1.

THE EQUATIONS OF MOTION FOR LINEAR DISCRETIZED SYSTEMS

1.1 The discretization issue

Structural dynamics deals with the vibrations that occur in structural systems due to a broad
range of exciting phenomena. These systems are supposed to be of continuous nature; both
inertia and deformability of materials are assumed, at the scale considered in structural
analysis, as properties which are continuously distributed throughout the structural elements.
In principle, an unlimited number
of coordinates would be necessary
for studying the vibrations of
systems of this type; on
mathematical grounds, this
amounts to introducing partial
differential equations into the
structural analysis process. In fact,
in analyzing continuous systems,
we must consider, as independent
variables, not only time but also
the spatial coordinates describing
the geometry of the body.
For a deeper insight into the matter γ
we shall analyze in the following
the example of the cantilever beam
shown in figure 1.1.1, subject to a
transverse distributed time-varying
load f(x,t).
The following hypotheses will be
adopted in the analysis:
- the beam undergoes small oscilla-
tions in the neighbourhood of the
undeformed configuration C0;
- the material is linear elastic;
- plane stress conditions can be assumed; figure 1.1.1: cantilever with distributed mass
- rotatory inertia of the beam
sections can be neglected;
- axial and shear deformations can be neglected.
To write the dynamic equilibrium equation of the beam we first consider the application of a
distributed static load p(x); in this case, a well known equation relating the applied load and
the deflection v(x) can be written, with reference to figure 1.1.1, as:

d2  d2 
2  EJ ( x ) 2
v( x) = p( x) (1.1.1)
dx  dx 

In the preceding equation, expressing the indefinite equilibrium condition (in the transverse
direction) for the beam, the Young modulus E, and the moment of inertia J(x) of the beam
section are introduced.

1
In light of D’Alembert principle we can subsequently replace the static load p(x) with the sum
of the dynamic load f(x,t) and the inertia force per unit length. By making explicit the time-
dependence of the deflection v(x,t) the following equation is obtained

∂2  ∂ 2 v ( x, t )  ∂ 2 v ( x, t )
 EJ ( x )  = f ( x , t ) − γ ( x ) (1.1.2)
∂x 2  ∂x 2  ∂t 2

where the linear density γ(x) appears, equal to the sectional area times the volume density of
the material. Re-arrangement of the terms in (1.1.2) leads to the following partial differential
equation (PDE)
∂2  ∂ 2 v ( x, t )  ∂ 2 v ( x, t )
 EJ ( x )  + γ ( x ) = f ( x, t ) (1.1.3)
∂x 2  ∂x 2  ∂t 2

To solve equation (1.1.3) boundary conditions must be specified, expressing conditions at the
edges of the spatial integration field, along with initial conditions, regarding the system state
at a specified time instant. The former, in the case of figure 1.1.1, must represent the effect of
the base clamp and the free-end top condition (zero bending moment and shear force), leading
to the following
[v( x, t )]x = 0 = 0 ;  ∂v( x, t )  = 0 (1.1.4)
 ∂x  x = 0

 ∂ 2 v ( x, t )  ∂  ∂ 2 v ( x, t )  
 EJ ( x )  = 0 ;   EJ ( x )  = 0 (1.1.5)
 ∂x 2  x =l  ∂x  ∂x 2   x =l

Finally, by assuming that initial conditions are specified in the time origin, the following
displacement and velocity distributions will be stated as known

 ∂v( x, t ) 
[v(x, t )]t =0 = r (x ) ;  = s( x ) (1.1.6)
 ∂t  t =0

being both r(x) and s(x) possible deflected shapes for the structure at study.
The closed form integration of differential problems of the type (1.1.3-1.1.6) is possible only
in very simple cases: for example, non uniform material properties (eg. tapered beams) or non
trivial boundary conditions are sufficient, in most cases, to make the exact solution of
equilibrium equations impossible. Looking at numerical procedures we can comment that
partial differential equations are hardly suitable for automatic computer-based solutions; the
latter, in fact, are intrinsically capable of treating large sets of data, but not functions defined
on a continuous domain, as is the case of the v(x,t) time varying deflection.
Some of the quoted aspects (eg. non uniform properties of the continuous medium) can be
treated by means of finite-difference methods, i.e. of procedures working “a posteriori” on the
PDE: it is worth noting, however, that such methods are often not practical when complex
boundary conditions are to be fulfilled, this being an usual situation for structural systems.
Summing up, the problem arises of discretizing continuous media characterized by distributed
flexibility and inertia, i.e. of their modelling as systems which can be studied by introducing a
finite number of degrees of freedom; the dynamic response of such systems, as it will be
shown in the following, can be analyzed by solving simultaneous ordinary differential

