Sei sulla pagina 1di 11

4

PMP Dynamical
Systems:
Matrix EOM Formulation

41

42

Chapter 4: PMP DYNAMICAL SYSTEMS: MATRIX EOM FORMULATION

TABLE OF CONTENTS
Page

4.1.
4.2.

4.3.

4.4.
4.

Matrix Equations of Motion


4.1.1. Approaches to EOM Formulation . . . .
EOM Derivation by Force Equilibrium
4.2.1. PMP Dynamic System Example . . . . .
4.2.2. Equilibrium Equations
. . . . . . .
4.2.3. Matrix Form . . . . . . . . . . .
4.2.4. Matrix Symmetry Check . . . . . . .
4.2.5. Center of Mass Motions Check . . . . .
4.2.6. PMP Equations of Motion for Numerical Data
EOM Derivation by the Finite Element Method
4.3.1. Element-Level Calculations . . . . . .
4.3.2. Assembly of Master Mass Matrix . . . .
4.3.3. The FEM Equations of Motion . . . . .
Comparison Between Formulation Approaches
Notes and Bibliography
. . . . . . . . . . . . . . .

42

. . . . . . .
. .
. .
. .
. .
. .
.

. .
. .
. .
. .
. .
. .

. .
. . .
. .
. . .
. .
. . .

. . . . . .
. . . . . . .
. . . . . .
. . . . . . .

43
43
43
43
45
46
46
47
47
48
48
49
49
410
411

43

4.2

EOM DERIVATION BY FORCE EQUILIBRIUM

4.1. Matrix Equations of Motion


This Chapter describe how to construct the dynamic equations of motion (EOM) in matrix form.
Several simplifying assumptions are made for clarity of exposition:
The dynamical model is a point-mass particle (PMP) system, with point masses linked by
collinear interaction forces.
Motions from the reference configuration are small in the sense that the initial geometry can be
used throughout
Interaction forces depend linearly on displacements and may be idealized as extenssional
springs. Damping effects are ignored although they are briefly covered in other Chapters.
The last two assumptions ensure that the EOM are linear in displacements and accelerations.
The first assumption (dynamic model is PMP) is partially lifted when considering the equivalence
between two ways of constructing the matrix EOM: by equilibrium or via FEM, as outlined next.
4.1.1. Approaches to EOM Formulation
Sections 4.2 and 4.3 cover two approaches to the formulation of matrix EOM:
Equilibrium Method. Writing down force balance expressions for each point-mass particle via Free
Body Diagrams (FBD) that account for dynamic effects modeled as forces.
Finite Element Method. Assembling element-level EOM provided by a FEM discretization.
The two methods are quite different in philosophy as well as procedure. The equilibrium method
reflects the direct application of Newtons laws to free bodies isolated from the rest of the system
by appropriate external and internal forces. Technically those are known as Free Body Diagrams
or FBD. By contrast, FEM equations are generally derived in a variational framework, such as the
Principle of Virtual Work, Lagranges equations, or Hamiltons principle.
Nonetheless, for the PMP systems considered in this Chapter, the two methods produce exactly
the same matrix EOM. Why then bother to go over both? The answer is that each has distinct
advantages as well as shortcomings. Those are contrasted in 4.4.
4.2. EOM Derivation by Force Equilibrium
The force equilibrium approach to deriving matrix EOM will be illustrated in the two-dimensional
PMP system pictured in Figure 4.1. A reader familiar with IFEM may note its similarity with the
example truss used in [106, Chapters 23]. That is not accidental: the stiffness matrix equations
derived there will be reused for part of the EOM derivation via FEM carried out in 4.3.
4.2.1. PMP Dynamic System Example
The two-dimensional PMP system of Figure 4.1(a) is referred to a fixed RCC frame {x, y}, with
origin as shown in the Figure. The geometry is defined by the reference position of the three point
masses, labeled 1,2 and 3.
43

Chapter 4: PMP DYNAMICAL SYSTEMS: MATRIX EOM FORMULATION

fy3 , uy3
m3 = 4

(a)

3(10,10)
45

k(3) = 20
(3)

y
1(0,0)
45o
m1 = 5 x

(b)

