applied

sciences
Article
Development of High-Fidelity Numerical
Methodology Based on Wavenumber-Frequency
Transform for Quantifying Internal Aerodynamic
Noise in Critical Nozzle
Garam Ku 1 , Songjune Lee 1 , Cheolung Cheong 1, * , Woong Kang 2 and Kuksu Kim 3
1 School of Mechanical Engineering, Pusan National University, Busan 46241, Korea
2 Division of Physical Metrology, Korea Research Institute of Standards and Science, 267 Gajeong-ro,
Yuseong-Gu, Daejeon 34113, Korea
3 Daewoo Shipbuilding and Marine Engineering, Geoje 53302, Korea
* Correspondence: [email protected]; Tel.: +82-051-510-2311




Received: 9 May 2019; Accepted: 16 July 2019; Published: 19 July 2019


Abstract: In industrial fields dealing with high-temperature and high-pressure gas such as chemical,
petrochemical, and offshore oil production plants, piping systems with valves are frequently used to
protect the relevant system and equipment from being damaged by such gases. However, excessive
noise is sometimes generated by the valve flow in the piping system, causing so-called acoustic
induced vibration in the pipe wall. Therefore, it is of great importance to design the related system to
avoid this phenomenon. In this study, a high-fidelity numerical procedure is proposed to assess the
acoustic power generated by pressure relief devices in a pipe. The method consists of three sequential
steps: high accuracy large eddy simulation, wavenumber-frequency transform, and duct acoustic
theory. The critical nozzle is selected as a target system since it is commonly used as a flowmeter and
thus there are a lot of relevant data for comparison. First, the steady Reynold-Averaged Navier–Stokes
(RANS) solver is used to predict the flow rate of the two-dimensional axisymmetric critical nozzles,
and its validity is confirmed by comparing the predicted results with the measured ones. There
is good agreement between the two results. Then, a high accuracy Large Eddy Simulation (LES)
technique is performed on the three-dimensional critical nozzle, and the steady-state RANS result is
used as the initial condition to accelerate the convergence of the unsteady simulation. The validity of
the unsteady LES results is also confirmed by comparing them with measured surface pressure data.
The wavenumber-frequency transform is taken on the LES results, and the compressible surface
pressure components matching the acoustical duct modes are identified in the wavenumber-frequency
pressure diagram. The inverse wavenumber-frequency transform taken on the compressible pressure
components leads to the acoustic power spectrum. These results reveal that the current numerical
procedure can be used to more accurately predict the acoustic power generated by pressure relief
device in the piping system.

Keywords: acoustic induced vibration (AIV); critical nozzle; wavenumber-frequency analysis;

aerodynamic noise

1. Introduction
In general, a complex pipe system is used to transport high-temperature and high-pressure
gas in industries such as chemical, petrochemical, and offshore oil production plants. When the
piping system is over-pressurized, the pressure relief valve is installed to prevent pipe failure due
to high-pressure. When the fluid passes through the pressure relief valve, high acoustic energy is

Appl. Sci. 2019, 9, 2885; doi:10.3390/app9142885 www.mdpi.com/journal/applsci

Appl. Sci. 2019, 9, 2885 2 of 15

generated and radiated in the downstream direction due to the sudden pressure drop through the valve.
Acoustic pressure waves induce the vibration of the pipe wall. This phenomenon is called “Acoustic
Induced Vibration” (AIV). AIV in a piping system is known to be involved in the structural and acoustic
fluctuations in a range of frequencies between 500 and 2000 Hz [1]. Carucci and Mueller [2] collected
the failure cases until 1980 to separate the failures due to AIV and suggested the simple procedure of a
pipe design to prevent AIV. Esinger [3,4] collected more pipe failure cases and modified the criteria of
Carucci and Mueller. NORSOK standard [5] adopted the method proposed by Carucci and Mueller to
assess the acoustic power level in a design stage of pipe. Bruce et al. [6] summarized the history of
empirical research about AIV in a piping system and suggested the modified empirical method newly
adopting the thickness of pipe wall as an additional parameter to the diameter of the pipe. Because
most of preceding studies on the AIV problems in piping systems depend on empirical approaches,
the fundamental mechanism of AIV does not come out into the open, and thus it is difficult to devise
effective measures for AIV even if the AIV problem is expected. To find the precise mechanism of AIV
and to devise its effective measure, the first-principle-based method is essential. Agar and Ancian [7]
used a fluid-structure one-way coupling method using ACTRAN software based on the finite element
method and suggested a methodology for AIV risk assessment. However, the method is still based
on an empirical formula of Carucci and Mueller for the prediction of acoustic power. Kim et al. [8]
estimated the internal aerodynamic noise due to valve flow in a simple constriction-expansion pipe,
by combining the large eddy simulation technique with a wavenumber-frequency analysis. However,
the results were not confirmed by using any measured data. The main goal of this study is to propose
the first-principle-based methodology to determine the acoustic power due to valve flow in a pipe.
The method consists of three sequential steps: accurate internal flow simulation using Large Eddy
Simulation (LES) techniques, the separation of compressible and incompressible pressure waves using
the wavenumber-frequency analysis and the assessment of acoustic power using the duct acoustic
theory. The critical nozzle is selected as a target system since it is commonly used as a flowmeter and
thus there are a lot of relevant data for comparison. First, the steady RANS solver is used to predict
the flow rate of the two-dimensional axisymmetric critical nozzles, and its validity is confirmed by
comparing the predicted results with the measured ones. There is good agreement between the two
results. Then, a high accuracy LES technique is performed on the three-dimensional critical nozzle,
and the steady-state Reynold-Averaged Navier–Stokes (RANS) result is used as the initial condition
to accelerate the convergence of the unsteady simulation. The validity of the unsteady LES results is
also confirmed by comparing them with measured surface pressure data. The wavenumber-frequency
transform is taken on the LES results, and the compressible surface pressure components matching the
acoustical duct modes are identified in the wavenumber-frequency pressure diagram. The inverse
wavenumber-frequency transform taken on the compressible pressure components leads to the acoustic
power spectrum.
The structure of the paper is organized as follows. In Section 2, the governing equations
and the numerical methods including the CFD techniques and the wavenumber-frequency analysis
are presented. In Section 3, the experimental setup is presented. In Section 4, the numerical
simulation results are presented with the accompanying experimental results for validation. Finally,
the acoustic power spectrum due to the critical nozzle flow in a pipe is predicted using the
wavenumber-frequency analysis.

2. Governing Equations and Numerical Methods

2.1. Target Internal Flow System: Critical Nozzle

Generally, in a nozzle, when the pressure difference between upstream and downstream is
larger than a specific value, the flow cannot be accelerated higher than the sonic speed at the nozzle
throat, and the mass flow rate is kept constant. This phenomenon is called a choked flow. When the
choke flow occurs, the mass flow rate through the nozzle can be determined using the only inlet
Appl. Sci. 2019, 9, 2885 3 of 15

flow field
Appl. Sci. information
2019, 9, x FOR PEER such as total pressure, temperature, and velocity upstream of the nozzle
REVIEW and
3 of 15
the nozzle area, regardless of the downstream flow conditions. The ratio of the theoretical mass
to therate
flow actual
to mass flow rate
the actual massis defined asisthe
flow rate efficient
defined ascoefficient, which
the efficient is closerwhich
coefficient, to one isas closer
the Reynolds
to one
number
as increases.number
the Reynolds The critical flow Venturi
increases. nozzle
The critical (CFVN)
flow Venturiis widely
nozzleused
(CFVN)as a is
flowmeter
widely usedby using
as a
the characteristics
flowmeter by usingof the
the nozzle. According
characteristics to ISO 9300,
of the nozzle. the CFVNs
According to ISO can
9300,bethe
classified
CFVNs intocan becylindrical-
classified
throat
into Venturi nozzle and
cylindrical-throat toroidal-throat
Venturi nozzle andVenturi nozzle depending
toroidal-throat on their
Venturi nozzle throat shape
depending [9]. In
on their this
throat
study,[9].
shape theIn toroidal-throat Venturi nozzles,
this study, the toroidal-throat which
Venturi are which
nozzles, commonly used in used
are commonly flowmeters, were
in flowmeters,
investigated. Figure 1 shows the shape of the toroidal-throat Venturi nozzle
were investigated. Figure 1 shows the shape of the toroidal-throat Venturi nozzle with its geometricwith its geometric
parameter, d.
parameter, d.

Figure 1.
Figure Toroidal-throat Venturi
1. Toroidal-throat Venturi nozzle
nozzle specified
specified by
by ISO
ISO 9300.
9300.

