Professional Documents
Culture Documents
www.elsevier.com/locate/engstruct
Received 13 July 2007; received in revised form 6 December 2007; accepted 6 December 2007
Available online 22 January 2008
Abstract
This paper presents an efficient algorithm for both limit and shakedown analysis of 3-D steel frames by the kinematic method using linear
programming technique. Several features in the application of linear programming for rigid-plastic analysis of three-dimensional steel frames
are discussed, as: change of the variables, automatic choice of the initial basic matrix for the simplex algorithm, direct calculation of the dual
variables by primal–dual technique. Some numerical examples are presented to demonstrate the robustness, efficiency of the proposed technique
and computer program.
c 2007 Elsevier Ltd. All rights reserved.
Keywords: Limit analysis; Shakedown analysis; Plastic hinges; Space frames; Linear programming
with variable repeated loadings; shakedown plastic design with where λi is the plastic deformation magnitude; ei is the vector
updating of elastic response in terms of the plastic capacity. of longitudinal displacement and two rotations of ith section;
NCi is a gradient vector of the yield surface Φ.
With the CEPAO, efficient choice between statical and
kinematic formulations is realized leading to a minimum The application of the LP techniques requires that the
number of variables; also there is a considerable reduction nonlinear yield surfaces must be linearized. In civil engineering
of the dimension of every procedure. The basic matrix practices, for bisymmetrical wide-flange shapes, several
of LP algorithm is implemented under the form of a Standards replace the curvilinear yield surface by a polyhedron
reduced sequential vector which is modified during each sixteen-facet:
iteration. An automatic procedure is proposed to build up
My
|N | |Mz | |N |
the common characteristic matrices of elastic–plastic or rigid- α1 + α2 + α3 = 1 for ≥ α0 ; (2a)
plastic calculation, particularly the matrix of the independent NP MPy M Pz NP
equilibrium equations. Application of duality aspects in the LP
|N | My |Mz | |N |
technique allows direct calculation of dual variables and avoids α4 + α5 + α6 = 1 for < α0 ; (2b)
NP MPy M Pz NP
expensive reanalysis of every problem.
At this time, the CEPAO is extended to the case of space where: M P y, M Pz are the plastic moment capacity with respect
steel frames with the following problems: rigid-plastic analysis to y and z axis, N P is the squash load, 0 ≤ α0 < 1, α1 , . . . α6
and design by LP; elastic–plastic analysis by the step-by-step are the dimensionless coefficients. The Eqs. (2a) and (2b) may
method in both first and second order (P-delta effects). The limit also be written
and shakedown analysis is based on the upper bound theorem |N |
≥ α0 ;
while the rigid-plastic design is supported by the lower bound a1 |N | + a2 M y + a3 |Mz | = S0 for (3a)
NP
theorem. It is important to indicate here that in the case of
|N |
frame structures both kinematic and statical methods lead to < α0 ;
a4 |N | + a5 M y + a6 |Mz | = S0 for (3b)
the same one and only solution. The geometrical nonlinearity is NP
completely ignored in this direct method (by LP). with S0 is a referential value, and a1 , . . . a6 are the nonzero
Because of the limitation of an article, the present work coefficients.
describes only details of the module of limit and shakedown At the ith critical section, the plastic admissibility defined
analysis where some original contributions are indicated in the by Eqs. (3a) and (3b) has the following form:
Section 3. We hope that other contents will be presented in our
next papers. Yi si ≤ si0 , (4)
where matrix Yi contains the coefficients a1 , . . . a6 ; si collects
2. Assumptions and modelling plastic hinges
the vector of internal forces; the column matrix si0 contains the
corresponding terms S0 .
The following assumptions have been made:
According to the definitions of the matrices NC , Y, we can
– Loading is quasi-static and service load domain is specified see that
by linear constraints;
– The torsional stiffness and the effect of the shear forces are
i
NC = YiT . (5)
negligible; The detailed form of matrix NC is below-mentioned by Eq.
– The rigid — perfectly plastic material is used in the limit (13).
analysis; the elastic — perfectly plastic material is applied
in the other problems. 3. Application of LP
– Plastic hinges are located at critical sections.
