3.1 Introduction
The codes of practice tackle the stability of steel structures by determining the effective buckling length of the structural members. Therefore, the problem of stability is very important. Underestimation of this effect may lead to disastrous results or unjustified factors of safety. Consequently, the Steel Construction Institute (SCI) suggested the previous approach for estimating the effective length factor presented BS 449: part 2: (1969) should be modified by recommending three methods for evaluating this factor. The first, termed extended simple design, also described in the previous chapter, starts by evaluating the relative stiffness coefficients of the surrounding columns and beams of the column under consideration. Then, using these stiffnesses, the effective length factor can be estimated from charts based on the study carried out by Wood (1974a). The second method, termed the amplified sway method, states that the bending moments due to horizontal loading should be amplified by a factor, as discussed in chapter 2. The third, a more accurate method, is to determine the elastic critical load factor f from cr
73
which the effective length ratios of individual members may be determined. The critical load factor f is defined as the ratio by which each of the factored loads would have to cr be proportionally increased to cause elastic instability. If this parameter is known, the axial load in every compression member Pi at instability is known as well. Then, the value of i =
moment of area and length of a column under consideration respectively. Consequently, the effective length ratio is evaluated as Leff Li = 1 i
Several attempts were suggested in order to overcome some shortcomings of the design chart procedure. Several methods, among them Hashemi (1993), Lokkas (1996), MacLeod and Zalka (1996) and Lokkas and Croll (1998), were suggested for the modification of the design procedure recommended by the British code of practice, but this may lead to a design procedure which is not accepted by a practising engineer. As well as the British code of practice, the American code of practice also suffers from the difficulty of evaluating the effective buckling length accurately enough. This is indicated in the studies by Duan and Chen (1988, 1989), Chen and Lui (1991), Kishi et al. (1997), White and Clarke (1997) and Essa (1997) who proposed modifications to the alignment charts recommended by the American Institute of Steel Construction (AISC). Virtually all methods of analysis that have been developed to improve the limit strength of structures are based upon a geometrically linear model of the structural response. In these methods, the stability concept, addressed in the following section, is used. The available methods of calculating the elastic critical load factor are subsequently described in chronological order in the section on historical background.
74
The compressed bar shown in Figure 3.2 can be similarly considered. In the state of stable equilibrium, if the column is given any small displacement by some external influence, which is then removed, it will return back to the undeflected shape. Here, the value of the applied load P is smaller than the value of the critical load Pcr. By definition, the state of neutral equilibrium is the one at which the limit of elastic stability is reached. In this state, if the column is given any small displacement by some external influence, which is then removed, it will maintain that deflected shape. Otherwise, the column is in a the state of unstable equilibrium.
75
P < Pcr
P = Pcr
P > Pcr
76
use of either the eigenvalue analysis or the second order elastic analysis. Unlike a first order analysis in which solutions can be obtained in a rather simple and direct manner, a second order analysis often entails an iterative type procedure to obtain solutions. Thus, the use of eigenvalue analysis to obtain the critical buckling load is the simplest way.
Load (Pi)
C (b) First order elastic analysis (a) Elastic critical load Eigenvalue Analysis (c) Second order elastic analysis (d) First order rigid-plastic analysis Mechanism load (e) First order elastic-plastic analysis B
Pcr
In order to study the buckling response on several possible idealised models, restricted or not against sidesway, let us consider the two structures in Figures 3.4 and 3.5. The framework, shown in Figure 3.4, is prevented from sidesway whereas in the framework given in Figure 3.5 there is a possibility of sidesway. Both frameworks have initially geometrically perfect members, which are subjected to a set of point loads Pi at
77
their joints. If the members remain elastic as loads are increased, there will be no flexural deformation until a particular level of loading is achieved. This load is known as elastic critical load, corresponding to which a bifurcation of equilibrium is possible (see Hashemi, 1993, Mahfouz, 1993 and Lokkas, 1996).
