A method of nondestructively assessing the integrity of structures using measurements of the structural natural
frequencies is described. It is shown how measurcments made at a single point in the structure can be used to detect,
locate and quantify damage. The scheme presented uses finiteelement analysis, since this method may be used on any
structure. The principle may, however, be used in conjunction with other mathematical techniques. Only one full
analysis is required for each type of structure.
Results are presented from tests on an aluminium plate and a crossply carbonfibrereinforced plastic plate. Excel
lent agreement is shown between the predicted and actual damage sites and a useful indication of the magnitude of the
defect is obtained.
Hence, equation (10) reduces to define the damage site uniquely or to reduce the number of
x~~Kx~Ax~Mx=O
possible sites to the minimum dictated by symmetry. How
ever, because the frequency changes tend to be small and
or
the model of damage is unsophisticated, it is desirable to
XT 6 K x use a larger number of mode pairs, so that some averaging
6A = ~
51
JOURNAL OF STRAIN ANALYSIS VOL 14 NO 2 1979 QIMechE 1979
P. CAWLEY AND R. D. ADAMS
It remains to decide how many grid points should be Thus, for the example shown in Fig. 2, W,,, ",0; W,,, z
used. The finite element used in the work described here 1.
gives a quadratic displacement variation across it, which The values of stresses at the grid points may be com
allows for a linear stress variation. The effect of damage at puted during the dynamic analysis using standard tech
a point in the structure is dependent on the stress at that niques (4). The values of the stresses obtained depend on
point, so it would be wasteful of computer time to use the position through the thickness at which they are calcu
more than two grid points along one side of the element. lated. In flexural vibration, the direct stresses would be
Values between the points could be computed by inter zero at the midplane. The stresses are therefore calculated
polation. The results presented here have been computed at a point away from the midplane, the precise position
using a 6 x 6 finiteelement mesh with two grid points per being immaterial, since only the direction of the stress
element side, giving 144 points in all. The accuracy vector is required, not the magnitude. The transverse shear
obtained has not warranted the use of interpolation stresses were sufficiently small to be neglected in all the
between the points. structures tested. Most of the results presented in this
An example of the chart output is given in Fig. 1, the paper were obtained on structures fabricated from layered
hole shown being the actual damage in this case. The materials. If the stressweighting procedure described
dotted lines show the finiteelement mesh. In this case, four above is to be used with these materials, it is necessary to
possible sites are indicated because of symmetry. compute a n average stress vector. This is because the
direction of the stress vector at a given position through
2.5 Problem of directional damage the thickness depends on the orientation of the layer. A
The model of damage used in the preceding sections has mean value may be computed by defining an average elas
assumed that the local stiffness of the structure is reduced ticity matrix for the ply orientations used. The derivation
by the same proportion in all directions. This would be the of this matrix is described in Appendix 1.
case with a hole in an isotropic material, but the assump In practice, in order to reduce the amount of data
tion is no longer valid with a crack or with a hole in an required to be output from the dynamic analysis and
anisotropic material. hence to save on the amount of disc space relquired, the
Consider the situation shown in Fig. 2. The crack stresses were computed at the midpoints of the elements.
shown will have negligible effect on the frequency of mode These values were then assigned to all the grid points
B, since there is little stress across it, whereas a hole at that within the element. This procedure is sufficiently accurate
site would have a significant effect. O n the other hand, the provided the stress directions d o not change rapidly with
crack would have a relatively large effect on modes A and position. (Rapid change of magnitude does not matter as
C. The damage location scheme described in the previous the stresses are normalized.) This is the case with the lower
section would, therefore, be expected to give erroneous modes of vibration, such as those used in this investi
results with mode pairs A and B or B and C. O n the other gation.
hand, modes A and C should give good results, since the The stressweighting procedure has been adopted in all
stress direction at the crack is similar in the two modes. the results presented here. The results obtained without
The damage location scheme could therefore be improved weighting were fairly similar to those presented, parti
by giving high weight to modes in which the directions of cularly in cases where more modes were used. This is
the stress vectors at the point of interest are similar, and probably because the results tend to averageout to give
low weight to modes where there is low similarity. This is the correct solution. The results with stressweighting
readily achieved by weighting the error according to the tended to be better defined and there was a marked
scalar product of the normalized stress vectors for the two improvement in some cases where only a few rnodes were
modes. Thus, equation (17) becomes used.
