Stability of Structural Elements: Book D

Book D
Stability of
Structural Elements
Introduction
Classical books in strength of materials deal with the elastic buckling of columns and
present the Euler equation. Things seldom go any further. Plate buckling might be occasionally
discussed, but in-depth analysis of the stability of structures is usually not part of the curriculum
because the classical designer will avoid those cases and, to steer clear of those problems, the Euler
equation is conservative enough.
In aircraft structures (although this is also valid for space-faring structures) the primary goal
is structural integrity with minimum weight. Thin elements introduce buckling modes which do not
exist in classical columns: so-called secondary instabilities in which the columns buckle as plates.
Compression panels made of thin sheets attached to stiffeners are largely used in the structures. In
contrast to columns, such plates can carry larger loads than their buckling load and still be in the
elastic range. Therefore, the designer working at minimal weight will very carefully investigate the
load-carrying capability of the buckled structure, which may seem nonsense to the classical
mechanical engineer designing at the proportionality limit.
Columns of short slenderness ratio are present in airframes; their buckling may take place in
the plastic range, so methods to evaluate the critical load must be provided taking into account the
material properties; otherwise, the weight penalty will be too large. In this case, we are not looking
at limit load design, for which no permanent deformation should occur, but we are checking the
structural integrity at the ultimate load, for which permanent deformations such as plastic bending
may be allowed – care must be taken to avoid unstable buckling of the elements, since the structure
cannot fail for three seconds…
The theory of elastic stability as found in airframe design books is really a theory only in the
first half of the stress range, up to buckling. Beyond buckling, no satisfactory theory exists, and
people rely heavily on approximate methods and test results established mainly during the 1940’s
and 1950’s. Among those are many NACA reports and a 7-volume compilation by Gerard and
Becker, called Handbook of Structural Stability and published as NACA technical notes.
Section D1:
Elastic column buckling
1 Euler buckling theory

1.1 Canonical case: the pin-ended column
1.1.1 Introduction
It has long been recognized that a form of y y
structural instability existed in beams subject to Px Px

compressive loads. In theory, a beam such as
shown on figure D1–1 should be stable since the z z
compressive loads act axially and there is full

x x
axial symmetry. Experience, however, shows
Unbuckled form Px Buckled form Px
that the beam “buckles” for a given value of the
axial load, which is far below the load which Figure D1–1 The buckling of a column
under compressive loads
causes the stresses to reach the elastic limit.
The value of the critical load leading to buckling has been first reported by Leonhard Euler
in 1744, hence the result is commonly referred to as the Euler critical load, Euler buckling load or
Euler equation. It is described in most textbooks dealing with structures [1-5].
1.1.2 Computation of the Euler buckling load

z
In order to compute the buckling load, let
My
us consider a pin-ended beam loaded as shown
P w
on figure D1–2. In this case, we do not consider
the non-deformed shape: if we did, we could not
compute buckling. We have to consider that the P P
beam has deformed as shown here, and is in x
equilibrium with the external loads This allows Figure D1–2 Computation of the Euler
column buckling load
us to derive the differential equation of the
buckling beam.
The insert shows the cut in the beam allowing to compute the internal bending moment
resultant. We obviously find:
My  P w (D1.1)
Now, according to beam theory, the differential equation providing the beam deflection is:
2 w |M
 y (D1.2)
x 2
EIyy
D1.1
Thus, we obtain the following differential equation:

2 w |P
 w0 (D1.3)
x 2
EIyy
This is an homogeneous differential equation, which admits a number of solutions. First of

all, it is clear that w  0 is a perfectly valid solution. This corresponds to the non-buckled solution
for which the beam remains perfectly straight.
In order to obtain the buckling load, we consider increasing P until we reach the buckling
load. Thus, P  λ P0 in which λ is a parameter and P0 is a unit load. The equation then turns into an
eigenvalue equation in λ.
We can consider solutions of the type:
|n π x
wn  An sin (D1.4)
L
These solutions satisfy the boundary conditions of no lateral displacement at x  0 and x  L.
Substituting in the differential equation, we find:
2
 |n π |λ P0
    w 0 (D1.5)
  L  EIyy  n
Clearly, if we are to obtain a solution with w  0, the term in the square brackets must then be zero.
Thence:
2
|n π
λ P0  EIyy   (D1.6)
 L 
This solution is only possible when λ reaches the lowest possible value corresponding to n  1. This
is the Euler column buckling load. For this value of the load, it is possible to simultaneously have
two solutions:
• a perfectly straight beam solution w  0;
• a solution with a sinusoidal beam deformation w1 given by (D1.4).
Such a situation is called a bifurcation point. If the beam were perfectly symmetrical with a
perfectly axial load, it would remain without deflection as the load would be increased beyond the
critical load. However, that corresponds to an unstable equilibrium. In real beams, imperfections
and asymmetries cause this equilibrium to be broken, and the beam then assumes the deflected
shape (which also corresponds to an unstable equilibrium: the beam collapses under the buckling
load).
Although we have considered a deflection in the (x,z) plane nothing constrains the beam to a
deflection in that particular plane. Thus, the first buckling load will be the lowest possible load for
which buckling may occur; from (D1.6) we see that it corresponds to a minimum value of I (the
smallest of the principal moments of inertia). Hence, the load which corresponds to the first
bifurcation is given by:
2
|π EImin
Pcr  (D1.7)
L2
D1.2
If, by putting a restraint in the middle of the beam, the beam cannot deform under the shape
w1, it remains straight for higher loads, until the bifurcation n  2 is reached at load 4 Pcr with the
deformed shape w2.
1.1.3 Critical stress

The buckling stress is given by the buckling load over the cross-section:
2
|π EIyy
σcr  (D1.8)
S L2
It is customary to define the radius of gyration of the section ρ as :
|Iyy
ρ (D1.9)
S
With this definition the buckling stress becomes:
2
|π Ec
σcr  (D1.10)
(L/ρ)2
The quantity λ  L/ρ is sometimes referred to as the slenderness ratio (élancement) of the
column.
NOTE: in this expression, we have added the subscript c to Young’s modulus since, in the analysis
of buckling, it is the compression Young’s modulus Ec which is the relevant quantity.
1.2 Other cases of column buckling

Buckling also occurs in beams with other restraints than pinned-pinned configurations and
with other loads that end loads. In such cases the formula is modified by introducing a so-called
fixity coefficient c:
2
|π Ec
σcr  c (D1.11)
(L/ρ)2
Alternatively, one can define an effective column length L’ such that:
2
|π Ec
σcr  (D1.12)
(L’/ρ)2
By comparing these two relations one obtains the relation between c and L’:
L
L’  (D1.13)
c
NOTE: some authors use C as (1/c1/2).
Values of fixity coefficients for some classical loading configurations are given in figure
D1–3 below. The values have been taken from Niu [6]. Both Bruhn [7] and Niu feature extensive
data for more complex load configurations (elastic supports, combined axial and distributed loads,
elastic restraints, …).
D1.3
1/2
Beam loading configuration c c
P P Pinned-pinned
Loaded at ends 1 1
p
P Pinned-pinned
Uniformly loaded
1.87 0.731
P P Fixed-fixed
Loaded at ends
4 0.5
p
P Fixed-fixed 7.5 0.365
Uniformly loaded
P P Pinned-fixed
Loaded at ends
2.05 0.7
p
P Pinned-fixed 6.08 0.406
Uniformly loaded
P P Free-fixed 0.25 2.000

Loaded at ends
p
P Free-fixed
Uniformly loaded
0.794 1.12
Figure D1–3 Fixity coefficients for common column loading configurations
1.3 Limit of the Euler formula

Let us consider a practical computation 100
Al 2023-T3 material data from Curtis - E = 9850 ksi
with e.g. an aluminium alloy 2023-T3. The

90
critical stress can be plotted against the
80
slenderness ratio, giving a column strength curve
as shown on figure D1–4. The Euler formula 70
Critical stress [ksi]
being of hyperbolic nature, we have a problem 60
applying it to short columns, since the critical 50
stress goes to infinity as the slenderness ratio 40

Ultimate compressive stress = 41.8 ksi
goes to zero. Since the above analysis is made 30

Yield stress (0.2% perm. strain) = 38 ksi
with elastic theory, we can use it only as long as Proportional limit = 24 ksi
20
the stress remains below the elastic (propor-
tional) limit. As figure D1–4 shows, this limits 10
the Euler formula to long, slender columns (ratio 0

0 10 20 30 40 50 60 70 80 90 100
above 65 in this case). This range (called the Slenderness ratio
long column range) does not correspond to ele- Figure D1–4 Column strength curve (Euler theory)
ments typically found within airframe structures.
Thus, in order to reasonably predict what happens to airframe members subject to
compressive loads, we need some theory able to predict the behaviour of columns above the
proportional limit. This is the short column range in which inelastic buckling of the column occurs.
This topic will be dealt with in section D2.
We should also note that, as the slenderness goes to zero, columns lose their tendency to
buckle. Very short columns (which include test material samples) fail either by fracture in
D1.4
compression or do not fail at all. As noted in MIL-HDBK-5, “most metals are so ductile that no
fracture is encountered in compression. Instead of fracturing, the material yields and swells out, so
that the increasing area continues to support the increasing load. It is almost impossible to select a
value for the ultimate compressive stress of such materials without having some arbitrary criterion.”
In figure D1–4, the ultimate compressive stress is marked as a red line. It is obviously
pointless to consider stresses above this level, which lead to fracture of the column. This stress level
is referred to as the column yield stress.
2 Buckling of non-uniform columns

2.1 Introduction
As engineers optimize structures to gain weight, it may well be that columns encountered in
aircraft structures do not have uniform properties. In this case the critical load cannot be predicted
by the simple relations obtained above and one needs to carry out deeper analyses of the problem.
At least three methods exist to analyse non-uniform columns. Bruhn [7] presents a method
based on the moment area principle. Niu [6] reproduces the numerical method of Newmark. A very
simple approach to the problem was presented by Martin [8] and will be explained here. Besides
these methods, of course, general eigenvalue algorithms can be used with a finite-difference form of
the buckling equation. This approach is briefly described in Rivello [1] but more details will
probably be found in books on numerical analysis.
2.2 The Martin method

2.2.1 General overview
This method consists in writing the beam buckling equation (D1.3) in finite difference form.
Assume that the beam length has been split in N equal parts of length ∆. Then, for any section n we
may write:
wn1  2 wn  wn1 |P
 w 0 (D1.14)
∆ 2
EIn n
For the sake of facilitating a tabulated solution, we refer each area moment of inertia to the one of a
reference section, say at the “left” edge 0. Then, (D1.14) is rewritten:
2
|(In/I0) |P ∆
(2 wn  wn1  wn1)  K (D1.15)
wn E I0
Martin’s method is thus as follows: in a first step, K is determined using successive
approximations. In a second step, the critical load is obtained from:
|K EI0
Pcr  (D1.16)
∆2
D1.5
This value of the critical load can be expressed by the classical Euler formula with a correction
factor C (similar to a fixity coefficient):
2
|π EI0
Pcr  C (D1.17)
L2
Considering that N ∆  L, it is obvious by comparing (D1.16) and (D1.17) that the correction factor
is given by:
2
|K N
C (D1.18)
π2
2.2.2 Application to symmetric columns

In this case, only half a column should be
considered. We may arbitrarily fix the displace-
ment of the middle point to 1 (see figure D1–5)
and use the condition of zero deflection at the
Figure D1–5 Martin method applied to a symmetric beam
pin end to compute the value of K.
Writing (D1.15) at node N requires to know wN1 which does not exist; however, because of
symmetry, this displacement is equal to the displacement wN1 so the equation can be solved for that
value:
 |K 
wN1  1  w (D1.19)
 2 (IN/I0) N
Then, for n  N1 down to 1, the following recursive formula may be used:
 |K 
wn1  2   w  wn1 (D1.20)
 (In/I0) n
The final relation for n  1 provides the value of w0 at the pin joint. The boundary condition being
no displacement, the value of K may then be adjusted until the recursion yields w0  0. If w0 is
positive, K must be increased. A first estimate of K can be derived from an assumed deflection,
either a sinusoidal shape or even a linear deflection: the N1 equations are solved for K and an
average value is used as a starting point.
We must note that, since only half of the plate was divided into N segments, the total
number of segments is 2N. Hence, (D1.18) should be written in this case:
2
|4 K N
C (D1.21)
π2
2.2.3 Application to asymmetric columns

More general columns are treated with N
segments over the full length. The scaling of the
deflection curve is done by setting the displace-
ment of node N1 equal to unity, as shown in
figure D1–6. Figure D1–6 Martin method applied to an asymmetric beam
D1.6
An initial value of K can again be obtained by assuming a deflection and computing an

average value. Then, for n varying from N1 down to 1, the recursive formula (D1.20) may be
used; it yields a value of w0 which serves to adjust K until w0  0 is obtained. The critical load is
then obtained from (D1.18).
2.2.4 Numerical stability

The limiting value of C should reach a constant as the number of segments is increased.
Thus, as N increases the value of K decreases quite abruptly. One then must be careful not to use
too large values of K, especially with large numbers of segments. Otherwise, what happens is that
the deflection curve crosses the axis, which indicates that a higher buckling mode has been found.
Thus, K is not single-valued. Care should be taken if the method is programmed in software code
form that the lowest possible value is obtained. This problem may easily be detected when using a
tabulated (spreadsheet) approach.
2.2.5 Application to columns of different materials

In the case of a composite column for which the material properties vary along the length, it
is possible to use the same technique by rewriting (D1.15) as:
2
|(En In/E0 I0) |P ∆
(2 wn  wn1  wn1)  K (D1.22)
wn E0 I0
The solution procedure remains the same.
Figure D1–7 Critical load Figure D1–8 Critical load

of a stepped asymmetric column [7] of a stepped symmetric column [7]
2.3 Non-uniform column charts

