N&O JOERNAAL VOL.

9, NR 3, 1993

Structural modification using frequency response functions: an experimental study

D.S. Masseyl and C'P. Constancon2

(First received Mry 1993; Einal versior June 1993)

Abstract the SIUURF technique. SMURF involves the direct ma-

nipulation of frequency response functions, or FRFs, of
The technique of Structural Modificalions Using Frequency component systems to yield the FRFs of the modified sys-
Response Functions, or SMURF, wts studied. The lech- tem. This is advantageous since FRFs from sources such
nique wts employed for the prediclion of the modified dy- as experimental modal tests, finite element models, or an-
namic characteristics of beam-like stru,ctures, with erper- alytical models may be combined to produce a modified
imenta,lly deriaed FRFs seruing as a basis. Pinned an,d set of FRFs for a system. However, one of the most signif-
rigid modif,cation types were employed, which made it nec- icant advantages of the SMURF technique is that it only
essary to consider FRFs which related rotational ercita- requires, as a minimum, a set of FRFs equal to the number
tion and response. Consequently an accelerometer capa- of degrees of freedom involved in the modification. This
ble of measuring rotational acceleration was utilised in the is advantageous since it enables the SMURF technique to
study. The results indicated that the SMURF technique be implemented in a short time and, if only experimental
uas able to predict the dynamic characleristics of the mod- data are utilised, then the need for a mathematical model
ified strtr,ctures with 0, high degree of accuracy. It was in the modification process is removed.
howeaer found that the accuracy of the predictions dimin- During the last few decades SMURF has been the fo
ished where noise was present in the component stru,ctures' cus of a wide range of research. One common finding is
FRFs. that while SMURF is sirnple to implement numerically, it
has been plagued with practical probleffis,lt;Z;3;a;51 and
Introduction it is therefore recommended that it only be used to pre-
dict the effect of simple modifications involving pinned re-
The ability to predict the effect of a structural modifica- straints, whether in the troubleshootitg environment, or
tion on a system's dynamic characteristics, without hav- as a method of validating a modal model.
ing to physically perform that modification, is of great Most of the experimental research conducted on the
advantage to the engineering community. Consequently, SMURF technique has been restricted to cases where
various techniques have been developed which are able translational mea^surements have been used. Translational
to produce these modified dynamic characteristics from measurements are sufficient for the implementation of pin
theoretical, experimental, or a combination of experimen- (or hinge) type modifications. However if rigid type mod-
tal/theoretical data. ifications are to be represented in which moment iransfer
Where the prediction of the effects of a modification occurs, then rotational, as well as translational, compat-
on a prototype that is still in the development stage is ibility must be satisfied at modification interfaces. The
sought, a modification technique of a purely theoretical neglect of the rotational information when implementing
nature must be implemented. However, in cases where the SMURF technique will lead to a modification that is
the proposed modification involves existing structures, it essentially hinged. [3]
may be desirable to include experimentally derived data The ability to measure rotations has always presented
for the modification prediction. This is especially the case problems for the experimentalist, and the general approach
for large complex structures, where it may be extremely in the past was to approximate a rotation from a set of
inaccurate and time-consuming to approximate quantities translational measurements.[6;7] Refinements of this ap-
such as damping distributions and boundary conditions proach have been published, by Larson,[8] in which an
theoretically. These quantities are inherent in experimen- improved formulation of constraints is presented which
tally derived data and so bring some degree of reality to utilises translational measurements on beams. However,
the basis of the modification process. recently transducers capable of accurately measuring rota-
One such technique which allows for the use of theo. tional acceleration have been developed and evaluated.[9]
retical, experimental, or experimental/theoretical data is The aim of this research is thus to illustrate the imple-
mentation of the SMURF technique in cases where pinned
l Lecturer, Applied Mechanics, School of Mechanical Engineering, as well as rigid modification types are to be performed.
University of the Witwatersrand, P.O. Wits, 2050 Republic of Souch Although analytical FRFs are applied in this research , dt-
Africa
tention is focused on the use of experimentally derived
2lecturer, Applied Mechanics, School of Mechanical Engineering,
FRFs as a basis for the SMURF technique. In the case
University of the Witwatersrand, P.O. Wits, 2O5O Republic of South
Africa of rigid modifications, rotational response measurement is
R&D JOURNAL VOL. 9, NO. 3, T993

