You are on page 1of 22

Pergamon

Int. J. Impact Engno Vol. 16, No. 1, pp. 149-170, 1995 Copyright 1994 Elsevier Science Ltd Printed in Great Britain. All rights reserved 0374-743X/(94)00039-5 0734-743X/95 $9.50+0.00

IMPACT DAMAGE PREDICTION IN CARBON COMPOSITE STRUCTURES


G. A. O. DAVIES and X. ZHANG
Depar:ment of Aeronautics, Imperial College of Science, Technology and Medicine, Prince Consort Road, London SW7 2BY, U.K. (Received 27 August 1993; in revised form 11 March 1994) Summary--This paper describes a strategy for predicting the extent of internal damage in a brittle carbon fibre laminated composite stucture, when subjected to low velocity impact by a single mass. The success of the predictions, which avoid expensive three-dimensional analysis, is validated by test for a wide range of structures from small stiffplates through to large flexible stiffened compression panels wl'tose residual strength is affected much more by internal delamination than tension structures It is shown that a numerical model needs to incorporate nonlinear behaviour due to both gross deformations and to in-plane material degradation.

1. INTRODUCTION Low velocity impact damage in laminated composite plate structures has been recognized as a debilitating threat for many years [1]. It is particularly damaging in the case of carbon fibre epoxy composites used in Aerospace (and increasingly in other high performance vehicles) where the fibres are elastic and brittle, and so is the matrix when compared with conventional ductile metals. If the impact threat is very low velocity, such as dropped tools and other manufacturing or handling accidents, then the damage can be a mixture of internal delamination~riven largely by interlaminar shear and tension and lamina matrix cracking--or back-face tension driven failure. The latter will be initially matrix cracking or splitting between fibres, and then fibres fracture and further delamination if the bending strains are high enough. These damaged modes are extremely debilitating, particularly to the compresslion strength of the structure, and can be completely invisible when viewed from the external impacted surface. Non-ductile structures have the disarming ability to recover elastically if there are no gross deformations, local indentation or performation. Barely visible impact damage (BVID) has become the hidden menace. Recognising this menace, the aerospace industry has traditionally used laboratory drop tests on coupon size specimens to explore the nature of impact damage and to seek ways of minimizing it. Countless papers have appeared over the years [1-12] and the tests are usually quoted as having a prescribed incident energy and the measure of success or failure being the residual compressive strength of the coupon--"compression after impact" or CAI. Valuable strategic information has been accumulated in a somewhat ad hoc fashion. Thus the importance of any form of energy absorption, in the absence of gross plasticity, has been confirmed. Glass fibre composites are superior in this respect to carbon since they have a low modulus, a comparable strength, and a weaker interface between fibre and matrix, but unfortunately their absolute performance in compression is poor anyway because of their flexibility. Hybrid structures can achieve an acceptable compromise [ 13,14] in which stiff carbon, or flexible glass, or intermediate aramid fibres are mixed. Improvement in damage tolerance can be achieved using tougher resins, particularly thermo-plastics such as PEEK and by using high strain fibres [2]. Metal matrix composites are much more tolerant but lack the specific strength and stiffness of carbon epoxy composites. It is fair to say that the problem of CAI remains and this can be seen in the current very crude design strategy of limiting the allowable compressive strains, often as low as 0.3% in structures whose matrix may also degrade when hot and moist. This is a punitive strategy when the undamaged material may be capable of withstanding strains of order 1%.
149

150

G.A.O. Davies and X. Zhang

A more realistic philosophy has been to recognise that the accidental impact threat is not the same over all parts of the structure; during its manufacture and its life, moreover the internal nature of a structure will make the dynamic response quite different to that of a laboratory specimen. The structure may be locally very stiff and massive or very flexible and light, factors which depend on whether an exposed surface is thick or thin, curved or fiat, and how it is supported internally. Thus the coupon test, although a very valuable screening test, may be very misleading, when compared with real structures. It is also clear that no manufacturer is likely to conduct hundreds of impact tests over (say) a complete aircraft wing or fuselage, and therefore a predictive strategy is crucial. With this in mind a collaborative project was started in 1990, the participants being British Aerospace, Deutsche Aerospace, Dornier, SAAB-SCANIA, Fokker, NLR, K U L and Imperial College. This project has just been completed. It was recognised that no strategy can predict directly the compressive residual strength simply from incident energy, so the project was split into predicting the damage extent and to predicting residual strength from a given damage. This paper summarises the damage prediction methods developed at Imperial College with the valuable collaboration of SAAB [15]. 2. THE P R O B L E M Laminated composite plates can only be treated as layered 2-D anisotropic plates when there are no significant through-thickness stresses. This is manifestly not true for normal point impact and we have to be prepared to consider even thin plates as internally 3-D structures where behaviour is dictated by the individual laminae and the sub-laminar stresses across them. The internal stress field is extremely complex but some insight can be obtained by considering first the static internal stress field due to a concentrated normal force. (In low velocity impact the plate inertia forces immediately adjacent to the impactor are relatively small anyway.) If the plate is isotropic, or quasi-isotropic, the bending strains very approximately linearly through the depth and vary like lnr away from the force 1-16]. The shear stresses vary approximately parabolically through the thickness and like 1/r away from the force. However these are really far-field approximations and near the force the near-field variation in shear stress azr in the mid-plane region will vary like Fig. l(a). From the equilibrium equations the normal stresses az~ will vary roughly as the derivatives of the shear as shown in Fig. l(b). This is a simplified description and as we move up through the thickness of the plate to the surface region where the force is applied, the stress variations become more acute. Post-impact inspections whether ultrasonic C-scanning, X-ray or microscopy sections do confirm failure patterns suggested by Fig. 1. The region on the centreline of the impactor is a low shear region, and in a state of through-thickness compression, so delamination will not start here, but be confined to an annular zone as Fig. 1 suggests. This has also been simulated in a detailed 3-D finite element model 1-15] where it is shown that the inner delamination boundary is very unstable for even small loads and will propagate inwards much faster than the outer boundary propagates away from the centreline. The delamination will not be symmetrical even in a quasi-isotropic lay-up since the interlaminar shear strength (or toughness) is a minimum along or close to the direcion of the figures, so a characteristic peanut shape [5,6,9] develops with different orientation at the various interfaces. There is also the narrowing of the damage zone towards the top surface, as mentioned, also the 'peanut' delamination looks rather like a spiralling surface contained within a frustum I-6]. We should also mention that in-plan e stresses cause transverse matrix tension cracks which usually trigger the delamination described. The problem of mixed mode failures, typically tension and shear, has been treated by looking at either Mode I and Mode II energy release values I-6] or by looking at several stress components and defining some empirical interaction curve governing a failure threshold. This has been done successfully by Choi and Chang I-9] who suggested peanut zones, but they use stress-based criteria, and so scale effects would not be exhibited as in a fracture model. In addition, the material is not degraded as the threshold front moves outwards.