?IFIF1
measurements, the appearance of a double resonance
which was not noticed, a relatively large change in mode
Displacement nodal lines
shape due to the damage (the location scheme assumes
that the mode shape is essentially unchanged) or the model
of damage not being good enough to predict the frequency
change accurately. A method was therefore sought to
automate the procedure of checking whether one mode
_ _ _ _  was causing the final result to be incorrect.
The technique developed uses the fact that if one mode
 is causing errors in the final results while the osther modes
I I
give correct results, then removing this mode will tend to
make the damage site more clearly defined on the location
Crack chart. The sum of the numbers on the location chart will
therefore be reduced. The modified location scheme com
Fig. 2. Illustration of the problem of directional damage putes the results with all the modes included and with each
mode neglected,in turn. The final solution is taken as the easily met, a variety of methods of exciting and detecting
one with the lowest sum of numbers on the chart. If less the vibrational response may be used. The best method to
than five modes are measured, then all the readings must use in industrial applications would probably be the tran
be used. sient test, in which the response of the structure to an
impulse is recorded and the natural frequencies computed
2.7 Estimation of the magnitude of damage by Fourier transform techniques. Unfortunately, the
The sensitivity analysis uses equations (11) and (15) to necessary equipment was not available when these tests
output the frequency change in each mode due to a hole of were carried out, so the results presented here were
area A centred on each of the grid points in turn. Compari obtained using steadystate measurements, which are
son of the measured frequency change in a given mode slower and more difficult to perform. It is hoped to report
with that given by the sensitivity analysis for a hole of area some results obtained using the transient test later.
A at the predicted damage site therefore gives an indica The system used in this investigation was identical to
tion of the severity of the damage in terms of the size of a that described in (5). The structures were suspended on
hole equivalent to the damage. The result varies consider nylon thread passed through small holes near the edge of
ably between the modes, particularly if the damage is the structure. Excitation was provided by a movingcoil
directional, so an average value is taken. loudspeaker fed by a highstability, variablefrequency
The method gives a rough indication of the magnitude oscillator and a power amplifier. Resonance was located
of the damage, but is clearly not accurate. The calculation by tuning to the maximum output of a piezoelectric strain
is performed at the grid point where the error is a mini gauge cemented to the structure at a point where the
mum, but the damaged area may be between grid points. stresses across it were predicted to be nonzero in the
This can introduce a substantial error into the calculation. modes of interest.
The model of damage used does not take into account the The tests were carried out in a constanttemperature
mass of material removed by a hole. This is reasonable enclosure, which kept the structure under test at 25 &
with most types of damage, since damage does not usually 1OC. This was found to be sufficiently stable to enable
reduce the mass of the structure. An error is, however, accurate frequency changes to be recorded. With metallic
introduced in the case of a hole. The sensitivity analysis is structures, it was found to be possible to dispense with the
probably only accurate for relatively small changes in the constanttemperature enclosure and to use a correction
local stiffness, since it does not take account of the effect factor based on the temperature coefficient of Young’s
of stress concentrations around the damage. This means modulus, as described in (2). This would be much more
that the effect of a hole is underestimated, resulting in the difficult with composite materials, since the different
area predicted from the measured frequency changes moduli have different temperature coefficients and the
being larger than the actual size of the hole. Nevertheless, coefficients of expansion would be directiondependent.
the method gives a useful indication of the severity of the Probably the easiest way to determine the correction
damage. The results show that the predicted severity factors with composite materials would be to measure
always increases as the amount of damage is increased. them experimentally from the variation of the natural
frequency with temperature. The signal from the oscillator
was stable to & 0.01 Hz and resonance could be found to
2.8 Final damage location scheme
within 0.1 Hz with most of the modes used. With metallic
The method implemented for these tests involved per
structures, the resolution was typically better than 0.05
forming the dynamic analysis, the sensitivity analysis and
Hz.
