Shape Optimization of Reactive Mufflers*

Robert J. Bernhard
School of Mechanical Engineering; Ray W. Herrick Laboratories; Purdue University; West Lafayette, IN
47907
*Received 25 November 1985; revised 5 April 1986

T e c h n i q u e s for c o m p u t i n g t h e p e r f o r m a n c e o f mufflers b y t r a n s m i s s i o n matrix t e c h n i q u e s

are well established. In this paper, t h e transmission matrix n u m e r i c a l t e c h n i q u e s are
r e f o r m u l a t e d to p e r m i t t h e c o m p u t a t i o n o f muffler d e s i g n sensitivity information as well.
W h i l e t h e d e s i g n sensitivity information has o t h e r applications, it is u s e d h e r e t o g e t h e r
w i t h a formal d e s i g n optimization algorithm to develop an a u t o m a t e d muffler d e s i g n
p r o g r a m . T h e p e r f o r m a n c e of t h e muffler optimization p r o g r a m is i l l u s t r a t e d for t h e
d e s i g n o f several s i d e b r a n c h and expansion c h a m b e r mufflers w h e r e b o t h one-
d i m e n s i o n a l a n d t w o - d i m e n s i o n a l w a v e p r o p a g a t i o n b e h a v i o r is c o n s i d e r e d .

Analytical procedures to predict the performance of acous- converge to an optimal solution. If the system identification
tical muffling and silencing systems have received consider- process is difficult, it is likely the designer will utilize a sub-
able attention, particularly in the last two decades. Several set of the actual design variables available. In such an itera-
recent survey articles summarize muffler analysis investiga- tive procedure, the optimal solution is also not guaranteed
tions. 1.2 While the propagation of sound in such mufflers is since the designer has poor information about whether a min-
complicated, and in some cases non-linear, in many cases it imum has been achieved with respect to all of the design var-
has been shown that classical linear acoustic models will give iables.
a reasonably faithful prediction of the muffler performance. When design sensitivity information is available, the mul-
While other techniques have been used, ~ transmission ma- tivariant optimization process can be simplified significantly.
trix techniques appear to be most popular because of their The design sensitivity information provides a great deal of
versatility and efficiency. 4 Considerable effort has been ded- system identification information. As a result, the system be-
icated to formulating typical complicated muffler geometry to havior can be understood more readily and convergence to an
the transmission matrix form. Transmission matrices for optimal solution can be achieved more efficiently. Further-
pipes with flow, temperature gradients, and perforated ele- more, when the design sensitivity is zero, the designer knows
ments have been developed. 5-1oIn addition, the finite element he has achieved at least a local optimum design solution.
and boundary element methods have been formulated to Thus, the optimal design is achieved more efficiently, and as
compute the proper transmission matrices when duct dimen- a by-product, the designer has more information available to
sions are sufficiently large to allow non-plane wave propaga- understand the muffler system better.
tion in the mufflers, lo-14 This paper will derive and demonstrate procedures for
However, while the transmission matrix methods can be computing design sensitivity information for mufflers and si-
used to give a reasonably faithful prediction of muffler per- lencers using transmission matrix analysis techniques. The
formance, in many cases the final objective of analysis is the techniques will be demonstrated for systems with one-di-
design of an optimal muffler system for a particular applica- mensional and two-dimensional components. The design sen-
tion. Unfortunately, current analysis programs give the de- sitivity information will also be incorporated into a design op-
signer very little information on how to proceed from the cur- timization program. Optimization to several typical muffler
rent configuration to an optimal solution. Such information performance criteria will be demonstrated.
will be referred to in this paper as design sensitivity infor-
mation. Thus, the designer must resort to an iterative pro-
cedure first to identify the behavior of the muffler system ver- Design Sensitivity Matrices
sus several design variables and then to find the optimal
configuration. Such a procedure requires a great deal of anal- Using transmission matrix theory for a muffler such as the
ysis, first to identify the behavior of the system and second to one shown in Fig. 1, the pressure and volume velocity at the

