UNIVERSITY OF CALIFORNIA BERKELEY Department of Civil Engineering Spring 2011

Structural Engineering, Mechanics and Materials Professor: S. Govindjee

Notes on Potential Energy Methods: Buckling1

In this section of the notes we will revisit the problem of buckling instability that you have studied in your previous courses. Here we will apply the concept of potential energy to reexamine this problem class. We will start with some simple systems composed of rigid bars and springs and then will turn our attention to continuously deformable systems and approximate solutions. Before beginning you should re-familiarize yourself with basic concepts of buckling; see for example Sections 12.1 12.3 in S.G. Engineering Mechanics of Deformable Solids or the relevant chapter in Hibbeler, etc. It is of importance to recall: 1. Both buckled and unbuckled solutions correspond to equilibrium states of a system. 2. As the load is increased one approaches the critical point. At the critical point one has the emergence of multiple equilibrium solutions. 3. An equilibrium solution that is stable has a positive second derivative of the potential energy. 4. If one only wishes to determine the buckling load and perhaps some general information about the buckled solution, then one can simplify the analysis by using small deection and rotation assumptions. This results in an eigenvalue type problem, where the eigenvalues correspond to the buckling loads and the eigenvectors to the buckled shapes.

Instability of discrete systems

As a bit of quick review, let us consider a simple mechanical system consisting of a rigid bar and a rotational spring as shown in Fig. 1. The bar is pinned at its base and further supported by a rotational spring with spring constant k; at the other end of the bar, it is subject to an axial (compressive) load P . Here we are interested in the equilibrium solutions of the system. Since these correspond to stationary points of the total potential energy, let us take the approach which employs the principle of stationary potential energy. The total potential energy for this system consists of the elastic potential energy and the potential energy of the load: elastic = load
1

1 2 k , 2 = P L(1 cos ) ,

Based on the original prepared by Dr. T. Koyama, Spring 2009.

L(1-cos#)

L k

Rigid bar

Figure 1: Rigid element and rotational spring system and thus is: total () = 1 2 k P L(1 cos ) . 2

The stationary points of the potential energy are obtained from dtotal = k P L sin = 0 . d The s which satisfy this equation for a given P are the equilibrium solutions. Note the use of the plural solutions, as here as in opposition to all other problems we have treated up to now, the problem does not necessarily have a unique solution. In fact, there are two possible solutions: 1. The trivial solution (straight) =0 true for any given load P . 2. The non-trivial solution P = k . L sin (2) (1)

A plot of the loci of points (, P ) which satisfy either equilibrium solution is given in Fig. 2. The plot exhibits an interesting transition in behavior as the load approaches Pcr = k L

from below. This load is called the critical load and denotes the load past which the trivial solution = 0 is no longer stable and no longer unique. For loads below Pcr , there is only the trivial solution. Above this load there are three solutions: = 0 and two solutions implicitly given by Eq. (2). 2

Stable

k L
Tension Compression Eqn.1

Figure 2: Solutions for varying loads P To actually ascertain the stability of the dierent equilibrium solutions, we need to look at the second derivatives of the potential energy as these tell us if the forces acting on the system serve to restore equilibrium after a perturbation or serve to move the system even further from the equilibrium state. For our example: d2 total = k P L cos . d2 The stability of the solution is: On solution path 1: d2 total = d2 On solution path 2: dtotal k L cos = k 2 d L sin = k (1 cot ) > 0 k P L > 0 for P < k P L < 0 for P >
k L k L

Thus we see that solution (1) is the stable solution for all loads P < Pcr but for loads P > Pcr it is unstable (negative second derivative of the potential energy). We also see that for this case that solution (2) is stable. Remarks:

3 solutions for

P >k L

Unstable Eqn.2

Stable

1. If one is only interested in determining the critical load and perhaps linearized information about the buckled state of the system, then one can make a small angle approximation. If we assume small values of , || 1, then 1 cos 1 1 such that, load = P L Therefore, 2 . 2 2 2 , = 2 2

1 2 total k2 P L . 2 2 Using this expression to compute the equilibrium equations gives: dtotal k P L = 0 d (k P L) = 0 k = 0 or P = L