In order to facilitate the computations, major aircraft manufacturers have made charts
available to their engineers through in-house manuals. Such charts are available in Niu [6] or Bruhn
[7]. Figures D1–7 to D1–10, from the North American Aviation Structures Manual (here taken from
Bruhn), show such charts for stepped and tapered columns. The graphs give the correction
D1.7
coefficient B to apply to the Euler equation. Note that according to (D1.17) we have B  π2 C. This
explains why the curves tend to 9.8696 (i.e. π2) rather than 1 at the left limit (which corresponds to
the uniform column of constant properties, C  1). In the charts, all columns are supposed pin-
ended.
Figure D1–9 Critical load of a tapered column Figure D1–10 Critical load of a prismatic column
of constant thickness [7] (also valid for cylinders tapering to cones) [7]
[1] Rivello, R.M. Theory and analysis of flight structures. McGraw-Hill, New York, 1969.
[2] Curtis, H.D. Fundamentals of aircraft structural analysis. Irwin, Chicago, 1997.
[3] Megson, T.H.G. Aircraft structures for engineering students, 3rd edition. Arnold, London,
1999.
[4] Timoshenko, S. Résistance des matériaux, 2e partie : théorie développée et problèmes, 2e
édition. Librairie Polytechnique Béranger, Paris et Liège, 1954.
[5] Timoshenko, S. Théorie de la stabilité élastique. Librairie Polytechnique Béranger, Paris et
Liège, 1947.
[6] Niu, M.C.Y. Airframe stress analysis and sizing, 2nd edition. Hong Kong Conmilit Press Ltd.,
Hong Kong, 1999 (republished with minor corrections, 2005).
[7] Bruhn, E.F. Analysis and design of flight vehicle structures. Jacobs Publishing, Indianapolis,
1973.
[8] Martin, G.H. A procedure for determining the critical load for a column of varying section.
Journal of the Aeronautical Sciences, 13(3), March 1946, pp. 135-140.
D1.8
Section D2:
Inelastic column buckling
1 Inelastic buckling theory

1.1 Loading and unloading beyond the proportional limit
This problem is developed e.g. in Rivello [1] and Megson [2]. Let us consider a column
which is loaded in compression beyond the proportional limit. We imagine that it remains perfectly
straight up to that point. The stress field is as indicated on figure D2–1 (left): pure compression.
Then, we allow the column to buckle. The stress field changes as shown on the right of the figure:
the concave side of the beam has an increase in compression stresses while the convex side has a
decrease in stresses. The stress distributions are triangular if the beam is homogeneous, but the
moduli are different on both sides. Indeed, beyond the proportional limit, the paths for loading and
unloading differ. Figure D2–2 shows the stress-strain curve with stress increase and decrease from a
starting point in the inelastic range: loading occurs along the curve with a local slope equal to the
tangent modulus Et while unloading occurs parallel to the original elastic line of slope equal to the
classical modulus E. Clearly, E > Et.
uniform compression uniform compression σ

(no buckling) and bending (buckling) loading
Et
unloading
ε
Figure D2–1 Typical stress distribution for inelastic buckling Figure D2–2 Loading and unloading paths beyond the
proportional limit
The situation is identical to that of an heterogeneous beam with two different moduli. For
such a beam, the “modulus-weighed” section properties are:
|E
A*  ⌠
 E dA (D2.1)
⌡| 0 A
1 ⌠| E
z*F  z dA (D2.2)
A* 
⌡| E0
A
D2.1
|2E
I*yy  ⌠
 z E dA (D2.3)
⌡|A
0
Hence, using l and u for the loading and unloading sides with E as reference modulus, we obtain:
|Et
A*  Au  A (D2.4)
E l
|Et
I*yy  Iu  I (D2.5)
E l
We can now derive the differential equation of the beam deflection:
*
2 w |1 |M y
 (D2.6)
x 2
E I*yy
This is identical to the equation for elastic beam buckling, as expected from the heterogeneous
beam analysis, provided we use the weighed area moment of inertia. Thus, the classical Euler
formula used in the case of elastic buckling may be retained in inelastic buckling provided we
change the moment of inertia and consequently the radius of gyration of the cross-section to take
into account the different properties inherent to the loading and unloading cases. However, having
to work out the weighed moment of inertia and radius of gyration may be tedious, so it has been
customary to try as much as possible to keep the actual moment of inertia of the column and modify
the modulus of elasticity to take inelastic effects into account.
1.2 Tangent modulus and reduced modulus formulas

1.2.1 Tangent modulus formula
The problem of buckling beyond the elastic range was addressed as early as in 1869 by
Engesser [3]. He seems to have been the first to consider loading in the plastic range by using the
tangent modulus. Engesser thus wrote the critical stress for buckling by a simple modification of the
Euler formula:
2
|π Et
σcr  (D2.7)
(L/ρ)2
1.2.2 Reduced modulus formula

It was pointed out by Wolford [4] that in 1889 Considère had developed a seemingly more
appropriate model. In the case of Engesser’s approach, the beam does not benefit from any
unloading due to bending (tensile) stresses, which is not appropriate. When allowing this effect to
take place (as presented above), we can derive a reduced modulus Er. Thus we define:
Er I  E I*yy  Et Il  E Iu (D2.8)
The critical buckling stress is then found to be:
2
|π Er
σcr  (D2.9)
(L/ρ)2
D2.2
Although simple, this is not a straightforward formula. Indeed, in order to find out Il and Iu
one must know the stress distribution: in contrast to heterogeneous beam theory, the part of the
cross-section with modulus Et and the part with modulus E are not known a priori. This is a section-
dependent problem. However, we know that the constraint zF*  0 must be applied, as per
heterogeneous beam theory; furthermore, the neutral axis in this case separates the zone with
loading from the zone with unloading. This condition allows determining the reduced modulus. The
expression of zF* is:
1  ⌠| |Et ⌠| 
z*F  *   z dA  z dA  (D2.10)
A  ⌡| E ⌡| 
 Au Al 
Setting the expression between brackets equal to zero provides a relation which provides the
distance to the neutral axis. In the case of a rectangular section, the reduced modulus is expressed
[1,4-6] as:
|Er |4 e
 (D2.11)
E
(1  e )
2
In this expression (which can be derived by the interested reader), e stands for Et / E. The other
extreme case, a beam with no web but two concentrated masses at maximal distance from the
neutral fibre, is shown [1,4,5] to have a reduced modulus:
|Er |2 e
 (D2.12)
E 1e
1.2.3 Comparison of results

Since the secant modulus is smaller than Young’s modulus, it is also smaller than the
reduced modulus; hence, equation (D2.7) always yields lower stresses than equation (D2.9).
Consequently, for design purposes it is advisable to consider that the inelastic buckling stress
corresponds to the tangent modulus model, since this is a conservative assumption.
The careful tests carried out by von Kármán in his doctoral dissertation [7] show good
agreement with the reduced modulus theory. In contrast, the tests of the 1940’s (such as Wolford’s
results [4]) show better agreement with the tangent modulus theory. It is ascertained that initial
imperfections in the columns led to failure under lower loads in the 1940’s tests, since those were
directly taking production pieces. These are more representative of the conditions of use typically
found in aircraft structures and are thus to be preferred, as indicated by all authors ([1,5,8]).
This apparent discrepancy in the results was explained by Shanley [9], who showed that
even a perfect column will buckle under the tangent modulus load when the axial load is allowed to
increase during buckling, which is the practical case.
1.2.4 Computation of the critical stress

The computation of the critical stress in the inelastic domain is iterative for a given
slenderness ratio, because the critical stress depends on the tangent or reduced modulus, which in
turn depends on the critical stress.
D2.3
Thus, it is easier to draw a column 50

Al 2023-T3 material data from Curtis - E = 9850 ksi
strength curve from tabulated material data by 45

computing the slenderness ratio which corres- 40 (reduced)
Column yield stress
ponds to a given critical stress (and modulus). Yield stress

35 (tangent)

Figure D2–3 shows a column strength 30
curve derived for a column of rectangular cross-
25
section of 2023-T3 material [5]. The inelastic Proportional limit
20
curves tend towards the Euler curve for slender
columns but deviate substantially for lower 15
ratios. 10
Bruhn [8] duly points out that if the 5
points lying above the compressive yield stress 0

0 10 20 30 40 50 60 70 80 90 100
and the column yield stress are to be used in Slenderness ratio
actual sizing, they must be substantiated by Figure D2–3 Column strength curve (inelastic buckling)
actual tests of the structure.
2 Practical computations of inelastic buckling cases

Several techniques exist to easily obtain the critical stress in the inelastic range without
working out the iterative procedure. Three are presented below: use of the Ramberg-Osgood
parameters, use of the tangent modulus chart and use of the Johnson-Euler curves.
2.1 Non-dimensional column curve

Extensive tabulated material properties might not always be available and test graphs may
be of bad quality to retrieve accurate data. Cozzone and Melcon [10] have devised a non-
dimensional column curve equation which uses the Ramberg-Osgood parameters in order to allow
inelastic buckling computations to be easily carried out.
The Engesser equation (D2.7) is rewritten as:
2
|Et |(L/ρ)
 (D2.13)
σcr π2
The tangent modulus is related to the Ramberg-Osgood parameters through the relation:
|E
Et  n1 (D2.14)
 |3 n  σ  
 1  
 7  σ0.7  
The left hand side of relation (D2.13) can thus be developed as:
|Et |Et |Ec |Et |σ0.7 |Ec |E
   B2 c (D2.15)
σcr Ec σcr Ec σcr σ0.7 σ0.7
D2.4
The value B is a function of stress and of the Ramberg-Osgood parameters:

|Et |σ0.7 |1
B2   n (D2.16)
Ec σcr σ |3 n  σ 

σ0.7 7  σ0.7 
It is thus possible to plot a chart of the stress ratio (σ/σ0.7) as a function of B. To find the value of
parameter B for a given slenderness ratio, (D2.15) can be substituted in (D2.13):
|(L/ρ) |σ0.7
B  (D2.17)
π Ec
This completes the solution. The relation can be plotted in a chart as given in figure D2–4 (the
figure is plotted according to the plot style found in Bruhn, which follows that of the original
reference).
1.5 2.8
1.4 2.6
2
3
1.3 2.4
5
8
1.2 2.2
Stress ratio
10
1.1
15 2
20 2
25
60
1 1.8 3
45
35 5
60 1.6
Stress ratio
0.9 35 45
25 8
20 1.4
0.8 15 10
15
1.2 25
20
0.7 10
45 35
60
1
0.6
8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
5
B parameter
0.5 3
2
0.4
0.3
0.2
0.1
0
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3 3.25 3.5 3.75 4
B parameter
Figure D2–4 Non-dimensional buckling inelastic column buckling curves based on the Ramberg-Osgood parameters
D2.5
The inelastic column strength curves of 60

Al 2023-T3 material data from Curtis - E = 9850 ksi - Column strength 41.8 ksi
figure D2–3 have been replotted as figure D2–5 50 Cozzone and Melcon

with an additional curve created using the 40
Column yield stress (41.8 ksi)
equation of Cozzone and Melcon. It can be seen Tangent modulus

30
that this curve follows the tangent modulus data
20
quite accurately. The Ramberg-Osgood parame-
ters used for 2023-T3 were extrapolated from the 10
Curtis data and are σ0.7  36 665 psi and n  8. 0

0 10 20 30 40 50 60 70 80 90 100
Slenderness ratio
Also on the figure is the horizontal line
which limits the critical stress in column Figure D2–5 Column strength curve
(Cozzone & Melcon curve)
buckling: the column yield stress.
2.2 Use of tangent modulus charts

MIL-HDBK-5 [11] suggests using a gra-
phical solution to determine the critical stress in
σ/Et  0.02
the inelastic range. Rewrite (D2.7) as:
2
|σ |π
 K (D2.18)
Et (L/ρ)2
It is possible to compute K from the column
geometry and plot σ  K Et on compressive
tangent modulus charts. The intersect provides
the value of the critical stress. Such a procedure
is shown on figure D2–6 for (L/ρ)  22.2, i.e.
K  0.02; the corresponding critical stress is
54 ksi in the case of a longitudinal grain and
Figure D2–6 Example of critical stress determination from a
57 ksi in the case of LT compression. tangent modulus chart (clad 2014 T6 Al from MIL-HDBK-5)
2.3 Johnson-Euler curves

It may also be desired to obtain an analytical description of the column strength curve from
simple column yield stress values. This can be done by the so-called Johnson-Euler or Johnson
curve formula, which uses a generalized parabolic shape which connects with the Euler curve at a
critical slenderness ratio [6-5,12].
The Johnson-Euler curve gives a critical stress equal to the column yield stress σcu at zero
slenderness ratio. The critical stress then decreases according to the following generic formula:
σ  σcu  K λn (D2.19)
Typical values of n are 1 for steel and 2 for aluminium [8,5]. In contrast, the Euler equation is
written:
σ  π2 E λ2 (D2.20)
The parameter K is found by matching the slopes of the two curves at a critical slenderness ratio λc.
D2.6
We thus obtain:
 K n λcn1   2 π2 E λc3 (D2.21)
Obviously, the two curves must have the same stress value σ at the junction point λc. Hence:
σcu  K λcn  π2 E λc2 (D2.22)
The solution of equations (D2.21) and (D2.22) provides the values of λc and K. First, from (D2.21)
we derive:
2
|2 π E
K (D2.23)
n λcn2
Then, from (D2.22) we compute the critical slenderness ratio:

 |2  |E
λc  π 1   (D2.24)
 n  σcu
The classical Johnson-Euler parabola, favoured for steel and ductile metal alloys, is obtained
for n  2. Its equation is:
2
σ  σ co 
|σcu
4 π2 E
(L/ρ)2
 |2 E
(D2.25)
(L/ρ) cr  π
σcu
For aluminium alloys the Euler-Johnson straight-line equation (n  1) is preferred. Its
equation is:
σ  σ co 
|2 σcu
3π
|σco
3E
(L/ρ)
 |3 E
(D2.26)
(L/ρ) cr π
σcu
Figure D2–7 shows a part of the same 60

