IFEM Ch04
IFEM Ch04
IFEM Ch04
The Direct
Stiness Method:
Additional Topics
41
42
TABLE OF CONTENTS
Page
4.1.
4.2.
4.3.
4.4.
4.5.
4.
4.
42
. . . . . . .
. . . . . . .
. . . . . . .
. .
.
. .
.
. .
.
. .
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
.
. .
.
. .
.
. .
.
. .
. . . . . . .
. . . . . . .
43
43
44
45
46
46
48
49
49
410
410
410
411
413
413
414
414
415
43
4.1
Chapters 2 and 3 presented the core steps of the Direct Stiffness Method (DSM). Those steps were
illustrated with the hand analysis of a plane truss structure. This Chapter covers some topics that
were left out from Chapters 23 for clarity. These include: the imposition of prescribed nonzero
displacements, and the treatment of thermal effects.
4.1. PRESCRIBED NONZERO DISPLACEMENTS
The support conditions considered in the example truss of Chapters 23 resulted in the specication
of zero displacement components; for example u y2 = 0. There are cases, however, where the known
value is nonzero. This happens, for example, in the study of settlement of foundations of ground
structures such as buildings and bridges, and in the analysis of driven machinery components.
Mathematically these are called non-homogenous boundary conditions. The treatment of this
generalization of the FEM equations is studied in the following subsections.
4.1.1. Application of DBCs by Reduction
We describe rst a reduction technique, analogous to that explained in 3.2.1, which is suitable for
hand computations. Recall the master stiffness equations (3.20) for the example truss:
20
10
10
10
10
10
10
0
0
10
10
10
0
10
0
0
0
0
0
0
5
0
5
10
10
0
0
10
10
u x1
10
10 u y1
0 u x2
=
5 u y2
10
u x3
15
u y3
f x1
f y1
f x2
f y2
f x3
f y3
(4.1)
Suppose that the applied forces are as for the example truss but the prescribed displacements are
u x1 = 0,
u y1 = 0.5,
u y2 = 0.4
(4.2)
This means that joint 1 goes down vertically whereas joint 2 goes up vertically, as depicted in
Figure 4.1. Inserting the known data into (4.1) we get
20
10
10
10
10
10
10
0
0
10
10
10
0
10
0
0
0
0
0
0
5
0
5
10
10
0
0
10
10
10
0
10 0.5
0 u x2
=
5 0.4
10
u x3
15
u y3
f x1
f y1
f y2
2
1
(4.3)
The rst, second and fourth rows of (4.3) are removed, leaving only
0
0.5
0
0
u
10 x2 = 2
0.4
1
15
u x3
u y3
10
0
10
10 10 0
10 10 0
0
0
5
0
10
10
43
(4.4)
44
Columns 1, 2 and 4 are removed by transferring all known terms from the left to the right hand side:
0
(10) 0 + 0 (0.5) + 0 0.4
0
= 2
= 3 .
(10) 0 + (10) (0.5) + 0 0.4
1
(10) 0 + (10) (0.5) + (5) 0.4
2
(4.5)
The matrix equation (4.5) is the reduced system. Note that its coefcient matrix is exactly the same
as in the reduced system (3.23) for prescribed zero displacements.
10
0
0
0
10
10
0
10
15
u x2
u x3
u y3
=
0
0.5 .
0.2
(4.6)
Filling the missing slots with the known values (4.2) yields the complete displacement
solution (listed as row vector to save space):
u = [0
0.5
0.4
0.5
fx3 = 2
0.2 ]T .
(4.7)
1
u y1 = 0.5
going down
u y2 = +0.4
going up
;;
;;
u x2
u x3
u y3
;;
;;
fy3 = 1
ux1 = 0
no horizontal motion
Going through the postprocessing steps discussed in 3.3 with (4.7), we can nd that the reaction
forces and the internal member forces do not change. This is a consequence of the fact that the
example truss is statically determinate. The force systems (internal and external) in such structures
are insensitive to movements such as foundation settlements.