Pi
Pi
Pi
Pi
Pi
Pi
78
79
which Eulers formula is inapplicable, and up to this value of slenderness ratio the strut fails, is due to direct compression rather than to instability. From that time, the elastic stability problems of braced and unbraced structural frameworks have been addressed by many researchers and a great wealth of literature exists in this field. A considerable amount of the literature is directed towards stability of plane frames within the plane of the frame. The elastic critical load can be evaluated for any symmetrical single-bay multi-storey rigid frame using the relaxation method with no-shear stability function as proposed by Smith and Merchant (1956). The analysis was extended to take account of axial deformation. Bowles and Merchant (1956) applied a more accurate method based on the same technique to the stability analysis of a five-storey two-bay steel frame. The results obtained were in good agreement with those previously obtained using a simpler version of the method. Subsequently, Bowles and Merchant (1958) proposed the conversion of a multi-storey multi-bay rigid plane frame, to an equivalent single bay frame so that it could be analysed by the method proposed earlier. Timoshenko and Gere (1963) treated the buckling behaviour and the buckling load of single-bay single-storey hinged base rectangular frame as well as closed frames. Waters (1964a, 1964b) presented, in two parts, direct approximate methods, involving no trial and error, for the elastic critical load parameter of plane rigid-jointed rectangular and triangulated frameworks. Two approaches were considered: equal rotations and the substitute frame, according to Bolton (1955), Bowles and Merchant (1956) and McMinn (1961). Goldberg (1968) was the first one to tackle the problem of lateral buckling load of braced frames. He did not consider the stability of the frame as a whole but he obtained the elastic critical load equations for a typical intermediate column in a multi-storey frame. He considered the effect of girder stiffness at the top and bottom of that column as well as the average
80
bracing stiffness of that storey. In the same year, Salem (1968) studied the problem of lateral buckling of rectangular multi-storey frames. These frames are loaded at intermediate floor levels and the column sections vary according to an arithmetic series. An investigation on the sway critical load factor of symmetrical and unsymmetrical frames, loaded with unequal and equal axial loads was carried out by Salem (1973), considering the effect of axial deformation variation in columns. Wood (1974a, 1974b, 1974c) adapted an approximate manual technique to be applied in conjunction with effective length and critical load factor charts. The method, which accounts for column continuity, is similar to moment distribution, and called stiffness distribution, involving no-shear stability functions. The elastic critical load factor for a particular storey can be estimated. The same procedure is followed for the rest of the stories and the lowest critical load is the elastic critical load of the original frame. This technique was recommended in BS 5950: Part 1 to be used in the design procedure. Horne (1975) recommended that a horizontal point load equal to 1% of the vertical load at that storey should be added at each storey level, and a linear elastic stability analysis be performed. Bolton (1976) proposed a single horizontal unit point load to be applied at the top of the frame, and the deflection at each storey to be calculated using an elastic analysis. Then, this deflection was multiplied by the total vertical applied load at that storey level, which was finally divided by the height of the storey, to yield the storey critical load factor. The lowest of all load factors corresponds to the critical load factor of the frame. Al-Sarraf (1979) adopted a computing method for predicting the lowest elastic critical load factor of sway and non-sway frames applying modified slope deflection equations based on no-shear stability functions. Anderson (1980) derived formulae, from slopedeflection equations which were used for yielding the storey sways based on sub-frames, assuming the point of contraflexure at the mid span of the elements. Then, sway angles
81
were computed from the storey sway, and the expression for the critical load factor by Horne (1975) was used. A direct calculation of elastic critical loads based upon the structural system concept involving no stability functions was also presented by Awadalla (1983). The computer aspect of this method was discussed and it was shown that the efficiency of the numerical solution can be improved by considering each column as a substructure. The results from this method consistently exceed those produced by the solution obtained by using the stability functions. Carr (1985) developed a computer program for the stability problem. The program also calculates the critical load factor of individual struts of varying cross-section, by defining a node at each change of cross-section. The effective length of each element is also computed while the actual critical load of frame is estimated. An elastic stability analysis was carried out by Simitses and Vlahinos (1986) for single-bay multi-storey frames with support of some rotational stiffness. The computer code implementing the analysis was applied to a two-storey single-bay in a parametric study, to investigate the effect of: (a) increasing number of stories, (b) proportional load, (c) the length and stiffness of beam variation, (d) the support rotational stiffness, and (e) the variation of the column stiffness of the second floor. Goto and Chen (1987) proposed a second-order elastic analysis that can be applied to any shape of structural frame. It takes into account the effect of axial deformation of a structural element. Since the stiffness matrices used were non-linear, iteration was necessary to arrive at the correct solution. Williams and Sharp (1990) used a substitute frame technique to obtain the critical load of multi-storey rigid jointed sway frames. Duan and Chen (1988) started their study by proposing a simple modification of the alignment charts in order to take into account the effect of the boundary conditions at the far ends of columns above and below the column being investigated in braced