2
equations which, in the case of small oscillations, can be made linear with constant
coefficients. The treatment of such equations is well suited for digital computation.
Conceptually, the simplest discretization procedure is the so called “lumped mass” method;
this is based upon the replacement of distributed inertia with a set of rigid masses, having
either negligible (point masses) or finite dimensions (rigid bodies) and with the parallel
replacement of external distributed forces with loads acting on the masses, which means that
for a system of point masses only concentrated loads would be admissible.
The dynamic response of a lumped mass system can be studied by assuming as lagrangian
coordinates the displacement components, and/or the rotation ones for the case of rigid
bodies; these are sufficient to obtain the system configuration at any time. The level of
approximation relies on the number of masses (as well as on the effectiveness of their
distribution), i.e. on the number of degrees of freedom of the model.
A second classical discretization method is represented by the adoption of generalized
coordinates. Let’s consider again, as an example, the flexural vibrations of the cantilever
beam of figure 1.1.1; the methods amounts to expressing, as an approximation, the deflected
shape of the beam as a linear combination of known, or “assumed” functions ψ k ( x) , i.e.

n
v ( x , t ) = ∑ ψ k ( x ) qk ( t ) (1.1.7)
k =1

The coefficients of the combination qk must vary with time and play the role of system
coordinates, since they allow, at any time, to obtain the structural configuration; these
coordinates, however, do not possess the physical meaning of displacement or rotation
components. The degree of approximation relies on a good choice of the ψ k ( x) functions,
called “shape functions”, and on their number, i.e., again, on the number of degrees of
freedom.
A third approach for the discretization process, which is tightly connected to the shape
function (or “assumed mode”) procedure, is the so called Finite Element Method (FEM). This
is based on the subdivision of the medium in a finite number of portions (called “finite
elements”) connected in a set of points (“nodes”); the displacements, and in some cases the
rotations, of the nodes are assumed as system coordinates. Within each finite element the
displacement field is again interpolated in the form (1.1.7); in this case, however, the
generalized coordinates in (1.1.7) are taken as the physical displacement components of the
nodes connecting the element. Note that mixed methods are often employed, in which inertia
forces are represented in terms of lumped masses, while the FEM is used to represent elastic
properties or, more generally, restoring forces.

1.2 The equations of motion of undamped n-DOF linear systems

The equations of motion for system whose configuration can be described by a finite number
of coordinates (n degree-of-freedom or n DOF system) can be derived by means of classical
Lagrange equations; the latter allows to describe the motion of any mechanical systems
subject to smooth and holonomic constraints. In structural dynamics, however, we focus
attention on the problem of small oscillations about a stable equilibrium configuration C0; the
case of conservative systems, i.e. with no internal dissipation, will be considered first.
We shall denote by q k (k = 1,..., n ) the Lagrangian coordinates of the system and assume that
they all vanish in the C0 configuration. The k-th equation of motion will take the form
3
 d ∂L ∂L
− = Qk (k = 1,...n ) (1.2.1)
dt ∂q
k ∂qk

where the generalized component Qk of external forces appears along with the Lagrangian
function L = T – V , this being the difference between kinetic and potential energy.
If we denote with q
the column vector listing the time derivatives of the system coordinates,
the kinetic energy can be in turn expressed, under the hypothesis of fixed constraints, as the
positive quadratic form
1
T = q
T m q
(1.2.2)
2

where m is a symmetric positive-definite square matrix whose coefficients can be regarded, in

the small oscillation setting, as constant and equal to the value assumed in the C0
configuration.
Potential energy can be expressed as well, in the case here considered, as a quadratic form; in
fact, if we expand it as a Taylor series in the neighborhood of C0, we can draw the following
considerations.
• The linear term is missing due to the fact that potential energy is stationary in
equilibrium points.
• The constant term does not affect the equations of motion, since derivation of the
function is performed in (1.2.1).
• Higher order terms are negligible due to the small oscillations hypothesis.
We can thus write
1
V = qT k q (1.2.3)
2
where k is a symmetric square matrix, listing the following coefficients

 ∂ 2V 
kij =   (1.2.4)
 ∂qi ∂q j  q= 0

Since we assume that C0 is a stable equilibrium configuration, the quadratic form (1.2.4) is
positive-definite, and so does the matrix k.
Finally, to define the generalized component Qk(t) of external perturbation forces we must
assign to the system, at time t, a virtual variation of configuration. By evaluating the virtual
work δ W performed by the forces we can write, by suitable manipulation, an expression of
the type
n
δ W = ∑ Qk (t ) δ qk (1.2.5)
k =1

which implicitly defines the Qk components. Note that the virtual work δ W depends, in
general, by the configuration of the system at time t: in the small oscillation hypothesis,
however, this dependency turns out to be negligible.
It is worth noting that, if the system is subject to a set of static loads superimposed to the
dynamic perturbation forces, C0 coincides with the static equilibrium configuration, assumed
to be stable; however, if C0 is characterized by small displacements, the coefficient of the m