2.2. Governing Equations

2.2. Governing Equations
For the numerical simulations of the flow through the critical nozzle, the Navier–Stokes equations
For the numerical simulations of the flow through the critical nozzle, the Navier–Stokes
are used as the governing equations. However, there are many physical and numerical issues to
equations are used as the governing equations. However, there are many physical and numerical
deal with in solving the Navier–Stokes equations numerically. The most difficult is due to the
issues to deal with in solving the Navier–Stokes equations numerically. The most difficult is due to
strong nonlinearity of the equations. Numerical approaches to nonlinear problems suffer from not
the strong nonlinearity of the equations. Numerical approaches to nonlinear problems suffer from
only poor accuracy and convergence but also high numerical cost. To solve these difficulties and
not only poor accuracy and convergence but also high numerical cost. To solve these difficulties and
to apply the Navier–Stokes equations to a broader range of fields by using a numerical method,
to apply the Navier–Stokes equations to a broader range of fields by using a numerical method, the
the Navier–Stokes equations are generally averaged. The most widely used averaged Navier–Stokes
Navier–Stokes equations are generally averaged. The most widely used averaged Navier–Stokes
equations in computational fluid dynamics are the Reynold-Averaged Navier–Stokes (RANS) equations.
equations in computational fluid dynamics are the Reynold-Averaged Navier–Stokes (RANS)
The behavior of fluids present in nature can be thought of as the sum of deterministic and random
equations. The behavior of fluids present in nature can be thought of as the sum of deterministic and
quantities. The RANS equations average over time, which eliminates the random behavior of fluids.
random quantities. The RANS equations average over time, which eliminates the random behavior
The compressible RANS equation is written as follows.
of fluids. The compressible RANS equation is written as follows.
∂ρ
𝜕𝜌 𝜕
∂
+ (ρu
𝜌𝑢i ) =
= 0 (1)
(1)
∂t
𝜕𝑡 𝜕𝑥∂xi

𝜕∂ 𝜕∂  ∂p
𝜕𝑝 𝜕∂ ∂u
𝜕𝑢i ∂u j 2 ∂u
𝜕𝑢 𝜕𝑢k ∂𝜕 
" !#
 
ρui ) +
(𝜌𝑢 + ρu
𝜌𝑢 i𝑢 j
u = −
− + 𝜇µ +
+ − δ
− 𝛿ij ++ −𝜌𝑢
−ρui u𝑢
j (2)
(2)
∂t
𝜕𝑡 ∂x j
𝜕𝑥 ∂x
𝜕𝑥i 𝜕𝑥∂x j ∂x
𝜕𝑥j ∂xi 3 ∂x
𝜕𝑥 𝜕𝑥k ∂x
𝜕𝑥j
∂ →  X 
𝜕 (ρE) + ∇· u (ρE + p) = −∇· hj Jj (3)
∂t 𝜌E + ∇ ∙ 𝑢⃗ 𝜌𝐸 + 𝑝 = −∇ ∙ ℎ 𝐽 (3)
𝜕𝑡
where ρ, u, p, and µ are the density of the fluid, velocity, pressure, and dynamic viscosity. δ is the
where ρ, u, p, and μ are the density of the fluid, velocity, pressure, and dynamic viscosity. δ is the
Kronecker delta function. E means energy per unit mass, h means enthalpy and J diffusion flux of
Kronecker delta function. E means energy per unit mass, h means enthalpy and J diffusion flux of
species j. Subscripts i and j mean x-, y-, and z-direction. ρui u j is a Reynolds stress tensor, which can be
species j. Subscripts i and j mean x-, y-, and z-direction. 𝜌𝑢 𝑢 is a Reynolds stress tensor, which can
obtained from turbulent modeling based on the assumption of Boussinesq.
be obtained from turbulent modeling based on the assumption of Boussinesq.
However, the RANS equations tend to over-predict the viscosity and are known to weaken the
However, the RANS equations tend to over-predict the viscosity and are known to weaken the
unsteady behavior of fluids such as vortex phenomena [10,11]. As another method for solving the
unsteady behavior of fluids such as vortex phenomena [10,11]. As another method for solving the
Navier–Stokes equation numerically, a spatial averaging operation is performed on the basic equation.
Navier–Stokes equation numerically, a spatial averaging operation is performed on the basic
Large Eddy Simulation (LES) separates the flow field into a component that can be resolved using
equation. Large Eddy Simulation (LES) separates the flow field into a component that can be resolved
the given grid and the smaller ones. The former is directly calculated using the governing equations,
using the given grid and the smaller ones. The former is directly calculated using the governing
and the latter is modeled using a subgrid-scale model. The governing equations are written as follows.
equations, and the latter is modeled using a subgrid-scale model. The governing equations are
written as follows. f = f + f0 (4)
𝑓 = 𝑓 ̅ + 𝑓′ (4)

𝜕𝜌̅ 𝜕
+ 𝜌̅ 𝑢 = 0 (5)
𝜕𝑡 𝜕𝑥
Appl. Sci. 2019, 9, 2885 4 of 15

∂ρ ∂
+ (ρe
ui ) = 0 (5)
∂t ∂xi

∂ ∂   ∂τij+ ∂  e 
(ρe
ui ) + ui uej + p+ δij =
ρe + µSij (6)
∂t ∂x j ∂x j ∂x j
 
∂ ∂ h  i
+ e ∂Qi ∂  ∂T  ∂  e 
(ρeet ) + ρe
et + p ui = + k  + µe
u j Sij (7)
∂t ∂xi ∂xi ∂xi ∂xi ∂xi
τ R 1 1
 
T+ T − kk , p+ = ρRT + +
= e − τkk , τij+ = τij − τkk δij = ρνt e Sij (8)
2cv ρ 2cv 3 3
LES is represented by the physical quantity that can be resolved by a grid and the physical quantity
that needs to be modeled, as expressed in Equation (4). The superscript ~ denotes the Favre- (or mass-)
averaged physical quantity and is used to consider the compressibility of the turbulent structure. et is
the total energy, and the physical quantity by the superscript + is given by Equation (8). cv is the static
specific heat, R is the gas constant, and ν is the kinematic viscosity [12].
Unlike the RANS equation using time averaging, LES does not use time averaging, so it is suitable
for solving unsteady flow with time variation. However, the computation cost is very high, and the
solution can be unstable according to initial conditions. In this study, to effectively and accurately
consider the influence of the unsteady flow, the steady-state flow field was calculated by solving
the RANS, and the LES was performed by applying the RANS results as the initial condition for the
unsteady flow field simulation.

2.3. Numerical Methods

First, the steady-state RANS simulation was performed to obtain an average flow field from
which the volume flow rate through the nozzle was predicted and compared with the measured data.
This comparison confirms the validity of the numerical method, and the RANS solution was used as
an initial condition for the subsequent LES simulation for fast convergence. In the RANS simulation,
the k-ω Shear Stress Transport (SST) turbulence model was used as a closure model. The second order
upwind scheme was used for the spatial discretization in both the equation for turbulent kinetic energy
and dissipation rate.
High-order numerical simulation techniques in time and space are required to accurately compute
the acoustic waves as well as the turbulent flow simultaneously. Therefore, the unsteady LES was
performed with the steady-state RANS results as an initial condition. The bounded second-order
implicit method and second-order upwind scheme were used for time and space discretizations,
respectively. The pressure and velocity coupled method was used. The Smagorinsky–Lilly model, which
is relatively simple and widely used, was employed as the subgrid-scale model for turbulence flow.
The numerical discretization scheme for time and space was the second order accuracy, and the time
interval was 5 × 10−5 s for a sampling rate of 20,000 Hz.
All of the numerical methods were realized using the ANSYS Fluent program.

