Fatigue of Engineering Marerinls and Strucrures Vol. 1, pp.

185-201
Pergamon Press. Printed in Great Britain.
Fatigue of Engineering Materials Ltd. 1979.

FATIGUE CRACK GROWTH IN LUGS

J . SCHIJVE and A. H. W. HOEYMAKERS

Department of Aerospace Engineering,
Delft University of Technology, The Netherlands

(Received 9 January 1979)

Abstract-Crack growth in aluminium alloy lugs was recorded during constant-amplitude

loading (R = 1/3). Observations were made both for artificial cracks started by a small saw
cut and natural cracks started by fretting corrosion between hole and pin. Scatter was low for
artificial cracks, whereas considerable scatter applied to the natural cracks as a result of
multiple crack initiation at different times and locations. The fastest crack growth was
observed for artificial cracks, which appear to be a worst case as compared to natural cracks.
K-values derived from crack growth results were in good agreement with K-values proposed
in the literature.

Nomenclature
a (Semi) crack length
daldn Crack rate
D Hole diameter
K Stress intensity factor based on oij = ( K 6 ). J j (0)
AK Kmax- Kmin
R Stress ratio = cmin/omax
r Hole radius = D/2
t Thickness
W Width of lug ,
Ua net stress based on (W- D)t
U." ... Mean stress
On Gross stress, based on Wt
kc loo0 cycles
Pm/c m/cycle
1 MPa megapascal = lo6 N/mz = 0.145 ksi = 0.1020 kgf/mmz
1 MN/m3I2 lo6 N/m3l2 = 0910 ksi ,/in. = 3.255 kgf/mm3/*

1. INTRODUCTION

INVARIOUS structures, especially in aircraft structures, lug type joints are adopted as a
connection between significant components of the structure. In a lug-type joint the lug is
connected to a fork by a single bolt or pin. Advantages are obvious: (1) the joint is simple
and it allows an easy mounting and dismounting; and (2) since clamping of the fork is not
applied the lug can act as a pivot, avoiding local bending moments. In several applications
the rotations will be small if they occur as a result of deformation of the structure. In this
185
186 J. SCHIJVE and A. H. W. HOEYMAKERS

latter situation, a joint fixed by clamping would still introduce undesirable bending
moments in the joint and the adjacent structure.
Disadvantages are also well known. (1) The fatigue strength of a lug is relatively low as
a result of fretting corrosion between the bolt (or pin) and the lug hole. There are several
means to improve the lug fatigue strength but they impair the simplicity of the joint [l]. (2)
A fatigue failure of a lug in a structure in many cases can have unacceptable consequences.
They will not always fail in a safe way.
As a result of the disadvantages the designer will adopt relatively low stress levels for a
lug. This implies that a fatigue crack will grow rather slowly and, if scheduled inspections
are applied, the joint may become a fail safe joint. Crack growth predictions will then be of
interest.
Damage tolerance requirements adopted by the U.S. Air Force [a] include fatigue life
predictions assuming that small cracks are present in the beginning of the fatigue life. It
implies that life predictions involve the prediction of crack growth. Such predictions require
the availability of stress intensity factors (K). These factors depend on the geometry of the
component and the crack length. K-values can be obtained by calculations and by
measurements and in both cases a variety of methods has been developed. For a designer it
is more satisfactory to see that published K-values and crack growth predictions based on
these values are in agreement with observed crack growth results.
Recently Hoeymakers [3] has carried out crack growth tests on aluminium alloy lug
specimens. A comparison between test results and prediction is presented in this paper.
Hoeymakers also tested lugs of laminated sheet materials, but these results will be
published in another paper.

2. EXPERIMENTS

Fatigue tests were carried out on lugs of 2024T3 Alclad material = 368 MPa,
gu = 478 MPa, elongation 20.4%) under constant-amplitude loading in an electro-
hydraulic Amsler closed loop fatigue machine. The dimensions of the specimen are shown
in Fig. 1. The diameter of the hole is 25 mm with tolerance limits 25.000 and 25.021 mm.
The hole was loaded by a low alloy steel pin, tolerance limits being 25-000 and 24979 mm
producing a neat fit.
Loading conditions applied in the tests are given below (Table 1).
Crack growth observations were made at both sides of the hole and at both sides of the
lug, i.e. at four locations of each specimen. For this purpose slots had to be made in the
steel fork (see Fig. 1). Around the hole of the lug a grid of concentric circles (spacing 1 mm)
was photographically etched on the specimen surfaces. The grid was used for visual
observations of crack length.