Al 2023-T3 material data from Curtis - E = 9850 ksi - Column strength 41.8 ksi
column strength curve as figure D2–3, with the 50

(Johnson-Euler, n=2)
Johnson-Euler equations plotted for both n  1 40

and n  2. It can be seen that the tangent
30
modulus curve lies closer to the straight line (Johnson-Euler, n=1)
20
formula although the parabola lies closer to the
experimental curve at very low slenderness 10
ratios. The straight line thus provides us with a 0

0 10 20 30 40 50 60 70 80 90 100
conservative estimate of the critical stress, espe- Slenderness ratio
cially in the very short column range. Figure D2–7 Column strength curve (Johnson parabolas)
D2.7
2.4 Non-uniform columns

According to the literature [8,12] there is no problem in transposing the techniques seen in
section D1 for non-uniform column buckling in the elastic range to the inelastic range. The elastic
modulus must be replaced by the tangent modulus, according to the practice already identified
above. Therefore, in most cases the problem remains iterative as the correction coefficient requires
to know the ratio of products Et I which are unknown until stress levels have been determined. The
classical method is to assume a critical load in order to compute the stresses in the various parts of
the column; from the stresses values of the tangent modulus can be computed, providing the
correction coefficient; the critical load can then be computed. The procedure can be restarted with
the new critical load until convergence is obtained.
1999.
[3] Engesser, Fr. Zeitschrift des hannoverschen Ingenieurvereines, 1869.
[4] Wolford, D.S. Significance of the secant and tangent moduli of elasticity in structural design.
Journal of the Aeronautical Sciences, 10(6), June 1943, pp. 169-179.
Liège, 1947.
[7] von Kármán, Th. Untersuchungen Über Knickfestigkeit. Inaugural-Dissertation zur Erlangung
der Doktorwürde der Hohen Philosophischen Fakultät der Georg-August-Universität zu
Göttingen, 1909.
1973.
[9] Shanley, F.R. Inelastic column theory. Journal of the Aeronautical Sciences, 14(5), May 1947,
pp. 261-267.
[10] Cozzone, F.P. & Melcon, M.A. Non-dimensional buckling curves – their development and
application. Journal of the Aeronautical Sciences, 13(10), October 1946, pp. 511-517.
[11] Metallic materials and elements for aerospace vehicle structures. Military Handbook MIL-
HDBK-5H, U.S. Department of Defense, December 1998.
D2.8
Section D3:
Theory of the buckling of
flat plates
1 Introduction
The problem of flat plate buckling is, in essence, similar to the problem of column buckling:
a flat plate buckles under the action of in-plane compressive loads for which the plate reaches a
state of neutral equilibrium with a slightly bent deformed shape.
Consider for example figure D3–1, which shows a long plate with two supported edges and
two free edges under the action of compressive in-plane loads along the supported edges (for
clarity, the internal resultants are shown to act compressively, which is in contradiction with our
positive sign convention for plates). It is obvious that, because of equilibrium, the flat plate should
remain flat.
z z
fy fy
x
x
y y
fy fy
Statically stable form Buckled form
Figure D3–1 Example of a plate buckling as a column
However, just as the beam deflection equation allows a neutrally stable condition with non-
zero deflection under compressive loads when large deformations are accounted for in the
equilibrium relations, the plate equations also allow a neutrally stable condition with non-zero
deflection under compressible loads, as shown on figure D3–1 (we studied the problem of the plate-
membrane in which large deformations of the plate were included in the equilibrium relations,
yielding the von Kármán equations). The critical load fcr at which the stability bifurcation occurs is
the plate buckling load. It is the primary instability mode of the plate.
A flat plate may buckle under a variety of modes, since it all depends on the way the plate is
supported [1]. In figure D3–1, the plate actually buckles following a column buckling mode which
is essentially one-dimensional. If the plate is supported along one free edge, a flange buckling mode
will occur as per figure D3–2. If the plate is now supported along its four edges a more complicated
pattern typical of plate buckling modes will occur as shown on figure D3–3.
D3.1
Theory of the buckling of flat plates
z z
fy fy
x x
y y
fy fy
Buckled form Buckled form
Figure D3–2 Example of a plate buckling as a flange Figure D3–3 Example of a plate buckling as a plate
In figure D3–3, the plate may be thought of being divided into a number of “pockets” in
which the surface is deflected in one direction or another (on the figure, we have three pockets). In
compression buckling, the pocket length (along y) is approximately equal to the plate width b
(length of the side on which the load is applied, x in this case).
Furthermore, a plate may buckle under other loads than compressive loads [2]. Figure D3–4
shows the buckling of a flat plate under shear loads. In that case, the plate still buckles in a
succession of peaks and valleys but the “pockets” are warped along y following the directions of the
shear loads on the long edges. The pocket length is longer, about 1.25 b in this case. Finally, one
must also consider bending buckling, when the in-plane loads are part-tensile, part-compressive due
to bending of the surrounding structure. Figure D3–5 shows the buckled form of the plate, where
peaks and valleys appear on the compressed side of the plate and almost no deflection appears on
the extended side. In this case the pocket length is about 2/3 b.
Figure D3–4 Shear buckling of a flat plate Figure D3–5 Bending buckling of a flat plate
The analysis of the buckling of flat plates is more complex than the corresponding column
problem. The skin of aircraft structures and the thin webs within the airframe are loaded under
combinations of compression, bending and shear. Each type of load has its own critical stress at
which buckling appears. However, it is to be expected that a plate subject to e.g. bidirectional
compression will buckle under a lower load than the lower of the two compression loads, because
the two loads strain the structure simultaneously. Thus, plate buckling problems also require the
determination of interaction parameters in addition to individual buckling loads.
D3.2
2 Buckling in the elastic range

2.1 Formulation of the elastic plate buckling problem
2.1.1 General principle
The plate buckling equations can be derived from the von Kármán plate equations. This
approach is presented by Rivello [3]. Let us recall the fundamental equation which yields the plate
deflection:
|  * | w | w   |  * | w  |  *  | w | w  
2 2 2 2 2 2 2 2
D  ν    2 (1  ν) D   D ν  
 x2    x2  y2   x y  x y   y2    x2  y2  
|  | w  |  | w  | F | w | F | w | F | w
2 2 2 2 2 2
 pz  ∆mT   V    V    2  (D3. 1)
x x  y y   y2  x2 x y x y  x2  y2
This equation can be simplified since the problem deals with determining the instability
point for a plate loaded in its plane. Consequently pz and mT are equal to zero since they correspond
to transverse loads. The stress function F must be determined from the applied in-plane loads.
In order to find the buckling loads, we assume that the loads increase until we reach
buckling. Thus, the procedure involves a multi-step method.
• We first consider prescribed edge load distributions X0 and Y0 as well as prescribed body forces
per unit area px0 and py0. If needed, a prescribed temperature distribution T0 is added.
• We then consider that edge loads, body forces and temperature all increase by a value λ such
that X  λ X0 etc. These values are plugged into equation (D3.1), yielding a homogeneous
differential equation for w with λ as a parameter. Note that, F being a linear function of the
loads, we have F  λ F0 in which F0 is the stress function under the prescribed basic loads. Note
also that the boundary conditions are homogeneous as well.
• The differential equation being homogeneous, it appears as the formulation of an eigenproblem.
It admits the trivial case w  0, which obviously corresponds to the statically stable solution of
the flat plate remaining flat (as shown on figure D3–1 left). A deformed solution is possible
only when λ is an eigenvalue of the equation. Buckling occurs for loads given by the smallest
value of λ.
2.1.2 The buckling load eigenproblem

Substitution of the λ relations in (D3.1) yields:
|  * | w | w   |  * | w  |  *  | w | w  
2 2 2 2 2 2 2 2
D
   ν   2 (1  ν)  D   D
  ν  
 x2    x2  y2   x y  x y   y2    x2  y2  
 |  | w  |  | w  | F0 | w
2 2
| F0 | w | F0 | w 
2 2 2 2
  λ  V0  V0   2   (D3.2)

 x   x   y   y   y2  x2 x y x y  x 2  y2 
This is indeed an eigenproblem formulation.

As was the case when we derived the plate-membrane equations, relation (D3.2) may be cast
in a form which is more suitable to prescribed displacements on the edges. In that case the in-plane
D3.3
displacements u0 and v0 are used in place of F. They must satisfy the following equations (taken
straight from our analysis of plate-membranes with small deflections):
  *  | u0 | v0   1  ν   *  | v0 | u0    fT
K  ν  K      px  (D3.3)
x  x y  2  y    x y  x
  *  | v0 | u0   1  ν   *  | v0 | u0    fT
 K   ν   K    p  (D3.4)
y  y x  2  x    x y 
 y
y
It follows that u0 and v0 are also proportional to λ. The internal resultants can be computed from the
following relations, which are simply the linearised forms of the plate resultant-displacement
relations:
 | u0 | v0 
fx  K *  ν  f
T
(D3.5)
 x y 
 | v0 | u0 
fy  K *  ν  f
T
(D3.6)
 y x 
(1  ν) K *  | v0 | u0 
fxy     (D3.7)
2  x y 
Hence, the internal in-plane load resultants are also proportional to λ. Thus, after finally replacing
the stress-function derivatives by the internal resultants, we obtain the following equation:
|  * | w | w   |  * | w  |  *  | w | w  
2 2 2 2 2 2 2 2
D  ν    2 (1  ν) D   D ν  
 x2    x2  y2   x y  x y   y2    x2  y2  
| w | w | w | w | w 
2 2 2

   λ   px  py  fx  2 f xy  f y  (D3.8)
0  x 0  y 0  x2 0 x y 0  y2
 
This is an alternate equation for the plate buckling eigenproblem.
2.2 Uniaxial compression of a supported plate

2.2.1 Derivation of the eigenproblem formulation
z
As an example of plate buckling X0
y
problem, consider a rectangular plate which is
simply supported along its four edges and which
is subject to uniaxial compression of intensity X0
along the x-axis. The problem is similar to the X0
example shown on figure D3–3 except for the
compressed axis. Figure D3–6 shows the a
updated problem definition. The plate is x
supposed to have a constant value of D*. The b
edges parallel to the x-axis are unrestrained Figure D3–6 Uniaxial plate buckling
against in-plane motion.
D3.4
This solution was first derived by Bryan (1891) to compute the buckling stress of ship hull
plates. It is derived e.g. in Bruhn [2], Rivello [3], Timoshenko [4] and Megson [5]. Results are
discussed in other sources, such as Curtis [1]. It forms the basis of more complex plate buckling
problems.
There are no thermal loads nor body forces in this problem. This is a simple case for which
the internal resultants are readily found to be:
  λ X0
fx0
fy0 0 (D3.9)
fxy0 0
In this simple case, the equation corresponds to the classical plate equation with a right-
hand-side term identical to the one that was derived in the section on long plates:
|X0 | w
2
∆2 w   λ (D3.10)
D*  x2
The boundary conditions are as follows. On the edges at x  0 and x  a we have:

| w
2
w 0 (D3.11)
 x2
On the edges at y  0 and y  b we have:

| w
2
w 0 (D3.12)
 y2
2.2.2 Derivation of the solution

In order to find non-trivial solutions we consider the deflection to be given by any of the
mode shapes corresponding to the general Navier solution for rectangular simply supported plates:
|m π x |n π y
wmn  Bmn sin sin (D3.13)
a b
The parameters m and n are integers which independently vary from 1 to ∞. As they change, the
plate takes on other modes of deflection, but all modes obviously satisfy the boundary conditions of
the problem.
In the Navier solution, all modes are considered simultaneously as terms of a double series
which coefficients are determined as a function of the boundary conditions. In the buckling
analysis, we consider each mode shape separately in the buckling equation. Substitution of (D3.13)
into (D3.10) gives us:
4 2 2 4 2
 |m π |m π |n π |n π  |X0 |m π
  2       wmn  λ *   w (D3.14)
 a   a   b   b   D  a  mn
This relation may also be rewritten:
2
  |m2 |n2  X m 
2
         λ |2 0 * |   wmn  0 (D3.15)
a b  π D a  
D3.5
A non-trivial solution wmn  0 is only possible if the terms in curly braces is equal to zero,
which provides us with critical values of (λ X0) for which buckling occurs under the deflected mode
shape wmn . Such eigenvalues are given by:
2
2 2
|π D  |m |n 
2 *
Xcr  2       (D3.16)
(m/a)   a  b 
The buckling load is given by the smallest of the critical loads for all possible values of m and n
since it will be smallest absolute load for which the shape is under (neutral) equilibrium with a
deflected shape.
2.2.3 Canonical form of the solution

Since the buckling load is applied along the edge of width b, it has been customary in the
literature to express (D3.16) in terms of the value b and of the plate aspect ratio (a/b). The critical
buckling load is thus defined as a function of the plate buckling coefficient k by the following
relation:
2 *
|π D
Xcr  k (D3.17)
b2
By comparing equations (D3.16) and (D3.17) the plate buckling coefficient can be
expressed in the case of buckling under uniaxial compression loads. It is the minimal value of:
2
2
 |m |(a/b) n 
k    (D3.18)
 (a/b) m 
Clearly, any value n > 1 produces a larger value of k than the value n  1. Hence, the plate
buckling coefficient is the minimum value of k from the following equation:
2
 |m |(a/b) 
k    (D3.19)
 (a/b) m 
The results of equation (D3.19) can be 10
plotted as a function of (a/b) for increasing

values of m, as on figure D3–7. The plate
m= 5
m= 4
8
buckling coefficient is the minimum value for a
m= 3
m= 2
Plate buckling coefficient
m=1
given (a/b) so it is given by the envelope curve

traced in bold. It is seen that for large values of 6
(a/b), i.e. for long plates, the envelope curve

tends to an asymptotic value of k  4. For small 4
values of (a/b) the buckling coefficient increases
and, since in that case m  1, tends to an
asymptotic value of 1/(a/b)2 (which goes to 2
infinity as (a/b) goes to zero. It is also seen that

the buckled shape of the plate depends on its 0
0 1 2 3 4 5
aspect ratio: short plates will tend to form only a/b
one wave but longer plates will form multiple Figure D3–7 Buckling coefficient of a plate
waves. under uniaxial compression
D3.6
2.2.4 Buckling stresses

