MECA0029-1: Theory of Vibration: Answers To Exam Questions
MECA0029-1: Theory of Vibration: Answers To Exam Questions
MECA0029-1: Theory of Vibration: Answers To Exam Questions
16th of December
Contents
2subsection.1.1
1.2 Starting from the description of the state of a system of N particles in terms of the equation
given by ( 1.1), describe the different types of kinematic constraints that may exist in
mechanical systems. Define and illustrate the concept of generalized co-ordinates using
examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Using time-integration of the virtual work principle for N particles in the form given by
1.4 Explain what means the Hamilton’s principle written in the form ( 1.13). Starting from the
general expressions of T and V in terms of displacement and velocity, deduce the Lagrange
equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Starting from the definition of the kinetic energy for a system of N masses (equation (1.24)),
and using the concept of generalized coordinates, uik (xjk , t) = Uik (q1 , ..., qn , t), explicit the
structure of the kinetic energy. Inserting this into Lagrange’s equations, describe then the
1.6 Describe how to apply the Hamilton’s principle in the case of constrained dynamical systems. 12
2.1 Define the different terms that appear into the Lagrange equations in the general case. . . 14
2.2 Introduce the concept of normal modes of vibration. Examine also the case of systems
of normal modes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Determine the response of a vibrating system to non-zero initial conditions (including the
expansion of the FRF matrix. Discuss the case of systems with and without rigid-body
2.5 Calculate the response to external loading through modal expansion. Show how each
2.6 Describe the effect of truncation of the response in the mode superposition method. Show
how this effect may be partially corrected using the mode acceleration method. . . . . . . 27
2.7 Show how to use the method of additional masses to approximate the response of a system
2.8 Explain why the concept of effective modal masses is useful when the support of a structure
undergoes a rigid body motion. Define what the different terms represented in equation
(2.76). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.1 Write the normal equations for a damped system. Introduce the assumption of modal
damping and discuss its validity in the case of a lightly damped structure. . . . . . . . . . 29
3.2 Calculate the forced harmonic response of a damped system. Derive the spectral expansion
3.3 Discuss the concept of identification using the force appropriation testing. . . . . . . . . 32
3.4 Explain what are the different terms of the FRF matrix written in the form (see below)
equation ( 3.34). How can this equation be used for modal identification? What is the
information that can be extracted from the identified variables? What is the difference
4 Continuous systems 35
4.1 Define the Green’s strain tensor. Apply Hamilton’s principle to a continuous system and
4.2 Write the Hamilton’s principle in the case of prestressed structures and discuss the case of
5.1 Introduce the concept of finite elements using the bar in extension as example. . . . . . . 40
5.2 Apply the finite element method to beam in bending without shear deflection. Extend the
6.1 Describe the concept of reduction of the size of an eigenvalue problem using the static
condesation method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.2 Describe the concept of mechanical impedance and how it can be used for substructuring
7.1 Introduce the general formula for direct integration of first-order systems. Define the
7.2 Newmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
7.2.2 According to Fig. 7.1 and the expressions of amplitude and periodicity error, discuss
the stability and accuracy of the method with respect to the choice of parameters
and time-step. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7.3.2 Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.4 Explain how Newmark’s algorithm can be extended to nonlinear systems of general form
8.3 Apply the harmonic balance method to the example of the Duffing oscillator to calculate
Introduction
Rules of writing
• Always recall the complete statement of the question as this file should work as a standalone for
the exam.
• After the statement, precise which set of slides are related to the question using the footnote
command.
• Answer in English.
• Include your graphics in the appropriate subfolders (inside the ’Figs’ folder).
• READBACK (relecture): Please write "EDIT" followed by the modification you do when you
want to modify something in a question (excepted for spelling mistakes). It will help if some people
• Send a message to Nicolas Guilluy on facebook when you’re done with your work.
δui = u∗i − ui
We suppose that the varied trajectory and the real one both pass through the same points at the ends
d
(Remark that dt (δui ) = δ u̇i ).
The dynamic equilibrium of the particle of mass m can be written in the form:
müi − Xi = 0 (i = 1, 2, 3)
where the first term represent the inertia force and the second one, external forces. We multiply this
equation by the associated virtual displacement and we sum over the components. The virtual work
expression results:
3
P
(m üi − Xi )·δui = 0.
i=1
which shows that: "The virtual work produced by the forces acting on the particle during a virtual
Each particle k of the system of N particles with mass mk satisfies the dynamic equilibrium:
where
• Xik are the force components composing the known external resultant force;
• Rik are the components of the unknown resultant reaction forces resulting from the kinematic
The virtual displacements for every particle k can be written in the form:
δuik = u∗ik − uik (i = 1, 2, 3)
The virtual work principle for the system of particles is obtained by projecting the dynamic equilibrium
However, for a system of N particles (let assume, as an example, a simple system with rigid bonds),
projecting the equations on the kinematically admissible displacement directions consist in summing the
equilibrium equations in the direction of the constraint such that the unknown forces of bonding
disappear.