Impact damage prediction


rce

151

quasi isotropic plate

~
tension compression tension I I-

zr (shear stress) shear threshold

Fig. l(a)

(Yzz (normal stress)

Fig. l(b)

Fig. 1. Average internal stress variations due to concentrated force.

Other attempts to model this 3-D failure field have been partially successful both statically [8] and in a dynamic impact mode I-9] using usually a stress-based criteria. The well-known DYNA 3-D code 1-17] for example uses the Chang-Chang model I-8] for failure and degradation. All these attempts undoubtedly add to our knowledge of the internal failure process, which is necessary because the only traditional way of measuring stress fields is to use surface', strain gauges. However the computational demands can be exorbitant even if the physics, is valid--that is stress based, fracture, or damaged mechanics? 1-18]. For example modern military aircraft can have composite skins of 8 mm thickness (64-ply) or more. If we try to use a finite element brick model, for each lamina, and we need at the very least to model in 3-D a zone of one plate thickness away from the impact point, then 64 x 128 x 128 = 1 million finite element bricks! This is not a feasible design tool yet. 3. A REALISTIC STRATEGY FOR REALISTIC STRUCTURES A new strategy has been developed which makes several gross simplifications, all of which have been validated by test and by numerical simulation. Firstly we propose to ignore the highly detailed and complex 3-D stress field in a thickness-wide zone immediately under the impactor, and the transient through-thickness stress waves which radiate out from the impact contact and are then reflected at the back-face and thereafter rapidly attenuate. Now if the impacted surface is impulsively moved at the indentor initial velocity V, and if these stress waves were confined to a small cylinder-like plug under the indentor, the induced transient strains would be of order V/C where V is the through-thickness speed of sound, governed mostly by the matrix density and modulus.. Failures attributed to these waves (like back-face spallation) would occur at V/C of order 1% and a typical value of C for epoxy is 2000 m s- 1, thus we expect such ballistic type failure modes to occur at velocities in excess of 20 m s- 1. This is not normally considered to be the demarcation between high and low velocity, but tests have shown this change in damage mode to occur at about 20-40 m s-1 [4]. For low velocity impact therefore, when V is less than 20 m s- 1, we can ignore through-thickness stress waves and

152

G . A . O . Davies and X. Zhang

take it that the plates have time to respond as simple engineering theory for plates and shells assumes, and we may consequently use analytical and finite element plate and shell models. We are therefore not able to model the local Hertzian deformation, and it is known that this can account for up to 15 %oof the energy absorption in extreme cases. We therefore use an empirical non-linear spring between the impactor and the plate surface, a successful ruse first advocated by Tan and Sun [19]. Secondly we recognise that local damage is primarily due to two effects. If the plate is flexible enough there will be large bending strains causing tension failure on the back-face and compression failures on the front. If the plate is not so flexible it will develop a significant impactor force and this will give rise to the internal shears, and normal stresses, and the variety of interlaminar and matrix damage mechanisms referred to earlier. We initially postulated that damage therefore is proportional to either the force, or to the local bending strain, and coupling between the two is weak. The way is now clear to eliminate entirely the effects of scale in absorbing elastic energy. This of course is affected by the size of the plate and the boundary conditions, and the balanace between mass and velocity contributing to the kinetic energy. All that is necessary is to solve the dynamic response of the structure--or possibly just the static response and equate incident energy to final strain energy at the maximum displacement. We find the maximum force and/or bending strain and then refer these to the force and strain histories in a series of small coupon tests whose damage signatures we map. Thus the strategy is to use coupons as calibrators, for the chosen laminar material and lay-up, and 'embed' them theoretically in a real structure and find the maximum induced force from which to deduce the internal delamination etc. To test this strategy a series of carbon composite plates made from HTA/6376 were made and tested. To cover a wide range of dynamic responses two sizes of plate were tested, two boundary conditions, three thicknesses, and also a series of stiffened compression panels.

4. THE TEST PROGRAMME AND INTERPRETATION OF RESULTS 4.1. The test rig and data acquisition Impact tests were conducted using an instrumented falling-weight rig of energy capacity 1-50 J obtained by adjusting the drop height (up to 2 m) and impactor mass (0.40-2.5 kg). The tup was caught after the first impact. A drawing of the tup is shown in Fig. 2. A commonly accepted standard hemi-spherical impactor (12.7 mm in diameter) was used. The impact force is measured by means of four active strain gauges mounted on a load cell cylinder above the impactor nose. They are all wired externally into a Wheatstone bridge. The load cell was calibrated in a tensile testing machine by compressing it against a soft material. An accelerometer is also mounted on the impactor to measure the acceleration and therefore the force, and then deduce the impactor velocity and displacement by integrating acceleration. The measured contact forces by load cell and accelerometer are remarkably close, but the output by load cell gives much less noise than that by accelerometer. Thus we used the load cell for most of the tests. A modest amount of filtering of force and strain gauge histories was done numerically on the recorded data by emulating a resistance/capacitor low-pass filter with a rise time equivalent to 4% critical damping. (This was later done for the undamped finite element predictions, making comparisons easier.) The incidence velocity of the impactor is measured by means of a ruled grid attached to the impactor side and which passes a photo-emitter/photo-diode device mounted on the fixed channel guides to give a pulse form of output every time a dark line is crossed. Knowing the spacing of the grid lines and time for each to pass, the velocity can be calculated before and after the impact event. The experimental data was recorded by a MicroLink transient data recorder connected to a P.C. The data recorder allowed acquisition of up to eight separate analogue-to-digital input channel, each with 8-bit accuracy and 1/~s sampling rate (1 MHz). The impactor velocity, contact force and up to six strain histories are recorded. The displacement is then deduced by integrating the acceleration, i.e. force/mass. In processing the data, the zero