2.4. Acoustic Power Due to Internal Flow Noise

As described in the Introduction, to find the precise mechanism of AIV and to devise its effective
measure, the first-principle-based method is essential. Agar and Ancian [7] used a fluid-structure
one-way coupling method for fluid-structure interaction based on ACTRAN software which uses the
finite element method and suggested a methodology for AIV risk assessment. However, the method
is still based on an empirical formula of Carucci and Mueller for the prediction of acoustic power.
In addition to this study, most of the previous studies for the design criteria to prevent the fatigue failure
caused by the AIV depend on the simple empirical formula by Carucci and Mueller. In this section,
the formula for acoustic power due to internal flow noise is presented using the wavenumber-frequency
transform of pressure field in time–space domains and the duct acoustic theory.
Appl. Sci. 2019, 9, 2885 5 of 15

As a starting point, the pressure fluctuation due to the flow on the pipe wall needs to be separated
into the incompressible and the compressible pressure fluctuations [13]. The incompressible pressure
fluctuation is not associated with the density perturbation of the fluid. It is also called the hydrodynamic
pressure fluctuation or the pseudo sound and propagates at the same speed as the convection eddy
velocity in the fluid. Compressible pressure fluctuation refers to acoustic waves due to perturbations
of density and velocity in a fluid and propagates at the speed of sound. A wavenumber-frequency
analysis can be used to separate the incompressible and compressible components from the total
pressure fluctuation components on the wall. Since the target problem of this study is a circular pipe,
the Fourier transform in the cylinder coordinate system is more suitable. From the previous study,
the wavenumber-frequency analysis is performed in the form [8],

∞ N t −1 N z −1 Nθ −1 mθ − lp )
1 X 1 X X X −2πi( Nko + N
S(m, p∆kz , o∆f) = wklj pklj e t θ Nz (9)
Nt Nz Nθ J m ( k r0 ,m,n r0 )
n=0 k =1l=0 j=0

where S is the spectral density, m is the circumferential mode, n is the radial mode, kz is the axial
direction wavenumber, and Jm is the Bessel function of mth order. N means the size of the discretized
signal over time and space. The Hanning function is used as a window. The axial wavenumber, kz ,
from duct acoustics is defined as:
h   i1/2
∓ Mk0 + k02 − 1 − M2 kr,m,n
2
±
kz,m,n = (10)
1 − M2
 
2
k02 − 1 − M2 kr,m,n > 0 (11)

The Fourier transform of the discretized pressure fluctuation signal in time and space can be used
to obtain a wavenumber-frequency diagram from Equation (9). The acoustic power generated in the
pipe can be obtained by defining the instantaneous sound intensity corresponding to the acoustic
power per unit area and integrating it over the area. Since the empirical approach uses acoustic
power as a criterion to determine the possible occurrence of AIV, the formula for the acoustic power is
derived and presented combining the wavenumber-frequency spectrum with the duct acoustic theory.
The sound pressure and particle velocity in the pipe can be deduced from the duct acoustic theory [14],
∞ ∞
+
X X  − 
p(r, θ, z, t) = Jm (kr,m,n r)e jmθ e jωt × C+,m,n e− jkz,m,n z + C−,m,n e+ jkz,m,n z (12)
m = 0n = 0

∞ ∞ +
X X kz,m,n +
uz (r, θ, z, t) = Jm (kr,m,n r)cos(mθ)C+,m,n ei(ωt−kz,m,n z) (13)
k ρ c
m = 0n = 0 0 0 0

The acoustic intensity can be expressed as the product of the compressible pressure fluctuation
obtained previously and the particle velocity. Finally, it can be expressed as

dW = Iz dA = puz dA
R RT (14)
W = IdA where I = T1 0 Iz dt

∞ ∞ + Z2π
X X kz,m,n 1 R2 n
Jm (kr,m,n R)2 − Jm−1 (kr,m,n R) Jm+1 (kr,m,n R)
o
W = C2 cos2 (mθ)dθ (15)
k0 ρ0 c0 +,m,n 2 2
m = 0n = 0 0

3. Experimental Setup
In order to measure the surface pressure fluctuation downstream of the critical nozzle causing
the AIV, the high-pressure gas flowrate standard system of the Korea Research Institute of Standards
Appl. Sci. 2019, 9, 2885 6 of 15

and Science (KRISS) was used as shown in Figure 2 [15]. The high-pressure gas flow standard system
is a blow-down system in which a gas is stored in a storage tank at a high pressure using a compressor,
and then the stored gas is adjusted to the specified pressure and discharged to a test pipeline. To keep
Appl. Sci. 2019, 9, x FOR PEER REVIEW 6 of 15
pressure and temperature stable during the test, the compressed air in the reservoir is adjusted in the
primary
is adjustedpressure
in theregulating valve upregulating
primary pressure to 50 bar and stored
valve up toin50
the temperature
bar and stored control loop. Compressed
in the temperature control
air passing through the temperature control loop is supplied to the test section with
loop. Compressed air passing through the temperature control loop is supplied to the test section specific pressure
regulation
with specific in the secondary
pressure pressure
regulation regulating
in the valve.
secondary pressure regulating valve.

Figure 2. High pressure gas flowrate standard system at Korea Research Institute of Standards and
Figure 2. High pressure gas flowrate standard system at Korea Research Institute of Standards and
Science (KRISS).
Science (KRISS).

To measure the surface pressure on the inner wall of pipe downstream of the critical nozzle,
To measure the surface pressure on the inner wall of pipe downstream of the critical nozzle, an
an experiment was conducted by installing a second pressure control valve and a critical nozzle to
experiment was conducted by installing a second pressure control valve and a critical nozzle to the
the straight pipe of length 50D after the U-band. The dynamic pressure sensor used for the pressure
straight pipe of length 50D after the U-band. The dynamic pressure sensor used for the pressure
measurement was the 102B15 model from PCB Piezotronic with a resonance frequency of 500 kHz
measurement was the 102B15 model from PCB Piezotronic with a resonance frequency of 500 kHz
and a noise measurement sensitivity of 3.6 mV/kPa. The low-frequency response (−5%), which is a
and a noise measurement sensitivity of 3.6 mV/kPa. The low-frequency response (−5%), which is a
low-frequency characteristic, was 0.5 Hz, and the sensitivity was maintained at the same sensitivity
low-frequency characteristic, was 0.5 Hz, and the sensitivity was maintained at the same sensitivity
throughout the frequency range of interest of this measurement. For accurate measurement of the
throughout the frequency range of interest of this measurement. For accurate measurement of the
AIV generated inside the pipe without any interference, the sensor was installed in the form of flush
AIV generated inside the pipe without any interference, the sensor was installed in the form of flush
mounting as shown in Figure 3, which shows a sensor installed inside the pipe to measure the AIV of
mounting as shown in Figure 3, which shows a sensor installed inside the pipe to measure the AIV
the critical nozzle. It was equipped with a pressure and temperature sensor to measure the reference
of the critical nozzle. It was equipped with a pressure and temperature sensor to measure the
flow rate in the pipe according to ISO 9300. A dynamic pressure sensor was installed at the position
reference flow rate in the pipe according to ISO 9300. A dynamic pressure sensor was installed at the
of 2D and 4D downstream of the critical nozzle. Each dynamic pressure sensor was first installed
position of 2D and 4D downstream of the critical nozzle. Each dynamic pressure sensor was first
in the jig which was manufactured for the easy installation on the pipe. The AIV signal detected
installed in the jig which was manufactured for the easy installation on the pipe. The AIV signal
through the dynamic pressure sensor was recorded by the Digitizer (NI PXI 5922) through the PCB
detected through the dynamic pressure sensor was recorded by the Digitizer (NI PXI 5922) through
conditioner Model 482C05. The PXI 5922 can record up to 10 M signals per second at a resolution of
the PCB conditioner Model 482C05. The PXI 5922 can record up to 10 M signals per second at a
18 bits using two-channel simultaneously and was installed on the NI PXI 1033-Chassis and connected
resolution of 18 bits using two-channel simultaneously and was installed on the NI PXI 1033-Chassis
to the personal computer (PC) via an Express card, which runs through the LabVIEW program. Signal
and connected to the personal computer (PC) via an Express card, which runs through the LabVIEW
acquisition was made at the sampling rate of 50 kHz, and the acquired sample size was 500,000.
program. Signal acquisition was made at the sampling rate of 50 kHz, and the acquired sample size
The related measurement devices are shown in Figure 4.
was 500,000. The related measurement devices are shown in Figure 4.
Appl.
Appl. Sci. 2019, 9,
Sci. 2019, 9, 2885
x FOR PEER REVIEW 77 of
of 15
15
Appl.
Appl. Sci.
Sci. 2019,
2019, 9,
9, xx FOR
FOR PEER
PEER REVIEW
REVIEW 77 of
of 15
15

Figure 3. Experimental setup for acoustic induced vibration (AIV) measurement.

Figure
Figure 3.
3. Experimental
Experimental setup
Experimental setup for
for acoustic
acoustic induced
induced vibration
vibration (AIV)
(AIV) measurement.
measurement.

Figure 4. AIV measurement system at KRISS.