crm-values R = 113
Normal fatigue tests 60,80, 100, 120 MPa Frequency = 20 Hz
Tests with artificial Through cracks om = 80 MPa
crack starters Corner cracks 6
, = 80 MPa
 Fatigue crack growth in lugs 187

.----
./-.7.-
.-.
/
I

t.--.Jk'
.-
I
-.7- I
I I
s t e e l fork \
I I \
I I /
I
I '1
I
I
I
I
I
I

-. -.
1LO

hardened ,I
. slots in fork
steel pin for crock growth
I observotions
I
I
L-

Kt N 2.8

LC
E
_.1

-
60
m i l l i meters

Fig. 1 . Dimensions of lug specimen.

3. CRACK GROWTH RESULTS

Artificial cracks
Cracks were initiated by a small saw cut made with a thin jeweller fret saw to a depth of
1 mm and a thickness of'02 mm. A saw cut was made at one side of the hole only in the
minimum section of the lug. For the corner crack the saw cut was made at 45" with the
sheet surface to a depth of 1 mm both along the surface and inside the hole (see Figs. 2 and
3).
I88 J. SCHIJVE and A. H. W. HOEYMAKERS

Through crocks Corner crocks

crack I ,
length
(I

lmml

PR

thickness
Of lug

sow cut 1
1 I , I
10 20 30 LO

Fig. 2. Crack growth curves of four specimens with artificial cracks.

Crack growth curves are shown in Fig. 2. For through cracks the crack length at both
sides of the lug was hardly different, thus the average value has been plotted. Corner cracks
initially grew in a quarter-elliptical shape and consequently it took some time before a
breakthrough to the other side of the lug occurred. If the crack lengths at both sides ( a and
b in Fig. 2) became approximately equal a normal through crack was obtained again and
the shape of the growth curves becomes similar to those for cracks that started as a through
crack (see also the crack rate curves in Fig. 7b).
Figure 2 also illustrates that the scatter between two similarly tested lugs was very
small. The crack growth life is significantly different for the two types of cracks, but the
difference is mainly in the first part of the life until the crack length is equal to the lug
thickness (Table 2).
Natural cracks
Cracks starting from a saw cut have been labelled as artificial cracks. If no saw cuts are
made crack nucleation in a lug is due to fretting corrosion inside the hole. Cracks initiated
in this way will be referred to as natural cracks.
For an artificial crack the location of the initiation point has been determined
beforehand. This is no longer true for a natural crack and the geometry of the preliminary
cracks may become fairly complex for several reasons :
(a) Cracks can nucleate at either side of the hole at different moments during the fatigue
life.
(b) At each side of the hole cracks can nucleate at different locations (see Figs. 3c, e and
f). However, an artificial crack starts from the saw cut over the full thickness right
from the beginning.
 Fatigue crack growth in lugs 189

l u g head

hole
l e
I crack I
I
I growth II
I
I
! I
(a] ARTIFICIAL
saw c u t a t corner

( b ) ARTIFICIAL
saw cut over f u l l thickness

[ cl NATURAL
corner crack

( d l NATURAL
unequal crack length
a t both sides of lug
(el NATURAL
crack initiation probably a!
i n s i d e of hole
( f lNATURAL
c r a c k nucleus inside hole c o n s i s t i n g
of some smaller nuclei at sliqhtly
staggered levels
( g ) NATURAL
irregular fracture surface due
to some larger crack nuclei (It
d i f f e r e n t locations

Fig. 3. Fracture surfaces.

 Fatigue crack growth in lugs 191

Table 2

Crack growth period Life(kc)

Through crack Corner crack

1-5 mm 101 33.1

5 mm to failure 7.1 7.8

Fig L o complete life I cycles I

Fig. L b crack growth period,

2 m m to failure

Fig.Lc crack growth period,

5 m m to failure

0 I natural c r a c k s
I = a r t i f i c i a l cracks
corner crock
I
, through crack

Fig. 4. Fatigue lives and crack growth periods of 2026T3 lugs.