The analysis reveals the critical load of the system Pcr = k/L but does not reveal too much about the buckled state. It says = 0 and P arbitrary is an equilibrium solution and that P = k/L and arbitrary is another solution. If we take a second derivative, we nd d2 total k PL . d2 This shows that the trivial (unbuckled) conguration = 0 is stable for P < Pcr and unstable above this load. It tells us that the second solution is neutrally stable near 0. As an example of this type of analysis for a multi-degree of freedom system, consider the two degree-of-freedom system in Fig. 3. The system consists of two rigid bars connected by torsional springs. Each bar has length L and the springs have stiness k1 and k2 . We will use the angle from the vertical to parameterize the possible congurations of the system; see Fig. 3. The total potential energy of the system can then be expressed as: (1 , 2 ) = 1 2 1 2 2 k1 1 + k2 (2 1 )2 P L 1 + L 2 2 2 2 2 .

Here we have already employed the assumption of |1 | 1 and |2 | 1 to simplify the expression for the potential energy of the load. Note that this will restrict the information which we can extract from the analysis. Notwithstanding, we will be able to determine the critical load and we will be able to make some assessment of the buckled state of the system. 4

k L k L

N O 

Figure 3: 2 rigid element and 2 rotational spring (2 DOF) system For equilibrium the potential energy must be stationary and this implies = 0, 1 = 0. 2 This leads to a linear system of equations consisting of two equations in two unknowns, k1 + k2 P L k2 k2 k2 P L 0 1 = 0 2 .

This equation is similar to the equation (k P L) = 0 in the one degree of freedom case. The combinations of (1 , 2 ) which satisfy this linear system of equations for a given load P are the solutions which satisfy equilibrium. One has two dierent cases of solutions: 1. = 0, which gives the trivial (straight) solution. This solution is valid for all loads P even if it is possibly an unstable equilibrium. 2. A non-trivial solution, , is only possible when det[ ] = 0. This will only occur for specic values of P assuming that the bar lengths and the springs are xed. These values are the eigenvalues associated with the generalized eigenvalue problem: k1 + k2 k2 k2 k2 P L 0 0 L 0 1 = 2 0 .

For our two degree of freedom problem there are two such load values P = P1 , P2 for which non-trivial solutions exist, 1 and 2 . These loads are called the buckling loads and the smallest buckling load is called the critical load : Pcr := min(P1 , P2 ) . Figure 4 displays the solutions using the norm of as the abscissa. For loads P < Pcr , 5

eigenvectors give shape

Stable

2
Stable

2
Tension Compression

Figure 4: Solutions for varying load P for the 2 DOF system

Mode shape 1 (eigenvector 1)

Mode shape 2 (eigenvector 2)

Figure 5: Mode shapes (eigenvectors) for the 2 DOF system

Unstable

Stable

||||

the trivial (straight) solution is stable. Above this load the trivial solution will be unstable. The two buckling solutions are depicted by the two sets of pitch-fork arms which are stable equilibrium branches of the solution. Note that our linear solution does not provide equations for the solution away from the small angle approximation. To obtain the details of the curves shown, one needs to perform a full non-linear analysis. What one can determine from the linear solutions is the general conguration into which the system buckles on each branch. This information is contained in the eigenvectors corresponding to the two buckling loads P1 , P2 . Figure 5 shows the eigenvectors and corresponding mechanical modes 1 and 2 .

Instability of continuous systems

Before investigating the stability of continuous mechanical systems via potential energy let us review some classical results for beam buckling that are typically derived using ordinary dierential equation methods. The classic results are associated with four basic sets of beam boundary conditions. These are: Clamped-Free, Pin-Pin, Clamped-Pin, and Clamped-Clamped with the addition of a compressive axial load as shown in Fig. 6. For all cases, the governing dierential equation of equilibrium is: EIv + P v = 0 . This equation has two classes of solutions: 1. v(x) = 0, which is the trivial (unbuckled) solution existing for any load P . 2. v(x) = 0, which is the non-trivial solution which exists only for certain loads P . One is faced with following questions: When can one have a non-trivial solution v(x)? What are the loads P for a non-trivial solution? One can nd the loads P corresponding to non-trivial solutions by solving Eq. (3) subject to the given boundary conditions of the problem. An outline of the solution method is as follows: (3)

EI L
Clamped-Free

P
Pin-Pin

P
Clamped-Pin

P
Clamped-Clamped

Figure 6: Various congurations for buckling of a beam 1. Assume v(x) = Cesx , and insert this into Eq (3) to yield the characteristic equation: C(EIs2 + P )s2 = 0 . The four roots of this equation are s = 0, 0, i following form for the homogeneous solution: v(x) = c1 sin P x EI P , where i = 1. This yields the EI P x EI

+ c2 cos

+ c3 x + c4 .