A systematic treatment of the application of LP in plastic
Modelling plastic hinges analysis can be found in [4,5]. In the present work, we restrict
Since the effect of both shear forces and torsional moments ourselves to describing some practical aspects of the CEPAO
are ignored, the condition of plastic admissibility at the critical package applied to the case of 3-D steel frames. They are:
sections becomes Φ(N , M y , Mz ) ≤ 0, with N is the normal the further reduction of the kinematic approach (Sections 3.2.2
force and M y , Mz are respectively bending moments about to and 3.3.2), and the direct calculation of the internal force (or
y and z axes. The plastic hinge modelling is described by the residual internal force) distribution (Sections 3.2.3 and 3.3.3).
choice of net displacement (relative) - force relationship at the
critical sections. In present work, the normality rule is adopted. 3.1. General formulation
∂Φ/∂ N
∆ In the CEPAO, the canonical formulation of the LP is
θ y = λ ∂Φ/∂ M y , considered:
θz ∂Φ/∂ Mz
Min π = cT x|Wx = b (6)
or, symbolically:
where π is the objective function; x, c, b are respectively
ei = λi NC
i
, (1) the vector of variables, of costs and of second member. W
H. Van Long, N. Dang Hung / Engineering Structures 30 (2008) 1895–1904 1897
is called the matrix of constraint. For the sake of simplicity, Therefore, the vector of variables, matrix of constraint, vector
the objective function has a state variable, and the matrix of second member corresponding to the problem of Eq. (8) for
formulation is arranged such that the basic matrix of the initial limit analysis are given below.
solution appears clearly as follows:
x∗T = π xT η = π d0T λT η
x
b∗T = 0 bT = 0 −Bd0 ξ + fT d0
−c1 1 −cT2 1
T
0
π = . (7)
W1 0 W2 b
1 0T −sT0 0
x2
W∗ = 0 −B NC 0
The basic matrix of the initial solution is 0 fT 0T 1
1 −cT2
X0 = . where η is an artificial variable which must be taken out of the
0 W2 basic vector in the simplex process.
Eq. (7) can then be written under a general form Because of the use of the simplex technique, finding an
initial admissible solution such that the initial value of any
W∗ x∗ = b∗ . (8) variable (except the objective function) must be nonnegative
The matrices W∗ , x∗ , b∗ and X0 for both limit and shakedown is needed. To satisfy this requirement, it appears that the
analysis problems will be accurately calculated in the following following arrangement leads to good behaviour of the automatic
sections: calculation.
The linearized condition of plastic admissibility for the ith
3.2. Limit analysis by kinematic method section (Eq. (4)) may be expanded as follows:
3.2.2. Further reduction of the kinematic approach The Fig. 1 describes the projection of 16 planar facets of the
In the kinematic method, the unknowns are the plastic polyhedral stress-resultant yield surface corresponding to the
deformation magnitude, λ, and the independent displacement, 16 inequalities numbered on the Eq. (11).
d (negative or positive). In LP procedure we need nonnegative According to Eqs. (4) and (5), we see that Eq. (11) can be
variables so that we adopt the change of the variables as in the written under symbolic formulation
following:
NC s ≤ si0 .
iT i
(12)
0 0
d = d + d0 so that d ≥ 0.
Put
The way to fix the value of d0 , which depends on the real
a1i a1i a1i
structure, such that d0 are always nonnegative is explained in i
the Appendix C. Now, the problem of Eq. (9) becomes ÑC = −a2i a2i a2i .
−a3i −a3i a3i
NC λ − Bd0 = −Bd0
i
Min φ = s0 λ fT d0 = ξ + fT d0
T Let us note that: ÑC is always nonsingular because a1i , a2i , a3i
(10)
λ, d0 ≥ 0. are certainly positive.
1898 H. Van Long, N. Dang Hung / Engineering Structures 30 (2008) 1895–1904
in which
i0 i i
λ̃ = 0.5(E + Si )λ̃ + 0.5(E − Si )λ̃+3 ;
i0 i i
λ̃+3 = 0.5(E − Si )λ̃ + 0.5(E + Si )λ̃+3 .
With the mentioned h 0 arrangement, and i if the case of initial basis
0 0
of variables is λ̃1 T λ̃2 T . . . λ̃n s T , the initial basic matrix
may be determined as follows:
is certainly non-negative.
in Eq. (14), yT = sT µ− ,
siT
iT iT iT i
s̃0 s̄0 = S0 · · · S0i ;
0 = s̃0
h i and h are the nonnegative slack variables:
λiT = λ̃iT λ̃iT+3 λ̄
iT
,
hT = 0T h1T h2T . . . hn s T , with:
where
h i
hiT = h̃iT h̃iT+3 h̄
i
.
iT
λ̃ = λi1 λi2 λi3 ;
It may be seen from the equality (14) that the internal forces are
iT related to the slack variables h:
λ̃+3 = λi4 λi5 λi6 ;
i iT i
λ̄ = λi7 . . . λi16 . si = (ÑC )−1 (s̃i0 − h̃ ).
