BEHAVIOUR OF IMPERFECT COLUMNS

[Reading – Megson, ch 18]

So far the cases we have considered have been rather idealised, overlooking a number of
important features or imperfections that will be present in most practical columns, such as
• initial lack of straightness
• material that is not perfectly linearly elastic
• non-slender cases (i.e. short, thick columns)

INITIALLY CURVED COLUMN

natural shape
P before loading P
z

vo

after loading

Note that the initial deflection (vo) and associated curvature corresponds to zero bm
throughout - it is only the extra deflection (v) and associated curvature that causes bm.
For equilibrium: M(z) = −EIv ′′ = P(v + v 0 )

EIv ′′ + Pv = −Pv0

To proceed we need to know (or assume) v 0 (z) . Could assume a Fourier series,
∞
nπz
v 0 (z) = ∑a
0
n sin
L
, (as Megson does) and get a more general result, but here we assume

πz
v 0 = a sin (the natural buckled shape of a pin-ended column). By doing this we make life as
L
tough as possible for the column.
πz
hence EIv ′′ + Pv = −Pa sin 1
L
P πz
subst = α2 v ′′ + α 2 v = −α 2 a sin
EI L
CF: v = A sin αz + B cos αz
πz
PI: try v = b sin and substitute in eq. 1
L
α2 a
This leads to b= 2
, and the complete solution,
⎛ π ⎞ − α2
⎜ ⎟
⎝L⎠
πz
α 2 a sin
CF + PI is thus v(z) = A sin αz + B cos αz + L
2
⎛ π ⎞ − α2
⎜ ⎟
⎝L⎠
ColumnNotes.doc p1 jwb 01/09/04
Applying boundary conditions: v(0) = 0 ⇒ B = 0
v(L) = 0 ⇒ 0 = A sin αL

If A ≠ 0, αL = nπ , giving solutions at a set of discrete P values only (the critical loads of the
π 2 EI 4π 2 EI
straight column), Pcr = , , etc.
L2 L2
π 2 EI
However, our interest is in what happens for values of P < , so we choose the other
L2
possibility and let A = 0 , so that the solution becomes
πz πz
α 2 a sin a sin P
v(z) = L = L
2 2
⎛π⎞ 1 ⎛π⎞
⎜ ⎟ −1
2
⎜ ⎟ −α
⎝L⎠ α2 ⎝ L ⎠
v0
=
Pcr
−1 a δ
P
v0
or total deflection, v0 + v = .
P
1−
Pcr
a
A plot of mid-span deflection δ = v(L / 2) = ,
Pcr
−1
P P

shows deflection increasing steadily with P and then increasing rapidly as

P → Pcr . The initial mid-span deflection, a, is “magnified” by the factor Pcr
1
.
Pcr
−1
P
SOUTHWELL PLOT

δ
a Pcr
Rewriting δ = as δ − δ = a,
Pcr P δ/P
−1
P slope=1/Pcr

δ ⎛ 1 ⎞ ⎛ a ⎞
finally = ⎜⎜ ⎟⎟δ + ⎜⎜ ⎟⎟ 1
P ⎝ Pcr ⎠ ⎝ Pcr ⎠
Pcr
giving a linear relationship between (δ/P) and δ.
Known as the Southwell Plot, it permits the experimental
a δ
determination of buckling loads from a series of load and
corresponding deflection measurements at loads less than the buckling load. Thus the
buckling load can be determined without actually buckling the column (and probably damaging
it).
The Southwell plot can be used for a wide range of buckling problems, not just columns. The
main requirement is that the measured quantity, δ, should exhibit a first order (major)
change with the primary buckling mode.

ColumnNotes.doc p2 jwb 01/09/04

ECCENTRICALLY LOADED COLUMN
e
The behaviour of an eccentrically loaded column is similar to that of an initially P
curved column.
For equilibrium in a slightly displaced position
EIv ′′ = −M(z) = −P(e + v)

leading to v ′′ + α2 v = − α 2 e .
1 − cos αL
Solving, v = e(cos αz + sin αz − 1)
sin αL P

αL 1 − cos αL αL
Mid-span deflection, δ = e(cos + sin − 1) ,
2 sin αL 2 Pcr

αL
simplifying to δ = e(sec − 1)
2
⎛ π P ⎞
or δ = e⎜ sec − 1⎟ .
⎜ 2 Pcr ⎟
⎝ ⎠
In both cases deflection starts to increase immediately any load is applied. δ
Consequently there is no distinct instant of buckling, merely a rapid increase in rate of
deflection as Pcr is approached. Simultaneously the bm at mid-height will be increasing
proportionately with a likelihood that the material at that location will yield or fail in some
way, initiating complete failure of the column.
CONSEQUENCES OF MATERIAL LIMITATIONS area A P

