303 Williams Ave Ste 123, Huntsville, AL 35801 Phone: 256-489-1480

http://www.simerics.com Fax: 256-489-1460

Simerics Inc.

White Paper 2014.04.16

Predicting Damage: Cavitation or Otherwise

Damage in fluid systems can occur for a variety of reasons. For any fluid system, with either gas or liquid,
jetting and pressure waves can potentially cause damage. For systems with moving parts, damage can be the
result of loads, creating solid to solid contact and erosion. And, for liquid systems, bubble formation and
collapse in fluid devices can degrade performance and severely damage components. Computational Fluid
Dynamics (CFD) can be used to predict the onset, location, and magnitude of these phenomena; indicate
potential damage; and help improve designs and operating conditions to improve performance and life.

Propeller with cavitation damage

INTRODUCTION

Damage in fluid systems is typically the result of one of

the following causes:

 Pressure waves
 Jetting
 Loads / Contact
 Bubble Collapse

This white paper discusses how these causes can be

identified and predicted using numerical simulation.

PRESSURE WAVES Figure 2: Linear function for the Bulk Modulus.

Fluid flow systems can generate pressure pulses, Figure 3 shows an example where the measured rate of
especially in systems where the forces driving the flow pressure rise in the cylinder of a piston pump depends
are themselves pulsating, as in a positive displacement on the rpm [2]. This rate dependence is attributed to the
pump. Figure 1 shows the pulsations being measured effects of dissolved vs. free gas. By using a gas model
and predicted in the outlet pipe of a piston pump. that computes the amount of free vs. dissolved gas
based on the equilibrium properties of the fluid, the CFD
simulation was found to product very good agreement
with the measured results as shown in Figure 4.

Figure 3: Measured Pressure rise vs rotation angle in a

piston pump at three rpms

Figure 1: Pressure ripple downstream of a piston pump:

predicted (PumpLinx) vs measured.

Compressibility: In order to realistically predict pressure

ripple, a CFD tool needs to accurately capture the
effective compressibility of the fluid. For liquids, this must
include the compressibility of the pure liquid, the
compressibility of any non-condensable gases, and any
vapor. For high pressure systems, the bulk modulus
should include the linear dependence on pressure [1], as
shown in Figure 2. All these effects were included in the
simulation shown in Figure 1 and the results show that
both the magnitude and wave shape are well captured.
Figure 4: Equilibrium Gas Model Predictions for the
Any non-condensable gas, such as air, should be pressure rise in a piston pump at a selected rpm [2].
modeled as compressible using the ideal gas law or real
gas properties. For highly dynamic systems, it may be
important to include the effects of both free and
dissolved gas.
Time-step size: Another critical parameter in accurately Mesh refinement: As with time-steps, the resolution of
capturing pressure ripples in a fluid system is the the mesh must be sufficiently fine to capture the details
selected size of the time-step. Simply stated, higher of any pressure waves traveling through the system. And
frequencies need smaller numerical time-steps to as with time-steps, a grid sensitivity study is strongly
resolve the higher harmonics. To resolve these recommended to determine at which point the solution is
frequencies, it is recommended to perform a sensitivity relatively insensitive to further refinement.
study by reducing the size of the time-step until both the
magnitude and wave shape no longer change to within Domain & Boundary Conditions: Finally, proper
an acceptable tolerance. selection and assignment of the location and parameters
associated with the boundaries in a CFD model is an
Figure 5 shows a sensitivity study of the effects of the absolute necessity for accurate results. If the specified
time-step size on the prediction of the pressure pressure at a model’s boundary does not correspond to
pulsations in a gerotor pumps. For this case, the default the physical values in the real system, the results will, at
time-step of 60 steps between tooth rotation (0.75 best, be the right answer to the wrong problem. For
degree of rotation per time-step) was increased to ten example, when modeling the outlet pipe of the piston
times that (corresponding to 0.075 degrees of rotation pump in Figure 1, the valve was modeled as an orifice
per time-step) before the results were qualified as time- with a fixed pressure downstream. This allowed pressure
step independent, with the results that the predictions upstream of the valve the freedom to vary. If instead, a
compare well with the measured pressure noise (Fig. 6). boundary upstream of the valve had been selected and
The cost of smaller time-steps is increased simulation specified as a constant pressure, it would have artificially
time. The ability to reduce the time-step size to get an damped the predicted pressure pulsations.
accurate results in a reasonable time requires a fast
CFD code. Pressure Pulse Damage: Given an accurate prediction
of the dynamic pressure in a system, the prediction can
subsequently be used to assess the potential for
damage. As an example, Figure 7 shows a simulation of
an external gear pump in which the pressure waves in
the high pressure port focus to a point on the housing.
This point corresponds precisely with observed damage.
The degree of damage depends on the concentrated
power of the pressure, relative to the strength of the
material. In PumpLinx, the instantaneous power in watts
can be output as a color map on the solid surfaces. In
the absence of an absolute threshold for damage, a
relative effect can be established and two designs can
be compared with the assumption that the one with a
lower wattage is less likely to exhibit damage.