So, since the reaction forces do vanish when projecting the equations of motion onto kinematically
admissible displacement directions, the virtual work principle for the system of particles takes the form
N P
P 3
(mk üik − Xik )δuik = 0.
k=1 i=1
"The virtual work of the forces effectively applied onto a system of particles is zero with respect to any
1.2 Starting from the description of the state of a system of N particles in terms of the equation given
by ( 1.1), describe the different types of kinematic constraints that may exist in mechanical systems.
displacement components uik (also called degrees of freedom). Mathematically, the instantaneous
configuration is given by
where,
But in most mechanical systems, the particles are submitted to kinematic constraints which restrain
their motion.
Different types of kinematic constraints may exist in mechanical systems. Here they are:
Holonomic constraints:
They are defined by implicit relationships of type: f (ξik , t) = 0. If there is no explicit dependance with
respect to time, the constraints are said to be Scleronomic constraints, otherwise they are rheonomic.
"A holonomic constraint reduces by one the number of degrees of freedom of the system".
Non-holonomic constraints:
They often take the form of differential relations: f (ξ˙ik , ξik , t) = 0 . Such relationships are generally
not integrable. They don’t allow reduction of the number of degrees of freedom of the system.
where the qi are these generalized coordinates. What we show in the red box is that we can express the
n
X ∂Uik
δuik = δqs
s=1
∂qs
n
N P
3
P P ∂Uik
(mk üik − Xik ) δqs = 0.
s=1 k=1 i=1 ∂qs
Here, the system kinematics is described by 4 instantaneous position components: ξ11 , ξ21 , ξ12 , ξ22 .
2 holonomic constraints:
2 2
= l12
ξ11 + ξ21 (Pythagore)
(ξ12 − ξ11 )2 + (ξ22 − ξ21 )2 = l22 (Pythagore)
Hamilton’s principle is no more than a time-integrated form of the virtual work principle for N particles
leads to
Zt2 "X
N X
3
#
(mk üik − Xik )δuik dt = 0 (1.2)
t1 k=1 i=1
where δuik are the arbitrary but compatible virtual displacements which verify the end conditions
The displacements of the system are expressed using the generalized displacements in the form:
n
P ∂Uik
δuik = ∂qs δqs
s=1
1) Let assume that the applied forces Xik derive from a potential V:
∂V
Xik = − (1.4)
∂uik
∂V
where Qs = − ∂q s
.
d
mk üik δuik = (mk u̇ik δuik ) − mk u̇ik δ u̇ik
dt
(1.6)
d 1
= (mk u̇ik δuik ) − δ( mk u̇ik u̇ik )
dt 2
Owing to the definition of the kinetic energy T of the system,
N 3
1 XX
T = mu̇ik u̇ik (1.7)
2 i=1
k=1
The time-integrated form of the virtual work principle may be rewritten in the form:
" N X
3
#t2 Zt2
X
− mk u̇ik δuik + δ (T − V )dt = 0 (1.8)
k=1 i=1 t1 t1
where the first term is equal to 0 because of the end conditions. Indeed the details of calculation are
presented herebelow:
5 slide 17-22 and Ref. book p.23-24
Zt2 X
N X
3 Zt2 X
N X
3 Zt2 XN X 3
d 1
(mk u̇ik δuik )dt − δ mk u̇ik u̇ik dt − Xik δuik dt = 0,
dt 2
t1 k=1 i=1 k=1 i=1
t1 | {z } t1 |k=1 i=1
{z }
= T = δV
n
∂Uik X ∂Uik
u̇ik = + q̇s (1.9)
∂t s=1
∂qs
and therfore,
any compatible virtual displacement, arbitrary between both instants t1 t2 but vanishing at the ends of
the interval."
