Elastic Buckling Behavior of Beams

CE579 - Structural Stability and Design

 ELASTIC BUCKLING OF BEAMS

• Going back to the original three second-order differential

equations:

Therefore,
 z  z
1
E I x v  P v    P x0  M BY   M TY  M BY    M BX   M TX  M BX 
 L  L
 z  z
E I y u   P u     P y0  M BX  (M  M TX  MBX
TY +M BY)    M BY  (M TX +M
M TY BX)
 MBY
2
 L  L
z
E I w    (G KT  K )    u  ( M BX  ( M BX  M TX )  P y0 )
3
L
z v u
 v ( M BY  ( M BY  M TY )  P x0 )  ( M TY  M BY )  ( M TX  M BX )  0
L L L
 ELASTIC BUCKLING OF BEAMS

• Consider the case of a beam subjected to uniaxial bending only:

 because most steel structures have beams in uniaxial bending
 Beams under biaxial bending do not undergo elastic buckling
• P=0; MTY=MBY=0
• The three equations simplify to:

z
E I x v  M BX   M TX  M BX 
1
L
z
2 E I y u    M BX   M TX  M BX  (-f)
L
 z  u
E I w    (G KT  K )    u    M BX  ( M BX  M TX )   ( M TX  M BX )  0
3
 L  L
• Equation (1) is an uncoupled differential equation describing in-plane
bending behavior caused by MTX and MBX
 ELASTIC BUCKLING OF BEAMS

• Equations (2) and (3) are coupled equations in u and f – that describe the
lateral bending and torsional behavior of the beam. In fact they define the
lateral torsional buckling of the beam.
• The beam must satisfy all three equations (1, 2, and 3). Hence, beam in-
plane bending will occur UNTIL the lateral torsional buckling moment is
reached, when it will take over.
• Consider the case of uniform moment (Mo) causing compression in the
top flange. This will mean that
 -MBX = MTX = Mo
 Uniform Moment Case

• For this case, the differential equations (2 and 3) will become:

E I y u    M o  0
E I w    (G KT  K )    u   M o   0
where :
K  Wagner ' s effect due to warping caused by torsion
K    a 2 dA
A

Mo
But ,   y  neglecting higher order terms
Ix
Mo
K   y ( xo  x) 2  ( yo  y ) 2  dA
A
Ix
Mo
K         2 yy0  dA
2 2 2 2
y xo x 2 xx0 y o y
Ix A

M  2 
K  o  xo  y dA   y  x  y  dA  x0  2xy dA  yo  y dA  2 yo  y dA 
2 2 2 2

Ix  A A A A A 
 ELASTIC BUCKLING OF BEAMS

Mo  
K    y  x  y  dA  2 yo I x 
2 2

Ix  A 
 y  x 2  y 2  dA 
    
K  Mo  A  2 yo 
 Ix 
 
    dA
2 2
y x y
 K  M ox  where,  x  A
 2 yo
Ix
 x is a new sec tional property
The beam buckling differential equations become :
(2) E I y u    M o  0
(3) E I w    (G KT  M o  x )    u   M o   0
 ELASTIC BUCKLING OF BEAMS

Mo
Equation (2) gives u    
E Iy
Substituting u  from Equation (2) in (3) gives :
2
M
E I w  iv  (G KT  M o  x )    o   0
E Iy
For doubly symmetric sec tion :  x  0
2
G K M
 iv  T
   2 o   0
E Iw E I y Iw
G KT M o2
Let , 1  and 2  2
E Iw E I y Iw
 iv  1    2  0  becomes the combined d .e. of LTB
 ELASTIC BUCKLING OF BEAMS

Assume solution is of the form   e  z

   4  1  2  2  e z  0
  4  1  2  2  0
1  12  42 12  42  1
  2
, 
2 2
1  12  42 1  12  42
   , i
2 2

 Let ,   1 , and  i 2

Above are the four roots for 

  C1e1z  C2e 1 z  C3ei 2 z  C4e i 2 z
 collecting real and imaginary terms
  G1 cosh(1 z )  G2 sinh(1 z )  G3 sin( 2 z )  G4 cos( 2 z )
 ELASTIC BUCKLING OF BEAMS

• Assume simply supported boundary conditions for the beam:

 (0)   (0)   ( L)   ( L)  0

Solution for  must satisfy all four b.c.
 1 0 0 1   G1 
 12 0 0  22  
G2 
    0
 cosh(1 L) sinh(1 L) sin( 2 L) cos( 2 L)  G3
 
12 cosh(1 L) 12 sinh(1 L)  22 sin( 2 L)  22 cos( 2 L)  
G4 
For buckling coefficient matrix must be sin gular :
 det er min ant of matrix  0
 
 12   22  sinh(1 L)  sinh( 2 L)  0
Of these :
only sinh( 2 L)  0
 2 L  n
 ELASTIC BUCKLING OF BEAMS
n
 2 
L
12  42  1 
 
2 L
2 2
 1  42  1  2
2

L
2
 2 2   2 2   2 2 
 2  1   1  2  21   2 
2

L L  L 
 2    
4 4
2   2 
 2   2  1   2   2E I y   2E Iw 
L  L  Mo    G KT 
L2  L
2

M o2   2 G KT   2 
 2  2  2   
E I y Iw  L E I w   L2 

  2 G KT   2 
Mo  E 2
I y Iw   2 
L E I
 2 
 w  L 
 Uniform Moment Case

• The critical moment for the uniform moment case is given by the
simple equations shown below.
 2
EI y  2 EIw 
M cr 
o
  2  GKT 
L2  L 
M cro  Py  P  ro2
• The AISC code massages these equations into different forms,
which just look different. Fundamentally the equations are the
same. 
 The critical moment for a span with distance Lb between lateral -
torsional braces.
 Py is the column buckling load about the minor axis.
 P is the column buckling load about the torsional z- axis.
 Non-uniform moment

• The only case for which the differential equations can be solved
analytically is the uniform moment.
• For almost all other cases, we will have to resort to numerical
methods to solve the differential equations.
• Of course, you can also solve the uniform moment case using
numerical methods
z
E I x v  M BX   M TX  M BX 
L
z
E I y u    M BX   M TX  M BX  
L
 z  u
  
E I w   (G KT  K )   u   M BX  ( M BX  M TX )   ( M TX  M BX )  0
 L  L
 What numerical method to use

• What we have is a problem where the governing differential equations are

known.
 The solution and some of its derivatives are known at the boundary.
 This is an ordinary differential equation and a boundary value problem.
• We will solve it using the finite difference method.
 The FDM converts the differential equation into algebraic equations.
 Develop an FDM mesh or grid (as it is more correctly called) in the structure.
 Write the algebraic form of the d.e. at each point within the grid.
 Write the algebraic form of the boundary conditions.
 Solve all the algebraic equations simultaneously.
 Finite Difference Method

f
Forward difference
Backward difference
f’(x)
Central difference

f(x-h) f(x) f(x+h)

x
h h
h2 h3 h 4 iv
f ( x  h)  f ( x) hf ( x)  f ( x)  f ( x)  f ( x)
2! 3! 4!
f ( x  h)  f ( x) h h2 h 3 iv
 f ( x)   f ( x)  f ( x)  f ( x)
h 2! 3! 4!
f ( x  h)  f ( x)
 f ( x)   O(h)  Forward difference equation
 h
 Finite Difference Method

h2 h3 h 4 iv
f ( x  h)  f ( x ) hf ( x)  f ( x)  f ( x)  f ( x)
2! 3! 4!
f ( x)  f ( x  h) h h2 h 3 iv
 f ( x)   f ( x)  f ( x)  f ( x)
h 2! 3! 4!
f ( x)  f ( x  h)
 f ( x)   O(h)  Backward difference equation
 h

h2 h3 h 4 iv
f ( x  h)  f ( x) hf ( x)  f ( x)  f ( x)  f ( x)
2! 3! 4!

h2 h3 h 4 iv
f ( x  h)  f ( x) hf ( x)  f ( x)  f ( x)  f ( x)
2! 3! 4!

f ( x  h)  f ( x  h) 2h 2
 f ( x)   f ( x)
2h 3!
f ( x  h)  f ( x  h)
 f ( x)   O(h 2 )  Central difference equation
 2h
 Finite Difference Method

• The central difference equations are better than the forward or

backward difference because the error will be of the order of h-
square rather than h.
• Similar equations can be derived for higher order derivatives of the
function f(x).
• If the domain x is divided into several equal parts, each of length h.