accomplished with the aid of a rotational accelerometer. Proposed modifications

Theory The structures used as a basis for the proposed modifica-

tions are illustrated in Figure 1. The first, &tr H-frame,
The theoretical formulation of the SMURF technique is was machined from a 25-mm thick mild steel plate, while
well established and has been documented on many occa- the remaining two beams were machined from a 25-mm
sions. [t;g ;4;7 ;8;10;11] However for the purpose of complete- square mild steel bar. All the structures had specially
ness this theory will be presented briefly. The formulation machined ends, enabling interconnection with the aid of
may be considered under two areas, namely, the formula- locating pins.
tion of constraints and the generalised stiffness method.
The formulation of constraints expresses a modified
FRF matrix in terms of a structures original FRF matrix
and a general constraint matrix. This may be written as:

H- (r- HC (c, HC)-' Cr) H

where I/ is the modified structure's FRF matrix and // is
the original structure's matrix. C is the general constraint
matrix f vector which represents the constraints i*posed at I soomm I
I
the modified degrees of freedom. Note that if it is possible
to express the constraints as a single vector, then the ex-
pression for H does not involve a matrix inversion. This
formulation applies directly to modifications implemented
on a single structure, such as tying a point(r) to ground or
connecting points together. The main features of this for-
o
mulation are that I/ must be evaluated at every frequency
line of interest, and that it requires a full FRF matrix f/. BOomm
This means that a point inertance must be available for
the point, or points, of modification. Figure 1. Structures used for the modifications.
The general stiffness formulation involves the matrix
inversion of the components structures' FRFs. If one con-
siders two structures A and B which are to be joined to- Three modifications were studied, namely:
gether at degrees of freedom X, to form a structure C ,
then FRF matrixes for structures A and B may be writ- 1. The addition of a beam via pinned joints across the
ten, respectively, as: end of the H-frame, a^s illustrated in Figure 2. This
beam was analytically represented by r translational
lHl" = constraint across the end of the H-frame and a mirss
addition of 1.61 kg at the point of modification.
and
2. The lengthwise joining of the two beams via a pinned
lHl" = joint, resulting in the two beams' essentially being
joined via a hinge. The FRFs of both beams were
where X are the degrees of freedom used in the coupling of
experimentally determined and only translational de-
the structures and Y are the remaining degrees of freedom.
grees of freedom at the modification point need be
If one writes[H];' = [I{]o and [H];' consider-ed.
matrix of the combined structu re C may be written
fa 3. The lengthwise joining of the two beams via a rigid
connection. The FRFs of both beams were experi-
t- ll{vvle [K, x]e mentally determined. However in this ca.se, rotational
lHlc =| [/{x v]o [Ii x x]o + IK x x]n measurements are required to account for a rigid con-
L tol lr{v xla nection.
This formulation also requires a full FRF matrix, and con-
sequently a point inertance involving the degrees of free- Modification 1 was implemented using a combination
dom at the point(s) of modification must be obtained. of the formulation of constraints (to apply the translational
Note that in both the formulation of constraints and the constraint) and the generalised stiffness formulation (to
generalised stiffness method, the component FR.F's may be apPly the mass). Modifications 2 and 3 were performed
generated analytically, or determined through experimen- by solely using the generalised stiffness method.
tal measurement. In general, both methods may involve These modification types were applied since it was en-
rotational, as well as translational, degrees of freedom. visaged that each provided a progressively more difficult
10 N&O JOERNAAL VOL. 9, NR 3, 1993

modification. Thus a steady progression in degree of diffi- The proposed modifications were then physically per-
culty is presented from the first, which considers a combi- formed and modal tests of each modified system were car-
nation of experimental and analytical FRFs, to the third ried out, so as to serve as a comparison with the SMURF
which not only uses experimental FRFs from both compo- predictions.
nent beams, but also includes rotational measurements. The natural frequencies of the components structures
are shorvn in Table I. Note thau Table 1 shows only the
natural frequencies for the H-frame and the larger of the
Ir
two component beams, since the shorter beam did not ex-
hibit any resonances in the examined frequency range.

