnu Four different flow rates were investigated at 60%, 80%, 100%,
164,000 1 1.19 and 120% of the target best efficiency point (BEP) for each of
332,000 0.99 1.05 the four designs. At each of these flow rates, the inlet total
590,000 1 1.01 pressure was gradually reduced to compute the head drop
p -2 -2 performance curves, essentially simulating a typical NPSH test
eh 5.33 -2.67 run. Figure 14 demonstrates a typical head drop curve
Hnu / NPSHrnu 509.3 13.83 predicted by the computational analysis.
Table 4: Mesh refinement sensitivity study

CFD SOLVER CRITERIA

The analysis of the four designs was performed utilizing Nss (nominal) = 15000 (290)
the ANSYS-CFX solver. The homogeneous two-phase mixture Q @ 100% BEP
model is employed to model cavitation. The cavitation model is
based on the Rayleigh-Plesset equation with source terms for 100%
the generation and destruction (vaporization and condensation)
of vapor bubbles [Bakir et al., 2004]. The model solves for two- 95%

TDH [ft]
phases, vapor phase (αvapor) and liquid phase (αwater), at each
control volume location, with the sum of both phases equal to
90% NPSH 3%
one (αvapor+αwater=1) at each location. The basic assumption of
the model is that all phases share the same velocity and a Break Point
mixture equation is solved for the conservation of momentum. 85%
High resolution fluxes are chosen for the discretization of mean
flow and turbulence equations. The shear stress transport (SST) 80%
turbulence model is used for modeling turbulence. 0 50 100 150
Suction Pressure [ft]
Simulations are performed for a single passage of the
Figure 14: Typical head breakdown curve
impeller geometry as shown in Figure 13. For the analysis, no
slip boundary conditions are applied at the hub, shroud and
blade; total pressure is set at the inlet with the volume fraction
PREDICTION OF RECIRCULATION BY THE FRASER
of water as 1.0 and vapor as 0.0; mass flow rate is specified at METHOD AND CFD
the exit; and rotational periodicity is applied at the periodic Warren Fraser (Fraser 1981), provides an estimate for the
interfaces (passage boundaries) as shown in Figure 7. onset of suction recirculation within centrifugal pumps based
on major dimensions within the impeller. The equation for this
Convergence for the velocity and momentum residuals is shown below. There is mention made in the paper that the
was determined below an RMS value of 10-4. Each of the trial equation was developed using observations of suction
runs required between 200 and 400 iterations to achieve recirculation in a special test pump equipped with a transparent
convergence. suction pipe. It is not clear from the paper as to exactly how
observations made on test pumps were correlated with the
resulting formula. Specifically there is no mention as to how
Outlet extensive the recirculation zone must be to assure experimental
observation. This makes it difficult to correlate with the CFD
determinations of the recirculation zones.
Periodic
#$ %#$ & '($ & )* 1
Interfaces ! " = ∗ for D1/D2 > 0.5
+,../ $

Thus for the purposes of comparison the impeller under

Inlet CFD analysis is deemed to be recirculating when the
Figure 13: Single-passage CFD model for analysis recirculation zone extends upstream of the leading edge of the

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 9

impeller vane, which presumably would have been observable
in Warren Fraser’s test pump.

For each impeller design, single phase CFD runs were

performed where the flow rate was reduced in 5% increments
from BEP. Figures 15a to 15d show samples of the resulting
output. The results were compared for each impeller and a
determination made regarding the flow at which recirculation
extended beyond the vane leading edge. This flow rate was
deemed to be “recirculation onset”

Predictions for the onset of recirculation are shown in

Table 5 for both methods. The flowrate at which suction side
recirculation occurs increases with increasing suction specific
Figure 15b.Recirculation cell ahead of vane at 65% BEP Flow,
speed. This is to be expected as the higher suction specific
11,000 Nss (S=213) design.
speed impellers have larger impeller inlet eye (D1) diameters.

It can be seen that the values predicted by CFD and

Fraser’s equation show substantive agreement. This appears to
validate the choice of CFD recirculation criteria.

Nominal Fraser CFD

Suction Suction Recirc. Suction Recirc.
Specific Speed (% of BEP) (% of BEP)
8000 (155) 48% ≈48%
11,000 (213) 60% ≈63%
13,000 (252) 66% ≈63%
15,000 (290) 75% ≈74%
Table 5: Recirculation predictions based on Fraser & CFD

Figure 15c.Recirculation cell ahead of vane at 65% BEP Flow,

13,000 Nss (S=252) design.