Figure 5: Time-step sensitivity study for the simulation

of pressure ripple in a Gerotor pump

Figure 6: Measured vs. Predicted pressure ripple at 600 Figure 7: Focused pressure in a gerotor pump causing
time-steps per pocket (See figure 5) damage
JETTING

Another potential cause of damage in fluid systems is

jetting which impinges on a solid surface. To capture this
effect requires many of the same modeling prerequisites
as for pressure waves. Specifically, it requires an
accurate representation of the fluid properties,
sufficiently small time-steps and sufficient grid resolution
to prevent numerical diffusion of the jet that would occur
with a course mesh or time-step.

As with pressure waves, the correlation between

predicted jetting and physical damage depends on the
hardness of the target material. And, as with pressure
waves, the potential for damage can be determined by
comparison with experimental data, if available.
Observed damage can be used to determine the
threshold for damage. In the absence of experimental
data, the simulations can be used to compare the
relative potential for damage between different designs.

LOADS / CONTACT

The integrated fluid loads on a part can cause damage if

those loads result in parts rubbing against each other.
To fully capture this effect requires a model that provides
two-way coupling between the fluid simulation and a
structural simulation. Furthermore, this requires that the
fluid model adequately includes all the key physics, such
as any cavitation, to provide accurate pressures to the
structural model. If the pressures are not correct, the
structural model is meaningless. However, even in the
absence of a structural model, an accurate fluid
simulation can still be used to evaluate the potential for
damage due to fluid loads. Figure 6 shows and example
where the tilting torque on the outer rotor in a gerotor
pump was predicted as a function of rotation angle. This
information allows the comparison of different port
designs to determine which design was best to reduce
the torque and potential for rubbing. Figure 7 shows a
related calculation and measurements for the torque on
a spool valve at different flow rates [3]. Figure 7: Measured vs. predicted torque for a valve

In summary, CFD can be used to predict the static and

transient loads and torques on components in a fluid
system. By comparing the forces for various geometries
and operating conditions, the potential for damage can
typically be reduced by selecting the design with the
lowest magnitudes.

BUBBLE COLLAPSE

The collapse of vapor bubbles and the rapid

compression of non-condensable gases can cause
severe localized damage and/or noise. Figure 8 shows
an example of vapor induced damage in the timing
groove of the valve plate of a piston pump. An accurate
Figure 6: Predicted torques on a gerotor gear as a aeration/cavitation model is essential for simulating this
function of angle phenomenon.
 .

Figure 8: Cavitation damage on a piston pump valve

plate
Figure 9: Predicted vapor pocket formation in a gear
pump
Physical models for the formation and transport of vapor
bubbles have been available for at least the past
decade. However, despite their critical importance for
pump modeling, only a few codes offer robust cavitation
models because of the difficulty of integrating them into
a numerical scheme. Gas and liquid mixtures are
inherently difficult to simulate due to 1) the large density
differences between the two, and 2) the strong coupling
between the gas void fraction and the pressure. This is
where strong numerics can mean the difference between
a reasonable solution, non-physical results, or no results
at all.

