You are on page 1of 7

Excerpt from the Proceedings of the COMSOL Conference 2008 Hannover

Analyzing Muffler Performance Using the Transfer Matrix Method


Author K.S. Andersen*
Dinex Emission Technology A/S
*Corresponding author: Fynsvej 39, DK-5500, Middelfart, Denmark, ksa@dinex.dk

Abstract: Exhaust noise must meet legislation equation in three dimensions for acoustic
targets, customer expectations and cost reduction pressure [1]
which call for design optimization of the exhaust ω2
systems in the design phase. One solution is to ∇2 p + k 2 p = 0 ,k = 2 (1)
c
use 3 dimensional linear pressure acoustics and
calculate the transfer matrix of the muffler. The
ω is the angular frequency, k is the wave
number, c is the speed of sound and p is the
transfer matrix is the basis for calculating either
acoustic pressure.
the insertion loss or transmission loss of a
The assumptions of linearity using a
muffler. The 3D simulations in Comsol of
frequency model is valid for sound pressure
different muffler configurations are verified by
measurements in a flow acoustic test rig using levels up to 150 dB re 20 µPa, and for exhaust
system lengths below 15 m, but nonlinearities
the two source method.
may still occur at local constrictions such as
perforated elements, due to high pulsating
Keywords: Muffler, Transfer Matrix,
Transmission Loss, Two Source Method. velocities [2]. In addition, a constant temperature
through out the system is also assumed. Finally,
the Mach number is set to zero, while the flow
1. Introduction induced losses are implemented through the
boundary conditions.
The noise from an exhaust system consists of
three components: Pulsation noise and flow
generated noise coming from the orifice of the 3. Methods
muffler outlet and shell noise coming from the
shell of the muffler. 3.1 The CAE Tools
Shell noise may be limited by using a stiffer
or damped shell, while flow generated noise, The available CAE tools for analyzing
such as turbulence and vortex shedding may be muffler performance includes 1D and 3D linear
limited by minimizing geometrical acoustic codes with and without mean flow,
discontinuities (edges, sharp bends etc.). where the 3D may be either BEM or FEM based
Minimizing pulsation noise, caused by the methods. Comsol Multiphysics offers a 3D
valves opening and closing inside the IC engine, linear acoustic code where the most important
effect of flow is included by altering the
is the focus of this paper and this is obtained by
designing the internal parts of the muffler in such boundary conditions, while the mean flow is not
a way that the most critical part of the frequency included.
spectrum is attenuated.
3.2 The Measurement Tools

The available measurement tools include


sound pressure based insertion loss measurement
by means of ISO 11820 in a reverberation room
using real engines, transmission loss
Figure 1. Comsol Multiphysics simulation of the
measurement using the three microphone method
sound pressure level at a frequency of 100 Hz. [3] and transfer matrix measurements by means
of the 2-load or the 2-source method [4]. The 2-
source method was chosen as it offers the highest
2. Governing Equations
accuracy for comparison with the simulations.
The governing equation of the sound field in
a muffler is the linear time harmonic wave 4. Theory
4.1 The Plane Waves The sound pressure, pi, and volume velocity,
qi , on the inlet side of a muffler may be fully
The wave equation (1) for a single transferred to the outlet side (po, qo) by using the
dimension, e.g. the x-direction in a duct, is transfer matrix, T, which consists of the four
∂2 p ω2 elements, T11, T12, T21, T22,
+k2p =0 ,k = (2)
∂x 2 c2  pˆ i   T11 T12  pˆ o  (7)
  =   
The solution to the equation above may be  qˆ i   T21 T22  qˆ o 
written in sine and cosine functions or in The two-port theory assumes a linear and
exponential functions as time invariant system, with continuity in sound
p ( x) = p + e − jkx + p − e jkx (3) pressure and volume velocity across transitions.
p(x) is a sum of two plane waves, one
travelling in the positive x-direction with qi qo
T11 T12
amplitude p+ and the other in the negative x- pi po
direction with amplitude p-. The plane wave T21 T22
decomposition is shown in the figure below.
Figure 3. The basic transfer matrix parameters.
+ − jkx
p e
One advantage of using transfer matrices to
p − e jkx describe a muffler is that you can connect
multiple sub mufflers to one final muffler by just
Figure 2. Plane wave decomposition in a duct. multiplying the matrices.