(1)
k(1) = 10

(2)

44

fx3 , ux3

45o

k(2) = 5
(3)
2(10,0)

fx1 , ux1

(2)

(1)

45o

fx2 , ux2
2
fy2 , uy2

1
fy1 , uy1

m2 = 3

External forces:
fx3ext=2H(t), fy3ext=H(t),
others zero. Here
H(t) is the Heaviside
unit-step function

Figure 4.1. Example PMP system to illustrate derivation of matrix EOM by force equilibrium:
(a) reference geometry and physical properties; (b) kinematic DOF and associate forces.

Point masses are connected by linearly elastic extensional springs, which are labelled as (1), (2)
and (3) for convenience.1 The springs possess extensional stiffnesses k (1) , k (2) and k (3) with values
indicated as in the Figure. The system only moves on the {x, y} plane. Consequently it has 6
kinematic degrees of freedom (DOF), which are taken to be the point mass displacements in the
coordinate directions. These DOF are collected in a system displacement 6-vector with a mass-bymass arrangement:
(4.1)
u = [ u x1 u y1 u x2 u y2 u x3 u y3 ]T .
The velocity and acceleration 6-vectors are configured accordingly as
u = [ u x1

u y1

u x2

u y2

u x3

u y3 ]T ,

u = [ u x1

u y1

u x2

u y2

u x3

u y3 ]T .

f x3

f y3 ]T

(4.2)

The 6-vector of generic forces associated to (4.1) is


f = [ f x1

f y1

f x2

f y2

(4.3)

Unlike displacements, several kinds of forces may be considered: external, internal, interaction,
constraint (also called reaction), inertial, effective, residual, and so on. When there is need to distinguish force type, the generic symbols of (4.3) are adorned with appropriate 3-letter superscripts.
For example,
ext
ext
ext
ext
ext
ext T
f ext = [ f x1
f y1
f x2
f y2
f x3
f y3
] ,
(4.4)
int
int
int
int
int
int T
f int = [ f x1
f y1
f x2
f y2
f x3
f y3
] ,
denote vectors of external and internal forces, respectively, for the example system. To distinguish
internal from interaction, the latter is identified with superscript iac, as in f iac . For consistency
with NFEM [107], residual forces may be denoted by r instead of f r es to reduce clutter.
1

This follows the element-identification convention of IFEM [106, Chapter 2], since in 4.3 the springs of Figure 4.1 are
shown to be equivalent to bar elements in a truss FEM model. Enclosing parentheses distinguish these from point-mass
labels, while avoiding confusion with exponents when placed as superscripts of spring properties.

44

45

4.2

EOM DERIVATION BY FORCE EQUILIBRIUM

fy3ext

(a)

3
f s(3)

(3)
f s = k(3) (ux3 cos

45 o + uy3 sin
o

m3

(b)

(2)

1
m1

f s(1)

ext

fs

f s(3)
f s(1)

(1)
f s = k(1)(ux2 ux1 )

..

m 3 u y3

f s(2)= k(2)(uy3 uy2 )

ux2 cos 45 uy2 sin 45 )

2
m2

..

m 1 u x1

fx3ext
f s(2)

f s(3)

f s(2)

45o
o

..
m 3u x3

ext
fx1

fy1 f (3)
s
f s(1)

1 m u..y1
1

..

m 2 u x2

f s(1)
ext

fy2

f s(2)
ext
fx2

2
..

m 2 u y2
Figure 4.2. Auxiliary diagrams in equilibrium EOM derivation for PMP system of Figure 4.1: (a) springinteraction forces in terms of displacements; (b) point-mass FBD diagrams. External, spring-interaction
and inertial forces are pictured in black, blue and red, respectively, for visualization clarity.