Figure 15a. Small recirculation cell ahead of vane at 50% BEP

Flow, 8000 Nss (S=155) design.

Figure 15d.Recirculation cell ahead of vane at 75% BEP Flow,

15,000 Nss (S=290) design.

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 10

EXPERIMENTAL RESULTS leading edge profile. From previous testing this
Each impeller was subject to a full performance and NPSH NPSHr improvement was believed to be
test based on the criteria set out in API 610. The results of each approximately 18%.
impeller performance compared to the design target are shown
in Figures 16 and 17. These results confirmed the prediction of In addition to comparing the target NPSHr values with the
the CFD. Head and efficiency agreed within 5% between CFD tested values, the tested NPSHr values were compared to the
and test when the calculated casing losses were combined with CFD simulation.
the CFD values.
For flow values higher than the onset of suction
recirculation, the values of NPSHr from CFD showed
Nss 15000 Nss 13000 Nss 11000
substantive agreement with the tested NPSHr values. NPSHr
Nss 8000 CFD 15000 CFD 13000
predictions obtained via CFD are typically 5-15% lower than
CFD 11000 CFD 8000 those attained on test. This can be accounted for with casting
150 and surface imperfections, unsteady flow features including
140 vane pass contributions and system response characteristics,
% Best Efficiency Head

and non-uniform inlet flow fields

130
120 At lower flow rates the tested NPSHr values diverged from
the CFD predicted NPSHr values due to impeller-casing
110
interaction in recirculation which the impeller only CFD is
100 unable to simulate. Results of predicted 3% head drop versus
flow are shown in Figures 17a and 17b below.
90
80
Nss - 11000 CFD Nss - 8000 CFD
20 40 60 80 100 120 140 Nss - 11000 Test Nss - 8000 Test
% Best Efficiency Flowrate 50
Figure 16: Experimental Testing and CFD analysis (note the 45
CFD results show the impeller performance only.) 40
35
NPSH3 (ft)

Tested NPSHr was 21% to 33% lower than the original

nominal target values. These values and the corresponding 30
suction specific speed are shown in Table 7. 25
20
Nominal Suction Target NPSHr Tested NPSHr Tested Suction
Specific Speed Specific Speed 15
@BEP ft (m) @ BEP ft (m)
8000 (155) 47.8 (14.6) 37.4 (11.4) 9568 (185) 10
5
11,000 (213) 31.3 (9.5) 21.1 (6.4) 14,776 (286)
0.0 0.5 1.0 1.5
13,000 (252) 25.0 (7.6) 17.6 (5.4) 17,066 (331) Flow rate relative to BEP
15,000 (290) 20.7 (6.3) 16.4 (5.0) 17,841 (346) Figure 17a: Computational & test results for 3% NPSHr (8000
Table 7: Nominal target vs. tested NPSH and Nss and 11,000 nominal Nss impellers)
There are two main reasons for this difference:

1. The acceptance tolerance for NPSHr allows no

positive tolerance. Thus impellers are designed to
achieve lower than the target NPSHr by
approximately 14% to allow for manufacturing
variances and uncertainty of design.

2. The design methods used did not take into account the
improvement achieved through use of the parabolic

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 11

 exceeded the acceptable vibration level at 86% of BEP due to
Nss - 15000 CFD Nss - 13000 CFD
Nss - 15000 Test Nss - 13000 Test vane pass.
25
The 13,000 nominal (17,066 actual) impeller design
exceeded the acceptable vibration level at 76% of BEP due to
20 vane pass.
NPSH3 (ft)

The 11,000 nominal (14,766 actual) and 8000 nominal

15
(9568 actual) impeller designs did not exceed any vibration
criteria throughout the entire operating range tested from 25%
10 to approximately 140% of BEP.