192 J. SCHIJVE and A. H. W. HOEYMAKERS

(c) Natural cracks can nucleate more or less simultaneously at adjacent locations,
which can be slightly staggered along the periphery of the hole. As a result ridges
occur where crack nuclei would overlap (Fig. 3g).
In all lugs fretting corrosion was easily observed and in general several small cracks
could be seen inside the hole. Consequently crack nucleation was a more chaotic process as
compared to crack initiation from a saw cut. Crack fronts can be more irregular and the
topography will be less flat than for an artificial crack. Thus it should be expected that
growth curves of natural cracks will show considerably more scatter and usually the curves
will be different from growth curves of artificial cracks. The complexity of natural cracks in
lugs was recently documented by Kirkby and Rooke [4].
Fatigue lives and crack growth periods are presented in Figs. 4a, b and c. Figure 4a also
shows the fatigue curve according to the prediction method for lugs developed by Larsson
[ S ] . Since his method is essentially based on multi-variable regression analysis of a large
amount of lug fatigue data the agreement between the present results and the prediction
shows that the complete lives obtained here are satisfactory.
The growth period data in Fig. 4b show one exceptionally long period at oa = 60 MPa.

crack
length
a
Irnrn)
through
cracks v/ corner
cracksw
I

10 20 30
Number of kilocycles

Fig. 5. Comparison between crack growth from natural cracks and artificial cracks.
 Fatigue crack growth in lugs 193

In this lug a crack was observed relatively early and the crack persisted in a growing
direction at 45" to the loading direction. In other lugs the tensile mode predominated in the
fatigue crack area.
Crack growth curves for two lugs tested at the same stress level as the four lugs with
artificial cracks are shown in Fig. 5 and one notes that curves for the natural cracks are
significantly different. Similar observations were made at the other stress levels. It should be
pointed out that values of the crack lengths at both sides of the lug and both sides of the
hole could be very different for the natural cracks, depending on when and where crack
nucleation started. The largest crack length measured was plotted in Figs. 5 and 6.
Differences between similarly tested lugs, primarily occur at small values of the crack
length. For larger crack length the curves become more or less parallel. The cracks then
have grown more or less to full through cracks, showing similar crack rates. The same is
true for the artificial cracks when comparing the through cracks and the cracks started as
corner cracks (Fig. 5).
Figure 5 further illustrates that artificial through cracks grew faster than the natural
cracks. Artificial corner cracks show an average curve somewhat similar to one of the
natural cracks.

4. STRESS INTENSITY FACTOR FOR A THROUGH

CRACK IN A LUG

In a previous study [6], Van Lipzig has tested centrally cracked specimens cut from the
same sheet under constant-amplitude loading with the same stress ratio. His results
obtained at 0,-values 30, 40, 50 and 60 MPa gave a very narrow scatter band in a
da/dn - AK. plot. The average curve is shown here in Fig. 7a. The present test data for a lug
with a through crack starting from a saw cut (specimens M16 and M22 in Fig. 2) have been
replotted in Fig. 7b as da/dn as a function of crack length. K-values for the lug crack can
now be read from Figs. 7a and 7b as shown by the dotted line. Values obtained in this way
are presented in Fig. 8, while a non-dimensional plot is given in Fig. 9 by plotting
K/(o,,,&) (a, is gross stress on lug) as a function of a/r (r is hole radius).
The curve in Fig. 8 shows that K is an increasing function of crack length, but the
increase is modest. This modest increase could be due to the fact that a pin-loaded hole has
some relation to crack edge loading which is known to show a decreasing K-value for
increasing crack length.
Figure 9 also shows theoretical K-values derived from publications by Impellizzeri and
Rich [7], Kirkby and Rooke [4] and Newman [8]. Since exact solutions cannot be derived
these authors developed approximation procedures starting from known solutions for more
simple cases.
A comparison of the test results with the theoretically predicted K-values indicates that
an anomaly exists in the experimental data for very small cracks, i.e. for a length between
1.2 and 1.8 m. Since the saw cut corresponds to a = 1 mm, the corresponding fatigue cracks
were 0.2-0.8 mm. In this initial phase of the test the crack growth rate was apparently lower
than predicted by the theoretical trend. This may well be attributed to the initiation from
the saw cut and a low accuracy of the crack growth rate derived in the beginning of the
growth curve. The experimental curve has been extrapolated to the theoretical value of K
194 J. SCHIJVE and A. H. W. HOEYMAKERS

crack
length
0
Im m )