1 2 3 i-2 i-1 i i+1 i+2 n

• At each of the ‘nodes’ or ‘section points’ or ‘domain points’ the

differential equations are still valid.
 Finite Difference Method

• Central difference approximations for higher order derivatives:

Notation
y  f ( x)
 1
yi  y i1  y i1 y i  f ( x  i)
2h
y i  f ( x  i)
 1
y i  2 y i1  2y i  y i1 y i  f ( x  i) and so on 
h 

 1
y i  3 y i2  2y i1  2y i1  y i2 
2h 
1
y i  4 y i2  4 y i1  6y i  4 y i1  y i2 
iv

h
 FDM - Beam on Elastic Foundation

• Consider an interesting problemn --> beam on elastic foundation

w(x)=w

EI y iv  k y( x)  w( x )
Fixed end Pin support
K=elastic fdn.
x L

• Convert the problem into a finite

 difference problem.

1 2 3 4 5 6 iv
EI y i  k y i  w
h =0.2 l

Discrete form of differential equation

 FDM - Beam on Elastic Foundation

iv
EI y i  k y i  w
EI
 4 y i2  4 y i1  6y i  4 y i1  y i2   ky i  w
h
write 4 equations for i  2, 3, 4, 5

0 1 2 3 4 5 6 7

h =0.2 l

Need two imaginary nodes that lie within the boundary

Hmm…. These are needed to only solve the problem
They don’t mean anything.
 FDM - Beam on Elastic Foundation

625EI
At i  2 : y 0  4 y1  6y 2  4 y 3  y 4   ky 2  w
L4
625EI
At i  3 : 4 y1  4 y 2  6y 3  4 y 4  y 5   ky 3  w
L
625EI
At i  4 : 4 y 2  4 y 3  6y 4  4 y 5  y 6   ky 4  w
L
625EI
At i  5 : y 3  4 y41  6y 5  4 y 6  y 7   ky 5  w
L4
• Lets consider the boundary conditions:
 y (0)  0  y1  0 (1)
y ( L)  0  y6  0 (2)
M ( L)  0 (3)
 (0)  0  y(0)  0 (4)
 FDM - Beam on Elastic Foundation

y (0)  0  y1  0 (1)
y ( L)  0  y6  0 (2)
M ( L)  0 (3)
 EI y( L)  0  y6  0
1
2  5
 y6  y  2 y6  y7   0
h
 y7   y5 (3)
 (0)  0  y(0)  0 (4)
1
 y1   y2  y0   0
2h
 y2  y0 (4)
 FDM - Beam on Elastic Foundation

• Substituting the boundary conditions:

kL4 wL4
At i  2 : 7y 2  4 y 3  y 4   y2 
625EI 625EI
kL4 wL4
At i  3 : 4 y 2  6y 3  4 y 4  y 5   y3 
625EI 625EI
kL4 wL4
At i  4 : y 2  4 y 3  6y 4  4 y 5   y4 
625EI 625EI
kL4 wL4
At i  5 : y 3  4 y1  5y 5   y5 
625EI 625EI
Let a = kl4/625EI

7  a 4 1 0 y 2  1
    
 4 6  a 4 1  y  
1
   
3  wL4

 1 4 6  a 4 y 4  1625EI
    
 0 1 4 5  ay 5  1
 FDM - Column Euler Buckling

w
P Buckling problem: Find axial load
x P for which the nontrivial
L Solution exists.

Ordinary Differential Equation

P w
y iv ( x)  y ( x ) 
EI EI

Finite difference solution. Consider case

Where w=0, and there are 5 stations


0 1 2 3 4 5 P 6

h=0.25L
 FDM - Euler Column Buckling
Finite difference method
P 
yiiv  yi  0
EI
At stations i  2, 3, 4
1 P 1
 yi 2  4 yi 1  6 yi  4 yi 1  y i2     yi 1  2 yi  y i 1   0
h4 EI h 2
Boundary conditions
y1  0 (1)
y5  0 (2)
y1  0
1
  y2  y0   0
2h
 y0  y2 (3)
M5  0
 EI  y5  0
 ( y6  2 y5  y4 )  0
 y6   y4 (4)
 FDM - Column Euler Buckling