Zt2
δ (T − V )dt = 0
(1.12)
t1
δq(t1 ) = δq(t2 ) = 0
1.4 Explain what means the Hamilton’s principle written in the form
( 1.13). Starting from the general expressions of T and V in terms of
displacement and velocity, deduce the Lagrange equations.6
1.4 Explain what means the Hamilton’s principle written in the form ( 1.13). Starting from the
general expressions of T and V in terms of displacement and velocity, deduce the Lagrange equations.7
Zt2
δ (T − V )dt = 0
(1.13)
t1
δq(t1 ) = δq(t2 ) = 0
Let us define a set of generalized coordinates q1 , q2 , . . . , qn , each point in this space corresponds to one
configuration of the system. Then, the time evolution of the system defines a curve in the configuration
space.
The integral of the equation (1.14) is called an action integral, as it depends of the path from t1 to t2 ,
In this context, the real trajectory of the system is such that the action integral I, defined in equation
(1.14) remains stationnary (i.e. is minimized), with respect to any compatible virtual displacement δq.
Those compatible virtual displacements are arbitrary between t1 and t2 , but equals 0 in t1 and t2 .
The word stationnary means that the difference of the action integral is zero to the first order of δq.
Then, the Hamilton principle reduces to the expression of equation (1.15), given that it respects the end
Summary:
Starting from the general expressions of T and V deduce the Lagrange equations.
The general expression of the kinetic and potential energy respectively, expressed in the generalized
T = T (q̇s , qs , t) (1.16)
V = V(qs , t) (1.17)
As δqs (t1 ) = δqs (t2 ) = 0, the first term of (1.21) vanishes, and (1.20) reduces to
n
!
Z t2
X ∂T d ∂T
+ Qs − δqs dt = 0 (1.22)
t1 s=1
∂qs dt ∂ q̇s
As the displacement δqs is arbitrary, the equation (1.22) reduces for each of the n generalized
where the two first terms are the generalized inertia forces associated to qs and the last one is the
1.5 Starting from the definition of the kinetic energy for a system of N
masses (equation (1.24)), and using the concept of generalized
coordinates, uik (xjk , t) = Uik (q1 , ..., qn , t), explicit the structure of the
kinetic energy. Inserting this into Lagrange’s equations, describe then
the classification of generalized inertia forces.8
1.5 Starting from the definition of the kinetic energy for a system of N masses (equation (1.24)), and
using the concept of generalized coordinates, uik (xjk , t) = Uik (q1 , ..., qn , t), explicit the structure of the
kinetic energy. Inserting this into Lagrange’s equations, describe then the classification of generalized
inertia forces.9
Structure of the kinetic energy
We defined previously the total kinetic energy for a system of N masses as in equation (1.24).
N 3
1 XX
T = mk u̇ik u̇ik (1.24)
2 i=1
k=1
8 slide 25-27 and Ref. book p.27-29
9 slide 25-27 and Ref. book p.27-29
Deriving the expression of the generalized coordinates with respect to time, we get
N 3 n
! n
!
1 XX ∂Uik X ∂Uik ∂Uik X ∂Uik
T (q, q̇, t) = mk + q̇s + q̇s = T0 + T1 + T2 (1.26)
2 i=1
∂t s=1
∂qs ∂t s=1
∂qs
k=1
Where the respective contributions Ti are homogeneous forms of degree i in the generalized velocities q̇s .
1. T0 :
N 3 2
1 XX ∂Uik
T0 = mk = T0 (q, t) (1.27)
2 i=1
∂t
k=1
It represents the transport kinetic energy, of the whole system, as it corresponds to the
2. T1 :
N X
n X 3
X ∂Uik ∂Uik
T1 = mk q̇s (1.28)
s=1 k=1 i=1
∂t ∂qs
3. T2 :
n n N 3
1 XXXX ∂Uik ∂Uik
T2 = mk q̇s q̇r (1.29)
2 s=1 r=1 i=1
∂qs ∂qr
k=1
It represents the relative kinetic energy, as it corresponds to what is left when the explicit
First of all, we must express the terms T1 and T2 with the use of the Euler’s formalism.
Euler’s theorem:
satisfied:
n
X ∂f
xf = mf (1.30)
i=1
∂xi
n
X ∂T1
T1 = q̇s (1.31)
s=1
∂ q̇s
n
1 X ∂T2
T2 = q̇s (1.32)
2 s=1 ∂ q̇s
Those are the forces obtained by assuming that the constraints do not depend explicitly on time
(i.e. ∂Uik /∂t = 0). Then, the remaining terms are only in T2 .