0 20 LO 60 8l
k i Io c yclc s

0 I 8 12 16
k i locvclcs

0 2 I 6 8
kilocycles

Fig. 6. Crack growth curves: natural cracks __ ; predictions for through cracks - - - -.
 Fatigue crack growth in lugs 195

10

Through crack starting

from saw cut ( 0 : 1 rnm I
Data from specbrnens M E

1 1

I
I
r

01 ’
8 10
1 15 20 25 30
0.1
1 2 3 L 6 8 10
AKI MN / rn oIrnm1
i C A G W

Fig. 7. Experimental determination of AK-values for fatigue cracks in lugs.

for a + 0 (dotted line) by considering a very small crack as an edge crack in a stress field of
magnitude K , .a and adopting an edge correction factor of 1.12 [9] :

K 1 12K,a&
* W
lim ~ - = 1.12Kt ___
a-+o a
,
& - an& W-D‘

with K, = 2.8, W = 60 mm and D = 25 mm, the limit value is 5.38.

Impellizzeri and Rich [7] started from a stress distribution obtained with the finite
element method for the uncracked lug. Stresses thus obtained were applied to the crack
edges and analytical solutions including a finite width correction were then used to
calculate the stress intensity factor for one particular lug dimension. The result was
published in graphical form shown here in Fig. 9 with numerical data (measured from the
graph in [7]) presented in Table 3 of this paper.
Newman [S] adopted a superposition of analytical solutions, making appropriate
assumptions to account for the dimensional variables involved. He thus arrived at formulas
for a semi-elliptical comer crack at a loaded hole, which allowed a conversion to a through
crack. In the formulas obtained the lug dimensions D (hole diameter), W (width) and a
(crack length) have to be substituted. For a single through crack Newman’s result is:
196 J. SCHIJVE and A. H. W. HOEYMAKERS

20
Derived from figs 7 a and b
AK
MN/m3/2

16 -

/
/
12- I
I
I ua = LOMPa
I I I I

I I
I -
D
- L..."
I I
I . - I- .----.) a present data I
I I I
saw cut
'/r
Fig. 9. The experimental stress intensity factor compared to theoretical predictions.

with

fb = 0707- 0.181 + 6.55L2- 1054L3+ 6~841~

 Fatigue crack growth in lugs 197

with
1
A=-
+
1 2a/r
With these formulas K was calculated for the present lug dimensions (see Fig. 9 and Table
3).
Kirkby and Rooke [4] employed a method developed by Cartwright and Rooke [lo]
which basically is a superposition of analytical solutions. They presented quantitative data
in graphical form for a lug with a WID ratio of 2.56 which is somewhat larger than for the
present lug. Their results are also presented in Fig. 9 and Table 3.
Figure 9 shows an excellent agreement between the test results and the curve of
Newman, while the curve of Impellezzeri and Rich is not very much different from the curve
of Newman, except for larger alr values. The curve of Kirkby and Rooke shows
systematically higher values. The specific reason for this is unknown. The W I D ratio for
Kirkby and Rooke's data is 2.56 which is larger than for the present lug. However,
introducing this value into Newman's formulas above still leaves a difference of some 20%.
The results of the present test were approximated by a polynomial equation shown in
Table 3 , which fits the test data better than 0.5%. This equation was used to calculate AK-
values for the crack growth data of all lugs. The result is shown in Fig. 10 together with the
curve for a centre cracked panel, also given in Fig. 7a. For each lug only a relatively small

Table 3 . Stress intensity factors for a single through crack in a lug

Theoretical values
Impellizzeri/
( r = D/2) Present tests Newman Rich Kirkby/Rooke
alr WID = 2.40 WID = 240 WID = 2.37 WID = 256 Eqn. (A)

0 538* 5.38 5.5 - 5.380

0.1 4.32* 4.16 419 4.74 4.327
0.2 3.58 3.44 3.48 4.07 3574
0.3 3.04 2.98 3.08 3.61 3.052
0.4 2.69 2.68 2.81 3.26 2.698
0.5 2.46 2.48 2-61 2.98 2.464
0.6 2.32 2.34 2.5 1 2.76 2.312
0.7 2.22 2.26 2.43 2.62 2.214
0.8 2.15 2.23 2.43 2,50 2.152
0.9 2.24 2.46 2.43 2.123
1 .o 2.31 2.57 2.41 2.130

* Extrapolated
K a
= 5.38- 12.3 -
r
+ 19.1
0.6
198 J. SCHIJVE and A. H. W. HOEYMAKERS

do
dn
(prn/c)
I I I

- o l
t

I
10
I
12 1L 16
I
18

Fig. 10. Crack growth rates of artificial cracks and natural cracks.