Figure
Figure 4.
4. AIV
AIV measurement
AIV measurement system
measurement system at
system at KRISS.
KRISS.
Results and Discussion
4. Results
4.
4. Results and
and Discussion
Discussion
In this section, the numerical
section, numerical simulation results together with the experimental results are
In
In this
this section,
section, the
the numerical
numerical simulation
simulation results
results together
together with
with the
the experimental
experimental results
results are
are
presented. First, the validity of steady numerical solvers was confirmed by comparing the mass flow
presented.
presented.
presented. First, the validity of steady numerical solvers was confirmed by comparing the mass
First, the validity of steady numerical solvers was confirmed by comparing the mass flow
flow
rate obtained from the compressible
compressible RANS simulations of the axisymmetric
axisymmetric critical nozzle with the
rate
rate obtained
obtained from
from the
the compressible
compressible RANS RANS simulations
simulations ofof the
the axisymmetric
axisymmetric critical
critical nozzle
nozzle with
with the
the
measured ones. Then,
ones. Then, the
Then, the compressible
the compressible steady-state
compressible steady-state RANS
steady-state RANS simulation
RANS simulation was performed on the 3D
simulation was
measured
measured ones. Then, the compressible steady-state RANS simulation was performed on the performed on the 3D
3D
critical nozzle, and the unsteady compressible LES was performed using the RANS results as an
critical
critical nozzle,
nozzle, and
and the
the unsteady
unsteady compressible
compressible LES LES was
was performed
performed using
using the
the RANS
RANS results
results as
as an
an
initial condition. The validity of the LES was confirmed by comparing the predicted surface pressure
condition. The
initial validity of the LES was confirmed by comparing the predicted surface
initial condition. The validity of the LES was confirmed by comparing the predicted surface pressure pressure
fluctuations with
fluctuations withthethemeasured
measuredones.ones.Finally,
Finally,
thethe wavenumber-frequency
wavenumber-frequency analysis
analysis waswas carried
carried out
out and
fluctuations
fluctuations with
with the
the measured
measured ones.
ones. Finally,
Finally, the
the wavenumber-frequency
wavenumber-frequency analysis
analysis was
was carried
carried out
out
and the acoustic
the acoustic power
power due due due to the critical
to the critical nozzle flow
nozzle flowflow was assessed.
was assessed.
and
and the
the acoustic
acoustic power
power due to to the
the critical
critical nozzle
nozzle flow was
was assessed.
assessed.
4.1. Two-Dimensional Axisymmetric
Axisymmetric Steady
Steady Simulation
Simulation
4.1.
4.1. Two-Dimensional
Two-Dimensional Axisymmetric
Axisymmetric Steady
Steady Simulation
Simulation
The compressible RANS simulation was performed for a two-dimensional axisymmetric critical
The
The compressible
compressible RANS simulation was
was performed
performed for
for aa two-dimensional
two-dimensional axisymmetric
axisymmetric critical
nozzle and the accuracyRANSof thesimulation
simulation was verified through the comparison of the predicted mass
critical
nozzle
nozzle and
and the
the accuracy
accuracy of
of the
the simulation
simulation was
was verified
verified through
through the
the comparison
comparison of
of the
the predicted
predicted mass
mass
flow rate with
with measured
measured data. critical nozzle
data. The critical nozzle targeted is the toroidal-throat Venturi nozzle as
flow
flow rate
rate with
with measured
measured data.
data. The
The critical
critical nozzle
nozzle targeted
targeted is
is the
the toroidal-throat
toroidal-throat Venturi
Venturi nozzle
nozzle as
as
specified in ISO 9300.
9300. Its geometry and the numerical mesh with the specified boundary conditions
specified boundary conditions
specified
specified in ISO 9300. Its geometry and the numerical mesh with the specified boundary conditions
are showninin ISO 9300.
in Figure
Figure 5.
5.
Its geometry and the numerical mesh with the specified boundary conditions
are
are shown
shown in in Figure
Figure 5.
5.

Figure 5. The geometry of the toroidal-throat Venturi nozzle and the computational grids with boundary
Figure 5. The geometry of the toroidal-throat Venturi nozzle and the computational grids with
Figure
Figure 5.
5. The
conditions geometry
applied
The of
of the
the toroidal-throat
for axisymmetric
geometry Venturi
Reynold-Averaged
toroidal-throat nozzle
nozzle and
and the
VenturiNavier–Stokes computational
(RANS)
the simulation.grids
computational grids with
with
boundary conditions applied for axisymmetric Reynold-Averaged Navier–Stokes (RANS)
boundary
boundary conditions applied for axisymmetric Reynold-Averaged Navier–Stokes (RANS)
conditions applied for axisymmetric Reynold-Averaged Navier–Stokes (RANS)
simulation.
The simulations were carried out for the two types of nozzle of which the throat diameter was
simulation.
simulation.
0.4 mm and 0.56 mm, respectively. The inlet and outlet pressure were specified on the inlet and
The simulations were carried out for the two types of nozzle of which the throat diameter was
outletThe
The simulations
boundaries,
simulations were
were carried
respectively, the out
carried out for
for the
adiabatic the two
and
two types
no-slip of
of nozzle
typeswall of
of which
conditions
nozzle werethe
which setthroat
the on thediameter
throat wall, andwas
diameter the
was
0.4 mm and 0.56 mm, respectively. The inlet and outlet pressure were specified on the inlet and outlet
0.4 mm and
axisymmetric 0.56
0.4 mm and 0.56 mm, respectively.
condition was applied
mm, respectively. The inlet
Thealong and
inlet the outlet
andcenterlinepressure were
axis ofwere
outlet pressure specified
the venturi on the
specifiednozzle. inlet
on the The and
inletgrids outlet
were
and outlet
boundaries, respectively, the adiabatic and no-slip wall conditions were set on the wall, and the
boundaries,
constructed
boundaries,by respectively, the
the adiabatic
using a structured
respectively, and
grid system,
adiabatic and y+ was
and no-slip
no-slip wall conditions
wallset to be around
conditions were set
five,
were on
setand the
onthe wall,
thetotal and
and the
number
wall, of
the
axisymmetric condition was applied along the centerline axis of the venturi nozzle. The grids were
axisymmetric
grids was 350 ×condition
100. The was applied
efficiency of along
the the
critical
axisymmetric condition was applied along the centerline centerline
nozzle axis
can be of the venturi
determined by nozzle.
observing The grids
the were
variation
axis of the venturi nozzle. The grids were
constructed by using a structured grid system, and y++ was set to be around five, and the total number
constructed
constructed by
by using
using aa structured
structured grid
grid system,
system, and and yy+ was
was set
set to
to be
be around
around five,
five, and
and the
the total
total number
number
of grids was 350 × 100. The efficiency of the critical nozzle can be determined by observing the
of
of grids was 350 × 100. The efficiency of the critical nozzle can be determined by observing the
grids was 350 × 100. The efficiency of the critical nozzle can be determined by observing the
Appl. Sci. 2019, 9, x FOR PEER REVIEW 8 of 15
Appl. Sci. 2019, 9, 2885 8 of 15
variation of mass flow rate according to the change of pressure difference between the inlet and the
outlet. In this study, the pressure of the outlet was varied to control the pressure difference between
of mass flow rate according to the change of pressure difference between the inlet and the outlet. In this
the inlet and outlet of the nozzles. The applied inlet and outlet pressure conditions are shown in Table
study, the pressure of the outlet was varied to control the pressure difference between the inlet and
1 [16].
outlet of the nozzles. The applied inlet and outlet pressure conditions are shown in Table 1 [16].
Table 1. Pressure ratio of outlet pressure (𝑃 to inlet pressure (𝑃
Table 1. Pressure ratio of outlet pressure (Pb ) to inlet pressure (P0 ).
Pressure ratio (𝑷𝒃 /𝑷𝟎 ) 𝑷𝒃 (Pa) Pressure ratio (𝑷 /𝑷 ) 𝑷 (Pa)
Pressure Ratio (Pb /P0 ) Pb (Pa) Pressure Ratio (P𝒃b /P𝟎0 ) Pb𝒃 (Pa)
0.954
0.954 429,300
429,300 0.954
0.954 429,300
429,300
0.931
0.931 418,950
418,950 0.934
0.934 420,300
0.887
0.887 399,150
399,150 0.911
0.911 409,950
0.843
0.843 379,350
379,350 0.888
0.888 399,600
399,600
0.798
0.798 359,100
359,100 0.844
0.844 379,800
379,800
0.666 299,700 0.778 350,100
0.666 299,700 0.778 350,100
(a) d = 0.4 mm (b) d = 0.56 mm
(a) d = 0.4 mm (b) d = 0.56 mm