4.1.2. *Application of DBCs by Modification
The computer-oriented modication approach follows the same idea outlined in 3.5.2. As there, the main
objective is to avoid rearranging the master stiffness equations. To understand the process it is useful to think
of being done in two stages. First equations 1, 2 and 4 are modied so that they become trivial equations, as
illustrated for the example truss and the support conditions (4.2):
1
0
10
0
1
0
0
10 10
10 10
0
0
10
0
0
0
0
0
0
1
0
5
0
0
0
0
10
10
0
u x1
0
0 u x2 0.5
0
u x2 = 0
0 u y2 0.4
u x3
10
2
u y3
15
1
(4.8)
The solution of this system recovers (4.2) by construction (for example, the fourth equation is simply 1u y2 =
0.4). In the next stage, columns 1, 2 and 4 of the coefcient matrix are cleared by transferring all known terms
44
45
to the right hand side, following the same procedure explained in (4.5). We thus arrive at
1 0 0 0
0 1 0 0
0 0 10 0
0 0 0 1
0
0
0
0
0
0
0
0
0
0
0
0
10
10
0
u x1
0
0 u x2 0.5
0
u x2 = 0
0 u y2 0.4
u x3
10
3
u y3
15
2
(4.9)
As before, this is called the modied master stiffness system. Observe that the equations retain the original
order. Solving this system yields the complete displacement solution (4.7).
Note that if all prescribed displacements are zero, forces on the right hand side are not modied, and one
would recover (3.30).
REMARK 4.1
The modication is not actually programmed as discussed above. First the applied forces in the right-hand
side are modied for the effect of nonzero prescribed displacements, and the prescribed displacements stored
in the reaction-force slots. This is called the force modication procedure. Second, rows and columns of the
stiffness matrix are cleared as appropriate and ones stored in the diagonal positions. This is called the stiffness
modication procedure. It is essential that the procedures be executed in the indicated order, because stiffness
terms must be used to modify forces before they are cleared.
4.1.3. *Matrix Forms of DBC Application Methods
The reduction and modication techniques for applying DBCs can be presented in compact matrix form. The
free-free master stiffness equations Ku = f are partitioned as follows:
K11
K21
K12
K22
u1
u2
f1
.
f2
(4.10)
In this matrix equation, subvectors u2 and f1 collect displacement and force components, respectively, that are
known, given or prescribed. On the other hand, subvectors u1 and f2 collect force and displacement components,
respectively, that are unknown. The force components in f2 are reactions on supports; consequently f2 is called
the reaction force vector.
On transferring the known terms to the right hand side the rst matrix equation becomes
K11 u1 = f1 K12 u2 .
(4.11)
This is the reduced master equation system. If the displacement B.C. are homogeneous (that is, all prescribed
displacements are zero), u2 = 0, and we do not need to change the right-hand side:
K11 u1 = f1 .
(4.12)
Examples that illustrate (4.11) and (4.12) are (4.5) and (3.23), respectively.
The computer-oriented modication technique retains the same joint displacement vector as in (4.10) through
the following rearrangement:
K11 0
u1
f1 K12 u2
=
,
(4.13)
u2
u2
0
I
This modied system is simply the reduced equation (4.11) augmented by the trivial equation Iu2 = u2 . This
system is often denoted as
Ku = f.
(4.14)
45
46
Solving (4.14) yields the complete displacement solution including the specied displacements u2 .
For the computer implemenmtation it is important to note that the partitioned form (4.10) is used only to
facilitate the use of matrix notation. The equations are not explicitly rearranged and retain their original
numbers. For instance, in the example truss
u1 =
u x1
u y1
u y2
DOF #1
DOF #2
DOF #4
,
u2 =
u x2
u x3
u y3
DOF #3
DOF #5
DOF #6
(4.15)
The example shows that u1 and u2 are generally interspersed throughout u. Thus, matrix operations such as
K12 u2 required indirect (pointer) addressing to avoid explicit array rearrangements.
Called initial stress or initial strain effects by many authors. These names reect what is viewed as the physical source
of initial force effects at the continuum level.
46
47
4.2
(a)
f yi , u yi
f xi , u xi
THERMOMECHANICAL EFFECTS
fyj , uyj
y
x
(e)
fx j , ux j
i
, T
ks = E A/L
(b)
d = d M + dT
L + d M + dT
Figure 4.2. Generic truss member subjected to mechanical and thermal effects:
(4.17)
An assumption justied if the temperature changes are small enough so that is approximately constant through the
range of interest, and no material phase change effects occur.