Impact damage prediction

153

""'~-'---! 3/8""--"-',,=
1/4" threaded bolt
FIFE or nylon bushes

]1"

:@
~ J
clearance hole

2"

@
3/4"

/
O l

tapped hole

1 1/2"

0
I

0
[~"~ ~::z~'~'St

PARTS 1. Impactor nose (removeable) steel 2.Load cell - aluminium 3.Main guide - aluminium 4.Packing piece(s) adjusting for mass- steel or aluminium S.Second guide - aluminium
Grid mounted on impactor body used to determine velocity

1/8"

"ngauges

1/2"

1/8"radius
1 ~

Aiuminium extruded

7/8"
~

Strapsfor holding channels

Fig. 2. Instrumentedtup.

time on the output graphs was set to the time of initial contact of impactor with target plate, so that the graphs are comparable with the numerical simulation.
4.2. Experimental results All the plates tested were quasi-isotropic having a symmetrical lay-up of equal numbers of 0 , + 4 5 , - 4 5 and 90 plies. The material used was carbon/epoxy HTA/6376, with mechanical properties summarised in Table 1. To cover a wide range of dynamic responses two sizes of plate were tested, each having two boundary conditions and three thicknesses. The test configurations are described in Table 2. After drop tests ultrasonic C-scan inspection was conducted to assess damage caused. The transmitter/receiver records the total attenuation of the signal passing through and reflected back to the probe. The area so recorded is therefore the envelope of all delaminations at several interfaces at different levels. This inspection technique is also combined with X-ray and optical microscopy to provide an understanding of the failure mechanisms that occur during impact. A typical C-scan map of a 32-ply 4 mm thick plate is shown in Fig. 3(a). Although it looks almost circular it is in fact the envelope of several elongated areas of delamination at a series of interfaces in the middle region of the plate. A corresponding X-radiograph in Fig. 3(b) shows the "peanut" delamination along the

54

G. A. O. Davies and X. Zhang Table 1. Material properties and laminate strength for HTA/6376composite*

Symbol
E11 E22 G12 $11 C11 $22 C22 $12 $13, $23

Value
140 GPa 9.5 GPa 5.8 GPa 2000 MPa 1650 MPa 70 MPa 240 MPa 105 MPa 85-105 MPa

Property~strength
Young's modulus in the fibre direction Young's modulus in the matrix direction In-plane shear modulus Tensile strength in the fibre direction Compressive strength in the fibre direction Tensile strength in the matrix direction Compressive strength in the matrix direction In-plane shear strength Interlaminar shear strength

*The parameters are obtained from material strength tests conducted by Deutsche Aerospace and British Aerospace. Table 2. Ply orientations and geometries of the test specimens (unit:mm)

Ply orientation
[45/-45/90/0]s [45/~,5/90/012 s [45/-45/90/0]4 s

Thickness Small coupon


1 2 4 100 75 100 x 75 100 x 75

Laroecoupon
200 200 200 200 200 200

Edge support
Clamp and simple support Clamp and simple support Clamp and simple support

Fig. 3. C-scan maps and X-radiograph of impacted plates. (a) Small, 4 mm thick plate, mass = 1.59 kg, energy= 13.2 J. (b) Small, 4 mm thick plate, mass= 1.58 kg, energy= 13.2 J. (c) Large, 2 mm thick plate, mass =0.65 kg, energy = 9.34 J.

Impact damage prediction

155

-45 o0 o 0o

0o

90

-45

impact face 45 -45

9o
0 internal J delamination NN~ 0 90 -45
I

J
X II IIIII I

45 back face

Fig. 4. Microscopy photograph showing multiple delamination level. (Test SS1-5, mass =0.42 kg, incident energy = 1.23 J).

local fibre directions. It also shows the very elongated delamination wihch can also occur when bending strains are high. To demonstrate this kind of damage another C-scan map for a higher incident energy drop test is presented in Fig. 3(c), which shows gross back-surface delamination. Figure 4 is a microscopy of the impact damage over the entire depth of a 1 mm thick plate. It was revealed that interlaminar delamination and matrix fracture occurred first, followed by tension back-surface fibre splitting (matrix cracking) at higher energy levels. A typical force vs time history for impact without damage is shown in Fig. 5(a). It is for a 4 mm thick small plate with clamped edges under 3.7 J incident energy. The response is almost sinusoidal, suggesting a fundamental mode characteristic of a large mass and light structure with no high frequency components. At a higher energy, 6.3 J, significant damage is caused, the force history shown in Fig. 5(b) now shows a truncation of the peak force as damage occurs, followed by a slower recovery indicating a decrease in the structural stiffness due to damage--a topic we return to. Damage w,~rsus maximum force maps for all combinations of plate sizes and boundary conditions are shown in Figs 6-8, for plate thicknesses of 1, 2 and 4mm, respectively.

156 7000

G . A . O . Davies and X. Z h a n g

Test SC4-2 6000 5000 coupon: 125x75x4 nun clamped edges [45/-45/90/014s

impactor: mass = 225kg


incident energy=3.7J

4000
30O0 2000

1000
0 .... , .... , .... , 0.000 0.001 0.002 0.003
,..-~.., ~ ,:...

0.005 Time (s)

0.004

0.006

0.007

0.008

(a) incident energy 3.69 J, no damage.

7000~
6000 "1 coupon:

Test SC4-5 125x75x4 mm [45/-45/90/014s

5000

clamped edges impactor: mass = 2.25kg


incident energy=6.24J

4000" 3000200010000 0.000

0.002

, .-r'Z"----"S'~, 0.004 0.006 Time (s)

0.008

(b) incident energy 6.3 J, significant damage. Fig. 5. Force vs time histories for 4 mm small plates with clamped edges. (a) Incident energy 3.69 J, no damage. (b) Incident energy 6.3 J, significant damage.