Derivations are carried out first for a floating PMP system, which has no motion constraints. In
Chapter 5 the EOM will be modified to account for kinematic constraints of various types.
4.2.2. Equilibrium Equations
Begin by disconnecting the point-masses. Replace the springs by interaction forces f s(1) , f s(2)
and f s(3) . These are depicted in Figure 4.2(a), in which their expressions in terms of point-mass
displacements is listed. The + sense of the interaction forces is governed by an easy-to-remember
convention: assume that each spring is in tension. To complete the Free Body Diagrams (FBD)
shown in Figure 4.2(b), two more force sets are added:
ext
ext
The external forces f x1
through f y3
.
ine
ine
The inertial forces2 f x1
= m 1 u x1 through f y3
= m 3 u y3 . Their displayed sense is dictated
by a simple rule: the pertinent acceleration is assumed positive. Since masses are positive,
these forces point in the opposite sense of a + acceleration. See Figure 4.2(b).

The three force sets: external, interaction and inertial, are pictured in different colors in Figure 4.2(b)
for easier visualization. Next, force equilibrium conditions for each degree of freedom (DOF)
are written down. Interaction and inertia forces are expressed in terms of displacements and
accelerations. For example, force balance of the motion of mass 1 along x gives

ext
ine
+ f s(1) + f s(3) cos 45 + f x1
f x1 = f x1

ext
= f x1
+ k (1) (u x2 u x1 ) + k (3) u x3 cos 45 + u y3 sin 45
(4.5)

u x1 cos 45 u y1 sin 45 cos 45 m 1 u x1


ext
= f x1
+ k (1) (u x2 u x1 ) + 12 k (3) u x3 +u y3 u x1 u y1 m 1 u x1 = 0.
2

Also known as DAlembert forces, pseudo forces, or effective forces in the literature.

45

Chapter 4: PMP DYNAMICAL SYSTEMS: MATRIX EOM FORMULATION

46

At this point it is convenient to revert the sign of the LHS of (4.5) so as to make the inertia term
m 1 u x1 positive. While doing this, common displacements are collected, and the external force
term passed to the RHS. The rearranged expression is
ext
m 1 u x1 + (k (1) + 12 k (3) ) u x1 k (1) u x2 + 12 k (3) u y1 12 k (3) u x3 12 k (3) u y3 = f x1
.

(4.6)

Following exactly the same pattern, five more balance equations are obtained. The end result is six
equilibrium equations, one for each DOF.
4.2.3. Matrix Form
Once the equilibrium equations for all DOF are obtained and appropriate arranged as discussed
above, several steps are carried out to put them in matrix form:
The external forces, which are givens, are kept in the RHS and arranged as a vector fext ,
configured as per the first of (4.4).
LHS terms that depend on accelerations are organized as a diagonal master mass matrix M
postmultiplied by the acceleration vector u configured as in the second of (4.2).
LHS terms that depend on displacements are collected and organized as a master stiffness matrix
K postmultiplied by the displacement vector u configured as in (4.1).
The resulting matrix EOM have the compact form
M u + K u = f ext .
For the example system the mass and stiffness matrices are

m1 0
0
0
0
0
0
0
0
0 m1 0

0
0
0 m2 0
0
M=
= diag [ m1, m1, m2, m2, m3, m3 ].
0
0
0 m2 0
0

0
0
0
0 m3 0
0
0
0
0
0 m3
(1) 1 (3)

1 (3)
k + 2k
k
k (1)
0
12 k (3)
12 k (3)
2
1 (3)
1 (3)

k
k
0
0
12 k (3)
12 k (3)

2
2
(1)
(1)
k

0
k
0
0
0

K=
(2)
(2)

0
0
0
k
0
k

1
1
1
1
(3)
(3)
(3)
(3)
k

k
0
0
k
k
2
2
2
2
1 (3)
1 (3)
1 (3)
1 (3)
(2)
(2)
2k
2k
0
k
k
k + 2k
2

(4.7)

(4.8)

(4.9)

If the RHS of (4.7) consists entirely of external (applied, given) forces, we will often drop the
superscript, and write
M u + K u = f,
being tacitly understood that f contains only given forces.
46

(4.10)