47
F = fxi
48
F = fxj
p = p M + pT
Now suppose that the member is also subject to mechanical forces, more precisely the applied axial
force F shown in Figure 4.2. The member axial stress is = F/A. In response to this stress
the length changes by d M . The mechanical strain is e M = d M /L. The total strain e = d/L =
(d M + dT )/L is the sum of the mechanical and the thermal strains:
e = e M + eT =
+ T
E
(4.18)
This superposition of deformations is the basic assumption made in the thermomechanical analysis.
It is physically obvious for an unconstrained member such as that depicted in Figure 4.2.
At the other extreme, suppose that the member is completely blocked against axial elongation; that
is, d = 0 but T = 0. Then e = 0 and e M = eT . If > 0 and T > 0 the blocked member
goes in compression because = Ee M = EeT = E T < 0. This thermal stress is further
discussed in Remark 4.2.
4.2.2. Thermomechanical Stiffness Equations
Because e = d/L and d = u x j u xi , (4.18) can be developed as
u x j u xi
=
+ T,
L
E
(4.19)
(4.20)
Here p M = A denotes the mechanical axial force, and pT = E A T , which has the dimension
of a force, is called (not surprisingly) the internal thermal force. The sum p = p M + pT is called
the effective internal force. The last relation in (4.20), F = p = p M + pT follows from free-body
member equilibrium; see Figure 4.3. Passing to matrix form:
F=
EA
[ 1
L
u xi
u
0 ] yi .
u yi
u y j
(4.21)
Noting that F = fx j = fxi while fyi = fy j = 0, we can relate joint forces to joint displacements
as
48
49
4.2
f xi
F
fyi 0
f = F =
xj
fy j
THERMOMECHANICAL EFFECTS
fM xi
1
1
E
A
f M yi
0
0
+ E T
=
1
fM x j
1
L
fM y j
0
0
0
0
0
0
1 0
u xi
0 0 u yi
,
1 0 u x j
u y j
0 0
(4.22)
u,
or
In compact matrix form this is f = f M + fT = K
u = f M + fT .
K
(4.23)
is the same member stiffness matrix derived in 2.6.3. The new ingredient that appears is
Here K
the vector
1
0
fT = E A T
(4.24)
,
1
0
This is called the vector of thermal joint forces in local coordinates. It is an instance of an initial
force vector at the element level.
REMARK 4.2
A useful physical interpretation of (4.23) is as follows. Suppose that the member is completely blocked against
joint motions so that u = 0. Then f M + fT = 0 or f M = fT . It follows that fT contains the negated joint
forces (internal forces) that develop in a heated or cooled bar if joint motions are precluded. Because for
most materials > 0, rising the temperature of a blocked bar that is, T > 0 produces an internal
compressive thermal force pT = AT = E AT , in accordance with the expected physics. The quantity
T = E T is the thermal stress. This stress can cause buckling or cracking in severely heated structural
members that are not allowed to expand or contract. This motivates the use of expansion joints in pavements,
buildings and rails, and roller supports in long bridges.
4.2.3. Globalization
At this point we restore the member superscript so that the member stiffness equations(4.22) are
rewritten as
(e) u (e) = f(e)
(e)
K
(4.25)
M + fT .
Use of the transformation rules developed in 3.1 to change displacements and forces to the global
system {x, y} yields
(e)
K(e) u(e) = f(e)
(4.26)
M + fT ,
where T(e) is the displacement transformation matrix (3.1), and the transformed quantities are
T (e) (e)
T ,
K(e) = T(e) K
(e) T (e)
f ,
f(e)
M = T
(e) T (e)
fT .
f(e)
T = T
(4.27)
These globalized member equations are used to assemble the free-free master stiffness equations
by a member merging process.
49
410
y // y
P = 90
(2)
(1)
x // x
;
;
;;
;;
;
;
;
(1)
=4
(2)
L =6
4.2.4. Merge
The merge process is based on the same assembly rules stated in 3.1.3 with only one difference:
thermal forces are added to the right hand side. The member by member merge is carried out
much as described as in 3.1.4, the main difference being that the thermal force vectors f(e)
T are also
4
merged into a master thermal force vector. Upon completion of the assembly process we arrive at
the free-free master stiffness equations
Ku = f M + fT = f.
(4.28)
4.2.5. Solution
The master system (4.28) has formally the same conguration as the master stiffness equations (2.3).
The only difference is that the effective joint force vector f contains a superposition of mechanical
and thermal forces.
Displacement boundary conditions can be applied by reduction or modication of these equations,
simply by using effective joint forces in the descriptions of 3.2.1, 3.4.1 and 4.1. Processing the
reduced or modied system by a linear equation solver yields the displacement solution u.