10 NOISECONTROLENGINEERINGJOURNAL/July-August1986
 sidebranch. If the sidebranch is modeled by transmission
matrices, the impedance at the mouth, Zb, is
AZr+ B
A , ,C I D E Zb = CZr + D (8)

I I' where Zr is the acoustic impedance at the termination of the

sidebranch, if an outlet exists. If the sidebranch is completely
Figure 1--Schematic of a typical automotive muffler closed, Zr * ~ and
A
Zb = ~ (9)

muffler system inlet, P1 and V1, are related to the pressure The relative extrema (i. e., maxima or minima depending
and volume velocity at an output location, P2 and V z, by the on the problem under consideration) of the performance met-
matrix equation rics are found by identifying the value of a design variable, ai,
where the partial derivative of the performance metric with
respect to that design variable is zero. For instance, for in-
sertion loss

where the composite transmission matrix T is the product of o', L',2 ( 0¢/
0ai - 2 Re ~t c3ai j (loa)
all the individual component transmission matrices
where
[T] = [Ta] [Tb] [Te] [Td] ... (2)
c3t* - Z r 0A*
Oai
+ ~0B* 0C* -4- Zs-~-al
q- Z s Z r " ' ~
0D* (10b)
Each of the component transmission matrices and the com- 0a~
posite transmission matrix are made up of four terms: and the superscript asterisks indicate the complex conjugate.
Thus, the derivatives of each of the composite transmission
[Ti] = Ci Di matrix terms with respect to the design variable must be
found. In a muffler where the composite transmission matrix
The evaluation of the terms of the component transmission is found from the multiplication of component transmission
matrices will be discussed later. matrices, the design sensitivity matrices are found by matrix
The terms in the composite transmission matrix, T, are multiplication. For example, in a two-component muffÊer sys7
used to evaluate various muffler performance metrics. For tern
convenience in later explanations each metric will be defined
in terms of a secondary variable. The logarithm of the sec- [T] = [T 1] [Tz] (11)
ondary variable is proportional to the metric. The secondary
variable must be minimized to maximize each metric. and
Transmission loss (TL) is computed from
[aT] = rL T,aa, 1j [T2] + [T,] [~T~ 1.I " (12)
TL = 101og10(1) , (4)
where The matrices, [ 0T ], can be assembled when T is assembled
t oa1 J
and used with equations such as Eq. 10 to give the sensitivity
T= I 2 12
, (5) of the performance index to variable a i. If this sensitivity is
(A+B S D) zero, the design has achieved at least a local extremum.
To illustrate the form of the sensitivity matrices, consider
and S is the area of the inlet and exhaust pipes. one-dimensional sound propagation in a lossless duct with no
For the case where the source and termination impedances flow. The component transmission matrix is
are not changed by the insertion of a muffler, the insertion
loss (IL) is [ coskl - ~ s i n k l ]
It 12
IL = - l o l o g I~ t ) (6) [T] = /jS sinkl coskl J (13)
I~o} LPoC
where
L = A Z r -Jr B + C Z s Z r -4- DZs (7) There are two design variables for this one-dimensional com-
ponent, the length i and the cross sectional area S. The sen-
where Zs is the acoustic impedance of the source, Z r is the sitivity matrix for length is
acoustic impedance at the muffler exit and Lois the secondary
-ksinkl Jk-~sc coskl"
variable for the case with no muffler in place.
For a complicated sidebranch resonator the important [-~-T] = jkS c o s k l - k s i n k l (14)
acoustical parameter is the impedance at the mouth of the poc

Volume27/Number I 11
The area sensitivity matrix is 18 19 20 21
17 A
Exhoust
--

Poc
0
J sinkl
-JP~Cosinkl ]
(15)

For muffler components of characteristic cross sectional di-

mension of half a wavelength or more, one-dimensional ana-
Inlet I9
14

6
"i-
I0
15

"tl
7
12
16'