∂T2 d ∂T2
− (1.36)
∂qs dt ∂ q̇s
n
X ∂ 2 T1 ∂T1
Fs = − q̇r + (1.37)
r=1
∂ q̇s ∂qr ∂qs
Considering that f = ∂T1 /∂qs , and that xi = q̇r , we can apply Euler’s formalism:
n
X ∂f ∂T1 X ∂ 2 T1
mf = xi ⇔ = q̇r (1.38)
i
∂xi ∂qs r=1
∂ q̇r ∂qs
n n
∂ 2 T1 ∂ 2 T1
X X
Fs = q̇r − = q̇r grs (1.39)
r=1
∂qs ∂ q̇r ∂qr ∂ q̇s r=1
We note that the coefficients grs do not depend on the velocities q̇s . They only depend on the
generalized displacements and time. The complementary inertial forces have the nature of
1.6 Describe how to apply the Hamilton’s principle in the case of constrained dynamical systems.11
Let’s remind that in the case of a system composed of N particles, we have 3N coordinates, such that
And we can express the displacement relatively to the generalized coordinates as follows:
fr (ξik , t) = 0 (1.42)
Now, let us suppose that a subset of m kinematic constraints12 isn’t explicitly satisfied by the choice of
We see that the constraints are verified if the δqs define a motion that is orthogonal to the direction
As the derivatives determine the directions of the reaction forces, we can express
∂fr
Rrs = λr (1.45)
∂qs
Where the coefficient λ of (1.45) denotes the intensity of the reaction associated to constraint fr , and is
called a Lagrange multiplier. Hence, substituting (1.45) in (1.44), we get the virtual work of the
reaction forces
m X
X n m
X X ∂fr
Rrs δqs = λr δqs
r=1 s=1 r=1 s=1
∂qs
Xm
= λr δfr (1.46)
r=1
We can add the term (1.46) to the general expression of the virtual work, as obtained by the Hamilton’s
Where the Lagrange multipliers λr are determined to satisfy the conditions given by Eq. (1.44). In the
end, the Lagrange equations together with the complementary conditions form a system of n + m
∂V ∗
d ∂T2 ∂T2 ∂D ∂ ∂T1
dt ∂ q̇s
− ∂qs
= Qs (t) − ∂qs
− ∂ q̇s
+ Fs − ∂t ∂ q̇s
(s = 1, ..., n) (2.1)
where we have:
• V ∗ = V − T0 , the potential modified by the transport kinetic energy (later, will be called
This last is simply the combination of, on the one hand, the Lagrange equation for a non-conservative
and on the other hand, the inertia forces which allow to rewrite the left part of the equation (2.2) in the
3 following terms:
2. Relative inertia forces obtained by assuming that the constraints do not depend explicitly on
∂Uik
time ∂t =0 :
d ∂T2 ∂T2
− + . (2.4)
dt ∂ q̇s ∂qs
n
X
Fs = q̇s grs . (2.5)
r=1
T0 , T1 and T2 are respectively the transport, the mutual and the relative kinetic energy.
Equilibrium configuration
Let us call the equilibrium configuration of a system, a time-independent configuration such as:
V ∗ (q, t) = V ∗ (q)
Qs (t) = 0.
Moreover, let us assume that the non-conservative forces acting on the system are zero when q̇s (t) = 0.
∂V ∗ ∂(V − T0 )
= =0 (s = 1, ..., n), (2.6)
∂qs ∂qs
where the imposed transport kinetic energy T0 is supposed to correspond to a uniform rotation or
Stable equilibrium position when the system doesn’t undergo overall motion
If the system does not undergo overall motion, its kinetic energy reduces to only quadratic term
T = T2 (q̇) in the generalized velocities and the equilibrium position is solution of:
∂V
=0 (s = 1, ..., n). (2.7)
∂qs
Since the potential energy is defined only to a constant, we choose it to be zero at equilibrium i.e.
V (0) = 0 when qs = 0 (s = 1, ..., n). Since the generalised displacements qs represent deviations from
equilibrium, the potential as well as kinetic energy could be expanded in the form of a Taylor series in
By using the two last linearized forms, the Lagrange equations provide the system of equations of
motion governing the free vibrations of a conservative system around a stable equilibrium position as
follow:
Mq̈ + Kq = 0. (2.8)
Let us now consider the more general case of a system undergoing steady motion (uniform rotational or
translational transport motion). Its equilibrium configuration defined by equation (2.6) corresponds to
∂V
equilibrium between restoring forces of potential origin ( ∂qs
term) and centrifugal forces ( ∂T
∂qs term). It
0
is an equilibrium configuration in the sense that the q̇s , which represent the relative velocities with
respect to the transport motion, vanish although the system itself is not at rest. The following results
will be useful for analysing the stability of systems in steady motion such as rotating systems. Let us
n n
∂ 2 T0
∗ 1 XX ∗ ∗
V (q) ≈ k qr qs where ksr = ksr − (2.9)
2 s=1 r=1 sr ∂qr ∂qs q=0
n n
1 XX
T2 (q̇) ≈ msr q̇r q̇s (2.10)
2 s=1 r=1
n
X n X
X n
T1 ≈ cs q̇s + fsr q̇s qr (2.11)
s=1 s=1 r=1
∂ 2 T1
∂T1
where cs = and fsr =
∂ q̇s q=0 ∂ q̇s ∂qr q=0
/!\ cs and fsr remain constant in a transport motion with uniform speed.