the stress calculations on the University ICL 470
The damage location scheme requires that the mode
computer, the results being output to disc or tape. The
shapes of the modes used be similar to those predicted by
time taken by this program was approximately 50 per cent
the dynamic analysis. Some symmetrical structures have
higher than that required for a single dynamic analysis, so
two or more modes at the same frequency, the mode
the saving in computer time over repeating the dynamic
excited in practice being a combination of the two depend
analysis with each element damaged in turn was very
ing on the position and nature of the excitation. In these
large.
cases, it is impossible to determine accurately the mode
These results, together with the experimental measure
shape. Modes of this type were therefore not used in the
ments, were then used as input to the program which com
location scheme.
puted the relative errors and produced the location chart.
The mode shapes of the structures tested were checked
This program could be readily adapted to run on a micro
roughly by determining the nodal patterns. The structure
computer. The .major computational effort is in per
was placed on spongerubber blocks and excited by an
forming the original analysis. However, this need only be
electromechanical shaker. Dry sand was spread lightly
performed once for a given type of structure.
over the surface of the structure. When the system was
tuned to resonance, the sand particles migrated to the
3 TEST PROCEDURE nodal lines, giving a check on the theoretical predictions of
their position.
The basic requirement of the method is to determine repro The test procedure used was to measure the initial
ducibly several natural frequencies of the structure under frequencies of the structure, remove it from the constant
consideration. It does not matter whether the boundary temperature enclosure, apply the damage and return it to
conditions are exactly the same as those assumed in the the enclosure. After allowing time for the temperature of
dynamic analysis provided that they are reproducible and the structure to stabilize, the frequencies were again
the mode shapes are not greatly altered from those pre measured. The frequencies quoted in Tables 1 and 2 are
dicted by the theoretical model. Since this criterion is the averages of two or more readings.

Undamaged frequencies (Hz)
Predicted area.
Damage 104.32 260.98 312.74 415.29 584.48 693.24 794.70 936.50 1062.25 1150.93 removed with respect
to virgin condition
Frequency reductions from virgin condition (Hz) (percentage of total)

First saw cut 0.06 0.39 0.29 0.14 0.07 0.14 2.70 0.62 0.20 0.28 0.19
Secondsawcut 0.85 5.46 4.50 2.08 1.78 4.32 5.29 3.70 1.20 1.68 2.17
Crushdamage 1.24 5.66 5.91 3.44 3.93 4.38 9.02 6.90 10.41 6.51 3.65

/ I
   Finiteelement mesh
Plate boundaries
Fig. 4. Location chart for CFRP plate with t w o saw cuts; frequency Fig. 6. Location chart for CFRP plate with t w o saw cuts and crush
changes computed with respect to virgin frequencies damage; frequencies after both saw cuts taken as 'undamaged'
values
/ _ _ _ _ Finiteelement mesh
Plate boundaries
Fig. 5. Location chart for CFRP plate with two saw cuts; Fig. 7. Location chart for CFRP plate with t w o saw cuts and
frequencies after first cut taken as 'undamaged' values crush damage; virgin frequencies taken as 'undamaged' values
quencies and those measured after the crush damage had of tests are set out in Table 2. It can be seen that the
been applied. The location chart for this case is shown in frequency changes produced by the first saw cut are much
Fig. 7. As expected, erroneous results were produced. The smaller than those produced by the second, in spite of the
presence of damage was, however, still detected from the fact that the cuts were of similar length. The predicted
frequency changes. equivalent area removed was 0.19 per cent for the first saw
The frequency measurements and the equivalent areas cut and 2.17 per cent after the second had been made. It
removed as predicted by the location routine for this series would appear to be difficult to reconcile the fact that two
be noted that the case given in Table 2 for crush damage where De is the mean elasticity matrix and Dk is the elasti
corresponds to Fig. 7 and not to Fig. 6. The predicted city matrix of the kth layer, the integral on the righthand
equivalent area removed corresponding to the location side of the equation being taken over the volume of the
chart shown in Fig. 6, which relates to the case where the layer and that on the left being over the volume of the
frequencies after the crush damage were compared with structure. Cartesian coordinates have been used in the
those after the application of the saw cuts, was 1.02 per work described here and, in this case, equati.on (24)
cent of the plate area. The actual area of the tube used in becomes
n
producing this damage was approximately 0.5 per cent of
the area of the plate, and it is fair to assume that the actual
111&'Dee d x dy dz = 111E'D~Ed x dy dz (25)
k=l
damaged area would spread somewhat beyond this. Thus,
the predicted damaged area of 1.02 per cent was of a The elasticity matrices are not dependent on x and y , so
similar order, bearing in mind the assumptions and equation (25) reduces to
simplifications used, to the actual damaged area. n
One of the modes tested showed a frequency increase 1,eTDe&dz = 1 E ~ Ddz~ c
k=l
after the first saw cut. This was probably due to an error in
the initial frequency measurement, so this mode was not where h, is the z coordinate of the top of the kth layer.