By using Euler’s equation for motion in a 4.4 Transfer Matrix Extraction


medium, the volume velocity is determined as
S
( p + e − jkx − p − e jkx ), (4) A plane wave is applied at the inlet side and
q( x) =
ρ 0c the muffler is terminated at both ends by an
where ρ0 is the density and S is the duct area. impedance match implemented by using the
radiation condition. It is assumed that the sound
4.2 The Plane Waves in Mean Flow field in the connecting ducts is only one
dimensional, which is true for [1]:
The wave equation changes a little when 1.84c (8)
f <
mean flow is included πD
∂2 p ∂p D is the duct diameter and f is the frequency.
2
( )
1 − M 2 + k 2 p − 2 jkM = 0, (5)
∂x ∂x
where M = u 0 c is the Mach number. p1 p2 p3 p4

The solution is similar to equation (3), but the pi + po +


T11 T12 Anechoic
wave numbers are changed Source termination
p( x) = p + e − jk + x + p − e jk− x (6) pi - T21 T22 po – = 0
k k
k+ = , k− =
1+ M 1− M
The mean flow is stretching the waves in the
x1 x2 x2 +L x3-L x3 x4
same direction as the flow direction, while it is
compressing the waves in the other direction. Figure 4. Parameters for extracting the transfer
matrix.
4.3 The Transfer Matrix
When the equation system is solved, the
When staying in the plane wave region for acoustic pressure at two specific points on the
the inlet and outlet of the muffler, the muffler inlet side (p1, p2) and on the outlet side (p3, p4) of
can be described by two port theory in form of the muffler are extracted for further calculations,
either the scattering matrix, or the transfer matrix where p2, and p3 has a numerical distance of L to
[4]. the muffler.
p1 = pi + ⋅ e − jkx1 + pi − ⋅ e jkx1 to use when comparing different methods and
designs. TL in decibels is defined as the
p2 = pi + ⋅ e − jkx 2
+ pi − ⋅ e jkx 2
(9)
difference between the incident sound power and
− jkx 3 jkx 3
p3 = po + ⋅ e + po − ⋅ e transmitted sound power, assuming a reflection
p4 = po + ⋅ e − jkx 4
+ po − ⋅ e jkx 4 free termination [3].
The left and right going pressure waves on  p 2
TL = 10 log i + 2  (15)
the inlet side (pi-, pi+) and outlet side (po-, po+) of  p 
the muffler are.  o+ 
( ) TL may also be calculated using the transfer
j p1 ⋅ e jkx2 − p2 ⋅ e jkx1
pi + = matrix elements [1].
2 sin k ( x1 − x2 ) 2
1 S ρc  (16)
po + =
(
j p3 ⋅ e jkx4 − p4 ⋅ e jkx3 ) (10) TL = 10 log  T11 + T12
4
+ T21 + T22 