4
and k matrices can be computed with reference to the undeformed configuration, i.e. as in the
case in which static loads are not considered.
Given the system linearity, the static loading effect can be analyzed separately and
subsequently superimposed to the one of dynamic forces. In addition, if static loads contribute
only to the linear variation of total potential energy and not to the quadratic term (1.2.3), they
will not affect the small oscillations about the C0 configuration; on practical grounds this
occurs when the absolute values of static loads are much smaller than those leading to loss of
stability.
To express the equations (1.2.1) we must perform the derivation of the Lagrangian function,
taking expressions (1.2.2) and (1.2.3) into account. We obtain, in deriving with respect to q
k

∂L ∂T 1  ∂q
T ∂q
 ∂q

= =  m q
+ q
T m = T
 ∂q
m q = ek m q


(1.2.6)
∂q
k ∂q
k 2  ∂q
k ∂q
k  k

where ek is a column unit vector having the k-th component equal to one and all the others
equal to zero. Analogously, the derivative of L with respect to qk can be expressed as

∂L ∂V
− =+ = ekT k q (1.2.7)
∂qk ∂qk

By subsequently deriving (1.2.6) with respect to time, equation (1.2.1) takes the form

ekT m q

+ ekT k q = Qk (k = 1,...n ) (1.2.8)

It can be observed that if we premultiply a square matrix by the row vector ekT we obtain, as a
result, the k-th row of the matrix; thus the two terms at the lhs of (1.2.8) are respectively the k-
th components of the column vectors m q

e kq. By ordering the Qk generalized components
in the column vector Q, the n equations of motions can be written, in compact matrix form, as

m q

+ k q = Q (1.2.9)

The equations (1.2.9) form a system of simultaneous linear ordinary differential equations
with constant coefficients.

1.2.1 Lumped mass systems

For lumped mass systems it is easy to demonstrate how the equations (1.2.9) can be directly
derived by simply imposing dynamic equilibrium, in light of D’Alembert principle.
Let’s consider a system of massless beams and point masses; assume that the latter are subject
to perturbation forces. The dynamics of such system is the one of the “material points”,
represented by the masses, under the effect of the perturbation forces and of the elastic
restoring forces exerted by the beams.
The Lagrangian coordinates of the system are (see the example system in fig. 1.2.1) the “free”
displacement components of the masses, i.e. the ones not subject to constraints. In case of
rigid bodies, rotations are also introduced as coordinates.

5
To express the generalized components of the perturbation forces F j (t ), j = 1,...m , assumed to
act on the masses, we shall denote with uj the displacement components of the points of
applications along the direction of the forces; virtual work will thus read as

m
δ W = ∑ Fj (t )δ u j (1.2.10)
j =1

The displacement components uj can be in turn expressed in terms of the Lagrangian

coordinates, i.e.

u j = u j (q1 ,..., q n )

so that (1.2.10) expression takes the form

m n ∂u j
δW = ∑ F j ∑ δqk (1.2.11)
j =1 k =1 ∂qk

By reversing the order of the summations and taking the definition (1.2.5) into account the
generalized components of the dynamic forces are

m ∂u j
Qk = ∑ F j (1.2.12)
j =1 ∂qk

It is worth noting that, in this case, the generalized components are simply the sum of the
projections of the dynamic forces onto the direction of the lagrangian coordinates.
Under these premises and in light of D’Alembert principle, let’s first analyze the system
loaded by a set of static forces; let P be the column vector listing the generalized components
of such forces with respect to the Lagrangian coordinates. By assuming the latter to be zero in
the undeformed configuration, the equilibrium position can be expressed, in the small
displacement hypothesis, via the classical relation

q=ηP (1.2.13)

in which η is the flexibility matrix.