Considering the case of a pin-ended column again, the critical load is given by
π 2EI
Pcr = σcr
L2
Pcr π 2E(I / A)
Mean stress at Pcr, σ cr = = .
A L2
I
But = r 2 , the radius of gyration.
A
L/r
π 2E
So σ cr = .
(L / r) 2
σ
L/r is known as the “slenderness ratio” and a plot of σcr against L/r will
appear as shown. This suggests that as L/r gets smaller the critical σy
(buckling) stress increases without limit. However, all real materials will
yield or fail as their stress increases.
For a simple elasto-plastic material (such as mild steel) the stress-strain
behaviour is approximately as shown to the right.
strain, ε
Hence, in our diagram showing buckling stress we should place a cut-off at a stress of σy:

ColumnNotes.doc p3 jwb 01/09/04

 σcr

material failure

σy
unstable
stable

L/r
‘long’
‘short’ columns
columns

yielding

Columns with a sufficiently small slenderness ratio (L/r) will fail by ‘squashing’ (yielding),
whereas more slender columns will fail (at least initially) by elastic buckling.
The dividing line between ‘short’ and ‘long’ columns depends on yield stress and elastic
modulus. For example, if σy = 300MPa and E = 200,000MPa,
π2E
σ cr = σ y =
(L / r) 2
π2E
L/r =
σy
= 81
The transition from short (yielding) columns to long (buckling) ones is not sharply defined in
practice. There is a transition zone in which failure involves a mix of buckling and yielding.
For example, a slight tendency to buckle may cause yielding and further deflection. Similarly
buckling may be precipitated by the first hint of yielding on one side of a column.
Actual failure loads (or the corresponding stresses) if plotted will follow a pattern as shown
in the next figure.
The curve followed will depend on factors such as initial imperfection (lack of straightness),
residual stresses resulting from manufacture and construction processes, etc.
Different curves are used for different classes of column and a considerable variety of
these ‘column design’ curves have evolved. We will look at just a few of these.

ColumnNotes.doc p4 jwb 01/09/04

 σfail

Euler

σy
locus of actual failure stress

L/r
‘long’
‘short’ columns
columns transition

RANKINE FORMULA
One of the oldest and simplest column design curves is the ‘interaction’ formula of Rankine:
1 1 1
= + σ
σ σe σy

where σ is the mean stress at failure

σe
σe is the Euler critical stress, π 2 EI / L2
σy
σy is the yield stress.
Rankine’s curve is asymptotic to the Euler and yield σ (Rankine)
stress lines at L / r → ∞ and 0 . It can also be written
1 1 1
in terms of loads: = + L/r
P Pe Py

σy
Rearranging the stress version σ=
1 + (σ y / π 2 E)(L / r) 2

σy
is replaced by a constant ‘a’ which is adjusted to take account of imperfections, giving
π2E
σy
failure stress σ=
1 + a(L / r) 2

and failure load, P = σA

By using Leff in place of L, the formula is applicable to columns with any type of end
restraints:
σy
σ=
1 + a(Leff / r) 2

ColumnNotes.doc p5 jwb 01/09/04

EXAMPLES
P
FAILURE LOAD OF A PROPPED CANTILEVER COLUMN
A 200mm x 200mm x 5mm hollow box section column is 5m long, fixed 5mm
at one end and propped at the other. σy = 300MPa. What axial load will
cause failure if Rankine’s a = 0.9(σ y / π 2 E) ? 200mm

Effective length: Leff = 0.7 L = 3.5m 5m

Section properties: A = 3.9x10 −3 m 2

I = 24.73x10 −6 m 4

r = I / A = 0.0796m

Slenderness ratio Leff / r = 43.95

Rankine ‘a’ a = 0.9(σ y / π 2 E) = 136.78x10 −6

300
Failure stress σ= = 237.3MPa
1 + 136.78x10 −6 × 43.95 2
Failure load P = σA = 237.3 × 3.9x10 −3 = 925kN

Note: Py = Aσ y = 1170kN (squash load)

π 2 EI
PE = = 3984kN (Euler load)
L2eff

STRESS IN AN INITIALLY CURVED TUBULAR COLUMN

A pin-ended steel tube 1.5m long, 25mm outside diameter, 2mm wall thickness, has initial
curvature v 0 = a sin( πz / L) , where a = 5mm. What is the maximum stress due to an axial load
of 3.5kN?
Section properties: I = πr 3 t = π × 11.5 3 × 29556mm 4
A = 2πrt = 144.5mm 2
v0
We had (on p2) v=
P
−1
Pcr
v0
or v + v0 =
P
1−
Pcr