AK-interval is covered as discussed before (see also Fig. 8). The result for the artificial
through crack fully coincides with the curve for the centre cracked panel. This should be
expected since these results were used as a calibration to arrive at the experimental K -
values.
The artificial corner cracks start with a lower du/dn which after some crack growth
approaches the artificial through crack result. This trend was noted before (Fig. 7b). The
natural cracks all start with low values, which rapidly increase to approach the curve of the
centre cracked panel. Most natural cracks were of the corner crack type, and as indicated
before the initiation phase was more complex than for the artificial comer crack. The more
chaotic initiation phase causes low apparent crack rates, considerably lower than predicted
 Fatigue crack growth in lugs 199

for a truly two-dimensional through crack. It should be pointed out once again that the
crack rates plotted in Fig. 10 are derived from crack length observations on the outside
surface and from the four possible cracks in each lug the crack rate applies to the largest
observed crack length. It should then be clear that a crack starting in the midplane of the
sheet as a semi-circular (or elliptical) crack could lead to apparent crack rates at the outside
surface which are initially higher than for the centre cracked panel curve.

5. STRESS INTENSITY FACTOR FOR A CORNER CRACK

A corner crack at a hole implies a three-dimensional configuration, while it should be

expected that K will vary along the crack front. In the literature several proposals for
arriving at some effective K-value were made, which were reviewed by Wanhill and Lof
[ll]. These authors also made a finite-elementcalculation for the K-value for a lug with the
same W and D as for the present investigation. The thickness, however, was considerably
larger, i.e. 20 mm instead of 5 mm in the present case. Wanhill and Lof further performed
crack growth tests to derive the K-value from growth rate data in the same way as outlined
in Fig. 7. They conclude to a reasonable agreement between calculated and measured K-
values.
The present data cannot be of great significance with respect to corner crack stress
intensity factors in view of the relatively small thickness. As Fig. 7b shows, the crack rate is
considerably smaller than for a through crack. However, if the crack length approaches the
sheet thickness (5 mm) the difference vanishes rapidly. The present corner cracks will
readily change from a quarter circular shape to an elliptical shape and later to a through
crack. At best for short cracks a comparison with other data seems worthwhile. The
following data were obtained (Table 4).

Table 4

Wanhill/Lof [11) Newman [8]

Crack length Present tests (experimental) (theoretical)
a=2mm 2.58 - 2.10
a=3mm 2.64 1.92 1.88

The agreement is not satisfactory, but in view of the low thickness in the present tests
larger K-values due to non-circular crack shapes should have been expected.

6. DISCUSSION

The present tests on lugs have shown significant differences between the growth of
artificial cracks (initiated from a small saw cut) and the growth of natural cracks (initiated
by fretting between the pin and the hole). The growth of artificial cracks showed little
scatter (Fig. 2) and the stress intensity factors derived from the growth rate of an artificial
through crack was in good agreement with predictions in the literature [7, 81.
200 J. SCHIJVE and A. H. W. HOEYMAKERS