 ρc S 
2 sin k ( x3 − x4 )
(
j p2 ⋅ e − jkx1 − p1 ⋅ e− jkx2 ) IL depends on the source and is better to use
pi − =
2 sin k ( x1 − x2 ) when looking at individual practical solutions. IL
is defined as the difference between the radiated
po − =
(
j p4 ⋅ e − jkx3 − p3 ⋅ e − jkx4 ) sound power with a reference exhaust system (no
2 sin k ( x3 − x4 ) muffler) and the radiated sound power with a
The sound pressures (pi, po) and volume muffler system [3].
velocities (qi , qo) at the inlet and outlet plane of PL, which is a gas dynamic parameter,
the muffler are (see figure 3). depends only on the muffler and is defined as the
pi = pi + ⋅ e − jk ( x 2+ L ) + pi − ⋅ e jk ( x 2+L ) difference between the static pressure before the
pi + ⋅ e− jk ( x 2+ L ) − pi − ⋅ e jk ( x 2+ L ) muffler and the static pressure after the muffler.
qi = S (11) TL will in this paper be used to evaluate the
ρc performance of the muffler.
− jk ( x 3− L ) jk ( x 3− L )
p o = p o+ ⋅ e + p o− ⋅ e
− jk ( x 3− L )
− po− ⋅ e jk ( x 3− L )
p o+ ⋅ e 5. Numerical Model
qo = S
ρc
The boundary conditions inside the muffler
The transfer matrix is symmetric [1]
are solid walls, coupling boundary conditions
T11 = T22 , (12)
and perforated plates. The coupling boundary
and reciprocity requires the matrix determinant conditions are used to couple two connecting
to be equal to one [1] sub-domains, where sound will propagate
T11T22 − T12 T21 = 1 (13) unaffected, e.g. a duct connected to another duct.
By using the two specific properties of the The perforated plates are implemented as a
transfer matrix mentioned above, the four complex impedance, Z, where θ and χ are the
elements of the transfer matrix becomes real and imaginary part respectively [5].
po qo + p i qi p i2 − p o2 Z (18)
T11 = T12 = = θ '+ jχ '
pi qo + po qi p i qo + p o q i (14) ρc
q i2 − q o2 p q + p i qi 1 8µk  
T21 = T22 = o o θ '= 1 +
t
 + θ f (19)
pi q o + po qi pi qo + po qi σ ρc  d h 
1 (20)
4.5 Performance Evaluation χ'= k (t + δ h )
σ
The flow resistivity, θf, and the Comsol
The acoustic performance of a muffler may parameter for the diameter of the hole, dh, are
be evaluated using either the transmission loss modified to account for flow induced losses. The
(TL) or the insertion loss (IL), while the flow modifications were done by comparing the real
performance is evaluated using the pressure loss and imaginary part of the impedance to the
(PL), also named the backpressure. model by T. Elnady [6]. The result were the
TL only depends on the muffler and not on
below parameters, where σ is the porosity of the
the source (inlet and outlet length and perforate and d is the actual diameter of the hole.
impedance) and is considered the best parameter
Table 1: Modified parts of the perforate impedance. The signal from the stepped sine generator
Comsol Dinex driving the loudspeakers, is used as reference for
Flow resistance θf =0 M the microphone transfer functions, using 100
θ f = 1.57 averages for enhancing the signal to noise ratio.
(bias flow) σ
Flow resistance θf =1
θ f = 0 .5
M
(grazing flow) σ 6.2 The Variable Standard Muffler
Hole diameter dh = d (
d h = 81000k ⋅ d )
1.5 −1
In order to verify the simulations, a variable
End correction d 1 standard muffler was constructed. Pictures of the
δh = h δh = d
4 4 basic three designs are shown below.

The medium in which the waves are traveling


is either air or absorptive material described by
Delany and Bazley [5]. The flow resistivity, Rf,
of the absorptive material is estimated using the
formula by Bies and Hansen [5], leaving only the
apparent density σap and the average fiber
diameter, dav to be determined.
(21) Figure 7. The reflective muffler.
R f = 3.18 ⋅ 10 −9 ⋅ ρ ap
1.53
⋅ d av−2

6. Results

6.1 Experimental Setup

The experimental setup based, on the two


source method, consists of a centrifugal fan
feeding cold air into the muffler and 6
neodymium loudspeaker units for sufficient Figure 8. The absorptive muffler.
sound pressure level compared to flow generated
noise.
Microphones Microphones Centrifugal fan

Test object
Loudspeaker Loudspeaker
s
Figure 5. Layout of the flow acoustic
s test rig for
experimental validation.
Figure 9. The plug flow muffler.
Six flushmounted ¼” B&K microphones for
double frequency span are connected to the 6.2 The Reflective Muffler
analyzer.
The reflective muffler reflects the low
frequency sound waves when the length of the
expansion chambers or quarter wave resonators
equals a multiple of a quarter of a wavelength.
720 mm