4.2.6. Postprocessing
The postprocessing steps described in 3.4 require some modications because the derived quantities of interest to the structural engineer are mechanical reaction forces and internal forces. Effective
forces by themselves are of little use in design.
Mechanical joint forces including reactions are recovered from
f M = Ku fT
(4.29)
To recover mechanical internal forces in member (e), obtain p (e) by the procedure outlined in
3.4.2, and subtract the thermal component:
(e)
E (e) A(e) (e) T (e) .
p (e)
M = p
(e)
The mechanical axial stress is then (e) = p (e)
M /A .
4
410
(4.30)
411
f(1)
T1
fT(1)
2
= E A T
(1)
(2)
1
150
=
, fT =
1
150
f(2)
T2
fT(2)
3
= E A T
(2)
1
60
=
.
1
60
(4.31)
3000
1 1
1 1
u (1)
1
u (1)
2
f(1)
M1
(1)
fM2
150
,
150
2000
1 1
1 1
u (2)
2
u (2)
3
f(2)
M2
(2)
fM3
60
, (4.32)
60
No globalization is needed because the equations are already in the global system, and thus we can get rid of
the local symbols: f f , u u. Assembling:
1000
3 3 0
3 5 2
0 2 2
u1
u2
u3
=
f M1
f M2
f M3
+
150
150 + 60
60
=
f M1
f M2
f M3
+
150
210
60
(4.33)
The displacement boundary conditions are u 1 = u 3 = 0. The mechanical force boundary condition is f M2 =
90. On removing the rst and third equations, the reduced system is 5000 u 2 = f M2 + 210 = 90 + 210 = 300,
which yields u 2 = 300/5000 = +0.06. The mechanical internal forces in the members are recovered from
E (1) A(1)
(u 2 u 1 ) E (1)A(1) (1) T (1) = 3000 0.06 12000 0.0004 25 = 60,
L (1)
E (2) A(2)
=
(u 3 u 2 ) E (2)A(2) (2) T (2) = 2000 (0.06) 12000 0.0004 (10) = 72,
p (2)
M
L (2)
(4.34)
(1)
(2)
whence the stresses are = 60/12 = 5 and = 72/12 = 6. Member (1) is in tension and member
(2) in compression.
p (1)
M =
411
412
The thermal forces for each member in global coordinates are obtained by using (4.25) and the third of (4.27):
f(1)
T
f(2)
T
f(3)
T
0 0 0
1
1
1 0 0 0
0
= 100 T
,
0 1 0 1
1
0 0 1
0
0
1 0
0
1
0
0 0
0 0
1
= 50 T
,
0 0 1 1
0
0 1
0
0
1
1 1 0
0
1
1
1
1 0
0 0
1
1
= E (3) A(3) (3) T (3)
1 = 200 T 1 .
0
0
1
1
2
0
0 1
1
0
1
1
0
0
(2) (2) (2)
(2) 1
= E A T
0
0
(4.35)
Merging the contribution of these 3 members gives the master thermal force vector
100 + 0 200
0 + 0 200
100 + 0 + 0
fT = T
0 50 + 0
0 + 0 + 200
0 + 50 + 200
300
200
= T 100
50
(4.36)
200
250
The master stiffness matrix K does not change. Consequently the master stiffness equations are
20 10 10 0 10 10 u = 0 f
300
x1
M x1
10 10 0 0 10 10 u y1 = 0 f M y1
200
10 0 10 0 0
u x2 f M x2 = 0
=
+ T 100
0
50
0
0
5
0 5 u y2 = 0 f M y2
10 10
10 10
0
0
0 10
5 10
f M x3 = 0
f M y3 = 0
u x3
u y3
10
15
(4.37)
200
250
in which f M x1 , f M y1 and f M y2 are the unknown mechanical reaction forces, and the known forces and displacements have been marked. Since the prescribed displacements are zero, the reduced system is simply
10 0
0 10
0 10
0
10
15
u x2
u x3
u y3
=
0
0
0
+ T
100
200
250
= T
100
200
250
(4.38)
Solving (4.38) gives u x2 = u x3 = u y3 = 10 T . Completing u with the prescribed zero displacements and
premultiplying by K gives the complete effective force vector:
20 10 10 0 10 10 0
300
10 10 0 0 10 10 0
200
10 0 10 0 0
10
0
T = T 100 .
f = Ku =
0
50
0
0
5
0 5 0
10 10
10 10
0
0
0 10
5 10
10
15
10
10
(4.39)
200
250
412
(4.40)
413
4.4
PSEUDOTHERMAL INPUTS
All mechanical joint forces, including reactions, vanish, and so do the internal mechanical forces. This is
a consequence of the example frame being statically determinate.5 Such structures do not develop thermal
stresses under any combination of temperature changes.