Figure
Figure 66 shows
shows thethe predicted
predicted iso-contours
iso-contoursof ofthe
theMach
Machnumber
numberinsideinsidethe thenozzles
nozzlesof ofdd== 0.4
0.4 mm at
at
the pressure ratio of 0.666 and d = 0.56 m at the pressure ratio of 0.778.
the pressure ratio of 0.666 and d = 0.56 m at the pressure ratio of 0.778. It is seen that shock waves are It is seen that shock waves
are generated
generated around
around thethe throat
throat in in
bothboth cases,which
cases, whichimplies
impliesthat thatthe
thechoked
choked flow flow appears.
appears. Figure
Figure 77
compares
compares the the predicted
predicted onesones with with the
the measured
measured data data in
in terms
terms ofof the
the variation
variation of of the
the non-dimensional
non-dimensional
mass
mass flow rate according to the pressure ratio between the inlet and outlet. The x-axis represents
flow rate according to the pressure ratio between the inlet and outlet. The x-axis represents the the
ratio
ratio of the outlet static pressure to the total inlet pressure, and the y-axis represents the ratio
of the outlet static pressure to the total inlet pressure, and the y-axis represents the ratio ofof the
the
mass
mass flowflow rate(q
rate(qmm)) to
to the
the mass
mass flow flow rate(q
rate(qm, ) when the outlet pressure is equal to the atmospheric
ref) when the outlet pressure is equal to the atmospheric
m, ref
pressure. In the case of the nozzle flow,
pressure. In the case of the nozzle flow, as the pressureas the pressure difference between
difference the inletthe
between andinlet
outlet becomes
and outlet
larger,
becomes thelarger,
flow isthe more
flowaccelerated up to the nozzle
is more accelerated up to the throat andthroat
nozzle the mass
andflow rate increases.
the mass However,
flow rate increases.
when
However, the outlet
when pressure
the outletbecomes
pressureequal becomes to orequalless than
to or the
lesscritical
than the pressure
critical at which the
pressure chocked
at which the
flow starts to form, the Mach number in the throat reaches one
chocked flow starts to form, the Mach number in the throat reaches one and does not exceed one and does not exceed one anymore,
and thus the
anymore, andmass
thusflow rate is
the mass kept
flow constant.
rate In FigureIn
is kept constant. 6, Figure
because 6, the outletthe
because pressure in both nozzles
outlet pressure in both
is lower than the critical pressure, the Mach number is one in the nozzle
nozzles is lower than the critical pressure, the Mach number is one in the nozzle throat. Then the flow throat. Then the flow is
accelerated to supersonic in the diffusion region through the nozzle
is accelerated to supersonic in the diffusion region through the nozzle throat, and shock waves are throat, and shock waves are
generated.
generated. This This phenomenon
phenomenon correspondscorresponds to to aa typical
typical process
process forfor creating
creating aa supersonic
supersonic flow flow in in an
an
internal flow. In Figure 7, it can be seen that the mass flow rate through
internal flow. In Figure 7, it can be seen that the mass flow rate through the nozzle increases as the the nozzle increases as the
outlet
outlet pressure
pressure decreases
decreases to to the
the critical
critical pressure.
pressure. As As the
the pressure
pressure ratio
ratio is
is lower
lower thanthan the critical ratio,
the critical ratio,
the choked flow occurs and the mass flow rate is kept constant. The numerical
the choked flow occurs and the mass flow rate is kept constant. The numerical results are in excellent results are in excellent
agreement
agreement with with thethe experimental
experimental results results [16].
[16].

Figure 6. Mach number contours; (a) d = 0.4 mm, nozzle pressure ratio (NPR) = 0.666; (b) d = 0.56 mm,
Figure 6. Mach number contours; (a) d = 0.4 mm, nozzle pressure ratio (NPR) = 0.666; (b) d = 0.56 mm,
nozzle pressure ratio (NPR) = 0.778.
nozzle pressure ratio (NPR) = 0.778.
Appl.Sci.
Appl. Sci.2019,
2019,9,9,x2885
FOR PEER REVIEW 9 9ofof15
15
Appl. Sci. 2019, 9, x FOR PEER REVIEW 9 of 15

.
.
Figure 7. Mass flow rate according to variation of exit pressure; left: d = 0.4 mm; right: d = 0.56 mm.
Figure7.7.Mass
Figure Massflow
flowrate
rateaccording
accordingtotovariation
variationofofexit
exitpressure; left:dd==0.4
pressure;left: 0.4mm; right:dd== 0.56
mm;right: 0.56 mm.
mm.
4.2.
4.2.Three-Dimensional
Three-DimensionalSteady SteadySimulation
Simulation
4.2. Three-Dimensional Steady Simulation
In the pipe
In the pipesystem
system used
used in thein offshore
the offshore plant, high-temperature
plant, high-temperature and high-pressure
and high-pressure gas is
gas is transferred,
In the pipe system used in the offshore plant, high-temperature and high-pressure gas is
transferred,
and a pressure and a pressure
relief valve is relief
used to valve is used
prevent to prevent
damage damage
to the pipe which tomay
the be
pipe which
caused bymay
strong bepressure
caused
transferred, and a pressure relief valve is used to prevent damage to the pipe which may be caused
by strong
waves. pressurethe
However, waves.
suddenHowever,
pressurethe suddenin
variation pressure variation
the pressure reliefinvalve
the pressure
emits high relief valveenergy
acoustic emits
by strong pressure waves. However, the sudden pressure variation in the pressure relief valve emits
high
in theacoustic
downstream energydirection
in the downstream
of the pipe. To direction
simulate ofthis
the sudden
pipe. Topressure
simulate this sudden
variation, pressure
the numerical
high acoustic energy in the downstream direction of the pipe. To simulate this sudden pressure
variation,
simulations theusing
numerical simulations
the prescribed using the
numerical prescribed
methods werenumerical
performedmethods were performed on
on the three-dimensional the
critical
variation, the numerical simulations using the prescribed numerical methods were performed on the
three-dimensional critical nozzle as shown in Figure 8. The computation
nozzle as shown in Figure 8. The computation grids are shown in Figure 9, and the boundary grids are shown in Figure 9,
three-dimensional critical nozzle as shown in Figure 8. The computation grids are shown in Figure 9,
and the boundary
conditions were set conditions
to be theweresame setastothose
be theused
sameinasthe those used in the
preceding preceding two-dimensional
two-dimensional axisymmetric
and the boundary conditions were set to be the same as those used in the preceding two-dimensional
axisymmetric
problem. The grids used for the flow simulation were structured grids, the number ofnumber
problem. The grids used for the flow simulation were structured grids, the grids wasof
axisymmetric problem. The grids used for the flow simulation were structured grids, the number of
grids was 6,470,000,
6,470,000, and the of
and the diameter diameter
the nozzle of the nozzle
throat wasthroat
8.5 mm.wasThe8.5overall
mm. The overall
grid grid
quality was quality was
confirmed
grids was 6,470,000, and the diameter of the nozzle throat was 8.5 mm. The overall grid quality was
confirmed using the commercial
using the commercial grid generation gridtoolgeneration
ICEM CFD tool ICEM
and CFDquality
the grid and the grid
index wasquality
greater index
than was
0.61,
confirmed using the commercial grid generation tool ICEM +CFD and the grid quality index was
greater than 0.61, which confirms the good quality of the grids.
which confirms the good quality of the grids. In addition, y based on the inflow In addition, y + based on the inflow
condition was
greater than 0.61, which confirms the good quality of the grids. In addition, y+ based on the inflow
condition
kept belowwas 1. Tokept belowthe
construct 1. initial
To construct
conditionthe initial
for the condition
unsteady LES,for the unsteady LES,
three-dimensional three-
steady-state
condition was kept below 1. To construct the initial condition for the unsteady LES, three-
dimensional
compressible steady-state
RANS simulation compressible
was performedRANSbeforehand.
simulation Figure
was performed
10 presentsbeforehand.
the numericalFigure
results10of
dimensional steady-state compressible RANS simulation was performed beforehand. Figure 10
presents the numerical results of pressure and temperature distributions
pressure and temperature distributions inside the pipe and the nozzle. It can be seen that the shock inside the pipe and the
presents the numerical results of pressure and temperature distributions inside the pipe and the
nozzle.
wave isItformed
can be inside
seen that
the the shock
critical wavedue
nozzle is formed inside
to the high the critical
pressure nozzleindue
difference the to theinlet
pipe highandpressure
outlet,
nozzle. It can be seen that the shock wave is formed inside the critical nozzle due to the high pressure
difference in the pipe inlet and outlet, and the sudden pressure variation occurs
and the sudden pressure variation occurs before and after the shock wave. The shock inside the nozzle before and after the
difference in the pipe inlet and outlet, and the sudden pressure variation occurs before and after the
shock
meets wave. The shock
the boundary layerinside
of the the
wallnozzle
surfacemeets
and the theboundary
boundary layer
layer of the wall
separation surface
occurs at thisand the
location.
shock wave. The shock inside the nozzle meets the boundary layer of the wall surface and the
boundary
This complex layerinteraction
separationgenerates
occurs ata strong
this location.
coherent This complex
turbulent flow interaction
which in generates
turn causes a intense
strong
boundary layer separation occurs at this location. This complex interaction generates a strong
coherent
aerodynamic turbulent
noise. flow which in turn
The corresponding causes phenomena
unsteady intense aerodynamic noise. inThe
can be identified corresponding
Figure 11, which is
coherent turbulent flow which in turn causes intense aerodynamic noise. The corresponding
unsteady
discussedphenomena can next
in detail in the be identified
section. The in Figure
similar11, which is discussed
phenomenon was also inobserved
detail in the
andnext section.
discussed in
unsteady phenomena can be identified in Figure 11, which is discussed in detail in the next section.
The similar phenomenon
the preceding study by Kim wasetalso observed and discussed in the preceding study by Kim et al. [8].
al. [8].
The similar phenomenon was also observed and discussed in the preceding study by Kim et al. [8].