8
lytical techniques are no longer appropriate. Finite element
analysis procedures are often used to evaluate the transmis-
sion matrices of such components, lO-12The procedure to eval-
l z 4 5
uate the transmission matrices involves the evaluation of two Figure 2--Typical finite element model of a muffler chamber
boundary condition, cases by finite element analysis, the case
with a closed termination and the case with a pressure re-
lease termination. For example, for the muffler shown in Fig. is solved at each frequency of interest for the two boundary
2, the inlet at node 6 and the exit at node 16 are small relative condition cases. The elements of the P-vector are the pres-
to the dimensions of the muffler and the acoustic wavelength. sures at each of the model node points. The elements of the
Thus, the boundary conditions can reasonably be applied at f-vector are calculated from known values of the volume ve-
the nodes. For case 1, the volume velocity at node 6 is set to locity at certain node points. Certain values of the P-vector
unity and the volume velocity at node 16 is set to zero. The solution are used to determine the four values of the trans-
muffler is analyzed to find the pressure distribution. The mission matrix as discussed in Eqs. 16 through 19.
transmission matrix parameters A and C of Eq. 3 are found The transmission sensitivity matrix terms of Eq. 10 are
from found using a matrix equation which is the derivative of Eq.
A- P6
P~6
(16) 21 to solve for the pressure sensitivity ~-,P: ooi
and
1 { of } , (22/
C - P]6
(17) [~__~__G
] {p} + [G] {a-~-i} = 0ai

For case 2, the pressure at node 16 is set to zero and the vol- where ai is the ith shape parameter. The P-vector is known
ume velocity at node 6 is set to unity. The transmission matrix from the solution of Eq. 21 and the G-matrix can be saved in
parameters B and D are found from decomposed form, shown here for convenience as [G]-1.
Thus, the pressure sensitivities are
B- P6 (18)
V16
and
{0-~i }= [G-X] [{0~. } - [ ~i ]{P}] , (23)
1
D- V16 (19) aG is found from the shape sensitivity
where the matrix [--37]
formulation. The { @ } vector is found by matrix multiplica-
The possible shape changes for such mufflers are probably
limitless. However, for most practical problems, manufactur- tion; thus, no matrix equation solution is necessary.
ing considerations leave a finite number of design variables. As an example of the formulation of the shape sensitivity
For this "constrained" optimization problem finite element matrices, consider an axi-symmetric geometry where the ra-
shape sensitivity matrices can be generated and evaluated to dial dimensions are allowed to vary by
determine the shape sensitivity of the transmission matrices.
Details of formulation of the shape sensitivity matrices are r 2 = air I + a2 (24)
discussed by Bernhard. 15.16Some further details of the devel-
opment of shape sensitive transmission matrices are pre- The stiffness matrix is
sented here for completeness.
For convenience, a G-matrix which is a function of fre- [K] = [K1] + a.__.g_2[K2]
al
(25)
quency and is a composite of the acoustic "mass" and "'stiff-
ness" matrices is defined such that and the mass matrix is

[G] = [K] - k 2 [M] , (20) [M] = a~ [M1] + ala2 [M2] (26)

where K is the "stiffness" matrix, M is the "mass" matrix and where K1, K2, M1 and M2 are partial stiffness and mass mat-
k is the acoustic wave number. The matrix equation rices as described by Bernhard and Kipp. 17Thus,

[G] {P} = {f} (21) [~_~1] = -a_______2a[K2]

~ + k2 [2al [M1] + a2 [M2]] (27)