82
frames. As reported by these authors, these far end conditions have a significant effect on the K-factor of the column under consideration. As an extension to their research on braced frames, Duan and Chen (1989) and Chen and Lui (1991) suggested another modification to the alignment charts to include the effect of far-end conditions of columns in unbraced frames. Essa (1997) derived expressions for the elastic effective length factors for columns in unbraced multi-storey frames. The model takes into account the effects of boundary conditions at the far ends of the columns above and below the column under consideration. He concluded that using the alignment charts to estimate the effective length factor for columns may be either overly conservative, or even unconservative, depending on the boundary conditions and the relative stiffness ratio of columns. Hashemi (1993) proposed a design methodology for beam-column. The methodology is based upon the following steps. First, an elastic critical load analysis is performed on an idealised model, this takes into account the stiffness interaction with the surrounding frame. Second, a total equivalent imperfection parameter is defined which accounts for the effects of both adopted geometric tolerances and all the loading based imperfections. Third, the non-linear elastic response is used to define the loads at which plastic failure is initiated. Lokkas (1996) extended the work done by Hashemi to circumstances where more than one mode contributed to the nonlinear elastic behaviour and consequently elastic-plastic failure. In 1998, the author continued the study by experimental work to investigate the simultaneous action between the sway and non-sway modes of rigid jointed frames. The experiments show the importance of taking care of the sway and non-sway critical modes exhibiting simultaneous or nearly simultaneous critical loads.
83
EI
= M X .
(3.1)
The quantity EI represents the flexural rigidity of the beam in the plane of bending, that is, in the X-Y plane, which is assumed to be a plane of symmetry. The general solution of equation (3.1) is
y = A sin ( x ) + B cos( x ) + C x + D
(3.2)
where =
P EI
y P x
Figure 3.6. Compressed bar
84
The constants A, B, C, D as well as the elastic critical load Pcr can be evaluated from applying the end boundary conditions of the member. Similarly, the elastic critical load can be obtained using the fourth order differential equation
4 y x 4 2 y x2
EI
+P
= 0.
(3.3)
The use of either the second order differential equation (3.1) or the fourth order differential equation (3.3) is not a simple task when dealing with the problem of elastic stability of either two or three-dimensional structural frameworks. That is due to the large number of boundary and compatibility conditions inherent in structural frameworks.
85
This deflection leads to an increase in the strain energy, known as U, of the system. At the same time, the applied load will move through a small distance L and does work equal to T. The system becomes stable in its undeflected form if
U > T
and unstable if
(3.4)
U < T
where U = 0.5 (L)2, T = 0.5 PL 2 and denotes the spring constant.
(3.5)
The critical load Pcr is obtained from equating the strain energy of the structural system due to a virtual lateral deflection with the work done by the loading pattern on that system. This can be expressed by
U =T.
(3.6)
The theoretical basis of the energy approach is described by Timoshenko and Gere (1963). At loads lower than the elastic critical load, the gain of strain energy in the elements is less than the potential energy of the loads. A condition of instability is defined, as the stage when the change of the above two energies is zero, that is, the stiffness of the structure is zero. Then the structure will not resist any random disturbance. Appeltauer and Barta (1964) applied an approximate energy method to obtain direct formulae for the elastic critical load depending on all the parameters of the problem. The point of contraflexure was assumed to be at the centre of all elements of the frame, so that an approximation to the deflected shape at neutral equilibrium could be obtained. It has been observed from the previous discussion that it is too difficult to use this method when dealing with the problem of elastic stability of a structural framework. The reason for this difficulty is as the number of framework elements
86
increases, the complications in formulae of the strain energy and work done increase too.
constructing the bending moments at each member end including the stability functions (see Galambos, 1968),
constructing the joint and shear equilibrium conditions from which the equilibrium equations are obtained, and
eliminating the unknowns from the equilibrium equations and obtaining the determinantal form of the critical load pattern and finally solving the determinantal form by a trial and error method. In order to explain the difficulty of using this method, Mahfouz (1993) studied the
framework shown in Figure 3.8 using two methods of analysis. One of them is the modified slope deflection method where the framework is subjected to the loading pattern given in Figure 3.8. It was also assumed that the distorted configuration of the framework is anti-symmetric as shown in Figure 3.8.