Impact tests on stiffened panels having the same 4 mm thick base skin are also included in Fig. 8. The more usual damage against incident energy maps are also displayed in the same figures. Note that the energy maps became increasingly more chaotic as we increase the thickness of the plates. This is to be expected as large flexible plates absorb considerable energy elastically compared with small stiff configurations. The force maps on the other hand virtually collapse the test data onto single bands where no particular configuration is separated from the rest. In Fig. 8 this includes large (1.0 x 0.85m) stiffened panels also. Thus the effects of dynamic response seem to have been eliminated allowing for a degree of scatter which is characteristic of impact tests. Apart from the success of the force maps in almost eliminating plate size effects the other point which emerges is the sudden increase in damage at a force threshold, particularly so in the case of the thicker plates It is virtually impossible to contrive to choose a force level so that a controlled increase in damage occurs near this threshold. It is of course easy to dismiss such indeterminacy as the scatter one has come to expect in composites, but it is better to establish whether a stress or fracture threshold actually exists. N o w instead of conducting a thorough 3-D finite element investigation we make the greatly simplifying assumption that the quasi-isotropic plate is perfectly isotropic and the

Impact damage prediction 350 300 ~250 125x75xlmm, Clamped [ [] 125x75xlmm, Simply-supported 200x200xlmm, Clamped [] 200x2OOxlmm, Simply-supported

157

I
m.'m

[]

150

lOO!
50
i

ammm

lid

i~

200

400

600 800 looo ]200 14oo 1600 Impact Force (N)

(a) Damage area versus maximum force.

400 350 o []

~ 300
,..~ 250" ~ 200 o ~ 150 tZl 100 50 0 0

125x75xlmm, Clamped [ 125x75xlmm, Simply-supported 200x2OOxlmm, Clamped 200x2OOxlmm, Simply-supported

[] []

~o m

~,
1

00

a~
[] !

Impact Energy (J)

(b) Damage area versus incident energy. Fig. 6. Impact damage maps for 1 mm thick plates. (a) Damage area vs maximum force. (b) Damage area vs incident energy. delamination iis perfectly circular. The latter is clearly not true as Fig. 3(b) indicated but at least the envelope is pretty circular until back-face tension splitting along fibres takes place. We also. j u m p to the stage where the delamination has become driven by trzr in the central region of the plate thickness. Thus we can a p p r o x i m a t e the m e a n shear stress by the statically determinate value in terms of impact force P, ignoring inertia of course, P
O'zr ~- 2zrt'

(1)

and the m a x i m u m value 3P

a~-4nrt"

(2)

This crude measure is not too far out as it happens. A full finite element solution is shown

158
650 600 550 500 450 400 v 350 o
o

G . A . O . Davies and X. Zhang

125x 7 5 x 2 m m , c l a m p e d e d g e s 125x 7 5 x 2 m m , s i m p l y s u p p o r t e d 200x2OOx2mm, clamped edges


200x2OOx2mm, simply supported

3o0~
250 2O0 150 100 50 0 0
i i -T~, i , i i i

<
000

tt1~0

500

1000

1500

2000

2500

3000

3500

4000

Impact Force (N)

(a) Damageareaversusmaximumforce.

650 -I-

t[" 55011.
600 0

50o

11

125x 75x2mm,clamped edges 125x 75x2mm,simplysupported 200x200x2mm,clampededges 200x200x2mm, simply supported


0

450 " t 4O0350 " 300250 " 200" 150 " 100 "
50" 0" O

0 e

;#m
2 4

o eO0

10

Impact Energy (J)

(b) Damage area versus incident energy.


Fig. 7. Impact damage maps for 2 mm thick plates. (a) Damage area vs maximum force. (b) Damage area vs incident energy.

in Fig. 9 for a 2 mm thick small plate under the incident energy of 2.2 J. The maximum impact force is 2000 N and the damage shape is almost circular with an area of 150 mm 2. The differences between FE prediction and Eqn (2) are small. N o w if the delamination envelope was due to exceeding the allowable interlaminar shear stress (T) we could insert this into Eqn (2) and deduce the circular damage area as
: A=xr 9 [,p'~2 .

= 161rt~-~)

(3)

Equation (3) gives a continuous map for damage versus force, unlike the threshold behaviour in Fig. 8. The values of interlaminar shear strength quoted in Table 1 are 85-105 MPa, but tests results do vary so we selected the absolute lower bound of 50 MPa in favour of Eqn (3) and show this in Fig. 10 for the three plate thicknesses. It has clearly little relevance with the reality except perhaps for the initial damage in the thinner plates, but only the onset. We therefore turn to a Mode II fracture possibility, again using the isotropic

Impact damage prediction


1800

159

1600

1400 1200 1000

O []

125x75x4mm, clamped edges 125x75x4mm, simple support 200x200x4mm, clamped edges 200x200x4mm, simple support Stiffened Panel I (type B) Stiffened Panel II (type A)

Ooo

,00

400 2
u

.:_-,
,l
u

0"~:----

~ 6000

2000

4000

8000

10000

Imoact Force (N~ (a) Damage area versus m a x i m u m force. 1600 O 1400
0

[]

1200

looo "00
t~

0
on

O[]

600-

"o:"

"

"
O [] 125x 75x4mm, clamped edges 125x75x4mm, simply supported 200x200x4mm, clamped edges 200x200x4mm, simply supported Stiffened Panel I (type B) Stiffened Panel II (type A)
I i i i

400"
2000l0
!

20

30 40 Impact Energy (J)

50

60

(b) Damage area versus incident energy. Fig. 8. Irapact damage m a p s for 4 m m thick plates and stiffened panels. (a) D a m a g e areas vs m a x i m u m force. (b) D a m a g e area vs incident energy.

0.12 Case SC2-TI, Small plate, 2ram thick O.lO


13 ;

FE77 (Linear analysis, 0 deg)

FE77(Linear analysis, 45 dog)


FE77 (Linear analysis, 90 deg) P Calculated by x = - 2~rtG

0.08
~4

"ii
O O9

0.06

0.04

0.02

0.00

l0

15

20

25

30

Distance from the impact centre (ram) Fig. 9. Distribution of radial shear strain.