Table 1 Natural frequencies of the component

structures
Mode ll-frame natural Large beam natural
I number frequencies (Hr) frequencies (Hr)
1 68.1 2r2.5
2 I44.6 592.7
3 167.7
4 203.5
5 600.9
6 703.5
7 871.9

SMURF implementation
Figu re 2. H-frame modification.
The FRF data which were obtained from the component
Experimen tal pro cedure structures could be directly used in the SMURF implemen-
tation for the modifications involving pinned joints. How-
Although the SMURF technique only requires a set of ever when considering the rigid modification of the beams,
FRFs which involve the degrees of freedom that are di- the rotational point inertance at the point of modification
rectly used in the coupling of structures (includittg point was required. This was synthesised in the following man-
inertances), additional FRFs were measured at various lo-
cations along each structure. This allowed for the pre-
diction of the mode shapes, &s well as the FRFs, of the
modified structure.
An impact hammer was used as an excitation source
for the measurement of the FRFs, and a translational ac-
celerometer was used to measure the response of all of the
structures at the required locations. However, in order
to implement the rigid connection for the third modifi-
cation with the SMURF technique, rotational FRFs were
required. A Kistler 8696 accelerometer was used to mea-
sure the rotational response at the point of modification
on each componen[ beam. In this way FRFs were di-
rectly measured which related an input force to a rota-
tional response (#), r pure couple input to a transla-
tional response (+, the reciprocal of the latter) and an
input force to a translational response (+ . The FRF
)
relating a pure couple input to a rotational (*) response
could nbt be directly measured and so an approximation
of this had to be made.
In all cases, the structures were tested in a free-free
conditi.on. FRFs were measured over a frequency range of
0 to 1024 Hz, with a resolution of 2 Hz. The FRFs were
averaged over 15 readings. Figure 3. Generation of a .rotational point inertance.
R&D JOURNAL VOL. 9, NO. 3, 1993 11

Consider a beam section a^s shown in Figure 3. It is of the measured resonant peaks. The FRFs also compare
required to approximate the frequency response function favourably for the case of the pin-jointed beams, where all
of the measured resonant peaks are again identified. How-
H,i,rw&) from the FRFs H,i,r,e) and Hii,r"(+), with ever, it is clear that there are spurious or false peaks that
the last two FRFs being measured directly. Each of the appear on the SMURF FRFs for these two cases. It was
two excitation forces may be represented by a force of equal found that these peaks occurred at frequencies that corre-
magnitude, and a couple at the centreline of the beam. In spond to the original natural frequencies of the unmodified
theory, the centreline forces will produce no rotation of the systems, but they did not exhibit a g0-degree phase change
beam. Thus one may write: displayed by the remainir,g peaks. The SMURF FRF does
not match the measured FRF as accurately for the rigidly
0 - Hii,rFr + Hii,u @td) modified beaffis, with the first measured resonant peak be-
ing omitted from the SMURF prediction.
Hii,r, = Hri,r + H6,Md (1)
0.0
and similarly - 180 .0
-360 .0
Hii,r, = Hti,p - H6,ud (2) 3.gg+01

Where H ,i,, is the FRF relating the angular motion

1 e[+01
of the beam produced by u centreline force, which is inde-
pendent of the magnitude of f'l or F2. Upon differencing
x
equations 1 and 2, the expression for the required point F
inertance becomes:
1.0E+00

H,i,* = (3)
2d
This expression wa^s applied to the FRFs of the com-
r .0E-0 1
ponent beams. The generated point rotational inertance (False Peaks :'*)
for the longer beam is illustrated in Figure 4. This FRF 3.8E-02
exhibits the characteristics of a point inertance. However, 0 .0 400.0 800 .0 to22 .o

it displays severe noise below 150 Hz. Frequency (Ur)

Figure 5. Comparison of measured t--1 vs SMURF

-
0.0
180 .0
{ \_ l' I -i-l
FRF : H-frame modification.