12 NOISE CONTROLENGINEERINGJOURNAL/July-August1986
 The Muffler Optimization Program

Design sensitivity information is helpful in any design

Input: experience. However, if the designer can write a valid per-
Partial K 8~ M Matrices
formance equation, it is possible to formulate the design
problem as an optimization problem. Design sensitivity in-
Input: formation permits the use of efficient gradient search optim-
- Wave No.
- Source Impedance ization algorithms. As an illustration of one of the uses of the
- Termination Impe~nce design sensitivity information, a muffler optimization pro-
-No, of Chambers
gram was developed using both one-dimensional and finite
element theory with a gradient search optimization algo-
rithm.
Clearly, a complete optimizer algorithm must have the ca-
pability to apply constraint equations and should determine
whether global minima rather than local minima are found.
On the other hand, in some instances a less complete optim-
izer or no formal optimizer is preferred in the interest of
Input: [ Input: economy. Such issues will not be addressed in detail here in
Design Voriabte Type -Matrix Assembly Info.
I = Length - Design Variable ]nfo. deference to the development of an efficient analyzer com-
2 = Area ponent of the optimization algorithm which supplies both
3= Length & Area
4 -" None performance and performance sensitivity information for
shape design of mufflers. For the illustrations in this paper a
l , conjugate gradient subroutine, ZXCGR, from the IMSL math
Input: J ( Input:
Initial Length ~, Area Initial Valuesof Design Variables subroutine library, was [Link] With conjugate gradient
1 I search algorithms there is no capability to impose design con-
straints or to check for global minima. In some cases, not hav-
ing these capabilities resulted in improbable acoustical
shapes or multiple optimal shapes, depending on the initial
shapes used. However, improved search algorithms exist
(Compute ReferenceCondition for IL)
which include the additional features to impose constraints or
search for global minima, if these features are desired.
(Det¢ t :nes (~ew VariabLes & Co,its ANAZR Until [Link]~
A flowchart of the program used for these studies is shown
in Fig, 3. The required input data includes information about
l
Ioutpu. Solo'ion I
the muffler system such as the number of chambers and the
source and termination impedance. For each chamber addi-
tional geometric information is required. If the chamber is
one-dimensional, the length or cross-sectional area or both
may be variables, In some cases, chambers with no allowed
variation may exist in the muffler system, in which case no
Figure 3--Flowchart of the muffler optimization program design variables are assigned to the chamber.
If the chamber is not one-dimensional and finite element
techniques are necessary to evaluate the transmission char-
and
acteristics and sensitivities of the chamber, information typ-
ical of a finite element model is required. In many such muf-
[_~_a2
]~G = a---~-I[K2] + k2al [M2] . (28)
fler r:,odels the muffler chamber can be divided into
substructures or superelements as shown in Fig. 4. These
Certain terms of the sensitivity vector [--~a ]are used to
create the transmission sensitivity matrices using a proce-
dure similar to that used in Eqs. 16 through 19. For example,
Inlet Exhaus'r
OP6
0A aai P6 0P16 (29)
Oai - ~ P26 dai '
where ~
0P6 ~ OP16 OP
ana --g~ are terms in the v ~ - vector from Eq. 23. / \
Equation 12 is used to find the composite sensitivity matrix. Superelement #1 Superelement32
The four elements of the composite sensitivity matrix are
used in an equation like Eq. l0 to find the metric sensitivity. Figure 4---Muffler chamber model using substructuring

Volume 27/Number 1 13
superelements share the same finite element and shape sen-
sitivity matrices. Thus, for the optimization program used in
this investigation, "partial" stiffness and mass matrices are
read to memory at the beginning of the program. The finite
element model of a chamber is found by supplying the pro-
gram with information on how to assemble the overall stiff-
ness and mass matrices and design sensitivity matrices from
the partial stiffness and mass matrices and the design varia-
bles. Initial values of the design variables are also provided.
When all the modeling information has been supplied, the
program calls subroutine ANAZR which evaluates the metric
coefficient of interest at the initial state. The initial value of
the metric coefficient is necessary when insertion loss cal-
culations (IL) are made and is also used by the optimization
algorithm ZXCGR to establish the first several design itera-
tions. The subroutine ANAZR is also called by ZXCGR dur-
ing the design optimization procedure to evaluate the metric
and sensitivity information at each design iteration. The
ANAZR subroutine is essentially an implementation of the
muffler performance and design sensitivity equations dis-
cussed in Eqs. i through 28.