(15)
H. Van Long, N. Dang Hung / Engineering Structures 30 (2008) 1895–1904 1899
h 0 0
i
It can be shown that the slack variables h are identified exactly And the initial basic variables: λ̃1 T λ̃2 T . . . λ̃n s 0T .
as the reduced costs c̄ of the primal problem (8): The problems of Eqs. (10) and (16) are similar except for the
h = c̄ = (Xop
-1
(1, :))W∗ choice of the initial admissible point in the permissible domain
and the shakedown analysis requires preliminary calculation of
where X−1 op (1, :) is the first row of the inverses basic matrix at elastic responses.
optimal solution.
The reduced costs c̄ necessary for the convergence test of 3.3.3. Direct calculation of the residual internal force
the simplex algorithm are variable in the output of the primal distribution
calculation. The automatic computation by (15) of the internal Again the dual form of Eq. (16) is written similarly to Eq.
forces distribution is independent of the type of collapse: (14) with:
partial, complete or over-complete.
yT = ρ T µs− ; hT = 0T h1T h2T . . . hn s T
x∗T = π d0 λ η ;
in which, n = N /N p is ratio of the axial force to the squash
load, m y = M y /M py and m z = Mz /M pz are the ratios of
b∗T = 0 −Bd0 ξ ;
the major-axis and minor-axis moments to the corresponding
1 0T −sT0 0 plastic moments. This yield surface is already used in several
Fig. 2. Example a— Six-storey space frame ((a) perspective view, (b) plan
view).
(a) Example a.
(b) Example b.
Fig. 4. Deformation at limit state given by CEPAO (From the left to the right: Elastic–plastic first order, Elastic–plastic second order; Limit analysis; Shakedown
analysis, load domain a; Shakedown analysis, load domain b. The points on Fig. (a) indicate the plastic hinges.).
in the CEPAO (kinematic and static methods) and the good – In the case of symmetrical horizontal loading (seismic load
correlation between the Orbison’yield surface and this in AISC- or wind load), the load multipliers determined by shakedown
LRFD. analysis are the smallest (alternating plastic occurs).
1902 H. Van Long, N. Dang Hung / Engineering Structures 30 (2008) 1895–1904
of the axial displacement and the net rotation of the member in the elastic analysis. The other procedures are identical to
ends (Fig. A.1(a)). Assemble for the frameworks (system of the those of the elastic analysis, even when the P-delta effect is
elements) we have the vector e. taken into account. Therefore, in a brief presentation, we only
Let dTk = [d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 dek ] present the construction of the elastic–plastic matrix.
be the vector of the member independent displacements in Let 1eC and 1eR be the vectors of relative displacement
the global coordinate system OXYZ, as shown in Fig. A.1(b). increments at the yielded sections (plastic hinges), and at the
Assembled for the frameworks we obtain the vector d. elastic sections. Elastic constitutive equation (the Hooke’s low)
In the sense of limit analysis, we may think that: d1 , d2 , for the structure may be written as follows:
d3 , d7 , d8 , d9 are the displacements corresponding to the
1sR
DRR DRC
1eR − 0
deflection mechanisms (beam and sideways); d4 , d5 , d6 , d10 , = p , (B.1)
1sC DTRC DCC 1eC − 1eC
d11 , d12 are the displacements showing the joints mechanisms;
p
dek displacement in the longitudinal direction of the element, in which, 1eC are the plastic strain increments at the plastic
describes the bar mechanisms (the bar translates along this hinges, 1s R and 1sC are the vectors of internal force
axis). Since the torsional stiffness of the elements is negligible, increments at the elastic sections and the plastic hinges.
we must eliminate the degree of freedom that only provokes Based on the Drucker’s normality rule, plastic strain
pure torsion in the bars. increments are normal to the yield surface and orthogonal to
H. Van Long, N. Dang Hung / Engineering Structures 30 (2008) 1895–1904 1903
(a) Relative displacements at critical sections. (b) Member’s independent displacements (global
axis).
the internal force increments. Thus, for the sections (element Based on the upper bound theorem of limit analysis, we have
ends) in the plastic state, we have
µ+ ≤ µ∗ , (C.2)
pT
1eC 1sC = 0 and, (B.2a) with µ∗ is a load factor of any licit mechanism. By giving any
p
1eC = NC λ, (B.2b) licit displacement field d∗ (for example, only one component
equals unity, and all other components are nil), µ∗ may be
(λ and NC are defined in Eq. (1)). easily obtained.
From Eqs. (B.2a) and (B.2b) and noting that λ is arbitrary, From the Eqs. (C.1) and (C.2), one has
we obtain
φ/ξ ≤ µ∗ . (C.3)
T
NC 1sC = 0. (B.3)
On the point of view of geometry (kinematic), with the real
Using (B.1) and (B.2b) and (B.3), the plastic deformation mechanism, d̄, there is at least a plastic deformation component,
magnitude may be deduced ē, such that:
λ = (NC DCC NC )−1 NC DRC (NC DCC NC )−1 NC
T T T T T
DCC ē ≥ d̄max /Hmax ,
1eR
, with Hmax is the maximum dimension of the structure.