The cavitation model in PumpLinx is based on the work

of Singhal, Athavale, Li, and Jiang [Singhal 2002]. The
PumpLinx adaptation includes improvements to enable
the introduction, transport and compressibility of non-
condensable gases. The strong numerics ( e.g. implicit
interface treatment, solvers, and binary mesh) in
PumpLinx make it sufficiently robust to model large
voids and dynamic collapse.
Figure 10: External Gear Pump Cavitation (Predicted
Figure 9 shows an example of the teeth meshing for a vs. Observed).
generic gear pump, with the predicted void formation
displayed with magenta corresponding to regions of
100% gas. The model predicts the formation of vapor Another example of predicting damage is given below for
pockets fractions between the teeth. As these pockets a piston pump. Piston pumps use one or more
collapse, they cause damage to the surface. These piston/cylinder assemblies to draw in and expel fluid. In
studies show a strong correlation between the predicted an axial pump, the pump rotates and each piston slides
vapor pocket formation/collapse and the observed in and out of its respective chamber. The ends of the
cavitation damage (Figure 10). cylinders opposite the piston faces are open and rotate
over a valve plate to regulate the flow through the pump.
Piston pumps are capable of creating very high
pressures and are often used in hydraulic equipment.
Figure 11 provides a schematic of an axial piston pump
showing the effect of swash plate angle.
 Cavitation Damage Bubbles can be displayed in
PumpLinx either as a color map or as an iso-surface, as
shown in Figure 14. However, the presence of bubbles is
not necessarily an indicator of potential damage, since
bubbles that do not collapse do not cause damage.
Animations can help visualize severe collapse, but a
better indicator of potential damage is the power
Figure 11: Axial Piston Pump Schematic associated with the localized water hammer effect that
occurs when a vapor bubble collapses or a void is
rapidly compressed. This wattage can be displayed as
For piston pumps, cavitation and rapid bubble collapse shown in Figure 15.
can lead to unacceptable noise and/or localized damage
as shown on the face of the valve plate in Figure 12.
This particular bubble growth and collapse is related to
rapid venting that occurs as a highly pressurized volume
first connects to low pressure. This type of damage has
been known to cause field failures in commercial pumps.
These venting effects are very three-dimensional and
cannot be modeled effectively with nodal type models.
Figure 13 shows the predicted instantaneous void
fraction for a piston pump using the 3-D CFD code
PumpLinx with its aeration and cavitation model. The
location of the predicted cavitation corresponds directly
with the observed damage. This type of correlation
would not be possible with a nodal model.

Figure 12: Cavitation Damage in a Piston Pump

Figure 14: Using Iso-Surfaces to display bubbles

Figure 15: Localized power concentration indicates

potential damage

The wattage shown in Figure 15 is an indication of the

power associated with the bubble dynamics on a
macroscopic level. However, even without resolving the
Figure 13: Predicted Vapor Formation vs. Observed bubble collapse at the microscopic scale, it provides an
indication of the localized power associated with the
bubble collapse and, when correlated with observed
damage, can serve as a useful tool for subsequent
prediction of damage. As with the other sources of REFERENCES
damage, sufficient refinement of the time-step and mesh
are a prerequisite to predicting the power 1. Nykanen, T. H. A., Esque, S., Ellman, A. U. (2000),
concentrations, as a coarse time-step or mesh would Comparison of different fluid models. Bath workshop
dissipate the power over time or space. on Power Transmission and Motion Control
(PTMC200), Bath 2000, p101-110.
SUMMARY:
2. C. Schleihs, H. Murrenhoff , Liebherr Machines
Damage in fluid systems is typically the result of one of Bulle - M. Deeken, E.Viennet. Simerics - Yanjun
the following causes: pressure waves, jetting, loads / Xia, Hui Ding, Sam Lowry, “3D-CFD simulation of a
contact and bubble collapse. These effects can be hydraulic axial piston displacement unit,” IFK 2014,
modeled using CFD, with careful adherence to the Aachen Germany
following prerequisites:
3. A. Senatore, D. Buono, E. Frosina, M. Pavanetto,
1) Inclusion of comprehensive physics, including
I.I. Costin, M. Olivetti, Improving the position
robust aeration and cavitation models when
control performance of a Proportional spool valve,
appropriate,
using a 3D CFD modeling, IFPE 2014.
2) Sufficient mesh and time-step refinement,
4. Singhal, A. K., Athavale, M. M., Li H., and Jiang,
3) Proper specification of boundary location and Y. 2002. Mathematical Basis and Validation of the
operating conditions. Full Cavitation Model, Journal of Fluid Engineering,
Vol. 124, Issue 3, pp. 617-624, 2002
Given an accurate simulation, the potential for damage
can be evaluated based on the magnitude of the various 5. JIANG, Y., ZHANG, D., FURMANCZYK, M., LOWRY, S. &
causes. In the event that experimental data are PERNG, C. (2008), “A three-dimensional design tool
available, the threshold for expected damage can be for crescent oil pumps,’ SAE-2008-01-0003, SAE
quantified. In the absence of calibration with World Congress, April 14 - 17, Detroit, Michigan,
experiments, designs can be compared and optimized USA.
by minimizing these potential sources of damage based
on the CFD calculations.