Results of the Optimization Studies

The muffler shape optimization program was used first to

design several sidebranch resonators. In order to demonstrate
and verify the program, the first test case was a closed pipe
sidebranch. The appropriate design length of the pipe should
be an odd multiple of a quarter-wavelength. Figure 5 shows
the convergence history for several cases. As expected, the
optimizer converged to different solutions (local minima) de-
\
pending on the initial length of the pipe. For some initial po-
sitions the design converged in one or two iterations. In all Figure 6--Concentric tube Helmholtz resonator geometry

7 cases, the solution converged within five iterations. The rate

Initial Length =4.40 of convergence was also found to be a function of starting po-
• Initial Length = 2 . 8 0 sition. The data shown is typical of performance one could ex-
6
¢-- • Initial Length = 1.25 pect for an optimization problem with one variable.
-4,"- The second case study was of a concentric tube Helmholtz
¢- 5 0-~-8 resonator as shown in Fig. 6. The radius of the Helmholtz re-
; • • sonator mouth opening was held constant. The radial and ax-
- • ial dimensions of the resonator volume were made the design
variables. The inner and outer radii of the resonator volume
--HI " are related in this problem. Figure 7 shows the convergence
lll
history for a muffler designed for wave number k = .25, and
o~
¢j}
the convergence of the square of the mechanical impedance
w 2 at the neck opening, ~tZmo~ 2. At iteration 3, the square of the
input mechanical impedance became quite high because of a
poor choice of the design variables. However, the program
i~ .A~_.- .._ ~ __
recovered quickly and within two more iterations had essen-
tially converged to the final solution.
I I I I I t
The muffler 6ptimization program was also used for the de-
0 I 2 3 4 5 6 7 sign of expansion chamber mufflers with both one-dimen-
Iteration Number sional models and two-dimensional finite element models us-
ing insertion loss as the optimization criterion. In each case,
Figure 5---Convergence history for closed pipe sidebranch optimi- the source impedance and radiation impedance are assumed
zation studies (k = ~r) to be equal to the characteristic impedance podS. The first

14 NOISE CONTROLENGINEERINGJOURNAL/July-August1986
 .... i ,A i! iiO ii i! iiiii i!ii!!iiiiiii!i!!iiiiiii!i!Yili!!ili
A ea S ;l: :;

Figure 8--Multiple chamber muffler

,~ : /=:/:/?

:' o?;2 . . . . :;;: !,&:

':::':2'2::; ::22222:-) : : : : : ....: : :

: :o

Figure 9--Convergence history of the multiple chamber muffler

When the performance index is relatively insensitive to de-

Figure 7--Convergence history for the concentric tube Helmholtz sign changes, as it is for expansion chambers near the opti-
resonator mum length, gradient optimizers tend to use more iterations
than necessary. Convergence tolerances can be adjusted to
case study is shown in Fig. 8. The muffler consists of three compensate and increase the efficiency of the optimizer. It
chambers with expansion and contraction area ratios of 10. might reasonably be argued for such problems that automatic
The design variables are the length of each of the chambers. optimization is less efficient than supplying design sensitivity
This case study uses one-dimensional plane wave theory as information to a human designer who will find an acceptable
discussed in Eqs. 13 and 14. Convergence is shown in Fig. 9. design alternative. On the other hand, there is some educa-
As expected, for the optimum muffler each chamber length is tional value in having the optimizer find the optimal solution
made a quarter wavelength. Note that after the second iter- so that the physics of the muffler can be understood.
ation the performance does not improve significantly but the The finite element analysis capabilities of the program
muffler dimensions change until the expected optimal solu- were used, first, for several classical problems to verify the
tion is found. There are positive and negative aspects of such program. The results are not shown but in each case the cor-
a result. The performance of the muffler does not improve rect solution was obtained. To investigate more interesting
significantly after the second iteration. Thus, the value of the muffler alternatives, several more complicated models were
additional design iterations from a design viewpoint is small. studied. The models were developed using a 21-node super-

Volume27/Number 1 15
 v T v . ical effects which have been ignored in the models. However,
the program has found what is likely to be an extremely high
Va - Exhaust
performance muffler chamber shape.
As a second design problem using finite elements, nine of
.- ÷-i-[~]~--'
- 1
the superelements were assembled as shown in Fig. 12. The
resulting model has six design variables. Both the muffler in-
V~ let and outlet are in the interior of the chamber. Figure 13
shows the convergence of the design. The solution does not
converge as quickly for the problem with six variables as it did
h
with four variables. However, it is evident that the program
Inlet
V3 is finding a relatively high performance solution in this case
as well.
Figure lO---Muffler chamber model using four finite element super- For the two finite element studies, the muffler perfor-
elements and four design variables mance is shown to be extremely sensitive to design changes
near the optimal values. For these problems, additional de-
sign iterations resulted in significant improvement in the
muffler. The gradient optimizer very intelligently converged