1eC
Therefore, a lower bound of the internal power may be
or: evaluated.
1eR
0 0 0 φ ≥ s p min d̄max /Hmax , (C.4)
= . (B.4)
λ R1 R2 1eC
in which, s p min is the smallest among the plastic capacity (N P ,
Eq. (B.2b) may be rewritten in the following form: M py , M pz ) of all the sections on the structure.
0
0 0
0 From the Eqs. (C.3) and (C.4), the maximum displacement
p = . (B.5) is constrained by an upper bound.
1eC 0 NC λ
Substituting (B.4) in (B.5), one obtains d̄max ≤ ξ µ∗ Hmax /s p min .
Then, any value of d0 that satisfies
1eR
0 0 0
p = . (B.6)
1eC NC R1 NC R2 1eC d0 ≥ ξ µ∗ Hmax /s p min ≥ d̄max ,
From (B.1) and (B.6), one finally obtains the elastic–plastic will lead: d0 = d̄ + d0 is always nonnegative.
constitutive relation: With similar argument, the value of d0 for the shakedown
1sR 1eR analysis may be obtained.
DRR − DRC NC R1 DRC − DRC NC R2
= .
1sC DTRC − DCC NC R1 DCC − DCC NC R2 1eC
References
Appendix C. Determination of the value of d0 [1] Hodge PG. Plastic analysis of structures. New York: McGraw Hill; 1959.
[2] Neal BG. The plastic method of structural analysis. London: Chapman &
Suppose that d̄ is the real displacement field (the real Hall; 1970.
[3] Massonnet Ch, Save M. Calcul plastique des constructions, vol. 1.
mechanism), in which, d̄max is the largest (absolute value). In Belgique: Nelissen; 1976.
case of limit analysis, the safety factor is determined by the [4] Cohn MZ, Maier G. Engineering plasticity by mathematical program-
equilibrium between the internal power and the external power. ming. Canada: University of Waterloo; 1979.
[5] Smith DL, editor. Mathematical programming method in structural
µ+ = sT0 λ/fT d = φ/ξ (C.1) plasticity. New York: Springer-Verlag; 1990.
[6] Jirásek M, Bažant ZP. Inelastic analysis of structure. John Wiley & Sons,
where the symbols are defined in Eq. (9). LTD; 2001.
1904 H. Van Long, N. Dang Hung / Engineering Structures 30 (2008) 1895–1904
[7] Parimi SR, Ghosh SK, Cohn MZ. The computer programme DAPS for the [13] Kim SE, Park MH, Choi SH. Direct design of three-dimensional frames
design and analysis of plastic structures. Solid Mechanics Division Report using advanced analysis. Engineering Structures 2001;(23):1491–502.
26. Canada: University of Waterloo; 1973. [14] Chiorean CG, Barsan GM. Large deflection distributed plasticity analysis
[8] Domaszewski M, Stanislawska EMS. Optimal shakedown design of of 3D steel frameworks. Computers & Structures 2005;(83):1555–71.
frames by linear programming. Computers & Structures 1985;21(3): [15] Cuong NH, Kim SE, Oh JR. Nonlinear analysis of space frames using
379–85. fiber plastic hinge concept. Engineering Structures 2006;(29):649–57.
[9] Dang Hung Nguyen. CEPAO-an automatic program for rigid-plastic and [16] Liew JYR, Chen H, Shanmugam NE. Inelastic analysis of steel frames
elastic–plastic, analysis and optimization of frame structure. Engineering with composite beams. ASCE Journal Engineering Structures 2001;
Structures 1984;(6):33–50. 127(2):194–202.
[10] Dang Hung Nguyen. Sur la plasticité et le calcul des états limites par [17] Jiang XM, Chen H, Liew JYR. Spread-of-plasticity analysis of three-
élément finis. Thèse de doctorat spécial. l’Université de Liège; 1984. dimensional steel frames. Journal of Constructional Steel Research 2002;
[11] de Saxcé G, Ohandja LMA. Une méthode automatique de calcul de l’effet (58):193–212.
P - Delta pour l’analyse pas - à - pas des ossatures planes. Construction [18] Orbison JG, McGUIRE W, Abel JF. Yield surface applications in
métallique. 1985 (3). nonlinear steel frame analysis. Computer Method in Applied Mechanics
[12] Liew JYR, Chen H, Shanmugam NE, Chen WF. Improved nonlinear and Engineering 1982;(33):557–73.
plastic hinge analysis of space frame structures. Engineering Structures [19] AISC. Load and resistance factor design specification for steel buildings.
2000;(22):1324–38. Chicago: American Institute of Steel Construction; 1993.