used in the damage location scheme. Flexural vibration has been used in the tests reported
here, so it is reasonable to assume a linear directstrain
variation through the thickness and a parabolic transverse
5 CONCLUSIONS shearstrain variation. The strain vector for the element
The test results show that the proposed method can be used here is therefore given by
used to detect, locate and roughly quantify damage in
structures. The technique is applicable to all systems
which are amenable to finiteelement analysis. The final
scheme has the advantage that only one dynamic finite
element analysis need be performed on a given type of
structure. The results of the dynamic analysis may be
stored on disc or tape and used as input to the damage
location program, along with the experimentallydeter
where k,, . . ., k , are functions of x and y.
mined natural frequencies. This program could be easily Substituting equation (27) into equation (26) and equa
run on a microcomputer. ting similar terms gives
The method requires access to only one point of the
structure and, if the measurement of the natural fre
quencies were carried out using transient techniques, the
test time would be very short. The presence of damage is
i
indicated immediately from changes in the natural fre 15 8
quencies without the need for any computation.
It is possible to use the method to monitor the growth of
0; = 
8t
1Dfj((hk
k=l
 hk 
3t 2
(hi  h i  (28)
The C F R P structure used here had a symmetrical ply (3) BISHOP, R. E. D. and JOHNSON, D. C. The mechanics of vibra
configuration about the midplane. This means that fion 1960 (Cambridge University Press).
(4) ZIENKIEWICZ, 0. C. The finite element method in engineering
De  D e  D e  D ; , = D e  D e 0 science 197 I (McGrawHill).
13 14 23 3s  4s 
(5) CAWLEY. P. and ADAMS. R. D. ‘The theoretical and experi
The average elasticity matrix, De, defined by equation mental natural modes of freefree C F R P plates’, Jnl of Composite
(28) has been used to compute the stress vectors for the M t l ~1978 12,336347.
C F R P plate used in these tests. (6) AHMAD, S., IRONS, B. M. and ZIENKIEWICZ, 0. C. ‘Analysis
of thick and thin shell structures by curved finite elements’, In!. J.
Numerical Methods in Engng 1970 2,419.
APPENDIX 2 (7) ZIENKIEWICZ, 0. C., TAYLOR, R. L. and TOO, J. M.
‘Reduced integration technique in general analysis of plates and
shells’, Int. J. Numerical Methods in Engng 1911 3, 275.
REFERENCES
(8) COURANT, R. and HILBERT, D. Methods of mathematical
ADAMS, R. D., WALTON. D.. FLITCROFT, J. E. and SHORT. phvsics 1953 1 (Interscience).
D. ‘Vibration testing as a nondestructive test tool for composite (9) CAWLEY, P. and ADAMS, R. D. ‘A vibration technique for
materials’, Composite Reliability, ASTM STP 580 1975, 159175 nondestructive testing of fibre composite structures’, Jnl of
(Amer. SOC. for Testing and Mtls). Composite Mtls 1978 (to be published).
ADAMS. R. D.. CAWLEY, P., PYE. C. J. and STONE, B. J. ‘A
vibration technique for nondestructively assessing the integrity of
structures’, J. mech. Engng Sci. 1978 20 (No. 2), 93100.
57
J O U R N A L OF S T R A I N A N A L Y S I S VOL 14 N O 2 1 Y 7 Y OIMecht 1979