Figure 8. Three-dimensional geometry of pipe with critical nozzle.

Figure 8. Three-dimensional geometry of pipe with critical nozzle.
Figure 8. Three-dimensional geometry of pipe with critical nozzle.
Appl. Sci. 2019, 9, 2885 10 of 15
Figure 8. Three-dimensional geometry of pipe with critical nozzle.

Appl. Sci. 2019, 9, x FOR PEER REVIEW 10 of 15

Figure 9. Entire
Entire computational domain, numerical meshes, and boundary conditions.

10. Steady RANS results: (top)

Figure 10. (top) static
static pressure (K).
pressure (Pa) and (bottom) temperature (K).

Figure Unsteady
11. Unsteady
Figure 11. Large
Large EddyEddy Simulation
Simulation (LES)(LES) results;
results; (a) pressure
(a) static static pressure
(Pa); (b) (Pa); (b)
vorticity
vorticity
magnitude.magnitude.

4.3. Three-Dimensional Unsteady Simulation

4.3. Three-Dimensional Unsteady Simulation
LES is preferred for high accuracy unsteady simulation but requires many grids and a long
LES is preferred for high accuracy unsteady simulation but requires many grids and a long
simulation time. Also, it requires proper initial conditions for stable numerical simulation. Therefore,
simulation time. Also, it requires proper initial conditions for stable numerical simulation. Therefore,
three-dimensional unsteady LES was performed with the initial condition of the preceding RANS
three-dimensional unsteady LES was performed with the initial condition of the preceding RANS
simulation results. The total number of the grid was the same as the RANS simulation. The simulation
simulation results. The total number of the grid was the same as the RANS simulation. The simulation
time interval was set to be 5 × 10−5 s. Figure 11 shows the iso-contours of pressure and vorticity
time interval was set to be 5 × 10−5 s. Figure 11 shows the iso-contours of pressure and vorticity
magnitudes in the zoomed plot as well as in the entire computational domain. It can be found that a
magnitudes in the zoomed plot as well as in the entire computational domain. It can be found that a
sudden pressure drop occurs at the nozzle throat, and strong vortices are formed downstream just
sudden pressure drop occurs at the nozzle throat, and strong vortices are formed downstream just
behind the throat, which causes a large amount of fluctuating turbulent disturbance components to
behind the throat, which causes a large amount of fluctuating turbulent disturbance components to
be convected in the downstream direction of the pipe. This turbulent disturbance generated acoustic
pressure inside the pipe.
In the ISO 9300 standards, it is required to measure the pressure at the positions 2D and 4D
downstream away from the end of the critical nozzle, where D is the pipe diameter. The pressure
fluctuations at the same locations denoted as Sensor 1 and Sensor 2 in Figures 3 and 11 are measured
Appl. Sci. 2019, 9, 2885 11 of 15

be convected in the downstream direction of the pipe. This turbulent disturbance generated acoustic
pressure inside the pipe.
In the ISO 9300 standards, it is required to measure the pressure at the positions 2D and 4D
downstream away from the end of the critical nozzle, where D is the pipe diameter. The pressure
fluctuations at the same locations denoted as Sensor 1 and Sensor 2 in Figures 3 and 11 are measured
using the experimental setup described in Section 3. The experimental devices and methods follow the
ISO 9300 and the experimental conditions are the same as the simulation conditions.
Figure
Appl. Sci. 2019,12
9, xcompares the 1/3 octave band spectrum of pressure between the measured and
FOR PEER REVIEW 11 ofthe
15
predicted ones at each sensor position. There is good agreement between two results at the position
of Sensor
of Sensor 22 while
while there
there is
is aa relatively
relatively large
large discrepancy
discrepancy between
between two
two results
results atat the
the position
position ofof Sensor
Sensor
11 where
where the numerical results results over-predict
over-predict in incomparison
comparisonwith withthe
themeasured
measureddata. data.This
Thisseems
seems to be
to
duedue
be to the factfact
to the thatthat
thethe
Sensor 1 is1located
Sensor in the
is located region
in the where
region the the
where strong vortex
strong is observed
vortex is observedin the
in
simulation
the simulation while
whilethethe
location of of
location Sensor
Sensor 2 is
2 isfarfarenough
enoughawayawayfromfromthe
the strong
strong turbulence
turbulence region.
However,the
However, thepredicted
predictedpressure
pressurelevels
levelsatatboth
bothlocations
locations become
become much
much lower
lower than
than thethe measured
measured oneone
in
in the
the high-frequency
high-frequency region
region above
above 2000 2000
Hz,Hz, which
which seems
seems todue
to be be due to poor
to poor resolution
resolution of grids
of the the grids in
in the
the corresponding
corresponding frequency
frequency range.range.
MoreMore precise
precise numerical
numerical simulations
simulations can becan be performed
performed by using
by using more
more resolved
resolved gridsfewer
grids and and fewer time intervals
time intervals but atbut
a muchat a much
more more expensive
expensive numerical
numerical cost. Since
cost. Since the
the AIV
AIV phenomenon
phenomenon is known is known
to occurto in
occur in the frequency
the frequency range ofrange
500 Hzofto5002000HzHz,to the
2000 Hz, the
current current
numerical
numerical was
resolution resolution
determinedwas determined to befor
to be sufficient sufficient for that purpose.
that purpose.

Figure 12. Total pressure level; (a) Sensor 1, (b) Sensor 2.

Figure 12. Total pressure level; (a) Sensor 1, (b) Sensor 2.
4.4. Wavenumber-Frequency Analysis
4.4. Wavenumber-Frequency Analysis
Based on the unsteady LES results, the wavenumber-frequency analysis was performed by
usingBased on the unsteady
the Fourier transformLESin results, the wavenumber-frequency
the cylindrical coordinate system as analysis was performed
described in Section 2.4by using
with
the in-house
the Fourier transform in thethe
code, of which cylindrical coordinate
validity was system
confirmed as previous
in the describedstudies
in Section 2.4The
[8,13]. withacoustic
the in-
house code,
power causedofby
which the validity
the sudden was variation
pressure confirmedinside
in thethe
previous
nozzle studies
provides [8,13]. The acoustic
the source power
of the AIV of
caused by the sudden pressure variation inside the nozzle provides the source of the AIV
the pipe wall. Therefore, for more precise assessment of AIV problems, it is necessary to extract the of the pipe
wall. Therefore,
acoustic for more precise
wave components assessment
propagating in theofdownstream
AIV problems, it is necessary
direction from thetofluctuating
extract thepressure
acoustic
wave components propagating in the downstream direction from the fluctuating pressure field
field in the pipe. Therefore, the pressure extraction area is selected to include the prescribed monitoring in the
pipe. Therefore, the pressure extraction area is selected to include the prescribed monitoring position
position as shown in Figure 13.
as shown in Figure 13.
The simulation time interval and the frequency interval used for the wavenumber-frequency
analysis were 5 × 10−5 s and 3.23 Hz, respectively, and the grid spacing was (∆θ, ∆z) = (2.5◦ , 0.0015 m).
The average speed of sound and Mach number in each region were 343 m/s, 0.064 in Region 1 and
0.011 in Region 2. The results of the wavenumber-frequency analysis are shown in Figure 14.