The growth of natural cracks was subjected to more scatter (Figs 5 and 6 ) , which is due
to differences in the initiation phase. As a result of fretting crack initiation will occur at
different locations in the hole. More cracks may nucleate more or less simultaneously at
locations slightly staggered along the circumference of the hole. Since the nucleation is not
a well-reproducible process, scatter should be expected.
The more complex geometry of a natural crack also causes a slower growth (growth
rate based on its maximum length) than observed for an artificial crack. This should be
explained by the fact that the natural crack is not a well-defined two-dimensional through
crack, contrary to the artificial through crack. In addition more cracks growing on
staggered levels, forming ridges at the overlap, will also cause a slower growth. If a crack
front is a single straight line, perpendicular to the surface (the shortest possible crack front)
the crack can more easily propagate. This applies to the artificial crack, but not to the
natural crack.
In the introduction reference was made to the practical problem of predicting crack
growth in lugs. It appears from the present results that a prediction for a well-defined two-
dimensional through crack under constant-amplitude loading is possible. The Newman
formulas given before account for the hole to width ratio ( D I W ) and the agreement with the
present results was good. It should be realised that the shape of the lug head does not occur
in these formulas and for large D / W values and larger cracks some influence of the lug head
should be expected. A parametric study of K-values in lugs of different dimensions should
be recommended.
With respect to corner cracks the present results are not particularly meaningful
because the thickness of the lugs was relatively small ( t / D = 0.2). For thicker sections the
problem of corner cracks and semi-circular cracks also should be studied in more detail,
both for obtaining crack growth and K-values. A major problem is the three-dimensional
character with a stress intensity that varies along the crack front.
In practice there are additional problems. In the first place a crack usually will not be a
two-dimensional through crack. However, the present investigation endorses the view that
such a two-dimensional crack is a “worst case”. In design studies it could be decided to
evaluate this worst case only. Another problem is that the prediction of constant-amplitude
crack growth is not sufficient if service loading is of the variable-amplitude type. Prediction
methods for variable amplitude loading must then be adopted, but this is beyond the scope
of the present discussion.

7. CONCLUSIONS

Observations on crack growth in aluminium alloy lugs (width 60 mm, hole diameter
25 mm) indicate:
1. Crack growth of artificial cracks (i.e. initiated by small saw cuts) shows little scatter.
However, considerable scatter is found for natural cracks (initiated by fretting inside
the hole). The scatter is caused by crack initiation at different locations and different
time intervals.
2. Crack growth of a natural crack is slower than for an artificial crack especially if the
crack is still small.
3. Crack growth of an artificial through crack may well be considered as a “worst case”
for natural cracks.
 Fatigue crack growth in lugs 20 1

4. K-values were derived from the growth of artificial through cracks. T h e values were
in good agreement with K-values proposed in the literature. As a result crack growth
predictions for the worst case seem desirable.

Acknowledgement-J. Snijder gave careful assistance in the fatigue tests

8. REFERENCES
[l] Schijve, J. “Fatigue of Lugs. Contributions to the Theory of Aircraft Structures. Professor A.
van der Neut Anniversary Volume”. Nijgh-Wolters Noordhoff Univ. Press, 1972, pp. 423-440.
[2] Military Specification Airplane Damage Tolerance Requirements. MIL-A-83444 (USAF), 2
July 1974.
[3] Hoeymakers, A. H. W. ‘Fatigue in lugs’. Thesis of Dept. of Aerospace Engineering, Delft, 1977
(in Dutch).
[4] Kirkby, W. T. and Rooke, D. P. ‘A fracture mechanics study of the residual strength of pin-lug
specimens’. In “Fracture Mechanics in Engineering Practice”. Applied Scientific Pub]., London,
1977, p. 339.
[5] Larsson, S. E.‘The Development of a Calculation Method for the Fatigue Strength of Lugs and
a Study of the Test Results for Lugs of Aluminium Alloys’. Fatigue Design Procedures, ICAF
Symp., Munich, 1969, pp. 309-339.
[6] Van Lipzig, H. T. M. ‘Reducing the growth of fatigue cracks’. Thesis of Dept. of Aerospace
Engineering, Delft, 1973 (in Dutch).
[7] Impellizzeri, L. F. and Rich, D. L. Spectrum fatigue crack growth in lugs. ‘Fatigue Crack
Growth mder Spectrum Loads’, ASTM STP 595, 1976, pp. 320-336.
[8] Newman, J. C., Jr. ‘Predicting Failure of Specimens with Either Surface Cracks or Corner
Cracks at Holes’. NASA T N D-8244, June 1976.
[9] Paris, P. C . Stress analysis of cracks. ‘Fracture Toughness Testing and its Applications’.
ASTM STP 381, 1965, pp. 30-81.
[lo] Cartwright, D. J. and Rooke, D. P. ‘Approximate stress intensity factors compounded from
known solutions’. Engng Fract. Me&. 1974 6, 563-571.
[l 11 Wanhill, R. J. H. and Lof, C. J. Calculation of stress intensity factors for corner cracking in a
lug. ‘Fracture Mechanics Design Methodology’. ARGARD CP 221, Feb. 1977, Paper 8.