~--o-- - --0-- -

I Exhaust
t~

! I " I

tN tN

T
---O- -t---Q- --4
I' q
L '
L
trn tN *,N
v, I
-*-÷-*
I
.... *-4--~-
I

I,
I I I

•~ - - - - - - V 4 ~V6~
Figure 12--Muffler chamber model using nine finite element super-
elements and six design variables

Figure 11--Convergence history of the four superelement muffler

model (k = .25)

element made up of four quadratic two-dimensional elements

as shown in Fig. 2. The models represent a three-dimen-
sional cavity with a depth dimension which is much smaller
than a wavelength. For these studies the overall length and
width of the superelements are the design variables.
Four of the superelements were assembled into a muffler
model with four independent design variables as shown in
Fig. 10. The muffler inlet is located at the lower left corner of
the muffler with the muffler exhaust at the center node. It is
assumed here that the inlet and exhaust pipe dimensions are
small compared to a wavelength. Figure 11 shows the con-
vergence of the design. Design was stopped after nine itera-
tions automatically. Note the program found an extremely
high insertion loss solution. It is possible that the predicted Figure 13---Convergence of the six superelement muffler rrugdel
performance would not actually be realized because ofphys- (k = .25)

16 NOISECONTROLENGINEERINGJOURNAL/July-August1986
to the optimal solution. There is strong possibility that these improved design cycle. A reasonable application of this op-
high performance solutions might not be found by traditional timization procedure could be made on a personal computer.
techniques. While analysis time would be significant, capital costs would
It should be pointed out for clarity that the example prob- be minimal.
lems shown are for illustration only and do not address some
of the practical aspects of finite element analysis of reactive
mufflers. For example, the finite element grids used in each
Acknowledgment
of the examples are relatively coarse and are only appropriate
to the frequency of the second or third mode in the muffler The author would like to thank Carl Kipp for his assistance
cavity. In addition, the models would accurately simulate the with some of the analysis.
local effects near the transition from chambers to pipes if the
pipes are very small. References
The objective functions in the illustrations in this paper are
constructed for a single frequency. However, in many muffler 1. A.D. Jones, "'Modeling the Exhaust Noise Radiated From Re-
design problems the objective function should measure the ciprocating Internal Combustion Engines~A Literature Re-
performance over a broad frequency band. Such problems view," Noise Control Engineering Journal, 23, 1, 12-31 (1984).
will be significantly more computationally intensive and con- 2. L.J. Eriksson, "A Review of Recent Progress in Exhaust Sys-
tem Design," SAE Paper No. 820622 (1982).
sideration should be given to all potential useful design strat- 3. D. Karnopp, J. Reed, D. Margolis, and H. Dwyer, "Computer
egies including those illustrated in this paper. Aided Design of Acoustic Filters Using Bond Graphs," Noise
One potentially very efficient and accurate strategy for Control Engineering Journal, 4, 3, 114-118(1975).
such problems is use of the shape sensitivity and optimization 4. M.J. Crocker, "Internal Combustion Engine Exhaust Muf-
techniques shown in this paper with modal superposition fling," Noise-Con 77 (Noise Control Foundation, New York), pp.
331-358.
procedures. The number of computational degrees of free- 5. M.L. Munjal, "VelocityRatio-Cum-Transfer Matrix Method for
dom in the muffler model would be significantly reduced but Evaluation of a Muffler with Mean Flow,"J. Sound. Vibr., 38,
the sensitivity information could still be computed and hence 1, 105-119 (1975.).