2. Apply the boundary conditions (there should be four) which yields four equations for the four unknowns c1 , c2 , c3 , and c4 . These equations have the form, F (P )c = 0 , c1 c2 c := , c3 c4

where F (P ) is a 4-by-4 matrix depending on P .

3. A non-trivial solution, c = 0, is only possible when det(F (P )) = 0. This condition yields a transcendental equation in P . By solving for the roots of this relation, one 8

obtains the buckling loads. In this setting there are an innite number of buckling loads; the smallest of these is the critical load for the system. For the four representative cases the results are: Pin-Pin (Euler load): Pcr = 2 Clamped-Pin: Pcr = (4.497)2 Clamped-Clamped: Pcr = (2)2 Clamped-Free: Pcr = (/2)2 EI L2 EI L2

EI L2 EI L2

2.1

Beam buckling using potential energy

The instability of continuous mechanical systems, e.g., beams, can be treated in a manner similar to the discrete mechanical systems consisting of rigid bars and springs. One must determine the elastic potential energy and the potential energy of the load. Elastic: elastic :=
0 L

1 EI(v )2 dx 2

Load: load := P displacement at load point L 1 2 dx, = P 0 2 L 1 2 = P (v ) dx . 0 2 The way that the displacement at the load point is derived is through the following argument. Consider an innitesimal segment of the beam of length dx. This segment rotates (during buckling) by an angle = v . The resulting vertical drop associated 1 with the segment is equal to 2 2 dx; see Fig. 8. For the total contribution from each segment of the beam, one must add up the contributions from each segment; i.e., one must integrate the expression over the interval [0, L]. 9

P P

L x

EI

Figure 7: Cantilever with compressive load

#L

dx

4ECE@>=H

Figure 8: Displacement drop of an innitesimal element

10

The total potential energy is dened as the sum of these two contributions: total (v(x)) = elastic + load
L L

=
0

EI(v )2 dx P

1 2 (v ) dx . 2

To approximately nd the loads at which instability occur (buckling loads) and to estimate the buckling modes, one can apply the Ritz method. As was done previously, we will assume an approximate form for the solution v(x) which satises the kinematic boundary conditions and contains free parameters. These will then be determined by nding the stationary conditions for the potential energy. Example 1: One parameter approximation for clamped-free beam buckling.
For the cantilever shown in Fig. 7, consider an approximation for v(x) of the form: v(x) Cx2 , where C is an undetermined coecient. The derivatives are computed as, v (x) = 2Cx, v (x) = 2C . Inserting these into the expression for the total potential energy yields, total (v(x)) total (C) = dtotal dC
v(x)=Cx2 L 1 1 2 2 EI4C 2 dx P 4C x dx 0 2 0 2 1 = 4EICL P 4C L3 3 4 3 = C 4EIL P L = 0 . 3 L

This yields two solutions: 1. C = 0, which is the trivial solution with v(x) = 0 corresponding to any load P . 2. C = 0, which is the non-trivial solution corresponding to a loading of P = 3EI . This value L2 for the load is the critical buckling load Pcr and is the load at which instability occurs.
2 exact = EI . The relative error in the The exact critical buckling load for this conguration is, Pcr 4 L2 approximation is:

relerr =

12 2 /4 3 = 1 2 = 21% . 2 /4

11

Example 2: Two parameter approximation for clamped-free buckling.

The advantage of the Ritz method is the ability to improve the approximate solution by expanding the possible solutions in an easy fashion. Consider the two parameter approximation: v(x) C x L
2

+D

x L

It is advantageous numerically and in terms of understanding the phenomenon to consider nonx 2 dimensionalized forms for the functions in the Ritz method, e.g., in this case the selection of L instead of x2 . Thus the coecients C, D have units of length and are often called the generalized displacements. The needed derivatives of the approximation are: v (x) = v (x) = 2C x + L L 2C 6D + 2 2 L L 3D x L L x . L
2

The total potential energy yields:

L

total (v(x)) total (C, D) =

1 6D 2C EI + 2 2 L2 L
L

x L

dx x L
2 2

1 2C 2 L

x 3D + L L

dx .