300 mm

320 mm 97 mm
Figure 6. Picture of the flow acoustic test rig.
Figure 10. An expansion chamber with a resonator.
Transmission Loss slowed during motion in the absorptive material.
30
Calculated
The effect of grazing flow in the perforate is
25 Measured limited compared to that of the absorptive
Simulated material.
720 mm
Transmission Loss [dB]

20

15 300 mm

ooooooooooooo
10

97 mm
5
Figure 13. The absorptive muffler.
0
Transmission Loss
60
-5
0 500 1000 1500 2000 Measured
Frequency [Hz] Simulated
50
Figure 11. Transmission loss for a basic expansion

Transmission Loss [dB]


chamber. 40

The measured and simulated transmission loss of 30

the expansion chamber (figure 11) correlates


well, besides a little offset in the peaks. This 20

might be due to inaccuracies in different 10


parameters (lengths, temperatures, densities etc.).
A simple analytical calculation is also shown, 0
0 500 1000 1500 2000
and it does not capture the higher order modes. Frequency [Hz]
The first axisymmetric higher-order mode will Figure 14. Transmission loss for a pure absorption
start to propagate above 1400 Hz. muffler.
Transmission Loss
60
Measured The effect of the absorptive material (Rf = 5000
50
Simulated
Rayls/m) sets in above 100 Hz, and supplies a
constant growing transmission loss (figure 14).
Transmission Loss [dB]

40 The correlation is good until 350 Hz, above


which the simulation at first overestimates, then
30
underestimates the transmission loss. The Delany
20
& Bazley model may be inaccurate when used in
a sub domain as large as this. Additional tests
10 with smaller sub domains could clarify this.

0
0 500 1000 1500 2000 6.4 The Plug Flow Muffler
Frequency [Hz]

Figure 12. Transmission loss for an expansion The plug flow muffler works at all frequencies
chamber with a 320 mm quarter-wave resonator. by dissipating sound energy when vortices at the
perforate interfere with the sound field. The flow
The measured and simulated transmission loss of in the perforate is now mainly bias flow in
an expansion chamber with a 320 mm quarter- opposition to grazing flow.
wave resonator also exhibits good correlation 720 mm
even though the peaks are also shifted a little.
300 mm
6.3 The Absorptive Muffler
ooooo

The absorptive muffler works at mid and high 280 mm 97 mm


frequencies by dissipating sound energy when
Figure 15. The plug flow muffler.
the particle velocity of the sound waves is
smoothing of the peaks and dips, except for the
peak at 1350 Hz.

Transmission Loss
60
v = 30 m/s
v = 20 m/s
50 v = 10 m/s
v = 0 m/s

Transmission Loss [dB]


40

Figure 16. The perforated pipes used to verify the


simulations. From left: Ø3 25.0%, Ø4 30.7%, Ø4 30

23.0%, Ø4 15.3%.
20
Transmission Loss
60
Measured
10
Simulated
50
0
Transmission Loss [dB]

0 500 1000 1500 2000


40 Frequency [Hz]

Figure 19. Simulated transmission loss for a plug


30
flow muffler. Ø3 mm holes with 25 % porosity at
different flow speeds.
20

10 Increasing flow velocities increases the


smoothing of the peaks and dips which is seen in
0
0 500 1000 1500 2000
figure 19.
Frequency [Hz] The flow in an automotive exhaust system is
Figure 17. Transmission loss for a plug flow muffler. pulsating (in opposition to a constant flow),
Ø3 mm holes with 25 % porosity at v = 0 m/s. hence the effect of the perforate increase in real
applications.
The simulation of the plug flow muffler (Ø3 mm The figure below shows that decreasing the
holes and 25 % porosity) without flow, shown in porosity has the same effect as increasing the
figure 17, captures the resonator effect at 800 Hz velocity, since they are both included as a real
due to the 80 mm extended inlet before the part of the impedance in the boundary condition
perforate section begins, though it for the perforate. Additional simulations and
underestimates the high frequency transmission measurements have shown that the diameter of
loss. the holes in the perforate may be varied between
3 mm and 12 mm without any significant
60
Transmission Loss influence.
Measured
Simulated
50 Transmission Loss
60
15.3 % porosity
Transmission Loss [dB]