muffler optimization is still possible. Regardless, the exam- 6. J.W. Sullivan, "A Method for Modeling Perforated Tube Muf-
ples shown in this paper illustrate how useful the design sen- fler Components. I. Theory," J. Acoust. Soc. Am., 66, 3, 772-
778 (1979).
sitivity information is for finding high performance optimal 7. M.G. Prasad and M.J. Crocker, "Evaluation of Four-Pole Pa-
solutions in relatively small numbers of design iterations. rameters for a Straight Pipe with a Mean Flow and a Linear
Temperature Gradient," J. Acoust. Soc. Am., 69, 4, 916-921
(1981).
Conclusion 8. PT. Thawani and R.A. Noreen, "Computer-Aided Analysis of
Exhaust Mufflers," ASME Paper No. 82-WA/NCA-10(1982).
This paper has formulated techniques for the computation 9. J.W. Sullivan and M.J. Crocker, "Analysis of Concentric-Tube
of design sensitivity information for mufflers. The techniques Resonators Having Unpartitioned Cavities," [Link]. Soc.
Am., 64, 1, 207-215 (1978).
are applicable to existing, popular one-dimensional and
10. D.E Ross, "A Finite Element Analysis of Perforated Compo-
three-dimensional muffler analysis procedures. The design nent Acoustic Systems,"J. Sound. Vibr., 79, 1, 133-143 (1981).
sensitivity information can be utilized by the muffler de- 11. C-I.J. Young and M.J. Crocker, "Prediction of Transmission
signer to select a more appropriate muffler configuration than Loss in Mufflers by the Finite Element Method," J. Acoust.
the current configuration or to determine whether the cur- Soc. Am., 57, 1, 144-148 (1975).
rent solution is close to optimum. If desired, the design sen- 12. Y. Kagawa and T. Omote, "Finite Element Simulation of Acous-
tic Filters of Arbitrary Profile with Circular Cross Section,
sitivity information can be incorporated into formal optimi- J. Acoust. Soc. Am., 60, 1003-1013 (1976).
zation procedures to develop an automatic muffler design 13. A. Craggs, "A Finite Element Method for Damped Acoustic
procedure. Several examples of muffler optimization have Systems: An Application to Evaluate the Performance of Reac-
been shown to illustrate the positive and negative aspects of tive Mufflers,"J. Sound Vibr., 48, 3, 377-392 (1976).
using the design sensitivity information for automated de- 14. T. Tanaka, T. Fujikawa, T. Abe, and H. Utsuno, "A Method for
the Analytical Prediction of Insertion Loss of a Two-Dimen-
sign. Regardless of the application of the design sensitivity sional Muffler Model Based on the Transfer Matrix Derived
information, however, such information is not difficult to From the Boundary Element Method," ASME Paper No. 84-
compute and will significantly enhance muffler design pro- WA/NCA-7 (1984).
cedures. 15. R.J. Bernhard, "'A Procedure for Explicit Determination of
Acoustical Shape Design by the Finite Element Method,'"
Current implementation of the muffler optimization pro- ASME Paper No. 84-WA/NCA-13 (1984).
gram is on a VAX780. The program is nicely sized for the VAX 16. R.J. Bernhard, "A Finite Element Method for Synthesis of
computer. The design sensitivity information requires more Acoustical Shapes,"J. Sound. Vibr., 98, 1, 55-65 (1985).
memory than typical acoustic finite element models. How- 17. R.J. Bernhard and C.R. Kipp, "Shape Optimization of Reac-
ever, large memory capability is widely available, particularly tive Mufflers by the Finite Element Method," Proceedings of
Inter-Noise 84, (Noise Control Foundation, New York)381-384.
on virtual memory machines. Furthermore, memory re-
18. IMSL Library Reference Manual, Edition 9, Vol. 4, IMSL Inc.,
quirements for any acoustic finite element model are not large Houston, Texas 1984.
by structural finite element standards. These extra require- 19. L.C.W. Dixson, E. Spedicato, and G.P Szego, Nonlinear Op-
ments should be more than compensated for by a shortened, timization: Theory and Algorithms (Birkhauser, Boston, 1980).

Volume27/Number 1 17