The flexibility matrix is symmetric due to the Maxwell theorem; in addition, if we express the
strain energy by equating it to the work done by the applied loads, we obtain, by Clapeyron
theorem
1 1
VE = P T q = P T ηP
2 2

If the system is entirely composed of deformable beams the previous expression is positive
for any choice of the P vector; thus, the η matrix is positive definite and non singular. By
inverting expression (1.2.13), we obtain

6
 figure 1.2.1: lumped mass system

P = kq ; k = η−1 (1.2.14)

In expression (1.2.14) applied loads are equated to elastic restoring forces exerted by the
beams; these forces are expressed in terms of the system displacements, by means of the
coefficients of the k matrix, i.e. of the stiffness coefficients.
For a deeper insight into the physical meaning of these coefficients let us re-write equations
(1.2.14) by setting qr = 0 for r =/ j and q j = 1 , obtaining

Pi = k ij ; i = 1,..., n

The coefficients kij, i=1,…,n, i.e. the j-th column of the stiffness matrix, are the loads to be
applied in the direction of the system coordinates to obtain a static deflection in which all
coordinates vanish, except the j-th, which is equal to one.
By applying D'Alembert principle we can now replace, in (1.2.14), each static component Pk
(acting on the j-th mass) with the sum of the component Qk of the dynamic force and of the
component -mj q

k of the inertia force, obtaining, in matrix form

Q − m q

= k q
or, by simply rearranging terms

m q

+ k q = Q (1.2.15)

The matrix m appearing in the preceding equations is a diagonal matrix listing the values mj
of the lumped masses (or the values of moments of inertia for rotational coordinates); each

7
value mj will appear r times if r is the number of degrees of freedom of the j-th mass. The m
matrix is positive definite when the Lagrangian coordinates are associated to the motion of
points where a non zero lumped mass is located.

figure 1.2.2: truss system with lumped masses

Let’s analyse, for example, the simple plane truss system in figure 1.2.2, where all truss
beams share the same sectional area A. The number of degrees of freedom (DoFs) of the
structure is equal to three; the mass m1 (which is free to move in the plane) has two DoFs,
while m2, due to the roller acting on it, has only one DoF. With reference to the coordinate
system shown in figure 1.2.2 the equations of motion are the following

3
m1 q

1 + ∑ k1 j q j = f1 (t )
j =1
3
m1 q

2 + ∑ k2 j q j = 0 (1.2.16)
j =1
3
m2 q

3 + ∑ k3 j q j = f 2 (t )
j =1

To compute the stiffness coefficients appearing in equations (1.2.16) we first apply suitable
additional constraints to the system coordinates (see figure 1.2.3); by prescribing, for
example, a unit displacement to the roller acting on q1, according to the upon definition, the
reactions in the additional constraints take the values of the coefficients k11, k21 e k31.
Computing these coefficients is, in this case, very simple; in fact, the elongation of all truss
beams, and thus of axial forces, is easily available by simple geometrical considerations. With
the symbols in the figure we get

EA EA 2 EA EA
Na = ∆la = = ; Nb = N d = 0 ; Nc = −
la l 2 2 2l l

8
 EA
By imposing equilibrium to nodes A and B we obtain, by denoting the ratio with k
l
2  2  2 2
k11 = N a − N c = k  + 1 ; k21 = N a =k ; k31 = N c = − k
2  4  2 4

If we repeat the procedure by applying unit displacements to q2 and q3, we obtain the
remaining stiffness coefficients.
The mass and stiffness matrices for the system in figure 1.2.3 are the following

 2 2 
 +1 −1 
m1 0 0  4 4 
m =  0 m1 0 
 k = k 0 
2 2
; +1
 4 4 
 0 0 m2   2 
 −1 0 + 1
 4 

figure 1.2.3: definition of stiffness coefficients

The reader will easily verify how the same matrices can be derived by means of Lagrange
equations: in fact, for the system in figure 1.2.2 both kinetic and potential energy can be
simply derived as functions of the system coordinates.
Note that the systems here considered were composed of elastic beams and point masses;
when rigid bodies of finite dimensions are present, internal constraints are applied to the
motion of the masses.
This situation arises, for example, when axial deformability of straight beams is neglected. In
such cases, kinematic analysis must be applied to correctly define the degrees of freedom of
the masses. To this aim it often useful to study the system motion by means of an auxiliary
system, composed by the masses and the rigid bodies; in the case of axially undeformable
beams, for example, rigid trusses should replace the beams in the auxiliary system.

9
Once the coordinate system is defined, the mass matrix can be derived by writing the system
kinetic energy: in fact, the direct evaluation of the coefficients of the matrix m (which can be
non diagonal) can be, in this case, far from simple. Stiffness coefficients, on the other hand,
can be rather easily computed, if geometrical effects are negligible, on the basis of the above
“static” definition and following the procedure illustrated for the system in figure 1.2.3.