There are two possible ways to satisfy these homogeneous equations:

The stationary points of total are the solutions which satisfy equilibrium. total 4EI P 4 6EI 3 =0 L3 P2 L L3 L3 C C = 0 . D 0 6EI 3 12EI 9 total P2 P5 L L L3 L3 =0 D

(4)

1. Take C = D = 0, which is the straight trivial solution with v(x) = 0 and is valid for any load P. 2. Assume that det[ ] = 0. This permits a non-trivial solution for two loads P1 and P2 (which are determined from the determinant condition). These two loads are the buckling loads and the smaller of the loads is the critical load. The buckling loads P1 , P2 and the corresponding non-trivial solutions [C, D]T can be found by rewriting Eq. (4) in the form: P L2 4 6 6 12 EI
4 3 3 2 3 2 9 5

C =0. D

12

By dening A := B := := x := these equations have the form: (A B) x = 0, which is known as a Generalized Eigenvalue Problem. Note that in the case that B is the identity matrix, one has Ax = x which is the standard eigenvalue problem.2 The two eigenvalues are computed by setting the determinant of A B zero which results in a quadratic polynomial for . The solutions are: = P L2 = 2.486, 32.1807 . EI 4 6 6 12
4 3 3 2 3 2 9 5

, ,

P L2 , EI C , D

Since the smallest buckling load is the critical load: Pcr = 2.486 Remarks: 1. With respect to the exact solution, this has a relative error of 0.75%. 2. If one computes the associated eigenvector [C, D]T , then one can plot the basic shape of the buckling mode as v(x) = C(x/L)2 + D(x/L)3 . EI . L2

Example 3: Buckling of a cantilever supported at mid-span by a spring support.

Consider computing the buckling load of a cantilever with a lateral spring support placed at midspan as shown in Fig. 9. The calculation involves only a slight modication to our last example. One only has to account for the additional contribution of the elastic potential energy associated with the spring. This is simply added the total potential energy as shown in the following formulation. total = elastic,beam + elastic,spring + load , 1 k v 2 L 2
2

elastic,spring :=
2

This generalized eigenvalue problem can be eciently solved in MATLAB by the function call eig(A,B).

13

x k EI L/2

P L/2

Figure 9: Cantilever supported by a spring at the center

For the case above with approximation v(x) = C one has: elastic,spring = 1 k C 2 1 2
2

x L

+D

x L

+D

1 2

3 2

Partial derivatives with respect to C and D of this added term yield new contributions to the equilibrium equations in terms of k. P L2 kL3 4 2 4 6 + 6 12 64EI 2 1 EI
4 3 3 2 3 2 9 5

C =0. D

Taking the determinant of the coecient matrix and setting it equal to zero yields a polynomial for the eigenvalues the smallest of which is the critical load. The corresponding eigenvector then denes the general buckled shape of the system.

2.2

Buckling with distributed loads

EI L x b(x)

Cut cantilever at height

b(x)dx

Figure 10: Cantilever with distributed load Consider a cantilever beam subjected to a distributed load as shown in Fig. 10. The total potential energy for this system is given as,
L L x

total :=
0

EI(v ) dx

{b(x)dx} 14

1 (v ())2 d x x 2

where the rst integral is the elastic potential energy and the second integral is the potential of the load. The expression for the potential of the load appears rather complex but can be justied if examined closely. The rst component under the integral represents the load for a small segment of material near x and the second component represents the displacement drop at point x. The expression for this displacement drop is necessarily an integral as was required in the earlier examples. Its limits are 0 to x to account for all rotation eects below the point in question. Recall that for a cantilever of length L with end load P , the contribution to the potential energy due to the load is
L

1 2 (v ) dx . 2

What has been done is a simple replacement of L with x and P with b(x)dx and the summation of this contribution from 0 to L. Example 4: Buckling behavior under a uniform distributed load.

EI A, L x
Figure 11: Tree collapsing problem
Let us consider the problem of a tall tree and determine how tall it can grow before it will collapse due to buckling. For this situation the loading is a distributed load which is constant b(x) = Ag = , where A is the cross-sectional area of the tree (assumed constant), is the mass density, and g is the gravitational constant. Thus is simply the weight per unit length of the tree. To solve let us determine the buckling condition for the system in terms of . We will then invert this relation at the end to nd a restriction of the height of the tree in terms of . For this case the potential energy takes the form,
L

total =
0

1 EI(v ) dx 2

L 0 0

1 (v ())2 d x x 2

dx.