87
P MBC MCB C
P
B
MBA
MCD
A MAB RA
HA
D MDC RD
HD
Figure 3.8. Single-bay single-storey framework: loading pattern and deflected shape
In this example, ten preliminary equations must be formulated. These equations are for MAB, MBA, MCD, MDC, MBC, MBC, RA, RD, HA and HD. These equations are then substituted into the three basic equilibrium equations to obtain their new form. It can be concluded that as the number of bays and stories increases, the number of preliminary equations increases too. This technique therefore cannot be used when dealing with more highly indeterminate frameworks such as multi-bay or multi-storey frameworks. In addition to, the technique mainly depends on the trial and error method which makes it difficult to link with optimization techniques.
88
at the critical load, there are no external moments or forces at the framework joints to keep it in its distorted configuration, the sum of moments at each joint of that framework should be equal to zero. This procedure will give many equations which are equal to the number of the framework joints. In rectangular frameworks other than symmetrical ones, another set of equations has to be obtained by equating the relative displacements of the framework columns. Finally, by eliminating the unknowns from these equations, a determinantal equation is obtained for the elastic critical load. This determinantal equation has a number of solutions from which the least is called the first buckling load. The solution of such determinantal form can only be done by the method of trial-and-error using a computer program. Figure 3.9 shows the basic simple operations of rotation and sway of an axially compressed isolated member for both cases of fixed and pin-ended bases. The principle of supperposition of any number of states of sway and rotation of an isolated axially compressed member is applicable so long as the axial compression is kept constant through all these states. Furthermore, the principle of resolution of any state of an isolated axially compressed member into any number of states of sway or rotation is also applied under the same condition. The no-shear stability functions m, n and O were introduced by Merchant (1955) to deal with the case of a member with fixed ends while Salem (1968) treated the hinged end case by introducing the stability function n for the no-shear sway for such members. Salem also decomposed the general state of sway into two components, which are the states of pure-shear sway and no-shear sway. These two states of sway are shown in Figure 3.10 for members with fixed and pin-ended bases. The non-dimensional stability functions S, C, S, m, n, O, and n (Appendix A) indicated in Figures 3.9 and 3.10 are all functions of the ratio of the axial load to
89
90
a) Rotation
H = S (1 + C )
K L
H = S
K L
M = CSK P
P M= L P P H H n L
M=
m 2
HL = S (1 + C ) K
HL
= SK
b) Pure-shear sway
H M= P m 2 HL
H P
M= nK
M= nK P
c) No-shear sway
L = m 2
L = 1 n
M = -OK P P
91
' " M1 = M1 + M1
' M 1 = m
HL 2
" M 1 = nK 1
P H
'
P H
"
1
L = ' L H P M2 + " = L H P
' M 2 = m
1
mHL " + L m 2
' L
2 S (1 + C ) K
= M '2 + M " 2
HL 2
" M 2 = OK
M1 = M1 + M1 P H
'
"
' M1 =
HL n
" M1 =
2 K
n
'
"
1
L ' L " = L ' L HL " S nK
1
" L
n 1
H P
H P P
' " M1 = M1 + M1
' M 1 = m
HL 2 H
M 1 = nK1 -OK2
"
P H
'
"
2
L = ' L + " = L ' L = mHL 2 S (1 + C )K +
2
" L = m 2 (1+2 )
1
H P
' " M2 = M2 + M2
1
H P
' M 2 = m
HL 2
M 2 = nK2 -OK1
"
State of sway
92
Start
Calculate the determinant value Increase Does the determinant change its sign?