1.2.2 Systems discretized by the assumed mode method

For systems discretized by means of shape functions and generalized coordinates the
equations of motion can be derived by Lagrange equations.
Let’s consider again the cantilever
beam studied at section 1.1 and assume
that, as anticipated, we express the time
variation of the deflection by means of
the following linear combination

n
v ( x, t ) = ∑ψ i ( x ) qi ( t ) (1.2.17)
i =1

Under the hypotheses at section 1.1 the

kinetic energy can be written as the
integral

1
l
γ
T = ∫ γ ( x ) v
2 ( x, t ) dx (1.2.18)
20

in which the following usual notation

for time derivatives has been adopted

∂v( x, t )
v
( x, t ) =
∂t
figure 1.2.4: axially loaded cantilever

Upon substitution of the time derivative of expression (1.2.17) in the kinetic energy we
obtain:
l n n
1
T = ∫ γ ( x ) ∑ψ i ( x ) q
i ( t ) ∑ψ j ( x ) q
j ( t ) dx (1.2.19)
20 i =1 j =1

By exchanging the order of integration and summations we get

l
1 n n
T= ∑
2 i =1
∑ ∫ γ ( x ) ψ ( x ) ψ ( x ) dx q
( t ) q
( t )
j =1
i j i j (1.2.20)
0

Equation (1.2.20) is a quadratic form in the time derivatives of the Lagrangian coordinates.
The coefficients of the mass matrix, see equation (1.2.2), have thus the following form

10
 l
mij = ∫ γ ( x ) ψ i ( x ) ψ j ( x ) dx (1.2.21)
0
From (1.2.21) symmetry of the mij coefficients is apparent; expression (1.2.18), on the other
hand, shows how kinetic energy is always positive, so that the mass matrix is positive
definite.
Applying the same reasoning to elastic potential energy we shall write:

l
1
2 ∫0
VE = EJ ( x ) v II 2 ( x, t ) dx =

l n n
1
= ∫ EJ ( x ) ∑ψ i ( x ) qi ∑ψ IIj ( x ) q j ( t ) dx
II

20 i =1 j =1

l
1 n n
VE = ∑
2 i =1
∑ ∫ EJ ( x )ψ ( x )ψ ( x ) dx q ( t ) q ( t )
j =1
II
j i
II
i j (1.2.22)
0

In the preceding expressions the following compact notation has been introduced

∂ 2 v ( x, t )
v II ( x, t ) =
∂x 2

If a constant (i.e. conservative) axial load P acts on the cantilever beam, as in figure 1.2.4, the
quadratic variation of the potential energy must be added to (1.2.22), due to the configuration
change from C0 (undeformed in this case) to the generic deflection v(x,t); by expressing this
variation as (minus) the work done by P we get:

l
1
VG = − P u = − P ∫ v I 2 ( x, t ) dx =
20
l n n
1
= −P
2 ∫0
∑ψ iI ( x ) qi ( t ) ∑ψ Ij ( x ) q j ( t ) dx
i =1 j =1

l
1 n n
VG = − P ∑ ∑ ∫ψ ( x ) ψ ( x ) dx q ( t ) q ( t )
i
I I
j i j (1.2.23)
2 i =1 j =1 0

The stiffness matrix, i.e. the matrix of the total quadratic variation of potential energy, can be
thus espressed as the sum of the two following contributions (respectively elastic and
“geometric”)
k = k E + kG
where
l
k E ,ij = ∫ EJ ( x )ψ iII ( x )ψ IIj ( x ) dx (1.2.24)
0
l
kG ,ij = − P ∫ψ iI ( x )ψ Ij ( x ) dx (1.2.25)
0

11
The symmetry of the stiffness coefficients is apparent from (1.2.24) and (1.2.25). In addition,
since we have postulated the system being stable in the reference equilibrium position, the
total quadratic variation of potential energy, i.e. the sum of (1.2.22) and (1.2.23), must be
positive, so that k turns out to be positive definite.
We finally analyze the effect of the distributed perturbation forces f(x,t) applied to the
cantilever beam; the virtual work performed by such forces due to a virtual change δv( x, t ) of
the beam configuration can be expressed as:
l
δW = ∫ f ( x, t )δv( x, t ) dx
0

Upon differentiation of (1.2.17) and substitution into the preceding integral we obtain:
l n n l
δ W = ∫ f ( x, t ) ∑ψ j ( x ) δ q j dx = ∑ ∫ f ( x, t )ψ ( x ) dx δ q
j j
0 j =1 j =1 0

By comparing the obtained expression to (1.2.5) we derive the following definition for the
generalized components of the distributed forces f(x,t) with respect to the Lagrangian
coordinates
l
Q j ( t ) = ∫ f ( x, t )ψ j ( x ) dx (1.2.26)
0

In applying the assumed mode method, an obvious question arises regarding the conditions
which are to be met by the shape functions υ/ j (x) which are introduced in the preceding
equations. The problem will be debated in chapter 5, where approximate methods for
vibration analysis of distributed systems will be treated to some more detail in analytical
terms. However, it seems to be reasonable to assume that those function must satisfy
continuity and derivability, sufficiently to allow for the computation of integrals (1.2.24), and
geometrical boundary conditions (1.1.4). This ensures that the approximate deflection (1.2.17)
will in turn satisfy such conditions; coming to the (1.1.5) conditions, i.e. the so-called
“natural" boundary conditions, it will be shown how their fulfillment is not strictly necessary,
even though obviously preferable in terms of approximation.