47

4.2

EOM DERIVATION BY FORCE EQUILIBRIUM

4.2.4. Matrix Symmetry Check


Observe that both M and K in (4.8) and (4.9) are symmetric. For M that is not a surprise, since
real diagonal matrices are necessarily symmetric. Can that be expected for K? Yes: it is not an
accident. It previews the equivalence of this approach with the Finite Element Method, as covered in
4.3. Since FEM equations are automatically symmetric when derived in a variational framework,
equivalence implies symmetry. Should K come out unsymmetric, two fixes can be applied:
Reverse sign of individual equations (which become matrix rows) as necessary so that diagonal
entries of K and M are positive. If they are not of the same sign, look for derivation errors.
If diagonal signs check out but unsymmetry persists, try row scaling by appropriate factors.
4.2.5. Center of Mass Motions Check
After verifying symmetry, reduction to center-of-mass motion provides another useful EOM check.
Assume that x and y translations are rigid by setting u x1 = u x2 = u x3 = u x and u y1 = u y2 =
u y3 = u y for all t. Hence u x1 = u x2 = u x3 = u x and u y1 = u y2 = u y3 = u y . Replace these into
vectors u and u of (4.10). Add equations 1, 3, and 5, factoring u x and u x , and add equations 2, 4,
and 6, factoring u y and u y . The result is
M u x = Fx ,

M u y = F y ,

ext
ext
ext
+ f x2
+ f x3
in which M = m 1 + m 2 + m 3 is the total mass of the system, whereas Fx = f x1
ext
ext
ext
and F y = f y1
+ f y2
+ f y3
are the resultants of the x and y external forces, respectively. These
are the well known equations for the translational motion of the center of mass, which is located
at xC = 10(m 2 + m 3 )/M and yC = 10m 3 /M, as may be easily verified. A similar check can be
performed for an infinitesimal rotation about the center of mass: the result is IC = TC , where
IC is the mass moment about C and TC the torque of the external forces about C.

In these reductions the contribution of the stiffness term K u cancels out. This results from the
rigid-body property of the stiffness matrix: K u R = 0 if u R is a rigid body motion. A direct
verification can be done by extracting the eigenvalues of K and checking that its rank is 3 [106,
Chapter 20], but symbolic eigenvalue verification becomes unwieldy for larger systems.
4.2.6. PMP Equations of Motion for Numerical Data
Numerical values for point masses and spring constants given in Figure 4.1(a) are: m 1 = 5,
m 2 = 3, m 3 = 4, k (1) = 10, k (2) = 5, and k (3) = 20. For the external forces, Figure 4.1(b) states
ext
ext
ext
ext
ext
ext
= f y1
= f x2
= f y2
= 0, f x3
= 2H (t), and f y3
= H (t), in which H (t) is the
that f x1
Heaviside unit-step function. Replacing into the foregoing matrix expression yields

5
0

0
0

0
5
0
0
0
0

0
0
3
0
0
0

0
0
0
3
0
0

0
0
0
0
4
0


0
u x1
20
0 u y1 10


0 u x2 10

+
0 u y2 0


u x3
0
10
u y3
4
10

10 10 0
10
0
0
0
10 0
0
0
5
10 0
0
10 0 5
47

10
10
0
0
10
10

10
u x1
0
10 u y1 0

0 u x2 0

5 u y2 0

u x3
10
2H (t)
u y3
15
H (t)
(4.11)

Chapter 4: PMP DYNAMICAL SYSTEMS: MATRIX EOM FORMULATION

fy3 , uy3
3
(3)

(3)

=
2 2,

(3)

y
fx1 , ux1
1
fy1 , uy1

45o

External forces:
fx3ext=2H(t), fy3ext=H(t),
others zero. Here
H(t) is the Heaviside
unit-step function

(2)

E (2) = 50, A(2) = 1,


(2)
L = 10, (2) = 1/5

fx2 , ux2

x (1)
(1)

E (1) =
(1)

fx3 , ux3

45o

E = 100, A
(3)
2, (3) = 3/20
L = 10

48

50, A = 2,
L = 10, (1) = 1/5

2
fy2 , uy2

Figure 4.3. The three-member example truss of IFEM [106, Chapters 2-3] reproduced for
convenience. The dynamic model is (for now) floating: it has no supports and thus may
move freely on the {x, y} plane. It is subject to the external forces indicated in the box.