To approximately solve this problem, consider a one-parameter approximation for the solution: v(x) Cf (x) , where f (x) is a known function and C is the undetermined coecient. Under this assumed form,
L

total (v) total (C) =

1 EI(Cf )2 dx 2

L 0 0

1 (Cf ())2 d x x 2

dx .

15

The stationary condition implies dtotal (C) =C dC

L 0

EI(f )2 dx

L 0 0

(f ())2 d x x

dx = 0.

For a non-trivial solution C = 0 one must have, = Let us now select the specic Ritz
L 2 0 EI(f ) dx x L x 2 x 0 (f ()) d dx 0 function f (x) = x2 . With L 2 0 EI(2) dx x x2 x 0 (2) d

. this approximation,

= = =

L 0 L 0

dx

4EIL
4 3 3x

dx

4EIL 1 4 3L 12EI . = L3 For < 12EI/L3 there should be no buckling and thus the height restriction is L3 < 12EI 12EI 12Er2 = = , gA g

where for the last equality we have introduced the radius of gyration r2 = I/A of the cross-section. Remarks: 1. From the result, we see that denser trees are necessarily shorter. Likewise a tree can grow taller, if it increases its radius of gyration. 2. We can assess the accuracy of our computation since there is a known reference solution. Up to 4 digits this is3 EI exact = 7.837 3 . L 12 7.837 Comparing we nd that the relative error for our simple approximation is 53%, 7.837 which is not too good. Adding additional polynomial terms would greatly improve the accuracy. 3. As an alternative choice one can pick f (x) = 1 cos = 8.2979
x 2L

. This yields,

EI , L3 which gives a relative error of approximately only 6%.

4. Every tree you see must know this relation!

This solution can be computed from the governing ordinary dierential equation. However, it requires knowledge of Bessel functions of fractional order.
3

16

2.3

Deection behavior for beam with compressive axial loads and transverse loads

EI f x L/2 L/2

Figure 12: Simply supported beam with intermediate transverse load Consider a simply supported beam with a xed load f applied at the middle as shown in Fig. 12. Additionally, the beam is subjected to a compressive axial load P . The total potential energy for this mechanical system is
L

total =
0

1 EI(v ) dx P 2

L 0

1 2 (v ) dx f v 2

L 2

If f = 0, we are looking at a classical buckling problem; viz., the beam remains straight until a critical load is reached after which the beam bends suddenly. The critical load for the conguration shown is Pcr = 2 EI/L2 . Let us investigate the behavior for f = 0. The stationary points of the potential energy still give the solutions v(x) which satisfy equilibrium. Let us compute an approximate solution using the form v(x) C sin The derivatives of this function are, x cos , L L 2 x v (x) = C sin L L v (x) = C Inserting these into the potential energy yields, total
L 4 2 2 1 2 2 x x 1 EI dx P C sin C cos2 = L L L 0 2 L 0 2 L 4 2 1 1 2x 1 EI cos dx C = L 2 2 L 0 2 L 1 2 2 1 1 2x P dx f C + cos C 2 2 L 0 2 L 4 2 1 2 2 1 EI C LP C L fC . = 4 L 4 L L

x L

dx f C sin

L/2 L

17

The stationary condition yields, 0= dtotal 1 4 1 2 = EI CL P CL f dC 2 L 2 L 1 4 1 2 = C EI LP L f =0 , 2 L 2 L

and thus C = f
4

2 EI P 2L3 2L 2 f 2L/ = EI 2 P L2 f 2L = . 2 P P cr

The approximate solution has the form v(x) f x 2L sin 2 Pcr P L .

The relationship between P and v(L/2) for a xed f is shown in Fig. 13. The relationship between f and v(L/2) for a xed P is shown in Fig. 14. Remarks: 1. Observe how the response curve in the v(L/2), P plane converges to the case for f = 0 as f decreases; see Fig. 13. The behavior is buckling-like for this system. 2. Note that for small axial loads, the system behaves like a standard beam in bending. However, as the axial load approaches the Euler load, then deection is greatly amplied by the factor 1/(Pcr P ); see Fig. 14.

18

P
f=0

Pcr
f=0

f increasing v(L/2)

Figure 13: Relationship between mid-displacement v(L/2) and the compressive load P

P increasing
2 (PP ) cr 2L

v(L/2)
Figure 14: Relationship between mid-displacement v(L/2) and the transverse load f

19