(4.42)
In these equations subscripts M and I identify mechanical and initial node forces, respectively. The
sum of the two: f at the member level and f at the structure level, are called effective forces.
A physical interpretation of (4.42) can be obtained by considering that the structure is blocked
against all motions: u = 0. Then f M = f I , and the undeformed structure experiences mechanical
forces. These translate into internal forces and stresses. Engineers also call these prestresses.
Local effects that lead to initial forces at the member level are: temperature changes (studied in 4.2,
in which f I fT ), moisture diffusion, residual stresses, lack of t in fabrication, and in-member
prestressing. Global effects include prescribed nonzero joint displacements (studied in 4.1) and
multimember prestressing (for example, by cable pretensioning of concrete structures).
As can be seen there is a wide variety of physical effects, whether natural or articial, that lead to
nonzero initial forces. The good news is that once the member equations (4.41) are formulated, the
remaining DSM steps (globalization, merge and solution) are identical. This nice property extends
to the general Finite Element Method.
4.4. PSEUDOTHERMAL INPUTS
Some commercial FEM programs do not have a way to handle directly effects such as moisture,
lack of t, or prestress. But all of them can handle temperature variation inputs. Since in linear
analysis all such effects can be treated as initial forces, it is possible (at least for bar elements) to
model them as ctitious thermomechanical effects, by inputting phony temperature changes. The
following example indicate that this is done for a bar element.
Suppose that a prestress force FP is present in a bar. The total elongation is d = d M + d P where
d P = FP L/(E A) is due to prestress. Equate to a thermal elongation: dT = TP L and solve for
TP = FP /(E A). This is input to the program as a ctitious temperature change. If in addition
there is a real temperature change T one would of course specify T + TP .
If this device is used, care should be exercised in interpreting results for internal forces and stresses
given by the program. The trick is not necessary for personal or open-source codes over which you
have full control.
5
For the denition of static determinacy, see any textbook on Mechanics of Materials.
413
414
B. A. Boley and J. H. Wiener, Theory of Thermal Stresses, Wiley, New York, 1960.
sik, Boundary Value Problems of Heat Conduction, Dover edition, 1989.
M. N. Ozi
J. S. Przemieniecki, Theory of Matrix Structural Analysis, McGraw-Hill, New York, 1968; Dover
edition 1986.
R. J. Roark, R. G. Budynas and W. C. Young, Roarks Formulas for Stress and Strain McGraw-Hill,
New York, 7th ed., 2001.
414
415
Exercises
EXERCISE 4.1
[N:20] Resolve items (a) through (c) omitting (d) of the problem of Exercise 3.7 if the vertical right
support sinks so that the displacement u y3 is now prescribed to be 0.5. Everything else is the same. Use
a reduction scheme to apply the displacement BCs.
EXERCISE 4.2
[N:20] Use the same data of Exercise 3.7 except that P = 0 and hence there are no applied mechanical forces.
Both members have the same dilatation coefcient = 106 1/ F. Find the crown displacements u x2 and u y2
and the member stresses (1) and (2) if the temperature of member (1) rises by T = 120 F above Tr e f ,
whereas member (2) stays at Tr e f .
Shortcut: the element stiffnesses and master stiffness matrix are the same as in Exercise 3.7, so if that Exercise
has been previously assigned no stiffness recomputations are necessary.
EXERCISE 4.3
[A:15] Consider the generic truss member of Figure 2.6. The disconnected member was supposed to have
length L, but because of lack of quality control it was fabricated with length L + , where is called the lack
of t. Determine the initial force vector f I to be used in (4.41). Hint: nd the mechanical forces that would
compensate for and restore the desired length.
EXERCISE 4.4
[A:10] Show that the lack of t of the previous exercise can be viewed as equivalent to a prestress force of
(E A/L).
EXERCISE 4.5
[A:40]. Prove that statically determinate truss structures are free of thermal stresses.
415