60 1800
16001400
=

G . A . O . Davies and X. Zhang

/
I- - stress criterion ) I.L.S.S.= 50MPa
/

/ /

/I "~ 1200' ,~ lOOO

I~

/'

//
i1

//
-t-

~U
/
/

//"

~
~

800.
600'

/
t=lmrn
[]

C~

400200 = o 0
. .

:/
/

t=2mm

...-"
.

.,s

.,.

"Q ../" %++Ot--4mm


U

q"

....
, , ,

: , t : ] , ~ -I

1000

2000

3000

5 0 0 0 6000 Impact Force (N)


4 ~

7000

8000

9000

Fig. 10. Stress-based predictions for delamination in a variety of plate dimensions and support conditions including stiffened panels 1800 16001400 1200 1000 80O E 6O0 400
200"

Fracturethresholdprediction P= 680 N (t=l ram) 1:)=1923N 0=2 ram) P=5645 N (t---4ram)

+e

IL

to
t=2mm
, - ,A t 4-

oa+
o t=4mm

t=lmm

"4-

I.

, .. 1000

. .

, ....

, I,-.,,--.

. . . . . . . . . . . .

20~

3000

4000 5 0 0 0 6 0 0 0 Impact Force (N

7000

8 0 0 0 9000

1800 1600
1400

i 1200
,~,
1000

/%

13

& SC1
n SSI

800 600

/~ /~ A I~ ] A /hA
.

+
o

+ + o

0 0

LC1 o LSI SC2 SS2 LC2 LS2 /~ SC4

+ []
SS4

~4m= 200"~lOo,,oo O=, ,

0 LS4 :Stiffl2 Stiff

10

20 30 Incident Energy (J)

40

50

Fig. 11. Fracture-based predictions for delamination in a variety of plate dimensions and support conditions including stiffened panels.

Impact damage prediction Bending Strainsin matrix

161

"14
~i0000~t 0.007 0.006 0.005 0.004" 0.003 -" 0.002 0.001 -

i
~ritieal strain=O.O075

........ *...... ........ * ......

Expt(SG 1) Expt(SG3) FE77 FE77

o.0002
-0.001 0.000 0.001 0.002 0.003 0.004 0.005

0.010 0.009"

BendingStrainsin fibre

........ *...... ........ *"....

o.oos0.007 " 0.006 0.005 -

Expt(SG2) Expt(SG4) FE77 FE77

0.004" 0.oo3-' 0.oo2-'


0.001 ~ 0.000" -0.001 0.000

,,a"~, N

0.001

0.002

0.003

0.004

0.005

2 (-45 o)

~ Y,

1 (45 o)

Strain gauge positions F i g . 12.

Comparison of surface bending strain histories between finite element predictions and experiments.

axisymmetric assumption. It has been shown [2] that the energy release at the boundary of a central circular delamination is given by 9p2(1 - v 2)
GII .=8~z2Et 3

(4)

This is a remarkable formula in that untypically the delamination radius does not appear. If we invert it, and substitute the critical energy release rate Gnc the critical threshold force for instant dehtmination is given by
p 2 _ 8rc2 Et3Gc

(5)

9(1 -- p2)" This can explain why the increase in damage is so sudden and indeterminate at a force threshold. For the quasi-isotropic lay-up, we use the mean flexural modules E of the laminate in the 0 and 90 direction, which is 60 GPa compared with the value of E l l = 1 4 0 G P a for a single lamina. Poisson's ratio v=0.3 and a propagation value of

162

G.A.O. Davies and X. Zhang

GHc=0.8Nmm -1 based on tests using the split end notched flexure specimens [20]. Equation (5) then becomes Pc = 680t3/2. (6)

The three thresholds for thicknesses of 1, 2 and 4 mm are shown in Fig. 1l(a) and are not too bad as predictions considering the greatly simplifying assumptions made. The complete energy maps are also shown in Fig. 11 for small (S), large (L), simply supported (S) and Clamped (C) plates of 1, 2 and 4 mm thickness. The map is again chaotic. The slight curving over the damage growth in Figs 6-8 and 10 for impact forces greater than Pc represents the change in mode from internal delamination to plate bending and back-face splitting and ultimately perforation. This splitting along the 45 figures in the outlet laminar, and shown in Fig. 3(c), is accompanied by delamination from the next sub-ply due to the consequent stress concentrations. Apart from the evidence of this back-face splitting in Fig. 3(c), strain gauges were also placed along the 45 direction at right angles to the fibres, and these gauges revealed the time and the strain level at which splitting took place Figure 12 shows strain histories for positions 6 and 16 mm away from the impact site. The gauges failed at values which were close to the known matrix failure strain of 0.75%. (See Table 1 for $22 and E22 etc.) Having demonstrated that the threshold of interior delamination damage can be directly related to the maximum force, it is necessary to show that forces can be reliably predicted so that coupon results can be carried over to other structures. It was hoped initially that the force and bending effects could be decoupled, but this turned out not to be possible, and we now find that a nonlinear simulation is necessary to include gross deformation and flexural degradation as surface fibres fail.

5. FINITE ELEMENT MODELLING The finite element impact code used in this project has been a version of FE77 [21] developed in the past 15 years and easily modified to add in extra modules or to refine existing features. The finite elements used in this case have been standard 8-noded Mindlin [22] curvilinear quads which allow for shear deformation, including a shape factor in this case 5/6. The shear modulus at 5.8 GPa compared with the in-plane value of 140 GPa does not allow us to ignore shear deformation. FE77 requires only the basic lamina properties to be specified, and the stacking sequence, the stiffness matrix is then assembled corresponding to in-plane membrane, bending, and coupled terms if necessary. The code will also model large deflections, and update the stiffness due to any membrane stretching induced when deflections exceed the plate's thickness. The impact version used here is the explicit solver on account of the need to use very small timesteps anyway; the rise time of impact force being less tha a millisecond typically. The central difference integration scheme was used for maximum stability and FE77 selects the timestep based on the highest eigenvalue, found automatically once the stiffness and mass matrices have been generated. The latter is row-lumped but the rotary inertia terms are scaled [23] to prevent an unrealistic and unnecessarily small critical timestep. The use of plate elements prevents us from modelling the local three dimensional Hertzian damage immediately under the impactor contact point. We use the now well-known artifice [19] of inserting a nonlinear spring between the plate and the indentor, which has a different nonlinear characteristic for loading and unloading, and therefore absorbs energy. The law used was for loading and unloading

F= ko~",
f ~-~o F= Fm~__~% \", )

(7) (8)

Impact damage prediction Small 4ram plate Clamped edges Impact energy 3.69 J Damage area = 0 ................ Expt (filtered) - -FE 77 (linear analysis)

163

,-, 5000 1
z .