-360 .0
I [,
r 3343 .6 I
0.0
-r80.0
-360.0
3.9f+01

1000.0
/\ 1.gf +0t

0
M x
F
t.0E+00
100.0

13 .3 r.0E-0t \
I
0.0 400 .0 800 .0 t022 .0 (False Peaks : *)
Ftequency (IIr) 3.0E-0e '-
0.0 400 .0 800.0 r0?2.0
Figu re 4. Generated point rotational inertance.
Frequency (IIz)

Results Figure 6. Comparison of measured t--1 vs SMURF

FRF : Pinned beam modification.
Typical comparisons between measured and predicted
FRFs for the modrfied structures are showu in Figures 5,
6, and 7. The FRFs compare favourably for the case of the Time domain polyreference parameter estimation was
H-frame modificatiotr, up to a frequency range of approx- performed on both the measured and SMURF FRFs, and
imately 500 Hz, with the SMURF FRF identifying all of natural frequencies and mode shapes were extracted. The
t2 N&O JOERNAAL VOL. 9, NR 3, 1993

extracted natural frequencies are shown in Tables 2, 3, in this region and hence it was difficult to produce an accu-
and 4. It was found that the spurious peaks exhibited by rate representation of the modified FRFs in this frequency
the first two modification predictions were not identified range.
as modes during parameter estimation. As can be seen,
with the exception of the omitted mode in the rigid modi-
ficatit-,u, t*e SMURF predictious compare favourably with Table 3 Comparison o-f rneasured vs SMURF raturat
the measured data. The extracted mode shapes for the frequencies : Pinned beam modification
SMURF and measured modifications were almost identi- Experimental SMURF
cal and hence are not shown. natural natural
Mode frequencies frequencies %
number (Hr) (H") error
0.0
'-trIJ)
-r80.0
-360.0
3.0Er0l
iY l) t f .l 1 183.6 182. 1 0.82
I 2 513.4 518 .2 0.93
t
3 887.0 9r2.4 2.86
ll'
t.0Er0l

X
F' ,i,
t
\ Table 4 Comparison of measured vs SMURF natural
t.0E{00
frequencies : Rigid beam modification
\
\ Experirnental SMURF
\
\
\
\
t natural natural
t.0E-0r
I :l Mode frequencies frequencies To
I
I number (H") (H") error
I
3 . OE-0e - 800.0
0.0 400.0 r02?.0 1 101.1
Frequeocy (IIz) 2 267.3 260.4 2.58
Figu re 7. Comprison of measu red t- -] vs S M U R F
a 3 558.7 563.3 0.82
FRF : Rigid beam modification. 4 925.7 920.2 0.59