1.3 Linear systems with viscous damping

The systems studied in the previous sections are internally conservative: this is obviously an
ideal assumption; in fact, during the motion of any real system energy dissipation occurs, due
to a number of causes. These can be tentatively summarized as follows.
• The hysteretic behavior of structural elements, i.e. the ones of the load carrying
system, and of the non-structural components (architectural, equipment, etc).
• The behavior of joints and connections between structural elements, which are
modeled as perfect constraints but, in reality, undergo phenomena such as friction,
hysteresis etc.
• The motion of foundation elements, resulting in the transmission of vibration to the
soil medium, which is responsible for hysteresis and dissipation phenomena due to
wave radiation.
• The introduction in the structural system of damping devices for vibration mitigation.

12
 • The effects of the relative motion between the structure and the surrounding fluid (air
or water), which can be of importance both in the case of high winds and in the
analysis of submerged structures.
Almost all the quoted phenomena give rise to dissipation forces which are markedly non
linear: nevertheless, the complexity which results from non linear dynamic analysis must be
considered. In addition, when damping forces arising in structural dynamics applications are
compared to inertia and elastic effects it can be observed that they are usually one order of
magnitude smaller and based on much less reliable experimental data. Summing up, the
introduction of linear damping forces to model structural dissipation is deemed to be
acceptable in practical design and assessment applications.
The damping forces will be introduced in the equations of motion as viscous terms,
proportional to the time derivatives of the system coordinates: in this light, we shall ideally
introduce in lumped-mass models suitable devices, called “dashpots”, which apply to the
system (see figure 1.3.1) two equal and opposite forces which are proportional to the relative
velocity between the connected points. The proportionality constant is termed as "damping
coefficient” of the dashpot; the sign of the forces is opposite to the one of relative velocity, in
order to produce negative power. With the symbols in figure 1.3.1 we thus write,

f D = −c z

To introduce viscous forces, the k-th Lagrange equation for the system can be initially written
in the following form

d ∂L ∂L
− = Qk + QD, k (k = 1,...n ) (1.3.1)
dt ∂q
k ∂qk

in which QD,k is the component with respect to qk of the generalized damping forces. To
express the QD,k components we can introduce the dissipation, or Rayleigh, function: this is
defined as half of the power dissipated by the dashpots. For s dashpots acting on the system it
reads:

1 s
D= ∑
2 r =1
cr z
r2 (1.3.2)

cr and z
r being the damping coefficient and the relative velocity between the points
connected by the r-th dashpot. Once the dissipation function is defined, the components of
generalized damping forces can be obtained by exploiting its fundamental property, expressed
as
∂D
QD , k = − (1.3.3)
∂q
k

To demonstrate the relationship (1.3.3), we first examine a single dashpot (r-th) acting in the
system. To express its contribution to the total dissipation function, we must define the
relative displacement across the dashpot as a function of the system coordinates, i.e.

13
 fig. 1.3.1: viscous dashpot

zr = zr (q1 , q2 ,..., qn ) (1.3.4)

It’s worth noting that, in writing (1.3.4), we implicitly assumed, consistently to the general
criterion of the lumped mass/force method, that damping forces as well are only applied to the
masses. From (1.3.4), which obeys to linearity, given the small oscillation assumption, we
obtain
n
∂z
z
r = ∑ r q
i (1.3.5)
i =1 ∂qi
Upon substitution of (1.3.5), the dissipation function (1.3.2) takes the following form

1 s n n
∂z ∂z
D= ∑ r∑c
2 r =1 i =1
∑ ∂qr q
i ∂qr q
j (1.3.6)
j =1 i j

By changing the order of the summations and by defining the coefficients:

s
∂zr ∂zr
cij = ∑ cr (1.3.7)
r =1 ∂qi ∂q j

the quadratic form (1.3.6) can be cast into the usual compact matrix form

1 T
D= q
c q
(1.3.8)
2

14
where c is a symmetric square matrix listing the coefficients (1.3.7). Derivation of the
preceding expression leads to