Figure 13. Acoustic pressure region.

The simulation time interval and the frequency interval used for the wavenumber-frequency
analysis were 5 × 10−5 s and 3.23 Hz, respectively, and the grid spacing was (Δθ, Δz) = (2.5°, 0.0015 m).
house code, of which the validity was confirmed in the previous studies [8,13]. The acoustic power
caused by the sudden pressure variation inside the nozzle provides the source of the AIV of the pipe
wall. Therefore, for more precise assessment of AIV problems, it is necessary to extract the acoustic
wave components propagating in the downstream direction from the fluctuating pressure field in the
pipe.Sci.
Appl. Therefore, the pressure extraction area is selected to include the prescribed monitoring position
2019, 9, 2885 12 of 15
as shown in Figure 13.

Appl. Sci. 2019, 9, x FOR PEER REVIEW 12 of 15

Figure 13. Acoustic pressure region.

The simulation time interval and the frequency interval used for the wavenumber-frequency
analysis were 5 × 10−5 s and 3.23 Hz, respectively, and the grid spacing was (Δθ, Δz) = (2.5°, 0.0015 m).
The average speed of sound and Mach number in each region were 343 m/s, 0.064 in Region 1 and
0.011 in Region 2. The results of the wavenumber-frequency analysis are shown in Figure 14.

Figure 14. Wavenumber-frequency analysis result; (a) Region 1, (b) Region 2. n = the radial direction
Figure 14. Wavenumber-frequency analysis result; (a) Region 1, (b) Region 2. n = the radial direction
mode; m
mode; m= =the
thecircumferential
circumferentialdirection
direction mode.
mode.

InIn
Figure 14, the black lines represent the modal cut-off lines in the pipe, and the circumferential
Figure 14, the black lines represent the modal cut-off lines in the pipe, and the circumferential
direction
direction mode
mode is is
represented
represented mm
byby and andthetheradial direction
radial directionmodemodeis represented
is represented bybyn. n.
TheThe dashed
dashed line
at line
the right
at the right bottom represents the wave components propagating at the constant speed V = 21.91in
bottom represents the wave components propagating at the constant speed V = 21.91 m/s
Region and V =13.87
m/s in1 Region andm/sV =in3.87
Region
m/s in 2, which
Regionis2,the incompressible
which pressure components
is the incompressible convected at
pressure components
theconvected
mean flow at velocity
the mean in flow
the downstream
velocity in the direction
downstreamand associated
directionwithand turbulence fluctuation
associated with turbulenceacting
asfluctuation
an aerodynamic noise source. Since the acoustic wave generated by
acting as an aerodynamic noise source. Since the acoustic wave generated by thethe aerodynamic noise sources
propagates
aerodynamic at thenoisespeed of sound,
sources the compressible
propagates at the speed pressure
of sound,fluctuation (planepressure
the compressible wave components
fluctuationin
the(plane
pipe)wave propagates
componentsin the in
downstream direction at
the pipe) propagates in the
the speed of c + udirection
downstream 0 , which at
is the
the speed
sum ofofthe c +sound
u ,
velocity
which and is thethesumflowofvelocity,
the sound and velocity
at the speed and ofthec − u0 in
flow the upstream
velocity, and at direction.
the speedInofthe c −case
u in of other
the
upstream waves,
non-planar direction. theInphase
the case of other
speed non-planar
is higher − u0 and
than c waves, + u0 inspeed
thec phase is higher than
the upstream c − u and
and downstream
c + u in respectively.
directions, the upstreamTherefore,and downstream
the pressure directions, respectively.located
wave components Therefore,
in thethe pressure
region betweenwave the
twocomponents
lines of which located theinphase
the region
speedbetween
is c − u0theand two
c +lines of whichthose
u0 denotes the phase speed is at
propagating c− auspeed c+
andhigher
u denotes
than the speed those propagating
of sound and thusat aare
speed higher thanacoustic
compressible the speed of sound andThe
components. thuscut-off
are compressible
line for each
acoustic
duct modecomponents.
can be defined Theascut-off line (10)
Equation for each
and isduct mode as
depicted can be defined
a curve as Equation
in Figure 14. In the (10) and
case of is
the
depicted as a curve in Figure 14. In the case of the incompressible
incompressible pressure fluctuation, the components propagating at the average mean flow velocity are pressure fluctuation, the
notcomponents
clearly visible, propagating
which isatdifferent
the averagefrommean flow velocity
the results shown are notstudy
in the clearlybyvisible,
Kim etwhich
al. [8].isThe
different
reason
for this is because the vortices of various sizes are formed and propagate at various speeds due toofan
from the results shown in the study by Kim et al. [8]. The reason for this is because the vortices
various
abrupt sizes in
change aretheformed and propagate
cross-sectional at various
area between the speeds
nozzle anddue the
to an abrupt change
downstream pipe.inInthethecross-
case of
sectional area between the nozzle and the downstream pipe. In the case of the compressible pressure
the compressible pressure fluctuation component, it can be seen that the spectral density is prominently
fluctuation component, it can be seen that the spectral density is prominently distributed in the
distributed in the vicinity of the cut-off line of each mode and becomes gradually weaker as the order
vicinity of the cut-off line of each mode and becomes gradually weaker as the order becomes higher.
becomes higher. As mentioned above, the predicted pressure level assures up to 2000 Hz, and high
As mentioned above, the predicted pressure level assures up to 2000 Hz, and high grid resolution is
grid resolution is required for higher frequency range. Therefore, the high-frequency components
required for higher frequency range. Therefore, the high-frequency components above 2000 Hz are
above 2000 Hz are rapidly attenuated.
rapidly attenuated.
The application of inverse wavenumber-frequency transform on the region of
The application of inverse wavenumber-frequency transform on the region of wavenumber-
wavenumber-frequency
frequency diagram including diagram onlyincluding
the compressibleonly the compressible
pressure pressure
fluctuations leads tofluctuations
the acoustic leads
powerto
thespectral
acoustic powereliminating
density, spectral density, eliminating the
the incompressible incompressible
pressure fluctuations.pressure fluctuations.
The acoustic power spectrum of the internal aerodynamic noise obtained in this way is shown
in Figure 15. It shows the power spectrum of three components: the incompressible pressure
fluctuation part due to turbulence fluctuation, the compressible pressure fluctuation part due to
acoustic waves, and the total pressure fluctuation including the incompressible and compressible
parts. From the spectrum and overall values of all parts, the power level in Region 1, which is the
sound, and is known to follow the −7 power law [17]. It can be found from Figure 15 that, in both
Region 1 and Region 2, the inertial subrange exists in the region between 100 Hz and 1000 Hz, and a
dissipation range exists in the frequency region higher than 1000 Hz. Especially in Region 1, where
larger coherent turbulent fluctuations exist and are followed by its energy cascade phenomena, the
inertial
Appl. subrange
Sci. 2019, 9, 2885and dissipation range are more prominent and distinct. Note that the decay rate 13 of of
15
the spectrum in the frequency range higher than around 2000 Hz to 3000 Hz is due to numerical
damping. The good agreement of the decay rate of the predicted spectrum confirm the validity of the
LES The acousticand
simulation powerthe spectrum of the internal aerodynamic
wavenumber-frequency noise
analysis. It is obtained in
noteworthy this
that way
the is shown in
magnitude of
Figure
compressible components is higher than that of incompressible ones in the frequency rangefluctuation
15. It shows the power spectrum of three components: the incompressible pressure belonging
part
to thedue to turbulence
dissipation rage. fluctuation, the compressible
This fact ensures pressure
that the energy fluctuation
of turbulent part dueisto
fluctuations acoustic waves,
transferred to the
and the total pressure fluctuation including the incompressible and compressible
compressible acoustic waves in the dissipation range. It can be also seen from the overall power parts. Fromlevel
the
spectrum and overall values of all parts, the power level in Region 1,
that both incompressible and compressible components are attenuated downstream, but the which is the region more upstream,
is greater than inpart
incompressible Region
is 2. The influence
further reducedofcompared
the incompressible part due to flow
to the compressible partis due
dominant in the
to more low
rapid
frequency region below
dissipation of the vortex. 1000 Hz, but the influence of the compressible part due to the acoustic wave is
dominant
Overall, in the region
these of 1000
results showHz to the
that 2000 Hz. However,
current numerical in methods
the higher frequency
predict range, itpower
the acoustic is rapidly
due
dissipated due to poor time and spatial resolution. It can be seen that the incompressible
to the valve flow in a pipe more precisely than the existing ones relying on the quasi-empirical part becomes
weaker
equations fromby Region
Carucci1andto Region
Muller,2 considering
because the more
vortexcomplex
rapidly dissipates
geometry and as shown in Figure 11b.
conditions.