π 2 EI
Need Pcr: Pcr = = 8.38kN
L2
5
At mid-height, total defl’n v + v0 = = 8.58mm
3. 5
1−
8.38
Bending moment M = P(v + v 0 ) = 3500 × .00858 = 30Nm

P My 3500 30 × 12.5 × 1000

Stress: σ= + = + = 63.5MPa
A I 144.5 9556

ColumnNotes.doc p6 jwb 01/09/04

BUCKLING LOAD FROM SOUTHWELL PLOT
A strut is subjected to axial force in a testing machine and the resulting increments in mid-
span lateral deflection measured as follows:
v(mm) 0.23 0.38 0.55 0.75 0.96 1.27
P(kN) 6.85 8.90 9.80 10.54 11.20 11.75
What is the predicted buckling load?
δ δ
The Southwell plot (p2) is a plot of against δ . We have δ and P so just need to calculate :
P P
δ(mm) 0.23 0.38 0.55 0.75 0.96 1.27
P(kN) 6.85 8.90 9.80 10.54 11.20 11.75
δ
.0336 .0427 .0561 .0712 .0857 .1081
P
0.12

0.1

0.08
δ/P

0.06

0.04

0.02

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
δ

Using the two points indicated calculate the inverse of the slope:
1.27 − .55
Pcr = = 13.85kN
.1081 − .0561
δ
(Alternatively one could plot δ against and calculate the slope)
P

NONLINEAR ELASTIC MATERIALS – TANGENT MODULUS THEORY

Many materials exhibit nonlinear elastic behaviour but with no clear yield point.
Engesser showed that for such materials buckling occurs (theoretically) when
π 2 Et I π 2 Et
Pcr = , or σ cr =
L2eff (Leff / r) 2

where Et is the tangent modulus at the critical stress (i.e. the slope of the tangent to the
stress-strain curve).
However, experiments generally revealed a higher buckling load than that given by the
tangent modulus theory.
Shanley, using a more accurate theoretical model showed that a better prediction is given by
the ‘reduced modulus’ or ‘double modulus’ load:

ColumnNotes.doc p7 jwb 01/09/04

 π 2 ER I
PR =
L2eff

4EEt
where ER = is the reduced modulus (for a rectangular cross-section).
( E+ Et )
2

The reason for the different result is that as buckling commences, strain on the concave side
of the column increases, with a corresponding stress change dictated by Et, whereas on the
convex side strain decreases slightly and stress follows the unloading curve governed by E.
Hence the need to use both moduli.
Measured buckling loads tend to lie between a lower bound given by the tangent modulus load
and an upper bound given by the reduced modulus load. (The simple Euler buckling load will of
course be higher than both.) Since the reduced modulus load errs on the unsafe side it is
more common to use the tangent modulus load which is also simpler to calculate.
PERRY-ROBERTSON FORMULA (BASIS OF COLUMN DESIGN TO NZ STEEL DESIGN STANDARD)
The formula underlying the NZ Steel Structures Standard is based on the Perry Robertson
formula which in turn is derived from the expression for the maximum stress in an axially
loaded initially curved column.
natural shape
P before loading P
z

vo

after loading

πz
For an initially curved shape v 0 = a sin ,
L
v0
we had on p2, max. deflection, δ = v0 + v = .
P
1−
Pcr

Pa
Hence max. bending moment M = Pδ =
P
1−
PE

P My max
Stress at mid-span = +
A I
P Paymax
= +
A ⎛ ⎞
⎜ 1 − P ⎟I
⎜ PE ⎟⎠
⎝
I P σay max
subst. r 2 = ,σ= , =σ +
A A ⎛ ⎞ 2
⎜1 − P ⎟r
⎜ PE ⎟
⎝ ⎠
Column assumed to be at its limit when this maximum stress reaches σy – i.e.

ColumnNotes.doc p8 jwb 01/09/04

 σay max
σy = σ +
⎛ ⎞
⎜ 1 − σ ⎟r 2
⎜ σ E ⎟⎠
⎝
ay max
Putting η = in the previous equation gives a quadratic in σ:
r2
σ 2 − σ[σ y + σ E (η + 1)] + σ y σ E = 0

σ y + σ E (η + 1) − (σ y + σ E (η + 1)) 2 − 4σ y σ E
with solution σ=
2
This is the Perry formula.
The corresponding load is given by P = σA.
η depends on initial imperfections. Test by Robertson were used to get suitable values for η.
Initially η was taken as 0.003(L/r) to give the Perry-Robertson formula. More recent tests
have been used to refine the value of η which now varies according to the types of steel
column. For example η = 0.00003(L / r) 2 .

ColumnNotes.doc p9 jwb 01/09/04