Table 2 Comparison of measured vs SMURF natural

Noise was not as noticeable in the pin-type modifi-
frequencies : H-frame modification
cations and consequently the problem of mode omission
Experimental SMURF did not arise. The major discrepancy between the pre-
natural natural dicted and measured results in these cases was the spuri-
Mode frequencies frequencies % ous peaks displayed by the SMURF FRFs. As has already
nurnber (H") (Hr) error been stated, these peaks corresponded to the original nat-
ural frequencies of the component systems and were not
1 106.7 105.8 0.84 identified as modes during parameter estimation. It is
2 182.3 191 .7 0.33 postulated that the false peaks originated from numeri-
3 472.6 473.3 0.15 cal ill-conditioning at the frequency of the original reso-
4 620.4 614.7 0.92 nant peaks. This ill-conditioning was amplified due to the
5 765.r 759.9 0.68 lightly damped nature of the component structures and it
6 964.1 947.9 1.68 is expected that the false peaks would decrease in magni-
tude if structures with higher damping were considered.
A visible trend in the results is apparent if one consid-
Discussion ers the comparison between the SMURF and experimen-
tal FRFs. Here the accuracy between the measured and
The implementation of the SMURF technique on the ex- SMURF FRFs decreases with each modification, which is
perimentally based FRFs provided acceptable results. The in agreement with the progressively more difficult nature
highest error encountered in the extraction of any natural of each modification. The H-frame modification displays
frequency was 2.86%. All modes, except the first mode of the highest correlation between measured and SMURF
the rigid modification, were identified using the SMURF FRFs, since it was generated by a combined experimen-
technique. It is believed that the reason for the omission talf analytical FRF data base. In contrast, the rigid modi-
of this mode lies in the fact that the approximated rota- fication of the two beams exhibited the lowest correlation,
tional FRFs exhibited a high degree of noise below 150 since rotational PRF approximations, as well as purely ex-
Hz. The first modified mode of the rigid modification lies perimentally based FRFs, were utilised.
R&D JOURNAL VOL. 9, NO. 3, 1993 13

In all cases, the predicted natural frequency shifts

rvere appreciable and it was not necessary to perform mi- t3] Brassard J & Massoud M. Identification of a complete
mobility matrix of a synthesized system from com-
nor modifications to illustrate the technique. In particular,
ponent mobility measurements . Proc. Fifth IMAC,
it should be noted that the use of the SMURF technique London, 1987, pp.319-323.
based on experimental FRFs eliminates the effect of modal
rruncation. This is shown in all three modifications where [4] Jones R & Iberle K. Structural modifications: a com-
modified modes were identified at higher frequencies than parison of techniques. Proc. Fourth IMAC, Los An-
any of the modes of the component structures. This was geles, California, 1986, pp.56-63
particularly so in the case of the combination of the two
beams, where the shorter of the two component beams did t5] Crowley JR, Crowley SM, Gorman TE & Rocklin GT.
not exhibit any modes in the examined frequency range. A comparison of SMURF and modal modeling. Proc.
Third IMAC, Orlando, Florida, 1985, pp.1053-1059.
Conclusions [6] Smiley RG & Brinkman BA. Rotational degrees of
freedom in structural modification . Proc. Second
This study has shown that the SMURF technique may IMAC, Orlando, Florida, 1984, p.934.
prove to be an efficient method of predicting the effects
cf a structural modification. The results obtained gave [7] Ewins DJ. Modal testing: theory and practice. Chich-
an acceptable indication of the modified dynamic char- ester, John Wiley and Sons Inc, 1986.
acteristics of the structures examined. Rotationally re-
Iated FRFs were incorporated into the SMURF formula- t8] Larson PO. Dynamic analysis of assembled structures
tion which must be accounted for if any rigid type mod- using frequency response functions : improved formu-
ification is to be implemented. It was also demonstrated lation of constraints . The International Journal of
that success of the SMURF technique is sensitive to the Analytical and Erperimental Modal Analysis, 1989,
quality of the FRF data used in the study. It is therefore 5(1), pp.l-12.
recommendedd that the SMURF technique be viewed with
some suspicion where noisy FRF data are encountered.
t9] Lang G. Including rotational terms in a modal model
An experimental prelude. Sound and Vibration,
1989, pp .24-32.
References
[10] Klosterman A. On the erperimental determination
[1] Crowley JR, Klosterman AL, Rocklin GT k Vold and use of modal representation of dynamic charac-
H. Direct structural modifications using frequency teristics. PhD dissertation, University of Cincinnati,
response functions. Proc. Second IMAC, Orlando, T971.
Florida, 1984, pp.58-65.
[11] Imregun M, Robb DA U Bwins DJ. Sructural mod-
[2] Wang BP, Chu FH k Clark G. Structural dynamic ification and coupling dynamic analysis using mea-
modification using modal analysis data. Proc. Third sured FRF data. Proc. Fifth IMAC, London, 1987,
IMAC Orlando, Florida, 1985, pp.42-45. pp.1136-1141.