It is a simple matter to obtain buckling stresses in the plate using the stress-internals
relations. We readily obtain:
2 *
E |fx E |π D
σxx  *   * k (D3.20)
1  ν2 K 1  ν2 K b2
In the case of homogeneous plates, this can be rearranged into:

2 2
|k π E  |t 
σxx     (D3.21)
12 (1  ν2)  b 
2.2.5 Long plates buckling

As already mentioned above, for large values of (a/b), the value of k reaches an asymptotical
value of 4. The buckling stress is thus given by:
2 2 2
|π E  |t   |t 
σxx   2     3.615 E   (D3.22)
3 (1  ν )  b  b
The numerical value has been obtained with ν  0.3.
2.2.6 Wide column buckling

We have noted that, when (a/b) is very small, the limiting value of k is 1/(a/b)2. Substituting
k in the relation (D3.17) yields:
2 *
|π D
Xcr  2 (D3.23)
a
The formula which gives the buckling load per unit width of plate is thus identical to the
formula giving the buckling load on a simply supported column (Euler column formula) provided
EI is replaced by D. This analogy has prompted the use of the term wide columns when referring to
compressively-loaded short wide plates.
2.2.7 Buckling shapes

The buckling shape depends on the value of m, which in turn depends on the aspect ratio
(a/b). As implied by figure D3–7, as the aspect ratio is increased from integer value to integer value,
so is the value of m. It thus appears that a square plate buckles with a shape of a “single pocket”
whereas longer plates buckle with as many “single pocket” shapes as one can fit the width in the
length. Figure D3–8 shows a visual rendering of the buckling of plates with values of (a/b) equal to
1 (square plate) and 5 (long plate).
In contrast, wide columns buckle with a “single pocket”: the break-up into small cells does
not happen when the loaded edges are the long edges (figure D3–9).
D3.7
square plate
rectangular plate with (a/b)=5

Figure D3–8 Buckling shapes of rectangular plates under uniaxial compression
Figure D3–9 Buckling shape of a wide column under uniaxial compression
2.3 Buckling under combined compressive loads

2.3.1 Introduction
The buckling of plates has been presented in a simple, uniaxial analysis which is not
representative of real load cases on plates. The vast majority of plates and plate elements are loaded
along different directions and, in addition, may be loaded by compressive, bending and shear loads
simultaneously on each edge. Indeed, combined extension-bending analysis of thin-walled beams
will lead to loads which combine constant compressive stresses with linearly-varying bending
stresses. Shear stresses also develop as shear flows in the skin panels.
It seems logical to expect that a plate subject to multiple types of loads will exhibit a
buckling strength much lower than in the case of uniaxial loads. Timoshenko [4] has addressed the
problem of a square plate under combined compressive loads and has drafted a method of analysis
which involves an interaction curve. This technique has been further used by other investigators to
determine other cases of interactions. We’ll analyse the combined compression case as an example
D3.8
of the methodology, since this simple case lends itself to a semi-analytical solution, as in the case of
uniaxial loading. For more complex cases, approximate methods of solution are usually sought.
2.3.2 Derivation of the eigenproblem formulation

z
Consider figure D3–10 in which a plate is X0 y
shown to be subject to constant compressive Y0
edge loads of intensity X0 in direction x and Y0 in
direction y. This case is the combination of two
uniaxial compressions as studied in the previous
section. The same method of analysis will be
used; as in the uniaxial case, the plate is Y0
a X0
supposed to have a constant value of D* and x
thermal loads will be neglected. This is a simple b
case for which the internal resultants are readily Figure D3–10 Combined compression buckling of a flat plate
found to be:
  X0
f x0
f y0   Y0 (D3.24)
f xy0 0
Substituting the values in the plate buckling equation yields:
|X0 | w |Y0 | w
2 2
∆2 w   λ  λ (D3.25)
D*  x2 D*  y
The boundary conditions are typical of simply-supported edges:

| w
2
x  0 and x  a w 0 (D3.26)
 x2
| w
2
y  0 and y  b w 0 (D3.27)
 y2
2.3.3 Derivation of the solution

As in uniaxial buckling, the Navier mode shape is assumed as buckled plate deflection, since
it satisfies all boundary conditions:
|m π x |n π y
wmn  Bmn sin sin (D3.28)
a b
Substitution in (D3.10) yields:
4 2 2 4 2 2
 |m π |m π |n π |n π  |X0 |m π |Y0 |n π
  2       wmn  λ *   wmn  λ *   w (D3.29)
 a   a   b   b   D  a  D  b  mn
This may be rewritten as:
2
  |m2 |n2  X m
2
Y n 
2
         λ |2 0 * |   λ |2 0 * |   wmn  0 (D3.30)
a b  π D a  π D b 
D3.9
Therefore, the characteristic equation which must be verified if a non-trivial solution is to

occur is:
2
2 2 2 2
 |m |n  |X0 |m  |Y0 |n
     λ 2 *  λ 2 *  0 (D3.31)
a b  π D a  π D b
2.3.4 Square plate solution

In the case of the square plate, a  b. Thus, equation (D3.31) becomes:
2
2 2 2 2
 |m |n  |X0 |m  |Y0 |n
     λ 2 *  λ 2 *  0 (D3.32)
a a  π D a  π D a
This can be written in the following form:
2 2
|a X0 2 |a Y0 2
(m2  n2)2  λ m  λ n (D3.33)
π2 D* π2 D*
Buckling occurs under the mode for which the 100
buckling loads will be minimal. In order to find

the mode, we graphically plot the linear relation 80
m= 3
, n= 1
between (λ X0) and (λ Y0) for fixed values of m

and n and observe whether or not a single value
of these integers always yields minimum 60
buckling loads. This is done by rewriting

y
(D3.33) as: 40
(m  n )  x m  y n
2 2 2 2 2
(D3.34) m=2, n= 1
Figure D3–11 shows a graphical plot of the 20

m=2, n=2
various lines obtained for the lower values of m

m=1, n=3
m=1, n=2
and n. It is clear from the figure that the lower 0

m=1, n=1
0 1 2 3 4
load values (i.e. lower x and y values) are x
obtained with m  1 and n  1. Equation (D3.33) Figure D3–11 Buckling curves for a square plate
is then rewritten using this result: under biaxial compression
2 *
|4 π D
 λ X0  λ Y0 (D3.35)
a2
This relation provides the limiting combinations of loads that lead to biaxial compression buckling
of a square plate. The derivation follows the approach of Rivello [3] and is also found in Bruhn [2].
2.3.5 Interaction equation

Let us recall that in the case of uniaxial buckling we had obtained the following limiting
value for Xcr  λ X0 at buckling:
2
2 2
|π D  |m |n 
2 *
Xcr  2       (D3.36)
(m/a)   a  b 
D3.10
Should the uniaxial analysis be carried out for the compression along Y, it would yield:
2
2 2
|π D  |m |n 
2 *
Ycr  2       (D3.37)
(n/b)   a  b 
It is known that uniaxial compression square plate buckling also occurs for m  1 and n  1.
Thus, the critical uniaxial loads are given by:
2 *
|4 π D
Xcr  Ycr  (D3.38)
a2
It therefore appears that equation (D3.35) can be rewritten in this final, simple form, which
relates the actual compressive loads for biaxial buckling to the uniaxial buckling loads.
|λ X0 |λ Y0
 1 (D3.39)
Xcr Ycr
This equation is the interaction equation for combined compressive buckling of the plate. It
relates the ratios of the critical buckling loads to the critical uniaxial buckling loads for the same
type of load. Such ratios are classically denoted R in the literature and are termed stress ratios.
Hence, the interaction equation becomes:
Rx  Ry  1 (D3.40)
Other combinations of loads will lead to other interaction equations, with two or more stress ratios
involved through non-linear combinations summed to unity.
2.4 Interaction curves

The interaction equations are usually 1.0
plotted as interaction curves. In the case of the

square plate, equation (D3.40) is fairly simply 0.8
plotted – the result is shown on figure D3–12.
For a given load case, the ratio between 0.6
X0 and Y0 is known, so as λ increases, the point
Ry
which is representative of the load moves on

figure D3–12 along a line passing through the 0.4
origin. The maximal load intensities at buckling

are then obtained at the intersection with the in- 0.2
teraction curve. For example, if the compressive
loads are equal in both directions, we draw a line
at 45° and find that at buckling Rx  Ry  0.5. 0.0
0.0 0.2 0.4 0.6 0.8 1.0
Rx
Thus, the plate will buckle at half the uniaxial
loads. Given that the uniaxial buckling coeffi- Figure D3–12 Interaction curve for a square plate
under biaxial compression
cient of a square plate is 4, the biaxial buckling
coefficient is 2 in this particular case.
D3.11
2.5 Margin of safety

Recall that the margin of safety is expres-
Ry
sed as:
|Allowable
M.S.  1 (D3.41)
Applied
From the structural engineer’s point of
view it is interesting to express the available
M.S. for a given load case.
path
Interaction curves can be drawn using load
R y0 B
M.S. values. Consider the interaction curve on
Ry A
figure D3–13. The applied loads lead to values
of Rx  Ry that define point A, well into the safe
O Rx R x0 Rx
domain of the interaction curve. The question is
then: how safe ? From the definition (D3.41) of Figure D3–13 Computation of the M.S.
from an interaction curve
the margin of safety, it would appear that:
||OB|
M.S.  1 (D3.42)
|OA|
However, it is clear from figure D3–13 that:
||OB|
Rx0  Rx (D3.43)
|OA|
and
||OB|
Ry0  Ry (D3.44)
|OA|
Summing these relations, we obtain:
||OB|
Rx0  Ry0  (Rx  Ry ) (D3.45)
|OA|
The left-hand-side is equal to 1 because the values of R0 correspond to point B, on the interaction
curve. Hence, substituting the length ratio in (D3.42), we obtain:
|1
M.S.  1 (D3.46)
Rx  Ry
3 Buckling in the inelastic range

3.1 Introduction
In the various buckling analyses of flat plates that have been carried out until now, no
reference was made to the yield stress of the material. It is thus possible that the buckling stress
predicted by the classical equation lies beyond the proportionality limit, in the inelastic domain.
D3.12
This is likely to happen for small values of b/t. The critical buckling stress will then be lower than
the corresponding elastic buckling stress and some correction must be taken into account.
The classical plate buckling equation (D3.21) may simply be modified into:
2 2
|π E t
σcr  η kc 2   (D3.47)
12 (1  νe )  b 
in which η is a plasticity correction factor which takes all inelastic effects into account.
3.2 Analysis of inelastic buckling

Inelastic buckling is a more difficult problem than elastic buckling. Theories have been
derived using either the deformation theory, which uses the secant modulus, or incremental theory,
which uses the tangent modulus. The problem is quite complex and cannot usually be treated by
exact analytical methods.
Rivello [3] presents an example of analysis based on the procedure published by Gerard and
Becker [6], which follow the theory initially developed by Stowell [7]. We will not reproduce the
development here.
The stability analysis of plate elements has been summarized in the handbook of structural
stability, a seven-part publication by the NACA. A selection of practical results (mostly coming
from that reference) will be given in the next section.
1973.
Liège, 1947.
1999.
[6] Gerard, G. and Becker, H. Handbook of Structural Stability, vol. 1: buckling of flat plates.
NACA technical note 3781, July 1957.
[7] Stowell, E.Z. A unified theory of plastic buckling of columns and plates. NACA Report 898,
1948.
D3.13
Section D4:
Practical charts for plate
buckling computations
1 Uniaxial elastic buckling charts

1.1 Compressive buckling coefficients
Charts and tables of compressive buckling coefficients can be found in NACA technical
note 3781 [1], which contains numerous data on flat plate buckling and behaviour in the plastic
domain.
The chart shown on figure D4–1 gives
the compressive buckling coefficient of a rectan-
gular flat plate under uniaxial compression for a
variety of edge restraint conditions. It is seen that
the magnitude of k strongly depends on the type
of restraint for the unloaded edges but weakly
depends on the type of restraint for the loaded
edges. Fully clamped plates have the highest
value of k and thus have the highest buckling
loads. Note that when an edge is free ν appears
in the boundary condition along that edge, so the
chart uses a mean value of 0.3 .
The buckling patterns always depend on
the number of buckling cells which form along
the plate, and thus of the value of (a/b). The
exception is flange buckling (curve E), for which
m  1 is always minimal, so the flange always Figure D4–1 Compressive buckling coefficients
buckles as a single pocket element, irrespective for a variety of edge restraints [1]
of aspect ratio.
Gerard & Becker [1] dutifully note that for long plates (a/b > 4) the effect of rotational
restraint on the loaded edges tends to vanish as the curves for the clamped and simply supported
cases converge one on the other. Recall the expression for the critical load:
2 *
|π D
Xcr  k (D4.1)
b2
It is readily seen that this load does not depend on the length of the long plate but only on its width
(i.e. the loaded edge size). As we have seen in the previous section, the plate tends to buckle by
D4.1
Practical charts for plate buckling computations
forming more or less square zones which characteristic dimension is the width. In aircraft
structures, the use of stiffeners creates long rectangular strips of plate which can be seen as long
plates, so this result is of importance to the design. The coefficients for long plate buckling are
summarised in table D4–1 at the end of the section.
Relation (D4.1) can also be expressed in terms of the buckling stress. Considering
homogeneous plates, we obtain:
2 2
|π E t
σcr  kc 2   (D4.2)
12 (1  ν )  b 
The buckling coefficient has been subscripted c to indicate that it is a compressive buckling
coefficient.
Figure D4–2 Compressive buckling coefficients for plates Figure D4–3 Compressive buckling coefficients for flanges
with a variable degree of edge restraint [1] with a variable degree of edge restraint [1]
1.2 Edge restraint effects