5 Allowable vibration 15000 Nss vibration

0.0 0.5 1.0 1.5
13000 Nss vibration 11000 Nss vibration
Flow rate relative to BEP
8000 Nss vibration
Figure 17b: Computational & test results for 3% NPSHr
(13,000 and 15,000 nominal Nss impellers) 0.18
0.16
Vibration values for the both overall vibrations and vane
0.14
pass are plotted on Figures 18 and 19. (Vibration levels due to Velocity (in/s RMS)
mechanical sources, specifically 1x and 2x running speed were 0.12
less than the 0.08 in/s (2.0 mm/s) allowable level for discreet 0.1
frequencies and were therefore not used as an acceptance
0.08
criteria).
0.06

Allowable vibration 15000 Nss vibration 0.04

13000 Nss vibration 11000 Nss vibration 0.02
8000 Nss vibration 0
0.50 1 1.5
0.18 Flow rate relative to BEP
0.16 Figure 19: Vane pass frequency vibration level on test.
0.14
It should be noted that the test setup was of average
Velocity (in/s RMS)

0.12 quality. The test pedestals were clamped to the test base and
0.1 were not specifically designed for the pump being tested.
Consequently the vibration levels achieved could be
0.08
meaningfully improved by refinement of the test setup.
0.06 Similarly for a pump permanently installed in the field on a
0.04 grouted baseplate, we would also expect a reduction in
vibration levels achieved on test.
0.02
0
0.5 0 1 1.5
Flow rate relative to BEP
Figure 18: Overall vibration level on test.

The results show that all impellers exhibit a rising

vibration level away from the impeller shock-less flow point.
The Nss 15,000 nominal (17,841 actual) impeller design

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 12

 The pump was also equipped with a suction tapping
adjacent to the eye of the impeller (see Figure 20). The Nominal Fraser CFD Test
Suction Recirc. Suction Recirc. Suction Recirc.
pressure at this tapping was logged and compared to the Suction
(% of BEP) (% of BEP) (% of BEP)
pressure at the suction tapping at a location 2D in front of the Specific Speed
suction flange. The purpose of this measurement was to 8000 (155) 48% ≈48% ≈62%
ascertain when suction recirculation occurred and compare this 11,000 (213) 60% ≈63% ≈65%
result with the values predicted by Fraser and CFD. 13,000 (252) 66% ≈63% ≈67%
15,000 (290) 75% ≈74% ≈64%
Table 8: Recirculation recorded on test & compared to the
Figure 21 shows a plot of suction pressure recorded at both
predictions from Table 5.
tappings, normalized for the area differences and corrected for
friction losses between the two locations. This shows an
apparent recirculation inflection point for all the designs
between 60% and 70% of BEP as shown in Table 8.
Nss - 8000 Nss - 11000
Nss - 13000 Nss - 15000
1.25

1.20

Pcasing / Ppiping
1.15

1.10

1.05

1.00

Figure 20: Location of the casing suction tapping. 0.95

0% 20% 40% 60% 80% 100%
While the tapping clearly records the presence of suction Pump Flow % of BEP
recirculation, the number of readings taken and their scatter Figure 21: Ratio of suction performance at casing to upstream
precluded this method from providing an accurate indication of in the suction pipe.
the onset of recirculation. As can be seen from Table 8, there
was significant disagreement between the both Fraser and CFD
as compared to the values obtained during this test.

With more time and refinement of the technique (in

particular a much higher density of measurement points), we
believe it could yield a more accurate indication, however for
the purposes of this paper the technique will not be discussed
further.

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 13

DISCUSSION Thus in order to achieve an acceptable operating range with
The test results can be converted into an operational range these impellers, the Nss target needs to be reduced.
chart similar in style to that used by Lobanoff & Ross. Figure
22 shows this in detail. Compared to Figure 3, the stable
Existing SGsT line Nss = 17841 (345)
operating window is substantively larger and only closes at
Nss = 17066 (330) Nss = 14766 (286)
suction specific speeds far above those typically specified by
Nss = 9568 (185) Future SGsT line
most users.
18000
55 Not attainable
16000
Stable Operation with

Attainable Pump Nss (US units)

Window 14000 acceptable
45 performance
12000
10000
35 8000
NPSHr (ft)