Figure 15. Overall acoustic power level (OAPWL); (left) Region 1, (right) Region 2.

For the incompressible part related to the turbulence fluctuation, during the convection of vortical
5. Conclusions
waves downstream, its energy is transferred to smaller-scale eddies oscillating with the high frequency,
In this study, a high-fidelity numerical procedure is proposed to assess the acoustic power
that is so-called energy cascade phenomena occurring in the inertial subrange. The decay rate of
generated by pressure relief devices in a pipe. The method consists of three sequential steps: high
its spectrum is known to follow the −5/3 power law. The even smaller eddy scale, known as the
accuracy large eddy simulation, wavenumber-frequency transform, and duct acoustic theory. The
Kolmogorov Scale, belongs to the dissipation range where energy is dissipated due to heat or sound,
toroidal-throat venturi nozzle specified in ISO 9300 was selected as a target pressure relief device,
and is known to follow the −7 power law [17]. It can be found from Figure 15 that, in both Region 1
and the numerical simulations were performed using the computational fluid dynamics technique to
and Region 2, the inertial subrange exists in the region between 100 Hz and 1000 Hz, and a dissipation
predict the flow through the critical nozzle. To verify the accuracy of the numerical method, the mass
range exists in the frequency region higher than 1000 Hz. Especially in Region 1, where larger coherent
flow rates through the two types of nozzles were computed by varying the outlet pressure. It was
turbulent fluctuations exist and are followed by its energy cascade phenomena, the inertial subrange
confirmed that the numerical results were in excellent agreement with the experiment. Then, a high
and dissipation range are more prominent and distinct. Note that the decay rate of the spectrum in
the frequency range higher than around 2000 Hz to 3000 Hz is due to numerical damping. The good
agreement of the decay rate of the predicted spectrum confirm the validity of the LES simulation and
the wavenumber-frequency analysis. It is noteworthy that the magnitude of compressible components
is higher than that of incompressible ones in the frequency range belonging to the dissipation rage.
This fact ensures that the energy of turbulent fluctuations is transferred to the compressible acoustic
waves in the dissipation range. It can be also seen from the overall power level that both incompressible
and compressible components are attenuated downstream, but the incompressible part is further
reduced compared to the compressible part due to more rapid dissipation of the vortex.
Overall, these results show that the current numerical methods predict the acoustic power due to
the valve flow in a pipe more precisely than the existing ones relying on the quasi-empirical equations
by Carucci and Muller, considering more complex geometry and conditions.
Appl. Sci. 2019, 9, 2885 14 of 15

5. Conclusions
In this study, a high-fidelity numerical procedure is proposed to assess the acoustic power generated
by pressure relief devices in a pipe. The method consists of three sequential steps: high accuracy large
eddy simulation, wavenumber-frequency transform, and duct acoustic theory. The toroidal-throat
venturi nozzle specified in ISO 9300 was selected as a target pressure relief device, and the numerical
simulations were performed using the computational fluid dynamics technique to predict the flow
through the critical nozzle. To verify the accuracy of the numerical method, the mass flow rates
through the two types of nozzles were computed by varying the outlet pressure. It was confirmed
that the numerical results were in excellent agreement with the experiment. Then, a high accuracy
LES technique was performed on the three-dimensional critical nozzle, and the steady-state RANS
result was used as the initial condition to obtain more stable and faster convergence. The validity of
the unsteady LES results was also confirmed by comparing them with measured surface pressure data.
The wavenumber-frequency transform was taken on the LES results, and the compressible surface
pressure components matching the acoustical duct modes were identified in the wavenumber-frequency
pressure diagram. The inverse wavenumber-frequency transform taken on the compressible pressure
components leads to the acoustic power spectrum. The estimated total acoustic powers are 128.3 dB
and 125.1 dB for two regions, respectively. However, the incompressible part is further reduced
compared to the compressible part due to the rapid dissipation of vortex. These results reveal that the
current numerical procedure can be used to more accurately predict the acoustic power generated by
the pressure relief device in a piping system.

Author Contributions: C.C. provided the basic idea for this study and the overall numerical strategies. G.K.
carried out the numerical simulations and worked on the analysis of numerical results. S.L. developed the code
of wavenumber-frequency analysis. W.K. and K.K. constructed the experimental environment and conducted
the experiments.
Funding: This research was funded by the Ministry of Education, grand number NRF-2016R1D1A1A099 18456.
Acknowledgments: This research was supported by the Basic Science Research Program through the National
Research Foundation of South Korea (NRF), funded by the Ministry of Education (NRF-2016R1D1A1A099 18456).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Cowling, J. Acoustic and Turbulence Flow Induced Vibration in Piping Systems: A Real Problem for LNG Facilities;
Perth Convention and Exhibition Centre: Houston, TX, USA, 2016.
2. Carucci, V.A.; Mueller, R.T. Acoustically Induced Piping Vibration in High Capacity Pressure Reducing Systems;
ASME 82-WA/PVP-8; The American Society of Mechanical Engineerings: New York, NY, USA, 1982.
3. Eisinger, F.L. Designing piping systems against acoustically induced structural fatigue. J. Press. Vessel Technol.
1997, 119, 379–383. [CrossRef]
4. Eisinger, F.L.; Joseph, T.F. Acoustically induced structural fatigue of piping systems. J. Press. Vessel Technol.
1999, 121, 438–443. [CrossRef]
5. Veritas, D.N. L-002 Piping System Layout, Design and Structural Analysis, 3rd ed.; NORSOK Standard: Lysaker,
Norway, 2009.
6. Bruce, R.D.; Bommer, A.S.; LePage, T.E. Solving acoustic-induced vibration problems in the design stage.
Sound Vib. 2013, 47, 8–11.
7. Agar, M.; Ancian, L. Acoustic-induced vibration: A new methodology for improved piping design practice.
In Proceedings of the Offshore Technology Conference Asia, Kuala Lumpur, Malaysia, 22–25 March 2016.
8. Kim, K.; Ku, G.; Lee, S.; Park, S.; Cheong, C. Wavenumber-Frequency Analysis of Internal Aerodynamic
Noise in Constriction-Expansion Pipe. Appl. Sci. 2017, 7, 1137. [CrossRef]
9. American National Standards Institute. Measurement of Gas Flow by Means of Critical Flow Venturi Nozzles;
Dansk standard DS/EN ISO 9300; American National Standards Institute: New York, NY, USA, 1987.
10. Johansen, S.T.; Wu, J.; Shyy, W. Filter-based unsteady RANS computations. Int. J. Heat Fluid Flow 2004, 25,
10–21. [CrossRef]
Appl. Sci. 2019, 9, 2885 15 of 15

11. Myong, H.K. Chapter 6: Turbulence and Turbulence model. In Computational Thermo-Fluid Dynamics;
Munundang: Seoul, Korea, 2002; pp. 338–373.
12. Hoffmann, K.A.; Chiang, S.T. Computational Fluid Dynamics Volume III, 4th ed.; Engineering Education System:
Wichita, KS, USA, 1993.
13. Lee, S.; Cheong, C. Decomposition of surface pressure fluctuations on vehicle side window into
incompressible/compressible ones using wavenumber-frequency analysis. Korean Soc. Noise Vib. Eng.
2016, 26, 765–773. [CrossRef]
14. Munjal, M.L. Chapter 1: Propagation of waves in ducts. In Acoustic of Ducts and Mufflers, 2nd ed.; John Wiley
& Sons: Chichester, UK, 2014; pp. 2–20.
15. Kang, W.; Lee, S.H.; Lee, S.J.; Ha, Y.C.; Jung, S.S. Effect of ultrasonic noise generated by pressure control
valves on ultrasonic gas flowmeters. Flow Meas. Instrum. 2018, 60, 95–104. [CrossRef]
16. Park, K.A.; Choi, Y.M.; Choi, H.M.; Cha, T.S.; Yoon, B.H. The evaluation of critical pressure ratios of sonic
nozzles at low Reynolds numbers. Flow Meas. Instrum. 2001, 12, 37–41. [CrossRef]
17. Bonelli, F.; Viggiano, A.; Magi, V. How does a high density ratio affect the near- and intermediate-field of
high-Re hydrogen jets? Int. J. Hydrog. Energy 2016, 41, 15007–15025. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).