The equilibrium configuration generated by steady motion remains stable as long as the condition
V ∗ ≥ 0 is verified, corresponding to the fact that effective stiffness matrix remains positive definite. In
the neighbourhood of such a configuration, the linearized equations of motion take the form:
n
X
∗
[msr q̈r + (fsr − frs ) q̇r + ksr qr ] = 0 (s = 1, ..., n), (2.12)
r=1
M q̈ + G q̇ + K∗ q = 0. (2.13)
The structure of this equation renders the analysis of systems undergoing overall motion is therefore
more difficult.
2.2 Introduce the concept of normal modes of vibration. Examine also the
case of systems with a neutrally stable equilibrium configuration.
Discuss the orthogonality relationships of normal modes.15
Another case which deserves attention is that of systems that can exhibit a behaviour corresponding to
a global motion but where either the global velocity is negligible or the additional kinetic energy
generated by the global motion remains constant. The determination of global motion and the analysis
of relative motion are then uncoupled problems, the relative motion occurring about a neutrally stable
(and thus non-unique) equilibrium position as suggested by figure 2.1. Examples of such systems are:
• any structure supported on springs which are very soft compared to its stiffness.
In these cases, non-zero displacement modes exist such that such that the associated potential
energy vanishes:
1 T
V (q) = q Kq = 0 for q 6= 0 . (2.14)
2
These solutions, denoted u, are called rigid-body modes. They correspond either to transport motion
In order to solve the free vibration linear equations (2.8) let us seek a particular solution such that the
generalized coordinates are governed to a common factor by the same temporal law: q = φ(t) x where x
is a vector of constants, known only to a common scale factor, which governs the shape of the motion.
Motions described by the temporal law are called synchronous since all degrees of freedom follow the
6 0)
1. Systems with a stable equilibrium configuration (dtm|K| =
The matrix K is then non-singular. Equation (2.16) can be put in the form:
φ̈(t)
Kx = − Mx. (2.17)
φ(t)
16 Good to know,M is non-singular by definition.
φ̈(t)
= λ = ω2 K − ω 2 M x = 0,
− and (2.18)
φ(t)
where λ is a real and positive constant with respect to time. For the differential equation, the
solution is:
where ω is called the circular frequency. For the eigenvalue problem (non-trivial solutions and for
r = 1, ..., n):
• ωr2 are called the natural frequencies, root of the algebric equation det(K − ω 2 M ) = 0.
• x(r) represent the synchronous free vibration shapes of the system and are called normal
modes
Remember that this system admit rigid-body displacement modes which are solutions of:
Ku = 0 (2.20)
q = η(t) u (2.21)
η̈(t) M u = 0 (2.22)
η̈(t) = 0. (2.23)
The number of rigid-body modes u(i) is equal to the degree of singularity of the stiffness matrix
(i.e. the number of rows which have to be removed to achieve to a non-singular K).
The normal modes verify the orthogonality relationships. Note that these last hold when several
eigensolutions degenerate and give rise to a multiple eigenvalue (which is the case when rigid-body
Its meaning is: the virtual work produced by the inertia forces of mode r in a virtual displacement
Its meaning is: the virtual work produced by the elastic forces of mode r in a virtual displacement
Let us note that very often, the modes are mass normalized (µr = 1) such that:
xT(s) K x(r) = ωr2 δrs
(2.27)
xT(s) M x(r) = δrs
In the absence of external loading, the response of the system may be determined by superposition of
Let us solve the system (2.29) by making the change of generalised coordinates:
M X η̈ + K X η = 0. (2.30)
Premultiplying by XT and taking into account the orthogonality relationship (equations (2.25)
and (2.26)) leads to the system of n uncoupled equations called the normal equations:
eigenmodes:
If we put all this in the q(t) form, the general solution is expressed in the form:
n
X
q(t) = (αs cos ωs t + βs sin ωs t) x(s) (2.33)
s=1
The 2n constants (αs and βs ) are obtained from the initial conditions such as:
It shows that αs and βs ωs are the modal coordinates of q0 and q̇0 respectively.