4000300020001000-

0 0.000

......
. . . . I . . . . I . . . .

......................
I . . . .

0.002

0.004 Time (s)

0.006

0.008

0.000

v~ -0.001 J

-0.002

.................. Expt (Integrated) FE77 (linear)


-0.003 , , ,

0.000

0.002

0.004 Time (s)

0.006

0.008

Fig. 13. Comparison of test and FE77 solutions of force and displacement histories (for small 4mm thick plate, without damage, linear analysis).

where n = 1.5 figr classic elastic Hertzian contact with no hysteresis; F is the contact force; a is the indentation and k the empirical contact 'stiffness' dependent on the geometry of the specimen, indentor and material properties. Fm is the contact force at the start of unloading; am !is the indentation corresponding to Fm (i.e. the maximum indentation); and ao is the permanent indentation All constants have been obtained experimentally [12]. The finite element predictions turned out to be remarkably accurate for the desired force histories. Figure 13 shows theoretical and experimental histories for a 4 mm plate. If the incident energy is large enough a flexible composite plate can easily undergo large deflections without undue damage and Fig. 14 for a thin 1 mm plate shows the need to include the nonlinear stiffening terms within the finite element model. Although no nonlinear iterations are needed in explicit dynamic codes, the computer times do increase by about 50% due to the necessary nonlinear updates. A further feature was unexpectedly found to be necessary, that is the need to degrade the structural stiffness matrix even though we seek only the force history. The recorded force history in Fig. 5(b) shows that damage has attenuated the force signature since it does not display the characteristic smooth simple harmonic response as did the other low energy tests in Fig. 5(a). The truncated force in Fig. 5(b) is due to damage, but is this in-plane fracture or delamination between plies?

164 1(~)0

G . A . O . Davies and X, Zhang

8(X)"

................. Expt (filtereddata) F'E77(Nonlinear,undamaged) ..... F'E77(Linear, undamaged) Small lmm thick plate Clampedexlges Impact energy0.6 J No ~ a g e d

600"

..

~-~ 400"

200'

'-,
i l !

0.000

0.002

0.004 Time (s/

0.006

0.008

0.000 V -0.001 E
/ I

// iI i
i

P> -0.002
o

-0.C03 -0.004 ....... ..... -0.005 0.000


i i

=
Expt (Integrated) FE77 (Nonlinear) FEW (Linear)
i

0.002

0.(X)4 0.006 Time (s)

0.008

0.010

Fig. 14. Thin plane simulation: gross deformations (small 1 mm thick plate, without damage).

At this point we are indebted to our colleague Anders Edlund [15] of SAAB-SCANIA who undertook a very refined finite element brick analysis, using DYNA 3D, and allowing elements to degrade at specified stress thresholds. Firstly the prospect of delamination between plies was tested. Because of the very computer-intensive nature of this level of modelling, only the thin ply 1 mm plates were modelled using only three "smeared" elements through the thickness--that is the possibility of two levels of delamination. An interlaminar shear strength criteria was used to unpick elements when this threshold was reached. The failure progression looked realistic, including the initial stage where propagation started some distance away from the impact site and then moved both inwards and outwards, however the size did not match the tests sufficiently accurately, probably due to the need for a fracture model as described before. However what was important was the very minor attenuation in the plate stiffness and the dynamic response as indicated in Fig. 15. Although delamination does indeed split the plate into separate components they do not behave completely separately at a much reduced stiffness since they have to rotate together at the ends of the delamination. Armed with this knowledge, the in-plane laminae stiffnesses were therefore degraded only in-plane using the combined tension-shear criteria of Chang and Chang [8] and the stress levels shown in Table 1. As soon as any lamina reached the critical equivalent stress, the modulus E 1z in the fibre direction was put to zero. The values in the table are obtained from material strength tests conducted by Deutsche Aerospace and British Aerospace. It has to be said that experimental evidence of fibre fracture was very slim and required very

Impact damage prediction

165

~ 2800" i
U

! - without d e l a ~ o n ///~

'

del

0 E-4

10

15

2'0

E5 Time

Fig. 15. Force response of a plate with and without delamination at stress threshold of 85 MPa. DYNA 3-D simulation (with acknowledgement to SAAB/SCANIA).

specialized and experienced interpretation of deplied specimens in a scanning electron microscope. However the changes in finite element predictions of force were dramatic. We have incorporated this routine into FE77 where it is possible to degrade any ply level and update the element stiffness, still using plate elements of course. Our results are virtually identical to [lt5] and agree remarkably well with the test histories as shown in Figs 16 and 17. Without this degradation the peak force can be seen to be 50% in error and the results of the embedding strategy seriously misrepresented as we show in Section 6. We also attempted to predict the back-surface splitting, or matrix failure in tension, and noted in Fig. 3(c) and Fig. 12. The test results indicated that this was happening at the expected matrix failure level of about 0.75%. Using this as a criterion, the f.e. model, including the loss of stiffness due to splitting, should be able to predict the extent of back-face damage. Figure 18 shows the predictions, for three incident energies, of matrix strain variations with distance from the impactor, and hence the point at which the threshold of 0.75% is c:rossed. The 9 mm prediction for 2.22 J compares with 6.5 mm actual, and for 4.25 J we predict 13 mm compared with 15 mm observed. This is disappointing. Having successfully predicted the actual strain levels, it is clear we need to improve the damage modelling, and allow for the fact that a surface lamina split will cause stress concentrations which precipitate delamination on either side of the radial split. This interlaminar damage propagates sideways from the split as indicated in Fig. 3(c) and 18. It almost certainly requires an energy release algorithm to predict it, and remains a challenging problem for these higher energy values.