∂D
= ekT c q
(1.3.9)
∂q
k

Let’s now turn our attention on the component QD ,k of generalized damping forces; to this
aim we shall express the virtual work done by such forces when a virtual configuration
change is applied to the system: this work is equal to

s
δ WD = −∑ cr z
r δ zr (1.3.10)
r =1

where a minus sign appears, since the work considered is the one done by the forces applied
to the system; the virtual relative displacement δ zr can be in turn expressed, upon
differentiation and in terms of the virtual changes of the system coordinates, as

n
∂zr
δ zr = ∑ δ qi (1.3.11)
i =1 ∂qi

By substituting the preceding expression and (1.3.5) into equation (1.3.10), we get, for the
virtual work done by damping forces

s n
∂zr n ∂zr
δWD = −∑ cr ∑ q
i ∑ δq j (1.3.12)
r =1 i =1 ∂qi j =1 ∂q j

In compact matrix form, and introducing the previously defined matrix coefficients (1.3.7),
we shall write

δ WD = −δ qT c q
(1.3.13)

where δq is the vector listing the virtual variation of the Lagrangian coordinates. By
comparing the preceding equation to (1.2.5), which can be re-written with reference to
damping forces, as

δWD = −δqT QD

we get the expression of the components of the generalized damping forces:

QD = −c q
(1.3.14)

For the k-th coordinate, this reads

QD , k = −ekT c q

From the last expression and equation (1.3.9), we can derive the dissipation function
fundamental property (1.3.3), allowing for the definition of generalized damping forces

15
through the definition of a suitable quadratic form; since the latter (dissipated power) is
always positive on physical grounds (and from 1.3.2), the damping matrix is positive definite
for a physically consistent dissipation model.

By substituting (1.3.3) into Lagrange equations (1.3.1) we obtain

d ∂L ∂L ∂D
− + = Qk (k = 1,...n ) (1.3.15)
dt ∂q
k ∂qk ∂q
k

Finally, by performing the derivatives in (1.1.15), turning Lagrange equations in the compact
matrix form (1.2.9) and adding the components of the generalized damping forces (1.3.14),
we get the following system of simultaneous linear ODEs

m q

+ c q
+ k q = Q (1.3.16)

where damping effects are introduced, once defined the viscous matrix c, in terms of forces
proportional to generalized velocities.

16
1.4 Eccentric rotating masses: equations of motion

figure 1.4.1: system under the action of an eccentric rotating mass

Let’s take the plane system of figure 1.4.1 as an illustrative example. A rigid disc rotates with
angular velocity ω about a pivot O. Design tolerances allow for an eccentricity e of the
centroid Gr of the rotating mass with respect to O. If we assume that the symmetry axes of the
rotating machine are parallel to those of the supporting structure plane, motion occurs in the
plane. In addition, assume that both the slab supporting the machine and the “fixed” (i.e. not
rotating) part of the machine can be modeled as rigid and rigidly connected; in this case their
motion can be described by the displacement components q1, q2 of the centroid G of the
resultant rigid body (slab+machine) and by its rotation q3 (see figure 1.4.1). Let’s denote with
M and IG the mass and moment of inertia of the rigid body, while the rotating disc has mass
mr and moment of inertia IGr.
The steady-state motion of the centroid Gr can be espressed as a function of the
motion of the pivot O. Denoting with u,v respectively the horizontal and vertical displacement
components we get:

uGr = uO + e cos(ωt )
 (1.4.1)

vGr = vO + e sin(ωt )

The displacement components of the pivot O can in turn be expressed in terms of the system
coordinates as:

17
 uO = q1 − aq3
 (1.4.2)
vO = q2

where a is the distance between the pivot O and the system centroid G. Upon substitution in
expressions (1.4.1) we obtain:

uGr = q1 − aq3 + e cos(ωt )

 (1.4.3)

vGr = q2 + e sin(ωt )

The Lagrange equations for the system

d ∂T ∂V ∂D
+ + =0 k = 1, 2,3 (1.4.4)
dt ∂q
k ∂qk ∂q
k

can be re-written by making the contribution of the rotating mass explicit and by assuming
that it doesn’t affect dissipation but only the kinetic energy Tr:

d ∂T f ∂V f ∂D f d ∂Tr
+ + + =0 k=1,2,3 (1.4.5)
dt ∂q
k ∂qk ∂q
k dt ∂q
k

In (1.4.5) Tf ,Vf, Df are respectively the kinetic energy, potential energy and dissipation
function of the fixed part. By neglecting, in the rotational terms, the fluctuation due to q
3 with
respect to the constant value ω , Tr is simply written as
1 1
Tr = I Grω 2 + mr (u
Gr
2 2
+ v
Gr ) (1.4.6)
2 2

which can be rewritten, taking expressions (1.4.3) into account, in the form:

Tr =
1
2
{I Grω 2 + mr ( q
1 − aq
3 − ω e sin(ωt ) ) +

2

(1.4.7)
( q
2 + ω e cos(ωt ) )
2
}


By performing the derivations in (1.4.5) we get:

d ∂T f ∂V f ∂D f
+ + + mr q

1 − amr q

3 = mrω 2e cos(ωt ) (1.4.8a)
dt ∂q1 ∂q1

∂q1


d ∂T f ∂V f ∂D f
+ + + mr q

2 = mr ω 2e sin (ωt ) (1.4.8b)
dt ∂q
2 ∂q2 ∂q
2
d ∂T f ∂V f ∂D f
+ + − mr aq

1 + mr a 2 q

3 =
dt ∂q
3 ∂q3 ∂q
3 (1.4.8c)
2
= − amrω e cos (ω t )

18
For a mechanical interpretation of the terms at the left hand side in equations (1.4.8) we can
note that they could have been obtained by regarding the rotating disc as a point mass located
in O. In this case its kinetic energy takes the form:
1
TO = [mr (u
O2 + v
O2 )] (1.4.9)
2
Taking equations (1.4.2) into account, the preceding expression leads to:
1
TO = mr (q
1 − aq
3 )2 + q
22  (1.4.10)
2 
Upon derivation of the kinetic energy (1.4.10) the terms at the lhs in (1.4.8) are obtained.
On the other hand, when the right-end-sides are considered, it can be easily verified that they
can be obtained as the generalized components of the horizontal force Fx = mrω 2e cos(ωt ) and
of the vertical force Fy = mrω 2e sin(ωt ) applied to the pivot O, as shown in figure 1.4.2.
Equations (1.4.8) can be thus rewritten as:

d ∂ (T f + TO ) ∂V f ∂D f
+ + = mr ω 2e cos(ωt ) = Q1 (1.4.11a)
dt ∂q
1 ∂q1 ∂q
1
d ∂ (T f + TO ) ∂V f ∂D f
+ + = mr ω 2 e sin(ωt ) = Q2 (1.4.11b)
dt ∂q
2 ∂q2 ∂q
2
d ∂ (T f + TO ) ∂V f ∂D f
+ + = −amrω 2 e cos(ωt ) = Q3 (1.4.11c)
dt ∂q
3 ∂q3 ∂q
3
Equations (1.4.11) describe the motion of a system in which the rotating part acts as a point
mass located in the center of rotation; the system is excited by two harmonic forces in
quadrature, having an amplitude which is proportional to the rotating mass, to its eccentricity
and to the square of the angular velocity of rotation (figure 1.4.2).

figure 1.4.2: forces transmitted by the rotating mass

19
1.5 The effect of the motion of the reference system: equations of motion

Let the system in figure 1.5.1 be subject to a motion of its constraints; if this motion
u(t) is uniform at all constraints, it can be seen as applied to the reference system.

figura 1.5.1: system subject to motion of the reference frame

The lagrangian coordinates describing the system motion can be written as

q (t ) = v (t ) + ru (t ) (1.5.1)

where q (t ) are absolute coordinates while v (t ) lists the coordinates in the moving reference
system. r is a constant vector listing zero and unit components, depending on the structure
geometry and on the direction of the motion of the reference system.
The kinetic energy reads, in terms of absolute velocities, as

1
T= q
(t )T m q
(t ) (1.5.2)
2

while potential energy depends only on relative coordinates, i.e.

1
V= v (t )T kv (t ) (1.5.3)
2

We assume that the dissipation function as well depends only on relative velocities, this being
a sound assumption for all the deformation-related phenomena:

1
D= v
(t )T cv
(t ) (1.5.4)
2

The related terms appearing in Lagrange equations are thus the following:

20
 ∂D ∂D
= = e Tk cv
(t ) (1.5.5)
∂q
k ∂vk

∂V ∂V
= = eTk kv (t ) (1.5.6)
∂qk ∂qk

When the kinetic term is considered we get, by introducing the relative and reference system
velocity:
∂T
= eTk m q
(t ) = eTk m (v
+ ru
) (1.5.7)
∂q
k

d  ∂T 
  = e Tk m (v

+ ru

) (1.5.8)
dt  ∂q
k 

Accordingly, the system of equations of motion takes the form:

m v

+ cv
+ kv = −mru

(1.5.9)

Equations (1.5.9) can be also derived by a non-inertial observer, i.e. in motion with the
reference system, who “sees” the kinetic energy as due to relative velocity only, provided that
the so called “apparent” forces (figure 1.5.2) are introduced to take the reference system
motion into account. These forces are parallel to the imposed motion and are equal to the
mass times the reference acceleration, with a minus sign, i.e. to − m j u

. For the case of lumped
mass systems the vector r lists null or unit coefficients ( r T = {1 0 1 0} in the case of
figure 1.5.2).

figure 1.5.2: apparent forces

21