This is a linear system of six second-order ODEs in time. The problem specification is closed by
providing 12 initial conditions: six displacements and six velocities at a start time, which is often
taken to be t = 0. For example, the rest condition at t = 0 is specified by
u x1 (0) = u y1 (0) = . . . , u y3 (0) = 0,

u x1 (0) = u y1 (0) = . . . , u y3 (0) = 0.

(4.12)

A rest initial condition would be appropriate for a structure hit by a transient event at t = 0, e.g.,
earthquake, explosion or impact.
4.3. EOM Derivation by the Finite Element Method
The FEM model for the example truss of IFEM [106, Chapters 23] is reproduced in Figure 4.3.
One additional property is now activated: the mass density of the three elements (truss members).
This is denoted by e .
4.3.1. Element-Level Calculations
As typical of FEM, computations proceed element by element. In this case two element-level
matrices are generated: stiffness and mass. The element stiffness matrices are exactly those derived
in IFEM [106, Chapters 23]. The reader is referred there for details.
The mass matrices are produced by a nodal lumping scheme that proceeds as follows. The total
mass of element e is m e = e Ae L e , in which L e and Ae are the length and cross section area,
respectively. Assign one half of this to each end node, in both the x and y directions. As a result
the so-called lumped mass matrix of the truss (bar) element is
1 e

m
0
0
0
2
1 e
0
m
0
0
2
= 1 e A e L e I4 ,
(4.13)
Me =
1
e
0
0
m
0 2
2

0
48

1 e
m
2

49

4.3

EOM DERIVATION BY THE FINITE ELEMENT METHOD

(3)

m3 =3

(a)

m3 =1+3=4

(2)

m3 =1

(b)
(3)

(2)

(3)

m1 =3
1

(2)

(1)
(1)

m1 =2

(1)

m2 =2

m2 =1
m1 =3+2=5

m2 =2+1=3

Figure 4.4. Mass lumping procedure for the example truss of Figure ?: (a) element-level
lumping; (b) addition at nodes to form master mass matrix.

in which I4 denotes the identity matrix of order 4. What happens to Me if the axes are rotated by
an angle to be {x,
y }? It becomes
e = (Te )T Me Te = e Ae L e (Te )T I4 Te = e Ae L e (Te )T Te ,
M
where Te is a 44 transformation matrix very similar to that used in [106, 2.8]. But (Te )T Te = I4
e = Me for any . It follows that Me is invariant with respect to
because Te is orthogonal, and M
rotation of axes. Consequence: the machinery for deriving FEM equations in a local system and
converting to global coordinates, as done for the truss element stiffness matrix, can be bypassed for
the mass matrix.
4.3.2. Assembly of Master Mass Matrix
The assembly of the master mass matrix follows exactly the same set of rules used for the master
stiffness matrix. Since Me is diagonal, so will be the master mass matrix. A simple hand calculation
proceed by inspecting which elements contribute to a given node. For the example truss:
Node 1

gets half of the mass of elements (1) and (3): m 1 = 12 m (1) + 12 m (3) .

Node 2

gets half of the mass of elements (1) and (2): m 2 = 12 m (1) + 12 m (2) .

Node 3

gets half of the mass of elements (2) and (3): m 3 = 12 m (2) + 12 m (3) .

These values are assigned to both x and y directions. Consequently the master mass matrix is
M=

1
2

diag [ m (1) +m (3) , m (1) +m (3) , m (1) +m (2) , m (1) +m (2) , m (2) +m (3) , m (2) +m (3) ]

= diag [ m 1 , m 1 , m 2 , m 2 , m 3 , m 3 ] = diag [ 5, 5, 3, 3, 4, 4 ].

(4.14)

This two-step construction of the master mass matrix is schematized in Figure 4.4. The node masses
are m 1 = 5, m 2 = 3 and m 3 = 4.
Remark 4.1. In later Chapters we will see that the lumped mass matrix is not the only choice; in fact there is an

infinite number of possible element mass matrices. Those other than the lumped one are no longer diagonal.
Consequently the assembly process becomes more involved, but can be carried out by the same rules followed
for the master stiffness matrix.