166
2600 2400 2200

G . A . O . Davies and X. Z h a n g

2OO0 1800 Z
o 0 t t

:-\ ' ~ ; ;
I |

................. Expt (filtered) ......... FE77 (undamaged, nonlinear) FE77 (degraded, nonlinear) ......... SAAB (degraded) ~ : ,~ ~ 125x 7 5 x l m m clamped [45/-45/90/0]s impactor: mass=0.418kg energy 3.1 1J coupon:

1600 1400 IO008OO" 6004002oo -

E 12oo

; j.~. ;,, ~.,~':, "'"

o 0.000

r -

--

0.002

0.004 Time (s)

0.006

0.008

0.0OO -0.OOl -0.OO2 -0.003 .

-0.004 ~ -O.OO5~, 4).006 ?5 - o . o o 7 -

-0.008 -0.009 -0.010 0.000


i

Expt (Integrated) FE77 (degraded, nonlinear) FE77 (undamaged, nonlinear)


| i

0.002

0.004 Time (s)

0.006

0.00S

C-scan m a p for test SC1-7 1 m m thick small plate incident energy = 3.1 J.

Fig. 16. Finite element simulation with gross damage included (test SC1-7).

6. C O U P O N S E M B E D D E D IN STRUCTURES We have already demonstrated that damage force maps are virtually independent of the size and dynamic response of the impacted structure, therefore the prima face case for inferring structural damage, from the impacted coupon force signature is proven. However it is pertinent to demonstrate how the damage map for one configuration can be deduced from coupon tests on the same laminate stacking sequence. We have selected not the best, but a typical test in the mid-range 2 mm thick. We select the stiffest of all of the coupon configurations, the smallest plate with clamped edges. The damage-force map is shown in Fig. 19 together with the theoretical threshold prediction of 1923 N. There is always scatter in damage curves so we have outlined a bandwidth which includes all the points bar the one shown where the damage is unrepresentative of all other trends. We now choose to predict the damage map for the much more flexible plate, 200 x 200 mm. There are eight points in the coupon map of Fig. 19 and at the three highest energy levels, when the

Impact damage prediction 5OOO

167

J 1 i ~,

4500 t
40001
3500
o

..........

EXl~(filtered data)
FE77 (undamaged, non-linear)

..... ' i

FE77 (aegnde tma-~r)


coupon:

3oo0 -t

4
f

) t

200x2OOx2mm
clamped

_, I
.~l,.

edges [45/-45/9o/o]2s
energy 9.34J

25oo -"

i { ~

impactor, mass=O.6OSkg

21111111500 1000500-

,j

,,\
0.002 0.004
Time (s~

0 0.000

/,
\

0.006

0.008

ooook
-o.ool -t ~

,,
I / /

/
i l
i

1 \
-o.oo5-

-0.009 -0.010 0.000


i

- ....
i

FE77(u~aged, nonlinear) FE77(degraded, non-linear)


i

0.002

0.004 Time (s]

0.006

0.008

C-scan map for test LC2-7 2 mm thick large plate incident energy = 9.3 J damage area = 390 mm2

Fig. 17. Finite element simulation for a badly damaged large plate (test LC2-7).

maximum impact force exceeds 2000 N, there is considerable damage and degrading of the flexural stiffness. However the f.e. model predicts these maximum forces well as Fig. 17 showed, so we can assume that the code can be used again to predict force levels for a larger more flexible plate, even when the plate laminae stresses exceed their known strengths. These damage/force points are now plotted, together with the band limits, in Fig. 20(a), where the test results are also shown. The agreement is well within the bounds that industry has come to expect for purely experimental evaluation. The incident energy levels for a given force are of course quite different for the larger plates, but having the f.e. predictions we know these energies and the damage/energy may be plotted for the large plates as shown in Fig. 20(b). This is probably a more useful demonstration for industrial use since it is the energy threat which is specified or known. The results are also mapped using the FE prediction without in-plane degradation and consequently predicting a much higher force when damage would take place. These results also shown in Fig. 20(b) are not so convincing clearly. The damage areas quickly peak at 300 mm 2 for energy levels 4, 5, 6 and 7 because the higher forces are not sustainable in the coupon test which has degraded to the point of perforation at this stage. It does seem

68
0.030

G . A . O Davies and X. Z h a n g

Numerical Impact Test Energy

0.025

\ \ \

SC2-18, SC2-11, SC2-9 \ \ 4.25J Expt

1.44J 2.22J 4.25J

t~

0.020

".. ':.

SC2-T2, 2.22J

0.015

0.010

rt~

o= 0.005

Critical strain = 0.0075

O.OO0 0

5 6.5 (SC2-1 I)

10 1: Distance from impact zor e (ram) 15.0 (SC2-9)

20

Fig. 18. Surface bending strain vs distance from the impact zone (for small, 2 m m thick plate with clamped edges).

400 I
350 300

C2

200-'

//
~
r ri |-rn

~5o"
~ 1~,~

. . . . . . . .

~ .....

0'" Im 500 1000 1500 2000

l o a c = 19[ 3 N
2500 3000 3500 4000 4500 5000

Impact Force (N) Fig. 19. Coupon calibrator for predicting damage vs m a x i m u m force (based on small, 2 m m thick plate with clamped edges, experimental).

Impact damage prediction

169

t
350 300 ~ 250150 10050"
0 , ..i ....

[...._~ Expt(200x200x2mm, clamped) FE77(degradedmodel)


E7 F_~

] ] ~ fracture delamination El, E2 U [ ~ load= 1923 N


i . . . ~ l - - . , u .... U .... e .... i . . . . i .... i .... i

500 1000 1500 2000 2500 3000 3500 4000 4500 5000 ImpactForce (N)

(a) Predicted damage versus force map using the calibrator.