40 23.0 % porosity
50
30.7 % porosity
Transmission Loss [dB]

30
40

20
30

10
20

0
0 500 1000 1500 2000 10
Frequency [Hz]

Figure 18. Transmission loss for a plug flow muffler. 0


0 500 1000 1500 2000
Ø3 mm holes with 25 % porosity at v = 30 m/s. Frequency [Hz]

Figure 20. Simulated transmission loss spectrum for a


When a flow speed of v = 30 m/s is added plug flow muffler Ø4 mm holes with different
(figure 18), the simulation captures the porosities at v = 30 m/s.
6.5 Simulation setup Finally the simulations will be used to
optimize production parameters such as minimal
Setting up a model takes about 20 minutes material consumption, simple construction,
from importing the CAD model to hitting the simple production and small.
solve button, while the solution time is 10
minutes when taking 100 frequency steps and 9. References
using 5 elements pr. wavelength and solving the
system using the PARDISO solver on a normal 1. A. D. Pierce, Acoustics – An Introduction to
desktop pc. Its Physical Principles and Applications, page
17, 317, 351, 353. Acoustical Society of
8. Conclusions America, (1989)
2. H. Bodén, M. Åbom, Modelling of Fluid
This paper shows how the transfer matrix Machines as Sources of Sound in Duct and Pipe
may be extracted using the complex wave Systems, Acta acustica, 3, p. 549-560 (1995)
amplitudes at two specific points on the inlet side 3. M. L. Munjal, Acoustics of Ducts and
and outlet side, while running only a single Mufflers, page 21, 55-59, 201-205. John Wiley
simulation at multiple frequencies. & Sons, Inc., (1987)
This method makes it possible to calculate 4. M. Åbom, Measurement of the Scattering-
the insertion loss in opposition to the direct matrix of Acoustical Two-ports, Mechanical
calculation of the transmission loss based on an Systems and Signal Processing, 5(2), 89-104
integration of the pressure across the inlet and (1991)
outlet of the muffler [5]. 5. COMSOL 3.4, Acoustic Module Model
The simulation results of the reflective Library, 74-89, 154-170. COMSOL AB (2007)
muffler and the plug flow muffler with perforate 6. T. Elnady, Modelling and characterization of
correlate well with the measurements, but work Perforates in Lined Ducts and Mufflers, doctoral
needs to be done in the field of absorption in dissertation. Dept. Aeronautical and Vehicle
order to trust the simulations of the absorptive Eng., Royal Institute of Technology, Stockholm
muffler. The model settings could be updated (2004)
using the measurements.
No frequency limitations, short setup time, 10. Acknowledgements
and easy redesign are among the advantages of
using 3D pressure acoustic simulation while the Mats Åbom and Hans Bodén at KTH for
down sides reads relatively long simulation time practical help during design of the flow rig.
compared to 1D acoustic codes and no “easy” Tamer Elnady at Ain Shams University for
pressure loss simulation. help with the experimental data analysis.
Still this approach is faster than using 1D gas Christian Svendsen at Danfoss Compressors
dynamic codes. GmbH for help with the automation of the
measurements.
8. Perspectives Mads Nygaard, Henrik Breiner Jensen, Mads
Juul Jacobsen and Carsten Thiel at Dinex
The confidence achieved in the simulation Emission Technology A/S for help with the
procedure makes it possible for Dinex to mainly construction of the flow rig.
use simulations in the early design phase. And finally Comsol support for continuously
Future work will involve simulation of helping with simulation problems.
pressure loss and mean flow distribution by
perhaps using impedance implementation of the
perforated elements.
The transfer matrices obtained will be of real
value, when the source impedance of a real
engine can be used together with the transfer
matrix to calculate the insertion loss, and later on
compared to direct measurements according to
ISO 11820.