1.2.1 Unloaded edge rotational restraint
It is possible to obtain similar charts for different amounts of lateral edge restraints. The
curves fall in between the values for clamped edges and simply supported edges. The partial
restraint is measured using a parameter  which corresponds to the ratio of rotational rigidity of the
plate edge stiffener to the rotational rigidity of the plate. A value of zero corresponds to a simply
supported edge (no edge rigidity) and a value of infinity corresponds to a clamped edge.
D4.2
Figure D4–2 shows the results in the case of a plate with two supported edges and figure
D4–3 shows a similar chart for a flange (plate with one free edge).
In the case of different degrees of restraint on each side, Lundquist and Stowell [2] have
checked that the following approximation is valid:
k  k1 k2 (D4.3)
The coefficients k1 and k2 correspond to the coefficients found on figure D4–2 for the value of
rotational restraint parameter corresponding to each edge.
1.2.2 Unloaded edges lateral restraint

Classical buckling analyses (such as the example calculation of the preceding section)
assume that the plate is free to develop in-plane movement along the unrestrained edges. This is
necessary because, as the plate is compressed, it expands laterally through the influence of
Poisson’s ratio. In aircraft structures, this lateral expansion is usually not allowed because the skin
is fixed to stiffeners. The lateral restraint changes the longitudinal stress and thus influences the
buckling behaviour of the plate.
According to the classical plane elasticity
relationships, if uniaxial stress is developed, then
the lateral strain is  (ν/E) σxx and since the
stress is negative the strain is positive, yielding
expansion. To prevent such expansion, one has
to introduce compressive stresses along the
unloaded edges. A free-body diagram is shown
on figure D4–4, which also shows that the ribs
which support the loaded edges must develop
tensile stresses to maintain equilibrium. Such
equilibrium of course requires that:
σr Ar  σy (t a) (D4.4)
We suppose for simplicity that the plate and the
Figure D4–4 Compressive buckling coefficients for plates
rib are of the same material and we express that with lateral restraint (ν = 0.3) [1]
the lateral strains in the rib and plate are equal:
|σr |ν |1
εr    ( σx)  σy  εy (D4.5)
E E E
Substituting (D4.4) we obtain the relation that must exist between σx and σy:
|σy |ν
 (D4.6)
σx |a t
1
Ar
Argyris and Dunne have recomputed the buckling coefficient of flat plates with lateral
restraint. The results are given in [1] and given here as figure D4–4. The values are smaller with
lateral restraint.
D4.3
1.3 Shear buckling coefficients

A flat plate can buckle when subject to shear stresses along its edges. Clearly, in the case of
constant shear loads, faces at 45° angles will develop either extension or compression stresses, the
latter leading to buckling of the plate. Two possible buckling modes co-exist: a symmetric mode
and an antisymmetric mode. The problem was first addressed by Timoshenko in 1915 and is
developed in [3]. The critical shear stress is still given by a formula similar to all plate buckling
formulas, i.e. (considering homogeneous plate properties so the equation simplifies):
2 2
|π E t
τcr  ks 2   (D4.7)
12 (1  ν )  b 
The buckling coefficient is here subscripted s for

shear.
Values of ks are given on figure D4–5 for
both clamped and hinged edges. It may be seen
that as the aspect ratio changes, the symmetric
and antisymmetric modes alternate.
The effect of partial rotational restraint 
can be obtained by using the charts of figure D4–
6. The limiting value of ks∞ for infinitely long
plates is given on sub-part (a) of the figure as a
function of the amount or restraint. The ratio of
ks /ks∞ is given on sub-part (b), where it is seen
that the amount of restraint has little influence on
that ratio. Hence, the product of both values
yields a reasonably accurate value of ks for a Figure D4–5 Shear buckling coefficients
given amount of restraint. for rectangular plates [1]
(a) Determination of ks,∞ (b) Determination of the ratio of ks/ks,∞

Figure D4–6 Shear buckling coefficients for plates with a variable degree of edge restraint [1]
D4.4
1.4 Bending buckling coefficients

Finally, a flat plate may also buckle
under in-plane bending loads, i.e. linearly-
varying loads along the edge b. This yields yet
another buckling coefficient, kb for bending
buckling. Figure D4–7 gives the value of this
coefficient for rectangular plates with a varying
degree of rotational unloaded edge restraint. The
figure of course also provides values for the
simply supported and clamped edges.
This is not the most critical buckling
mode, as the values of kb are very large when
compared to the values of kc for similar boundary
conditions. Thus, for equal stress levels, bending
buckling cases will be much rarer than compres-
sion or shear buckling cases. However, these
stresses add up with other stresses (which are
lower) since bending is seldom encountered
alone in the structure, so the effect of bending
buckling needs to be taken into account when
cases of combined buckling are investigated.
Figure D4–7 Bending buckling coefficients for plates
with a variable degree of edge restraint [1]
1.5 Summary of buckling coefficients for long plates

In the case of long plates (a/b  3) the values of all buckling coefficients tend to an
asymptotic value. Table D4–1 hereunder lists the limit values for the most important practical cases.
Loading Edge support Coefficient Reference
Compression All edges supported 4.00 NACA report 733 [2]

All edges clamped 6.98 NACA report 733 [2]
Supported flange (one free edge) 0.43 NACA report 734 [4]
Clamped flange (one free edge) 1.28 NACA report 734 [4]
Shear All edges supported 5.35 NACA TN 1222 [5]
All edges clamped 8.98 NACA TN 1223 [6]
Bending All edges supported 23.9 NACA TN 1323 [7]
All edges clamped 41.8 NACA TN 1323 [7]
Table D4–1 Common buckling plate coefficients for long plates [1]
D4.5
2 Uniaxial inelastic buckling charts

2.1 Equations for computing the plasticity correction
According to the results published by Gerard and Becker [1], the plasticity correction factor
η is given by the following relations. First, define j as:
|Es |1  νe
2
j (D4.8)
E 1  ν2
The Poisson ratio in the inelastic range is related to the elastic Poisson ratio νe through the relation:
|Es
ν  νp  (νp  νe) (D4.9)
E
The fully plastic Poisson ration νp may be taken as 0.5 for isotropic materials.
The relations giving η are given below. All are for compression buckling except the last one,
which is for shear buckling:
• long plate with simply supported unloaded edges:
 |3 Et 
η  j 0.500  0.250 1  (D4.10)
 Es 
• long plate with clamped unloaded edges:
 |3 Et 
η  j 0.352  0.324 1  (D4.11)
 Es 
• long flange, one unloaded edge being clamped:
 |3 Et 
η  j 0.330  0.335 1  (D4.12)
 Es 
• long flange, one unloaded edge being simply supported:
ηj (D4.13)
• long column (a/b  1):
|Et
ηj (D4.14)
Es
• short plate loaded as a column (a/b  1):
 |3 Et
η  0.250 j 1   (D4.15)
 Es 
• square plate loaded as a column (a/b = 1):
 |E 
η  j 0.114  0.886 t (D4.16)
 Es 
• shear buckling of a rectangular plate, all edges being elastically restrained:
 |E 
η  j 0.83  0.17 t (D4.17)
 Es 
D4.6
The determination of the critical buckling load in inelastic plate buckling is, like in inelastic
column buckling, an iterative process since the tangent and secant moduli are determined from the
given stress level. It is possible to make a direct computation by using a Ramberg-Osgood
representation.
2.2 Inelastic buckling charts using Ramberg-Osgood stresses

Gerard and Becker provide diagrams which express the following relationship:
|σcr |σcr 
 f  el (D4.18)
σ0.7  σ0.7 
in which σ0.7 is the stress for 0.7 E secant modulus. Several curves are provided with the Ramberg-
Osgood parameter n as a free parameter. Thus, the calculation can be made directly as a correction
to the elastic buckling stress.
Figure D4–8 Nondimensional compressive buckling Figure D4–9 Nondimensional compressive buckling
stress for plates and long clamped flanges [1] stress for long hinged flanges [1]
In the case of plates with either simply supported lateral edges or rotationally-restrained
edges, figure D4–8 can be used to get the critical stress. This figure can also be used for long
clamped flanges. The results are an approximation since the plasticity correction η is slightly
different between those cases (see equations D4.10 to D4.12 above). For long hinged flanges the
curves are given as figure D4–9. A different curve is needed due to the quite different form of
(D4.13). Finally, the practical chart to be used for inelastic shear buckling of panels with edge
rotational restraint is given as figure D4–10.
It is also possible to use the method of Cozzone and Melcon [8] as outlined in section D2.
By analogy with the Euler-Engesser buckling formula, one has to use:
|12 (1  νe )
2
b
L/ρ  (D4.19)
t kc
However, this method seems to give more inaccurate results than the charts presented above, which
are in quite good agreement with test data.
D4.7
Figure D4–10 Nondimensional shear buckling stress for panels with edge rotational restraint
3 Cladding reduction factors

3.1 Definition
Clad aluminium plates are widely used in cladding material f t/2
aircraft structures. These are a kind of sandwich
design in which a covering of pure aluminium is t core material
placed on each side of the alloy plate, such as
shown on figure D4–11. The resistance of clad cladding material
plates is less than the resistance of the alloy. Figure D4–11 Cladding geometry
Hence, a cladding reduction factor  η has been introduced by Gerard [1] and the critical

stress σ of the clad plate is related to the critical stress of an equivalent plate made of core material
with the same thickness t by the relation:

σcr  
η σcr (D4.20)
The cladding reduction factor is based on the value f (ratio of total cladding thickness to
plate thickness, see figure D4–11) and on the value β which is the ratio between the cladding
proportional-limit stress and the core proportional-limit stress.
3.2 Results
The following results are reported in [1]. In all cases, if the critical stress is larger than the
core proportional-limit stress, set β  0 in the equation.
• Long simply supported plates in compression or shear:
 |1  3 β f
η  (D4.21)
13f
D4.8
• Short plate columns:
 |1  0.75 β f
η  (D4.22)
13f
• Long plate columns:

 |1
η  (D4.23)
13f
4 Interaction curves for various combined loads

The interaction curve method presented in section D3 may be used when two or more
simultaneous load conditions tend to cause buckling. In the case of two-load condition, an
interaction curve is obtained. In the case of three-load conditions, an interaction surface is obtained,
which is rendered as a set of curves for constant value of the third parameter. Not all relations are
linear, and interaction curves do not exist for all combinations of loads.
4.1 Biaxial compression

As already noted, the M.S. for biaxial compression is:
|1
M.S.  1 (D4.24)
Rx  Ry
This equation is valid for square plates or long plates that buckle in square patterns. Figure D4–12
shows a chart for determining the margin of safety. For plates of arbitrary a/b, it is possible to
deduce the interaction curve from the three-parameter curves of biaxial compression and bending
presented below.
1.2 1.2
M.S. buckling chart M.S. buckling chart
Biaxial plate compression Plate bending and shear
-10
Square or long plate Supported plate, all values of a/b
1 1 0
0 10
0.8 -1 0.8 -1
0 0 0
30
10 20
50
30
RS
Ry
0.6 40 0.6 30
10
20
0 40
50
-1
0
0.4 0.4
0
-10
10
20
50
30
40
0.2 0.2
20
0
10
30
40
-1
0
50
0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
Rx RB
Figure D4–12 Margin of safety curves Figure D4–14 Margin of safety curves
for combined biaxial compression for combined bending and shear
D4.9
4.2 Compression and shear

In that case the interaction curve has been determined by Stowell & Schwartz [9]. It is
expressed as:
R C  R S2  1 (D4.25)
with RC the stress ratio for the compressive load and RS the stress ratio for the shear load. The
relation is valid for simply supported plates with a/b  1 and for long plates with elastically
restrained edges. The associated margin of safety is then given by:
|2
M.S.  1 (D4.26)
RC  RC2  4 RS2
The interaction curve and M.S. curves are presented on figure D4–13. Note that, according
to Bruhn [10], the plate can buckle under combined tensile and shear loads, which explains the
range of negative values for RC (tension) in the chart.
M.S. buckling chart

Plate compression and shear
1.8 Supported plate, a/b > 1
1.6
1.4
1.2
-1 0
RS
1 10 0
50 30 20
40
0.8
0.6
0.4
0.2
Plate under compression loads
0
-2 -1.8 -1.6 -1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
RB
Figure D4–13 Margin of safety curves

for combined compression and shear
4.3 Bending and shear

Gerard & Becker [1] report that this problem was addressed by Timoshenko. They deduced
the following interaction equation:
RB2  RS2  1 (D4.27)
D4.10
with RB the stress ratio for the bending load and RS the stress ratio for the shear load. The relation is
valid for simply supported plates for all values of a/b. The associated margin of safety is then given
by:
|1
M.S.  1 (D4.28)
RB  RS2
2
The interaction curve and M.S. curves are presented on figure D4–14.
4.4 Bending and compression