500 450 FE77 (undamaged model)

4oo
350

~r7 (degta~ mo0ea)


E7

o,o ///..
100

50

.,:,z4~
1 2 3

E1 E2

,t. . . . . . . . . . . . . . . . . . .
4 5

Impact Energy (J)

10 11 12 13 14

(b) Predicted damage versus energy map using the calibrator. Fig. 20. Use of the finite element model (FE77) to predict damage in large plates. (a) Predicted damage vs forcemap using the calibrator. (b) Predicted damage vs energymap using the calibrator.

therefore that our hope of decoupling force-driven behaviour and bending stresses has not been completely realised, but if we include the in-plane degradation the strategy does work remarkably well. The price paid is the computing cost for nonlinear simulation which cannot be avoided as this work demonstrates. A typical run on a simple small plate (1 m m thick with nonlinear deformation) takes 1.2 h (CPU) on a 4 M F l o p workstation. With degradation this same problem takes 27.5 h (CPU). Of course the cost/power ratio of workstations is decreasing almost monthly. The technique described here has also been applied to stiffened panels impacted over stiffener flanges where many of the simplifying assumptions made herein may become even more doubtful. The results were unexpectedly accurate and have recently been reported 1-24]. 7. C O N C L U S I O N S An attempt has been made to predict low velocity impact damage in real structures, without resorting to three-dimensional analysis, using instead a single coupon test series as calibrators. The technique works well for the carbon fibre composites tested here.

170

G. A. O. Davies and X. Zhang

Moreover the onset of delamination has been shown to be critical and theoretically predictable. The use of simple finite element plate models, to predict the necessary force histories, worked well but in-plane degradation is needed for high incident energy. The success of the strategy relied on a force-driven delamination, uncoupled from bending strains. Whilst this worked for carbon composites, we have found it not to do as well for other materials. Kevlar for instance fails in compression and delamination is driven solely by bending strains. Glass reinforced plastics do behave like carbon but the toughness depends on strain rate so damage maps for small and large plates may not be the same.
Acknowledgements--The authors would like to thank British Aerospace and all members of the consortium (BE3159) including of course the European Commission who partially funded this damage tolerance project Thanks are also due to the DRA (Farnborough) who supported the initial development of the FE77 software and the impact test facility.

REFERENCES
1. G. Dorey, Impact damage in composites~development, consequences and prevention. Proc. of ICCM 6: Sixth Int. Conf. on Composite Materials & ECCM 2: 2nd European Conf. on Composite Materials 3, 3.1-3.26. Elsevier Applied Science (1987). 2. G. A. O. Davies and P. Robinson, Predicting failure by debonding/delamination, AGARD: 74th Structures & Materials Meeting, Debonding/Delamination of Composites (1992). 3. Z. L. Gu and C. T. Sun, Prediction of impact damage region in SMC composite. Composite Structures 7, 179-190 (1987). 4. E. W. Godwin and G. A. O. Davies, Impact behaviour of thermoplastic composites. CAD in Comp. Mat. Tech. (Eds Brebbia, de Wilde & Blain), Springer-Verlag (1988). 5. R. C. Tennyson and S. Krishna Kumar, Delamination damage and its effect on buckling of laminated cylindrical shells, AGARD: 74th Structures & Materials Meeting, Debonding/Delamination of Composites (1992). 6. S.A. Salpekar, Analysis of delamination in cross ply laminates initiating from impact induced matrix cracking, NASA Contractor Report 187594 (1991). 7. O. Ishai and A. Shragai, Effect of impact loading on damage and residual compressive strength of CFRP laminated beams. Composite Structures 14, 319-337 (1990). 8. F. K. Chang and K. Y. Chang, A progressive damage model for laminated composites containing stress concentrations. J. Composite Materials 21, 834-855 (1987). 9. H.Y. Choi and F. K. Chang, A model for predicting damage in graphite/epoxy laminated composites resulting from low-velocity point impact. J. Composite Materials 26, 2134-2169 (1992). I0. H.Y. Choi, R. J. Downs and F.-K. Chang, A new approach toward understanding damage mechanisms and mechanics of laminated composites due to low-velocity impact: Part I--Experiments. J. Composite Materials 25, 992-1011 (1991). 11. H. Y. Choi, H.-Y. T. Wu and F.-K. Chang, A new approach toward understanding damage mechanisms and mechanics of laminated composites due to low-velocity impact: Part II--Analysis. J. Composite Materials 25, 1012-1038 (1991). 12. S. W. Khoo, Low velocity impact of composite structures, PhD Thesis, Imperial College, University of London (1991). 13. G. Dorey, P. Sigerty, K. Stellbrink and W. G. J. 'r Hart, Impact damage tolerance of carbon fibre and hybrid laminates, Garteur TP-037, Royal Aircraft Establishment technical Report 87057 (Oct. 1987). 14. G. Marom, E. Drukker, A. Weinberg and J. Banbaji, Impact behaviour of carbon/Kevlar hybrid composites. Composites 17, 00~0 (1986). 15. A. Edlund, Finite element modelling of low velocity impact damage in composite laminates. Proc. o f l C C M 9, Vo. V, p. 334, Madrid (July 1993). 16. S. Timoshenko, Theory of Plates and Shells McGraw-Bill (1940). 17. J. O. Hallquist and R. G. Whirley, DYNA 3-D Users Manual, Nonlinear Dynamic Analysis in Three Dimensions, Lawrence Livermore Nat. Lab. Rep. UC1D-19592, 5 (1991). 18. P. Ladeveze A damage mputatina mde fr cmpsite structures. Cmputers Structures 44 79-87 ( 992). 19. T. M. Tan and C. T. Sun, Use of statical indentation laws in the impact analysis of laminated composite plates. J. Appl. Mech. 52, 6-12 (1985). 20. I. Verpoest and V. Efstratiou, BE 3159-89 MTM Progress Report, Kath. University Leuven, June (1992). 21. D. Hitchings, FE77 general purpose modular finite element system for static and dynamic, linear and nonlinear analysis, Dept. Aeronautics, Imperial College, University of London. 22. R. D. Mindlin, Influence of rotary inertia and shear on flexural motion of isotropic elastic plates. J. AppL Mech. 18, 31-38 (1951). 23. A. Pica and E. Hinton, Efficient transient dynamic plate bending analysis with Mindlin elements. Earthquake Engng Struct. Dyn. 9, 23-31 (1981). 24. G. A. O. Davies, X. Zhang and A. Edlung, Predicting damage in composite aircraft structures due to low velocity impact, Aerotech 1994, Structures & Materials Seminar, Birmingham, UK (Jan. 1994).

You might also like