6000
Attainable &
25 4000 acceptable
performance
2000
0
15
0 1000 2000 3000 4000
Pump Ns (US units)
Figure 23: Trade off line (SGsT line) for Ns vs. Nss
5
0 20 40 60 80 100 120 140
Pump Flow % of BEP SUMMARY AND CONCLUSIONS
The testing results confirm that substantive improvements
Figure 22: Stable operating window for the impellers tested
in stable operating range vs. suction specific speed are
This result was compared to an existing internal guideline achievable utilizing modern impeller design and pump
(known as the SGsT line), used by the author’s company construction standards. The realizable performance (Nss =
governing achievable end suction impeller designs with 14,776 (S = 286) with acceptable vibration characteristics), is
acceptable vibration and stability characteristics. This so far removed from what most users consider attainable that it
guideline is reproduced in Figure 23 with the impellers should give pause for consideration as to whether Nss =11,000
designed for this paper plotted on it. The main solid blue line is always the appropriate choice for medium to low specific
indicates the dividing line between acceptable and non- speed impellers.
acceptable performance.
The authors would recommend that users consider
For the specific speed of the test pump, the existing SGsT adopting their own version of the SGsT line. The specific
line value of approximately 13,000 nominal suction specific speed of the pump is an important determinant of the attainable
speed would be appear to be easily attainable with the currently Nss (with reliability) and needs to be recognized.
available design tools. As the state of the art continues to
improve it may be possible to revise the SGsT line upwards in If correctly applied the use of such enhanced designs allow
the future as shown on Figure 23. the designers of processes utilizing pumps, increased flexibility
and the potential to realize a lower first cost with equal or even
It should be further noted that the SGsT line dips below improved reliability (if high specific speed pumps are specified
11,000 Nss for higher specific speed impellers. This is because with conservative SGsT limits), than is possible with current
at these higher specific speeds the ratio of impeller outlet one size fits all suction specific speed limit.
diameter to impeller inlet diameter (D2/D1) is significantly
reduced. As Warren Fraser (Fraser 1981) demonstrated, this Of equal importance is action from the users of pumping
ratio strongly determines when suction recirculation will occur. equipment to build on Hallam’ s work and provide an updated
large scale study of pump reliability for the 21st century.

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 14

NOMENCLATURE ACKNOWLEDGEMENTS
BEP = best efficiency point (flow rate) of the pump Special thanks to John Salerno (Jr.) for his help with all the
CFD = computational fluid dynamics necessary parts, Susan Sullivan for her successful impeller
API = American petroleum institute designs and: Denny Fenner, Marty Temple and Patricia
NPSHA = available net positive suction head BabowiczWebb for all their help with the extensive testing and
NPSH3 = net positive suction head at 3% head drop data logging.
NPSHr = net positive suction head required ( = NPSH3 )
Ns = specific speed (RPM, USGPM, ft)
Nss = suction specific speed (RPM, USGPM, ft)
S = suction specific speed (RPM, m3/hr, m)
Q = pump flow rate USGPM (m3/hr)
MCSF = minimum continuous stable flow
D1 = impeller eye diameter
β1 = impeller vane inlet angle
D2 = impeller outlet diameter
B2 = impeller outlet width
β2 = impeller vane outlet angle

REFERENCES
1. ANSYS-14.5 CFX Solver Theory Guide, 2012.
2. Hallam, J. L., Centrifugal Pumps: Which Suction Specific
Speeds are Acceptable?, Hydrocarbon Processing, April 1982
3. Lobanoff, V.S., Ross, R.R., Centrifugal Pumps: design &
application 2nd Edition, Figure 8-7
4. Gülich, J.-F., Selection Criteria for Suction Impeller of
Centrifugal Pumps, Parts 1 to 3 World Pumps January, March
and April 2001
5. Hergt, P., Nicklas, A., Mollenkopf, G., and Brodersen, S.,
1996, The Suction Performance of Centrifugal Pumps
Possibilities and Limits of Improvements, Proceedings of the
13th International Pump Users Symposium, 1996
6. Stoffel, B., Jaeger, R., 1996, Experimental Investigations in
Respect to the Relevance of Suction Specific Speed for the
Performance and Reliability of Centrifugal Pumps, Proceedings
of the 13th International Pump Users Symposium, 1996
7. Hirschberger, M., James, I., 2009, A Review of Nss
Limitations – New Opportunities, Proceedings of the 25th
International Pump Users Symposium, 2009
8. Balasubramanian, R., Bradshaw, S., Sabini, E., 2011,
Influence of Impeller Leading Edge Profiles on Cavitation and
Suction Performance, Proceedings of the 27th international
Pump Users Symposium 2011
9. Schiavello, B., and Visser, F. C., 2008, Pump Cavitation –
Various NPSHr Criteria, NPSHa Margins, and Impeller Life
Expectancy, Proceedings of the 24th International Pump Users
Symposium, 2008, Turbomachinery Laboratory, Department of
Mechanical Engineering, Texas A&M University, College
Station, TX.
12. Güilich, J-F., Centrifugal Pumps, Second Edition

Copyright© 2013 by Turbomachinery Laboratory, Texas A&M Engineering Experiment Station 15