This problem was also addressed by
Timoshenko. Bruhn [10] provides the following
interaction equation:
RB1.76  RC  1 (D4.29)
with RB the stress ratio for the bending load and
RC the stress ratio for the compression load.
Bruhn has plotted this equation and its associated
M.S. The chart appears as figure D4–15. There is
no closed-form solution of the M.S. equation in
that case.
Figure D4–15 Margin of safety curves

for combined bending and compression [10]
4.5 Bending, compression and shear

Johnson & Buchert [11] have studied the case of flat plates with the three combined loads:
bending, shear and transverse compression (i.e. not in the direction of the bending stresses). They
derived interaction surfaces, which are provided in [1] and reproduced as figure D4–16. The curves
apply for infinite aspect ratio (long plates).
(a) Upper and lower edges simply supported
D4.11
(b) Upper edges simply supported, lower edges clamped
Figure D4–16 Interaction surfaces for combined bending, lateral compression and shear [1]
Figure D4–17 Interaction curves for combined bending, compression and shear [10]
Another presentation is given by Bruhn [10], which allows to evaluate the margin of safety.
The figure, reproduced as figure D4–17, contains two parts. The left part is used if RS  RC and the
right part is used in the other case. To use the chart, enter with either RC or RS (the smallest of the
two) and RB to locate point 1 in the chart. Extend a radial line passing through 1 until it reaches the
correct ratio (RCRS or inverse). The values of RBA, RCA or RSA give the margin of safety.
4.6 Bending and biaxial compression

Noel [12] has studied the case of biaxial compression combined with bending. The
interaction surfaces, which are also provided in [1], are reproduced as figure D4–18. The curves
apply to simply supported plates with the specified aspect ratios.
D4.12
Figure D4–18 Interaction surfaces for combined bending and biaxial compression [1]
[1] Gerard, G. and Becker, H. Handbook of structural stability, part I: buckling of flat plates.
NACA technical note 3781, July 1957.
[2] Lundquist, E.E. and Stowell, E.Z. Critical compressive stress for flat rectangular plates
supported along all edges and elastically restrained against rotation along the unloaded edges.
NACA report 733, 1942.
D4.13

Liège, 1947.
[4] Lundquist, E.E. and Stowell, E.Z. Critical compressive stress for outstanding flanges. NACA
report 734, 1942.
[5] Stein, M. & Neff, J. Buckling stresses of simply supported rectangular flat plates in shear.
NACA Technical Note 1222, 1947.
[6] Batdorf, S.B. & Stein, M. Critical combinations of shear and direct stress for simply supported
rectangular flat plates. NACA Technical Note 1223, 1947.
[7] Schuette, E.H. & McCulloch, J.C. Charts for the minimum-weight design of multiweb wings in
bending. NACA Technical Note 1323, 1947.
[8] Cozzone, F.P. & Melcon, M.A. Non-dimensional buckling curves – their development and
application. Journal of the Aeronautical Sciences, 13(10), October 1946, pp. 511-517.
[9] Stowell, E.Z. & Schwartz, E.B. Critical stress for an infinitely long flat plate with elastically
restrained edges under combined shear and direct stress. NACA Wartime Report L-340, 1943.
1973.
[11] Johnson, A.E.Jr & Buchert, K.P. Critical combinations of bending, shear and transverse
compressive stresses for buckling of infinitely long flat plates. NACA technical note 2536,
1951.
[12] Noel, R.G. Elastic stability of simply supported flat rectangular plates under critical
combinations of longitudinal bending, longitudinal compression and lateral compression.
Journal of the Aeronautical Sciences, 19(12), December 1952, pp. 829-834.
D4.14
Section D5:
Buckling instabilities in
composite elements
1 Secondary instabilities in columns

1.1 Introduction
Sections D1 and D2 dealt with the buckling of columns, sections D3 and D4 with the
buckling of plates. These are two extreme cases in the range of buckling phenomena. Most columns
used in aircraft structures are not of the classical, closed hollow-cylinder type, but open or closed
sections of polygonal shape. These cross-sections can be qualified of “unstable” because they can
feature local buckling (or secondary buckling) modes. In contrast to a “stable” column, for which
buckling develops according to sections D1/D2 (Euler-Johnson curves), an unstable column can
develop a secondary buckling mode in which one of or all the elements composing the column
buckle as plates (typically, as plate flanges).
Regarding plates, they are not used alone in aircraft structures. Indeed, the critical stresses
for the various buckling modes of plates are relatively low, so that using plates is usually fairly
inefficient. Hence, plates are turned into composite elements1 through the use of stiffeners made of
formed or extruded thin-walled members, which are columns with unstable cross-sections. The
question then arises whether the element buckles as a whole (primary buckling) or through the
secondary buckling of an element.
1.2 Column strength curve of stiffeners

As a column, stiffeners can buckle under a column primary buckling load, such as discussed
under beam buckling (Euler theory). In that case the column buckles with a semi-wavelength equal
to the column height. In this primary instability mode, the cross-sections remain undistorted, as in a
thick beam. Figure D5–1 (left) shows2 a T-beam having buckled under primary instability.
However, thin-walled members can also be seen as plates attached one to another by a side
wall; thus, the member can be broken down into individual web or flange elements, each providing
some lateral support to its neighbour – this is a kind of plate assembly in which each element can
buckle locally as a long plate, i.e. with a half-wavelength of the order to the column width. This is
called the second instability mode, or local buckling, of stiffeners. In that mode the cross-section of
the column will be distorted due to individual buckling of some or all elements. Figure D5–1 (right)
1
Composite here doe not refer to a composite material but to a composite structure made of metallic elements.
2
All the drawing from the figure are for illustrative purposes only and not the solution of an actual buckling problem.
D5.1
Buckling instabilities in composite elements
shows the same T-beam viewed as an assembly of three flanges which have each buckled under the
same load (buckling of equal flanged elements, see subsection 2.1).
Primary buckling occurs for large values of L’/ρ (effective length over radius of gyration of
the column section) and follows the Euler equation for values of L’/ρ of 80 or more. Secondary
buckling occurs for values lower than 20. In the intermediate range a combination of the primary
and secondary modes will occur. Thus, the Euler-Johnson type of curve as derived in sections D1
and D2 do not apply: for very low values of L’/ρ the maximum stress is not the column yield stress,
but rather is the crippling stress, which occurs when the elements composing the stiffener
completely fail. Crippling failure involves a study of the post-buckling behaviour of plates, which
will be dealt with in a later section. This section will deal with buckling loads only.
Z Z
X X
Y Y
Primary instability: column buckling mode Secondary instability: flange buckling mode
Z Z Z Z
X X
Y X Y X
Y Y
Figure D5–1 Illustrative drawings of the primary and secondary instability modes of a T-shaped thin-walled beam
1.3 Sturdy stiffeners

Consider a stiffened plate under compression. It may buckle under a variety of modes: local
instability of the plate alone, global buckling of the stiffener, local buckling of the stiffener or
general instability of the plate-column composite.
In order to be fully effective, a stiffener must be “sturdy”, i.e. it must not buckle locally
under the applied load; in that case, its axial, bending and torsional rigidities influence the buckling
behaviour of the composite structure.
D5.2
Hence, it is necessary to first determine the local buckling stress of the stiffeners to ensure
their sturdiness. Then, the buckling stress of the stiffened plate will be found using the classical
relationship:
2 2
|π E t
σcr  kc   (D5.1)
12 (1  ν2)  b 
in which b is a characteristic dimension.

The data presented in this section has been taken from the handbook of structural stability
(part II) of Becker, published by the NACA [1].
2 Local buckling of stiffeners

2.1 Equal flanged elements
The simplest analysis pertains to equal flanged elements, such as corners, T or cruciform
sections in which all flanges have the same width and thickness (figure D5–2). In that case, the
member is divided into individual flanges which are all equal. If treated as individual plates, all
flanges will therefore buckle under the same load and no flange can provide restraint to its
neighbour (they buckle simultaneously). Hence, the boundary condition between adjacent flanges
may be assumed to be simply supported.
simply supported edges
angle
 T-section
cruciform
Figure D5–2 Local buckling analysis of equal flange elements
It has been shown previously that the compression buckling coefficient for simply supported
long plates has a value of 0.43. This may be used in equation (D5.1) to determine the local buckling
stress of equal flanged members.
D5.3
Figure D5–3 Buckling coefficient Figure D5–4 Buckling coefficient

for I-shaped stiffeners [1] for Z-shaped and channel-shaped stiffeners [1]
2.2 Elements with webs

In the case of members with webs and flanges the preceding analysis cannot be used
because all individual elements do not have the same boundary conditions. Becker [1] recommends
the method of Lundquist, Stowell and Schuette [2] and provides charts prepared by Kroll, Fisher
and Heimerl [3] and Van Der Maas [4]. In each case, the critical stress is given by equation (D5.1),
with t and b depending on the geometry. The charts cover:
• I-sections (figure D5–3), with tw and bw used in the stress equation;
• Z- and channel sections (figure D5–4), with tw and bw used in the stress equation;
• Rectangular-tube sections (figure D5–5), with th and h used in the stress equation;
• Top-hat sections (figure D5–6) with t and bT used in the stress equation.
These critical stresses are given for the elastic buckling range. If a plasticity correction is
needed, Becker [1] suggests to use plasticity reduction factors derived using a clamped flange
boundary condition, which seems conservative.
3 Buckling of stiffened plates

3.1 Integral composite web-flange shapes
When the stiffeners and skin are designed to buckle at approximately the same stress level,
separating buckling analysis into skin and stiffener is not accurate, because of the interaction of the
stiffeners rigidity with the plate rigidity.
D5.4
Figure D5–6 Buckling coefficient for hat-shaped stiffeners [1] — Note that t  tf  tw  tT
Figure D5–7 Compressive buckling coefficient

Figure D5–5 Buckling coefficient for plates with plate stiffeners [1]
for rectangular-tube stiffeners [1]
In the case of idealised integral composite web-flange shapes, Becker [1] reported the results
of Boughan and Baab [5] and Gallaher and Boughan [6]. Figure D5–7 shows a chart for analysing
plates with plate stiffeners, figure D5–8 shows charts for Z-shaped stiffeners and figure D5–9
provides data for T-shaped stiffeners. In all cases, the dimensional parameters to be used in
equation (D5.1) are ts and bs.
We should note that the analysis carried out by these authors assumes monolithic
construction, which is applicable to integrally-machined stiffened skin panels. In the case of panels
D5.5
assembled from skin and extruded or formed members, the theory remains valid as long as the panel
does not develop an inter-rivet buckling instability, which will be discussed in the next section.
(a) tw/ts  0.50 and 0.79 (a) tw/ts  0.63 and 1.00
Figure D5–8 Compressive buckling coefficient for plates with Z-shaped stiffeners [1]
(a) tw/tf  1.0; bf/tf > 10; bw/bs > 0.25 (b) tw/tf  0.7; bf/tf > 10; bw/bs > 0.25
Figure D5–9 Compressive buckling coefficient for plates with T-shaped stiffeners [1]
D5.6
4 Buckling instabilities of stiffened panels

4.1 Introduction
Not all compression panels are machined out of a solid block of metal; this is costly and
sometimes not efficient for practical reasons (one of which might be the size of the elements). It
may be more cost-efficient to assemble the panels by riveting or spot welding. If this is the case, the
panel may experience two other modes of instability, the so-called inter-rivet buckling (quoted by
Rivello [7] as inter-fastener buckling since not all assemblies are riveted) and sheet wrinkling.
An assembled stiffened panel may thus buckle under monolithic crippling, inter-rivet
buckling or sheet wrinkling; the failure mode corresponds to the lowest crippling stress.
Figure D5–10 Schematic representation Figure D5–11 Schematic representation

of the inter-rivet buckling mode of the sheet wrinkling mode
4.2 Inter-rivet buckling

4.2.1 Buckling stress
Inter-rivet buckling is the buckling of the skin in wide-column mode between the rivets (or
fasteners) that hold it to the stiffeners. This buckling mode does not deform the flange to which the
skin is attached [8]. In that mode, the wavelength of the skin is equal to the inter-rivet spacing p,
called pitch. Figure D5–10 shows a sketch of the inter-rivet buckling mode of a band of skin riveted
to a stiffener. It was observed soon after all metal airframe construction was introduced. The
problem was first analysed by Howland [9]. The results of Howland are expressed in [10] and other
sources using the classical buckling stress equation:
2 2
 |π E  |ts 
σi  η η C 2   (D5.2)
12 (1  ν )  p 
The stress σi is the inter-rivet buckling stress, η is the plasticity reduction factor and  η the
cladding reduction factor (for clad plates), determined from the corresponding flat plate cases. The
thickness to be considered is obviously the skin thickness and the characteristic dimension is the
pitch. The particular constant C for inter-rivet buckling is called the end fixity coefficient. It depends
on the type of rivet or fastening used. Typical values are given in table D5–1.
D5.7
Type of fixation C
Flat-head rivet 4
Brazier-head rivet 3
Machine-countersunk rivet 1
Dimpled-countersunk rivet 1
Spot welds 3.5
Table D5–1 Values of end fixity coefficient for inter-rivet buckling [10, 7]
4.3 Sheet wrinkling