• Final Equations
1 P 1
 7y 2  4 y 3  y 4    (2y 2  y 3 )  0
h4 EI h 2
1 P 1
4 
4 y 2  6y 3  4 y 4    ( y 2  2y 3  y 4 )  0
h EI h 2
1 P 1
( y 2  4 y 3  5y 4 )   ( y 3  2y 4 )  0
h4 EI h 2

 Matrix Form
7 4 1 y 2  2 1 0 y 2  0
 
  2
PL     
4
 6 4 
 3y  1 2 1 y 3  0

   16EI 0 1 2y  0
1 4 5  y 4    4   


 FDM - Euler Buckling Problem

• [A]{y}+[B]{y}={0}
 How to find P? Solve the eigenvalue problem.
• Standard Eigenvalue Problem
 [A]{y}={y}
 Where,  = eigenvalue and {y} = eigenvector
 Can be simplified to [A-I]{y}={0}
 Nontrivial solution for {y} exists if and only if
| A-I|=0
 One way to solve the problem is to obtain the characteristic polynomial
from expanding | A-I|=0
 Solving the polynomial will give the value of 
 Substitute the value of to get the eigenvector {y}
 This is not the best way to solve the problem, and will not work for
more than 4or 5th order polynomial
 FDM - Euler Buckling Problem

• For solving Buckling Eigenvalue Problem

• [A]{y} + [B]{y}={0}
• [A+  B]{y}={0}
• Therefore, det |A+  B|=0 can be used to solve for 
7 4 1  2 1 0 
   
A  4 6 4 B  1 2 1 

1 4 5 
 
0 1 2

PL2 PL2
and     1.11075
16EI 16EI
7  2 4   1 EI
4   6  2  4    0
 Pcr  17.772 2
L
1 4   5  2 EI
Exact solution is 20.14
   1.11075 L2
 FDM - Euler Buckling Problem

• 11% error in solution from FDM

• {y}= {0.4184 1.0 0.8896}T

0 1 2 3 4 5 P 6
x
 FDM Euler Buckling Problem

• Inverse Power Method: Numerical Technique to Find Least

Dominant Eigenvalue and its Eigenvector
 Based on an initial guess for eigenvector and iterations
• Algorithm
 1) Compute [E]=-[A]-1[B]
 2) Assume initial eigenvector guess {y}0
 3) Set iteration counter i=0
 4) Solve for new eigenvector {y}i+1=[E]{y}i
 5) Normalize new eigenvector {y}i+1={y}i+1/max(yji+1)
 6) Calculate eigenvalue = 1/max(yji+1)
 7) Evaluate convergence: i+1-i < tol
 8) If no convergence, then go to step 4
 9) If yes convergence, then = i+1 and {y}= {y}i+1
Inverse Iteration Method
Different Boundary Conditions
 Beams with Non-Uniform Loading

• Let Mocr be the lateral-torsional buckling moment for the case of

uniform moment.
• If the applied moments are non-uniform (but varying linearly, i.e.,
there are no loads along the length)
 Numerically solve the differential equation using FDM and the
Inverse Iteration process for eigenvalues
 Alternately, researchers have already done these numerical solution
for a variety of linear moment diagrams
 The results from the numerical analyses were used to develop a simple
equation that was calibrated to give reasonable results.
 Beams with Non-uniform Loading

• Salvadori in the 1970s developed the equation below based on the

regression analysis of numerical results with a simple equation
 Mcr = Cb Mocr
 Critical moment for non-uniform loading = Cb x critical moment for
uniform moment.
Beams with Non-uniform Loading
Beams with Non-uniform Loading
 Beams with Non-Uniform Loading

• In case that the moment diagram is not linear over the length of the
beam, i.e., there are transverse loads producing a non-linear
moment diagram
 The value of Cb is a little more involved
 Beams with non-simple end conditions

• Mocr = (Py P ro2)0.5

 PY with Kb
 P with Kt
 Beam Inelastic Buckling Behavior

• Uniform moment case

 Beam Inelastic Buckling Behavior

• Non-uniform moment
 Beam In-plane Behavior

• Section capacity Mp
• Section M- behavior
 Beam Design Provisions

CHAPTER F in AISC Specifications