49

Chapter 4: PMP DYNAMICAL SYSTEMS: MATRIX EOM FORMULATION

410

4.3.3. The FEM Equations of Motion


Combining the master mass matrix derived above with the master stiffness derived in IFEM [106,
Chapter 3] and the external forces specified in Figure 4.3, we obtain

20 10 10 0 10 10
u x1
0
u x1
5 0 0 0 0 0
0
0 10 10 u y1 0
0 5 0 0 0 0 u y1 10 10

10 0
0
0 u x2 0
0 0 3 0 0 0 u x2 10 0
+
=

0
0
5
0 5 u y2 0
0 0 0 3 0 0 u y2 0

u x3
u x3
10 10 0
0 10 10
2H (t)
0 0 0 0 4 0
u y3
u y3
10 10 0 5 10 15
H (t)
0 0 0 0 0 4
(4.15)
Comparing(4.15) with (4.11) shows these matrix EOM to be identical. Is this a fluke? No. It is a
consequence of two modeling choices.
The mass matrices are the same because of the lumping choice in FEM: conservation conditions
forces diagonal matrix entries to agree. If another choice had been made, the FEM mass matrix
would be nondiagonal and clearly different from that of the PMP model.
Stiffness matrices coalesce because we have set the spring constants of the PMP model of Figure 4.1
to be
(4.16)
k e = E e Ae /L e , e = (1), (2), (3),
where E e , Ae and L e are member properties of the FEM truss model of Figure 4.3. As shown in
Mechanics of Materials textbooks, as well as IFEM [106, Chapter 2], (4.16) provides the equivalent
axial stiffness of a truss (bar) element.
Given that the same EOM are obtained, downstream tasks such as
Application of constraint conditions
Modal analysis
Direct time integration
Internal force recovery
need not distinguish between the two derivation methods. Those topics are covered in the following
Chapters.
4.4. Comparison Between Formulation Approaches
Having gone through the two EOM formulation methods, it is appropriate to summarize their
relative advantages and shortcomings.
The key advantages of the force equilibrium method are simplicity and physical transparency. These
emanate from the use of just one modeling tool: Newtons laws, as well as basic linear algebra. This
makes it the method of choice in undergraduate instruction. Students in those courses have been
exposed to those laws, as well as trained in static FBD and basic linear algebra through introductory
lower-division courses, but have not yet encountered the more advanced mathematical machinery
used in FEM.

410

411

4. Notes and Bibliography

For modeling dynamic systems in practical projects in science or engineering, the equilibrium
method is recommended in the following scenarios:
Preliminary design. The nature of the method encourages the use of highly simplified dynamic
models. For example, an experienced engineer designing a car suspension would likely lump
the car and the wheels as link-connected point masses, thus bypassing the overkill ingrained in
using FEM models too early. Likewise, seismic design of a multistory building often starts with
a point-mass stick model. Keeping-it-simple can significantly reduce design flowthrough,
while avoiding costly modifications.
Particle model fits problem best. There are some scientific applications where PMP models are
proper and natural. The classical example is Astronomy, which is precisely where Newtonian
mechanics emerged as system of the world. The key advantage is the ability to discard ab initio
internal particle structural details, which may not be known anyway. This enforces retention of
key effects, such as inertia and gravitational interaction in that application. A FEM model of
the Solar System at astronomical scales (or the trajectory of a space vehicle) would be not only
absurd but a total waste of time.
The key advantages of FEM are generality and automation. Once system details need to be included,
the equilibrium force method rapidly loses its attractive simplicity. FEM is naturally adapted to
distributed systems modeled by field (continuum) theories within a variational framework. Its
computer implementation favors abstraction and modularity, reducing formulation errors while
concealing internal processing details.
Summarizing: in system design that involves dynamics, the two approaches form a natural hierarchy.
The PMP approach may capture gross effects during preliminary design, while FEM can zero in
details in subsequent stages. At the other extreme: verification and validation of existing (in situ)
systems, FEM is typically the preferred choice.
Notes and Bibliography
(To be completed)

411

Potrebbero piacerti anche