Sheet wrinkling occurs for more closely-spaced rivets than inter-rivet buckling. The skin
and the stiffener flange buckle together, as schematically shown on figure D5–11. The wrinking
wavelength takes several rivets. This linked deformation of the flange and the skin creates tensile
loads in the rivets. Semonian and Peterson [11] have studied an idealised model. Specific
parameters are bw , which is the distance from the flange to the support provided by the unattached
flange, and the effective fastener offset f, which is the distance from the stiffener web to the line
along which the fasteners clamp the flange to the skin. It is an empirical distance determined from
figure D5–12.
The wrinkling stress is computed from:
2 2
|π E  |ts 
σwr  η 
η k 2   (D5.3)
12 (1  ν )  bs 
The buckling coefficient k is obtained from figure D5–13. If needed, plasticity and cladding
reduction factors can be included – the configuration to be used is that of a long, simply supported
flat plate.
Figure D5–12 Determination of effective fastener offset [10] Figure D5–13 Buckling coefficient for sheet wrinkling [10]
D5.8
[1] Becker, H. Handbook of Structural Stability, part II: buckling of composite elements. NACA
technical note 3782, July 1957.
[2] Lundquist, E.E., Stowell, E.Z. and Schuette, E.H. Principles of moment distribution applied to
the stability of structures composed of bars and plates. NACA wartime report WR L-326,
1943.
[3] Kroll, W.D., Fisher, G.P. and Heimerl, G.J. Charts for calculation of the critical stress for local
instability of columns with I-, Z-, channel and rectangular-tube section. NACA wartime report
WR L-429, 1943.
[4] Van Der Maas, Ch.J. Charts for the calculations of the critical compressive stress for local
instability of columns with hat sections. Journal of the Aeronautical Sciences, 21(6), June 1954,
pp. 399-403.
[5] Boughan, R.B. & Baab, G.W. Charts for calculation of the critical compressive strength of
local instability of idealized web-T-stiffened panels. NACA wartime report WR L-204, 1944.
[6] Gallaher, G.L. & Boughan, R.B. A method of calculating the compressive strength of Z-
stiffened panels that develop local instability. NACA technical note 1482, 1947.
1973.
[9] Howland, W.L. Effect of rivet spacing on stiffened thin sheet in compression. Journal of the
Aeronautical Sciences, 3(12), October 1936, pp. 434-439.
[10] Gerard, G. Handbook of structural stability, part V – Compressive strength of flat stiffened
panels. NACA technical note 3785, August 1957.
[11] Semonian, J.W. & Peterson, J.P. An analysis of the stability of short sheet-stringer panels with
special reference to the influence of riveted connection between sheet and stringer. NACA
Technical note 3431, 1955.
D5.9
Section D6:
Post-buckling behaviour,
crippling and failure
1 Post-buckling behaviour in columns and plates

1.1 Introduction
The wartime period has seen the publication of many studies regarding the buckling
behaviour of semi-monocoque structural elements. It was found out that the buckling loads did not
necessarily coincide with the failure loads. In contrast with a buckled beam, a buckled plate may
still carry loads larger than the buckling loads, because the post-buckled shape is stable. It takes a
greater load to cause plastic failure of the buckled plate. This load and its corresponding average
stress are sometimes referred to as the crippling load and stress (but also as the crushing or local-
failure load and stress). The same is true of thin-walled stiffening members used in structural
design. Thin-walled columns also exhibit a post-buckling load-carrying capability: although the
flange elements suffer from secondary buckling, the corners constitute stabilizing zones where
significant compressive stresses can still be balanced. It takes a greater load (again called the crip-
pling load) to obtain the failure of such a column. Of course, this crippling load is smaller than the
column yield load (described in sections D1 and D2) which would exist if the section was stable.
Hence, a post-buckled structure can be capable of carrying some greater load, provided the
crippling load has not been reached. Thus, determination of the crippling loads and stresses is of
interest to the designer. This is especially true in aerospace, where mass is critical. Certification
allows distinguishing between limit and ultimate loads. Hence, elastic buckling with stable
deformation (such as in plates) could be allowed at the limit load. The design at the ultimate load
may be such that elements are brought to the crippling loads. Furthermore, the need for redundancy
(fail-safe structures) implies that some knowledge of the load-carrying capacity of the structure is
required in case one element has to stand when the other element is supposed to have failed.
1.2 Schematic difference between plates and columns

Gerard [1] has shown summary figures presenting the post-buckling differences between
columns and plates (see figure D6–1). The stress ratio is plotted versus the ratio of wave depth over
thickness, which represents the buckling deformation occurring in the structural member.
A perfect elastic column has an indeterminate deformation once the buckling stress has been
reached. Thence, it fails when plasticity effects lead to structural collapse. The curve is essentially a
horizontal line. Real-life structures will almost always feature an initial imperfection shown here as
a small wave depth existing at zero stress. However, as the stresses increase, this initial
D6.1
Post-buckling behaviour, crippling and failure
imperfection remains essentially constant up to stresses close to the buckling stresses, where the
member deformation is shown to be essentially identical to the perfect member. The impact of such
initial imperfection is thus negligible on the computation of the critical load.
The graph shows that the buckling and
failure stresses in a column are essentially iden- flat plate
tical. Even if buckling occurs in the plastic (elastic buckling)
σ
range, the conclusion remains the same. The σcr flat plate
representative curve dips down, indicating an (plastic buckling)
stress over buckling stress

unstable behaviour.
In contrast, a flat plate buckling in the perfect column
elastic range features a totally different beha- columns with (elastic buckling)
initial imperfection
viour: the post-buckling deformation is stable,
since it requires an increase in stress to obtain a flat plates with
initial imperfection
further increase in deformation. It is thus
perfect column
possible to increase the load past the critical plasticity effects
(plastic buckling)
schematic point of failure
load, until the plate fails and yields plastically.
The same behaviour exists for a plate which wave depth over thickness w
t
buckles in the plastic range; of course the stress Figure D6–1 Post-buckling behaviour of
increase is lower in that case since the buckling columns and plates (after [1])
stress is closer to the yield stress.
1.3 Post-buckling stresses in flat plates

Approximate post-buckling analyses of
flat plates have been carried out by some
authors. This is a complex endeavour, since
nonlinear equations must be used to include the
large deflections of the plate. Plasticity need not
necessarily be included, provided that buckling
occurs in the elastic range.
Coan [2] has shown typical stress distri-
butions in post-buckled plates. The figure is
given here as figure D6–2. Compared with free
lateral edges (b), it is clearly seen in case (a) that
the lateral restraint causes lateral tensile (hoop)
stresses in the plane of the plate; the curved plate
acts thus as a shell and can carry membrane (a) (b)
loads after buckling. This leads to stable post-
buckling configurations with failure loads Figure D6–2 Post-buckling stress distributions in plates
under compression loads
largely in excess of buckling loads, explaining (a) straight unloaded lateral edges
the behaviour shown by figure D6–1. This effect (b) stress-free unloaded lateral edges
is less important in plastic buckling.
D6.2
2 Effective skin widths

2.1 Effective skin width of plates
Any post-buckling stress analysis requires taking
σe σe
into account the actual stresses such as shown on figure
D6–2. While this is possible in accurate theoretical
works it proves impractical to downright impossible in
routine sizing computations. Hence, the concept of
be be
effective plate width has been developed for post-
buckling analyses.
The effective width concept [3-6] simply
assumes that only part of the width will still be able to b
carry out loads in the post-buckled regime. Figure D6–3
shows the concept graphically: the load is carried
through constant stresses σe across two effecttive widths σe σe
be close to the edges rather than through the real stresses σx
σx across the full width b. The stress σe corresponds to Figure D6–3 Effective width concept
the real stress level on the unloaded edges.
We can therefore write:
|1 ⌠ b |
be  σxx dy (D6.1)
2 σe 
⌡ |
0
For small post-buckling loads (σe/σcr < 3), Gerard [1] provides the following expression for
the effective width:
|2 be |σcr  |σe 
 1  β  β  (D6.2)
b σe  σcr 
where β is a coefficient given in table D6–1 and σcr is the critical (buckling) load. The value of β is
influenced by the lateral restraint provided by ribs. In table D6–1, β is a function of (Ar/at) with Ar
the rib area, a the rib spacing and t the skin thickness (refer to section D4 for more information).
Ar/at Unloaded edges held straight Unloaded edges stress-free
0 0.500 0.408
0.25 0.548 0.458
0.5 0.580 0.494
1 0.621 0.540
2 0.665 0.590
4 0.696 0.613
∞ 0.746 0.684
Table D6–1 Values of β for the effective width of long simply supported flat plates [1]
Ar: rib area — a: plate length along unloaded edge — t: plate thickness
D6.3
For long plates at larger post-buckling loads Gerard [1] and Rivello [6] quote the Koiter
equation:
0.4 0.8 1.2
|2 be  |σ   |σ   |σ 
 1.2  cr   0.65  cr   0.45  cr  (D6.3)
b  σe   σe   σe 
The effective width for a simply supported square plate is given by:
0.5 1.2
|2 be  |σ   |σ 
 0.19  0.81  cr   0.45  cr  (D6.4)
b  σe   σe 
Both equations assume that the unloaded edges are held straight but are free to move laterally. This
is identical to setting the rib area to zero (Ar/at  0).
2.2 Effective skin width of stiffened plates

In the case of stiffened plates (also referred to as compression panels), stiffeners provide an
additional support against buckling. Figure D6–4 now presents the situation. The skin plate in
between stiffeners cannot carry more than the critical buckling stress σcr but the skin attached to the
stiffeners can carry the same stress σst as the stiffener. The true distribution (bottom of drawing) is
replaced by an equivalent distribution (top of drawing) with constant stress at stiffener level acting
on an equivalent width w (corresponding to 2 be of the flat plate).
The derivation of equations giving w seems to have originated from von Kármán and
Sechler [3]. Assuming that the boundary restraints provided by the stiffeners corresponded to
simple support, they derived w from the plate buckling equation by setting w as the plate width and
σst as the stress. This stress can be either the yield stress for the stiffener or, if plastic damage is allo-
wed (which seems sensible since the skin has buckled anyway), the crippling stress of the stiffener.
σst
w w
σcr
σst
Figure D6–4 Effective width of a stiffened plate
D6.4
The generic equation is thus given as [3] :

E
wCt (D6.5)
σst
The values of C are determined as follows:
• Flanges with a single line of rivets: C  1.90 (or 1.70 if the stiffener is “light”). If the stiffener
has two flanges attached to the skin, count 2 w under the stiffener.
• Flanges with two lines of rivets: C  2.52 (again, double w per stiffener if two flanges are
attached to the skin). The crippling stress in the stiffener needs to be computed using an
equivalent thickness of ¾ the sum of the skin thickness and flange thickness.
• Integrally-machined panels: C  1.90 but the thickness t is taken as the average of the thickness
of the skin alone and the thickness of the panel in its stiffened part.
• Sheet with a free edge: C  0.62.
Niu [5] provides another criterion, based on (b/t). The condition C  1.90 is acceptable up to
40. This corresponds to heavy skins, capable of twisting the stiffener; the condition C  2.52 is
acceptable for 110 and beyond, corresponding to thin skins which are practically cantilevered with
respect to the stiffener. An intermediate value may be used in between.
If the two materials are different, the condition of equal strain implies different stress levels,
because the stress-strain curves of the skin and stiffeners will not superimpose. This effect is more
markedly seen in the plastic range and modifies the above results. Bruhn [3] proposes to correct
equation (D6.5) by multiplying by the stress ratio σsh/σst in which σsh is the stress that would exist in
the skin if it was strained at the strain level existing in the stiffener:
|σsh E
wCt (D6.6)
σst σst
3 Crippling failure of structural members

3.1 Gerard’s plate failure theory
The theoretical analysis of plate failure is even more difficult than post-buckling analyses
since most of the plate is loaded in the inelastic range. Hence, semi-empirical methods have been
preferred to theoretical approaches. Gerard [1, 7] has published an equation, based on the one hand
on the theoretical work of Mayers & Budiansky [8] and on the other hand on many different tests on
plates of different materials and geometric proportions by various authors – references are given in
[1]. Experimental evidence has shown that plates tend to fail when the effective stress equals the
compressive yield stress of the material, σcy. Therefore Gerard introduced:


n
|σf  |σcy 
α  σcr  α1/n σcy

σcr  σcr 
(D6.7)

 |σf
σcr
1 σcr  α1/n σcy
D6.5
In these equations,  σf is the average failure or crippling stress of the plate. The values of α
and n must be determined from experimental data. Table D6–2 provides the values recommended
by Gerard. The equation shows the existence of a “cut-off stress”. For critical stresses above the
cut-off stress, the crippling stress equals the cut-off stress.
Condition α n
Theory (straight unloaded edges) 0.78 0.80

Test data (edges free to warp) 0.80 0.58
Test data (three-bay plates) 0.80 0.65
Table D6–2 Values of α and n for the crippling of flat plates [1]
A more practical alternative form can be obtained, which uses the buckling coefficient and
thickness-to-width ratio of the plate as main parameters. First, equation (D6.7) is rewritten:


1 n
|σf  |σcr 
α  σcr  α1/n σcy

σcy  σcy 
(D6.8)

 |σf |σcr

σcy σcy
σcr  α1/n σcy
The critical buckling stress is replaced by its value:

2 2
|π E t
σcr  kc 2   (D6.9)
12 (1  ν )  b 
This provides the relationship:



m
|σf  |t |E  |t |E α1/n
β  
 b σcy  b γ

σcy σcy
(D6.10)
 
2
|E 
 |t |t |E 1/n
|σf α
β  
σcy  b σcy  b σcy γ
with
m  2 (1  n) (D6.11)
2
|kc π
γ (D6.12)
12 (1  ν2)
β  α γm/2 (D6.13)
These relations predict crippling stresses
with an uncertainty of maximum 10%, as shown
by a graphical rendering with experimental data
points on figure D6–5.
Figure D6–5 Comparison of prediction of plate crippling

stresses with experimental data [1]
D6.6
3.2 Crippling of thin-walled columns

Gerard’s theory for the crippling stress of columns has been widely reported in the literature
(Bruhn, Rivello, Megson, Curtis) [3,4,6,9]. It consists in using a form similar to equation (D6.10)
for plate crippling. Gerard’s equation is written [10]:
 n m
|g t2  |E  
|σf
β   (
σf  κ σcy) (D6.14)
σcy  A  σcy  
The ratio (t/b) has been replaced by the ratio (t2/A) in which t is the wall thickness and A the cross-
sectional area. The values of β, m and n must be determined from experimental data. As was the
case with plates, a cut-off stress also exists, below which equation (D6.14) applies but above which
the crippling stress is equal to the cut-off stress, which is given as κ σcy. Table D6–3 summarizes the
recommended values of the coefficients. If the material is clad, replace E by  η E in (D6.14).
Finally, g is the Gerard coefficient. It is equal to the number of cuts necessary to break the
column in single flange or web elements plus the number of individual flanges which make up the
column after the cut. Figure D6–6 shows how to determine g for many typical basic shapes. The top
three shapes are termed “angle-type” and the three bottom shapes are termed “T-type”.
angle plate
tube
0 cut 1 cut 4 cuts