No
Yes Stop
Figure 3.11. Flowchart for computer program based on the direct method
93
method, are usually thought of as being based upon overall functions (i.e. those defined over the entire region of the structural system). In the finite element method, each member of the structure is subdivided into a series of fairly short elements. The deformation over each element may be defined by a simple polynomial function. The coefficients of these polynomial functions may be determined if the displacements of each node are known. As a result, the individual displacements of the entire structure may be calculated and consequently the behaviour of the structure may be fully described in terms of the displacements of the nodes. For equilibrium the increment in total potential energy must be stationary with respect to these nodal displacements. This leads to a set of linear homogeneous equations, where the dependent variables of these equations are the nodal displacements , i.e. the following eigenvalue problem:
f [K CG ] { } = [K CE ] { }
(3.7)
where f is the load factor, K CE is the global elastic stiffness matrix corresponding to the connecting joints (nodes), K CG is the geometric stiffness matrix. The first eigenvalue, i.e. the smallest value of f at which the structure becomes unstable is termed the critical load factor f . cr This classical eigenvalue approach discussed by many authors among them Prezemieniecki (1968), Allen and Bulson (1980), Graves Smith (1983), Brebbia and Ferrante (1986), Coates and Kong, (1988), Galambos (1988) and Bathe (1996). The eigenvalues and eigenvectors can be obtained by applying several techniques, among them vector iteration methods i.e inverse iteration, forward iteration and Rayleigh quotient iteration, transformation methods such as Jacobi method and generalised Jacobi
94
method, and the subspace iteration method. Subroutines, written in FORTRAN 77, are available in Bathe (1996).
M C = M CB + M CD = 0 ,
K2 K1 4K b
K1
K1
1 S 2
2K b
K1
K1
S1 +
2K b
2 = 0 , and
(3.8)
2 = 0 .
(3.9)
95
Eliminating the unknowns ( 1 and 2 ) from (3.8) and (3.9), the elastic critical load
equation becomes:
P Ib B Ib Kb = L K1= I1 L K2= C
P Ib Ib Kb = L I2 L I1 K1= I1 L K2= I2 L
I1
I2 L
L a) Loading pattern
4 K b 1 -2 K b 2 2 K b 1 -4 K b 2 -S2K22
S1 K1 1
C1S1K11
b) Distorted configuration
c) Operation of rotation
Figure 3.12. Single storey single-bay fixed base framework prevented from sway
K1
K1
K1
S2
2K b
K2
4K b
det
K1
K1
S1 +
4K b
2K b
= 0.
(3.10)
I2
Rigid bracing
- C2S2K22
96
Following the same previous procedure, the elastic critical load equation (3.11) of a single-bay single-storey fixed base framework permitted to sway (Figures 3.13), is obtained:
where the third equation is obtained by equating the sidesways at joints B and C. In this equation, the unknowns are 1 , 2 and HL K1 . For a single-bay single-storey hinged base framework permitted to sway (Figures 3.14), the elastic critical load equation is
# #
In order to verify the developed program, the previous described framework models have been analysed and the results obtained are compared with established theoretical results are given in Table 3.1.
"
"
n1
"
n2
n1
n2 K 2
" S1
S "2
1
! !
K1
"
K1
# "
K1 1
K1
n2
"
det
n" 2
2K b
!
K2
4K b
K1
"
"
K1
n1 1
" n1 +
# ! "
4K b
2K b
=0.
m1
m2
m2 K1 + 2 S1 (1 + C1 ) 2 S 2 (1 + C 2 ) K 2
m1
K1
K1
K1
det
n2
2K b
K2
4K b
K1
K1
2 m2
n1 +
4K b
2K b
m1
= 0 (3.11)
(3.12)
97
P Kb
4 K b 1 2 K b 2 n1 K 1 1 + m1 2 HL
2 K b 1 4 K b 2 n2 K 2 2 m2 HL 2 -O2K22 m2 HL 2
1
K2
K1
L -O1K11 m1 + HL 2
H L
4 K b 1 2 K b 2 n1 K1 1 1 + HL n1
"
2 K b 1 4 K b 2 n 2 K 2 2 1 HL n2
"
Kb K2
K1
H L
98
Table 3.1. Comparison of the theoretical and developed code in plane buckling loads. Model
P P
Obtained results
I I I L
Pcr = 25.2
EI L2
=
L
Pcr
2 EI
L2
= 2.55329
= 2.554
Pcr = 7.34
I
L
EI L2
Pcr
2 EI
L2
= 0.7437
= 0.7475
L P P
I I I L
Pcr = 1.82
EI L2
Pcr
2 EI
L2
= 0.1844
= 0.1843
L P P
I b= 0 I I L
Pcr =
2 EI
4L2 = 0.25
=
L
Pcr
= 0.2499
2 EI
L2
99