2 flanges 2 flanges 8 flanges
g2 g3 g  12
T section cruciform H section
0 cut 0 cut 1 cut

3 flanges 4 flanges 6 flanges
g3 g4 g7
Figure D6–6 Determination of Gerard's coefficient for simple column cross-sections [10]
Condition β n m κ
Angle-type elements, angles 0.558 1/2 0.85 0.7

Multiple corner and hollow tube sections 0.558 1/2 0.85 0.8
Plates, edges free to warp 0.558 1/2 0.85 1.0
Plates, straight unloaded edges 0.67 1/2 0.40 1.0
T-type elements, cruciforms, H sections 0.67 1/2 0.40 0.8
Two-corner sections (Z, J, channel) 3.2 1/3 0.75 0.9
Table D6–3 Recommended values for the parameters of Gerard's equation [10]
D6.7
In case of variable wall thickness of the column cross-section, a mean wall thickness 
t

should be used in equation (D6.14), with t given by:
 |Σ bi ti
t  (D6.15)
Σ bi
The sums are carried over the various single-thickness elements which make up the section.
The reason for an increased crippling stress in columns subject to secondary instabilities is
the following: the stress distribution after the flanges have buckled is much alike to that of a
stiffened plate; the stiffening members in this case are the corners of the column, because they
provide mutual restraint. Thus, it is possible to increase compressive stresses in the corners beyond
the buckling stresses, up to crippling, i.e. the failure of the material in the corners [11].
3.3 Crippling of monolithic stiffened panels

Monolithic stiffened panels can fail under compression loads as well. Gerard has applied its
method for columns to the case of stiffened panels. Equation (D6.14) is therefore used for stiffened
panels, but the parameter g must be re-evaluated. One repeating element (skin and one stiffener) is
isolated from the panel and analysed using the g counting method shown on figure D6–6. The skin
and stiffener must both be cut into single flange elements which are then counted. The value of g is
the sum of all cuts and all flanges. The cuts made to isolate a single repeating element are not
counted in the analysis but the skin must be cut into single-supported elements, which means two
skin cuts for stiffeners which have two fixation points on the skin.
Z stiffener Y stiffener
Skin: 1 cut and 2 flanges

Stiffener: 1 cut and 4 flanges
g358
Skin: 2 cuts and 4 flanges
Stiffener: 3 cuts and 10 flanges
Top hat stiffener g  6  13  19
Skin: 2 cuts and 4 flanges

Stiffener: 3 cuts and 8 flanges
g  6  11  17
Figure D6–6 Determination of Gerard's coefficient for stiffened panels [10]
The skin-stiffener combination requires several modifications to the basic approach used for
columns. First of all, let tw be the stiffener thickness and ts the skin thickness. They may be
different, so the ratio (tw/ts) enters the equation as a parameter.
D6.8
Second, if the stiffener is made out of parts with different thicknesses, a mean wall thickness
is computed using (D6.15):
 |Σ bi twi
tw  (D6.16)
Σ bi
and used in place of tw.
Third, because of the difference in thicknesses between skin and stiffener, the yield stress
used in the Gerard equation is an averaged panel yield stress computed from:
 |σcy  σcyw (tw /ts  1)

σcy  s (D6.17)
tw /ts
Finally, the general form of the Gerard equation in that case is:
 |g t t
m
|σf |E 
β w s (D6.18)
  A 
σ cy  σcy 
As we see, parameter n has a value of 0.5 in this case. Table D6–4 provides the
recommended values for the parameters β and m as well as for the value of cut-off stress that may
be used as a fraction κ of the yield stress.
Condition tw / t s β m κ
Formed Z stiffeners — 0.558 0.85 1.0

Formed top hat stiffeners 1.25 0.591 0.85 0.8
1.00 0.561 0.85 0.8
0.63 0.499 0.85 0.8
0.39 0.483 0.85 0.8
Extruded Y stiffeners 1.16 0.562 0.85
0.732 0.505 0.85
0.464 0.478 0.85
General angle-type stiffeners — 0.558 0.85
General T-type stiffeners — 0.670 0.40 0.75
Table D6–4 Recommended values for the parameters of Gerard's equation for stiffened panels [10]
3.4 Crippling of assembled stiffened panels

3.4.1 Crippling stress for inter-rivet failure
We have already seen before what happens when a stiffened panel buckles. The buckled
skin cannot withstand higher loads than the buckling loads, because it has buckled in a column
mode and not in a plate mode. It will, however, continue to carry the buckling load. Load
redistribution occurs for higher loads, with the stiffener taking more load, until the stiffener itself
also buckles as a column, at which point the crippling stress is reached since the structure will not
take higher loads.
D6.9
We can compute the effective width of the skin w at the buckling load (σcr  σir). The load
that the skin carries after buckling is therefore limited to w σir t. We can also compute the crippling
 and the corresponding load A σ
stress for a single stiffener acting alone; let this be σ f st st f st
The average crippling stress in inter-rivet buckling corresponds to the stress achieved at the
total crippling of the panel, so it is given by:

 |w σir ts  Ast σf st
σfi  (D6.19)
w ts  Ast
3.4.2 Crippling stress for sheet wrinkling

Wrinkling failure often occurs at stresses that are close to the wrinkling buckling stresses,
because the lateral forces on the stringers become very large and force their failure. However, it is
still possible for sturdy stiffeners to carry additional loads before failure of the riveted panel.
Therefore, a variety of failure cases must be considered.
In the model of Semonian and Peterson [12], wrinkling failure of the skin may be addressed
by an equation similar to that for the buckling stress
2 2
 η |π E  |ts 
σ w η kw 2   (D6.20)
12 (1  ν )  bs 
The failure coefficient kw is obtained from figure

D6–7. It can be seen that indeed, the values of k
for buckling and failure are quite similar (refer to
section D5), except for small values of f/bw . We
now consider the following crippling stresses:
•  σ : crippling stress of stiffener alone
f st

• σf : crippling stress of the monolithic panel

• σ : skin wrinkling stress
w
The fate of the panel depends on the relative

magnitude of the four crippling stresses (the
three above plus the skin wrinkling stress). By
comparing the stresses, a value of the crippling
stress of the riveted panel, 
σf r , is obtained. Figure D6–7 Failure coefficient for sheet wrinkling [10]
3.4.2.1 Stable stringer

This corresponds to  σf st  
σw . The skin fails before the stiffener, so the panel fails in skin
 
wrinkling mode. Thus, we have in that case σ fr σw . If the stiffener is unusually sturdy, it could be
able to carry the load, in which case the panel failure is given by the following stress:

 |Ast σf st
σfr  (D6.21)
bs ts  Ast
The failure stress should be considered as the larger of those two values.
D6.10
3.4.2.2 Unstable stringer

This corresponds to σf st  
σw . The assumption of Semonian and Peterson is that the stringer
will not carry any additional load above  σf st . The skin will still carry an increased load until the
value 
σw. Thus, the average crippling stress for the panel is given by:
 
 |σw bs ts  Ast σf st
σfr  (D6.22)
bs ts  Ast
This relation gives satisfactory results over the practical range of applications.
3.4.3 Alternate failure equation

The relations and coefficients provided by Semonian and Peterson pertain to aluminium
alloys. Gerard found that this theory might be optimistic by as much as 230% when used on other
alloys, including other aluminium alloys not considered in the original tests [10]. Consequently he
proposed an alternative equation in the usual Gerard form:
 4/3 1/6 
|σfr  |tw   |tw  |ts |η E
 17.9     (D6.23)
  f   bw  bs 
σ
cy σcy
4 Column curves in case of crippling

4.1 Introduction
When the member is subject to crippling, it is quite difficult to determine the allowed
maximal load on the member in the transition range of slenderness ratio (L/ρ). Indeed, for very
short columns (L/ρ  20) it is extremely likely that failure by crippling will be observed, while for
very long columns the classical Euler parabola will be retrieved. The relations obtained above allow
us to compute the failure loads of the different failure modes and predict, for a given load case,
which failure mode is likely to occur and at which stress. However, since many different load cases
should be considered in design, it is desirable to obtain a representation of the full range of
slenderness ratio for a given structural member.
4.2 Use of the Johnson parabola

A classical approach to that problem [3] is simply to use the Johnson parabola, replacing the
column yield stress (for stable columns) by the column crippling stress (for unstable columns). In
that case, the equation is:
2
σ  σ cs 
|σcs
4 π2 E
(L/ρ)2
 |2 E
(D6.24)
(L/ρ) cr  π
σcs
D6.11
This relation has been derived in section D2. In this form, σcs is the crippling stress
(supposed to exist at zero slenderness ratio).
4.3 Use of the Gerard parabola

Gerard [10] provides the following alternate parabolic equation:
2
|σ  |σ  |σ (L/ρ)
 1  1  cr cr 2 (D6.25)
σcs  σcs  π E
In this expression, σcs is the crippling stress and σcr is the buckling stress for the column
cross-section. This equation applies for σ  σcr. It also supposes that the buckling stress is below the
proportional limit (if not, Bruhn [3] suggests using the proportional stress rather than the buckling
stress in the formula). It connects to the Euler curve at σ  σcr.
Although Gerard proposes the preceding relation, he also notes that most crippling stress
data was obtained by tests on specimens with L/ρ  20. The equation is thus corrected to terminate
at that slenderness ratio, with a constant stress assumed for lower ratios:
|σ  |σ (L/ρ)  σ20  σEu  2

2
|σ 
 1  1  cr cr 2 (D6.26)
σcs  σcs  π E  σ20  σcr 

In this relation, σEu is the stress given by the Euler formula.
[1] Gerard, G. Handbook of structural stability, part IV – Failure of plates and composite elements.
NACA technical note 3784, August 1957.
[2] Coan, J.M. Large deflection theory for plates with small initial curvature loaded in edge
compression. Journal of Applied Mechanics, 18(2), June 1951, pp. 143-151.
1973.
[7] Gerard, G. The crippling stress of compression elements. Journal of the Aeronautical Sciences,
25(1), January 1958, pp. 37-52.
[8] Mayers, J. & Budiansky, B. Analysis of behaviour of simply supported flat plates compressed
beyond the buckling load into the plastic range. NACA technical note 3368, 1955.
1999.
[10] Gerard, G. Handbook of structural stability, part V – Compressive strength of flat stiffened
panels. NACA technical note 3785, August 1957.
D6.12
[11] Niu, M.C.Y. Airframe structural design, 2nd edition. Hong Kong Conmilit Press Ltd., Hong
Kong, 1999 (republished with minor corrections, 2002).
[12] Semonian, J.W. & Peterson, J.P. An analysis of the stability of short sheet-stringer panels with
special reference to the influence of riveted connection between sheet and stringer. NACA
technical note 3431, 1955.
D6.13

Stability of Structural Elements: Book D

Uploaded by

Copyright:

Available Formats

Stability of Structural Elements: Book D

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stability of Structural Elements: Book D

Uploaded by

Copyright:

Available Formats

Book D

1 Euler buckling theory

structural instability existed in beams subject to Px Px

compressive loads act axially and there is full

1.1.2 Computation of the Euler buckling load

Thus, we obtain the following differential equation:

This is an homogeneous differential equation, which admits a number of solutions. First of

1.1.3 Critical stress

1.2 Other cases of column buckling

P P Free-fixed 0.25 2.000

1.3 Limit of the Euler formula

with e.g. an aluminium alloy 2023-T3. The

being of hyperbolic nature, we have a problem 60

applying it to short columns, since the critical 50

stress goes to infinity as the slenderness ratio 40

goes to zero. Since the above analysis is made 30

the Euler formula to long, slender columns (ratio 0

above 65 in this case). This range (called the Slenderness ratio

2 Buckling of non-uniform columns

2.2 The Martin method

2.2.2 Application to symmetric columns

2.2.3 Application to asymmetric columns

An initial value of K can again be obtained by assuming a deflection and computing an

2.2.4 Numerical stability

2.2.5 Application to columns of different materials

Figure D1–7 Critical load Figure D1–8 Critical load

2.3 Non-uniform column charts

1 Inelastic buckling theory

uniform compression uniform compression σ

1.2 Tangent modulus and reduced modulus formulas

1.2.2 Reduced modulus formula

1.2.3 Comparison of results

1.2.4 Computation of the critical stress

Thus, it is easier to draw a column 50

strength curve from tabulated material data by 45

ponds to a given critical stress (and modulus). Yield stress

Critical stress [ksi]

Bruhn [8] duly points out that if the 5

points lying above the compressive yield stress 0

and the column yield stress are to be used in Slenderness ratio

2 Practical computations of inelastic buckling cases

2.1 Non-dimensional column curve

The value B is a function of stress and of the Ramberg-Osgood parameters:

The inelastic column strength curves of 60

Critical stress [ksi]

equation of Cozzone and Melcon. It can be seen Tangent modulus

Curtis data and are σ0.7  36 665 psi and n  8. 0

2.2 Use of tangent modulus charts

2.3 Johnson-Euler curves

Then, from (D2.22) we compute the critical slenderness ratio:

Figure D2–7 shows a part of the same 60

column strength curve as figure D2–3, with the 50

Johnson-Euler equations plotted for both n  1 40

ratios. The straight line thus provides us with a 0

2.4 Non-uniform columns

2 Buckling in the elastic range

2.1.2 The buckling load eigenproblem

  λ  V0  V0   2   (D3.2)

This is indeed an eigenproblem formulation.