When the system possesses m rigid-body modes, the solution found in equation (2.33) for the
response to initial conditions is not applicable, some of the frequencies being zero. However a
similar reasoning based on modal superposition can be followed provided that the contributions of
rigid-body and elastic modes are handled separately. So we express the solution as:
The dynamic influence coefficient hkl (ω 2 ) called also admittance or dynamic flexibility of a system
represents the forced vibration amplitude of the degree of freedom qk for a unitary amplitude harmonic
loading applied on degree of freedom ql at frequency ω. The dynamic influence coefficient is nothing
else than the inverse of the impedance matrix and it is possible to get it by the following steps.
Forced vibration of an n-degrees-of-freedom oscillator is defined as the motion resulting from the
where ω is the excitation frequency and s describes the spatial distribution of the excitation amplitude.
(2.36)
⇔ x = ( K − ω 2 M )−1 s.
| {z }
Impedance matrix
At last, we find the dynamic influence matrix such as it’s equal to the Frequency Response Function
matrix:
This last equation highlight the influence of such coefficient in the harmonic domain.
18 Ch2, part 2, slide 4-14 and Ref. book p.83-90
It is possible to get this spectral expansion solving the equation (K − ω 2 M)x = s by an eigenmode
series expansion:
m
X n−m
X
x= αi ui + βs xs , (2.38)
i=1 s=1
where m are the possible rigid-body modes and αi and βs are (thanks to to the eigenmode
uTi s xTs s
αi = − βs = (2.39)
ω 2 µi (ωs2 − ω 2 )µs
So:
m n−m
!
1 X ui uTi X xs xTs
x= − 2 + s (2.40)
ω i=1 µi s=1
(ωs2 − ω 2 )µs
m n−m
1 X ui uTi X xs xTs
H(ω) = − 2 + (2.41)
ω i=1 µi s=1
(ωs2 − ω 2 )µs
m n−m
2 1 X uki uli X xks xls
Hkl (ω ) = − 2 + 2 − ω 2 )µ
(2.42)
ω i=1 µi s=1
(ωs s
In case with no rigid body modes (u = 0), the FRF is reduced to:
n
X xks xls
H(ω 2 ) = (2.43)
s=1
(ωs2 − ω 2 )µs
An important property of the principal coefficient Hkk is that are always increasing with excitation
This property implies that between each two successive eigenfrequencies ωr and ωr+1 there is an
anti-resonance frequency ωrk (ωr < ωrk < ωr+1 ), see figure 2.2. To have the resonances and
anti-resonances we can rewrite the diagonal element Hkk (ω 2 ) in the form of a ratio between two
polynomials in ω 2 of order n for the denominator and of order n − 1 for the numerator, so we have n
Figure 2.2: Principal dynamic influence coefficient for a system with no rigid-body mode.
where ωsk are the anti-resonance frequencies and ωs are the resonance frequencies, while gkk is the static
In this case, we have m rigid body modes. The principal coefficients are in the form:
m n−m
1 X u2ki X x2ks
Hkk (ω 2 ) = − 2
+ (2.47)
ω i=1 µi s=1
(ωs − ω 2 )µs
2
m n−m
dHkk 1 X u2ki X x2ks
2
= + >0 (2.48)
dω ( ω 2 )2 i=1 µi s=1
(ωs − ω 2 )2 µs
2
So when ω 2 → 0 then the slope of Hkk tends to infinity as it is possible to see in figure 2.3. In this case
Figure 2.3: Principal dynamic influence coefficient for a system with rigid-body modes.
we can notice that the system presents n − m resonance frequencies (excluding ωr = 0) and n − m
anti-resonances.
Pseudo-resonance
possible to obtain an harmonic type response of finite amplitude even when the excitation frequency
becomes equal to a resonance frequency ωl and this is possible when the loading produces no work on
eigenmode x(l) , i.e. the external loading verifies the following condition:
xT(l) s = 0 (2.49)
m n−m
!
1 X ui uTi X xs xTs
x= − 2 + s (2.50)
ω i=1 µi s=1
(ωs2 − ω 2 )µs
Let’s consider an undamped n-degree-of-freedom system subjected to an external force p(t). Its
M q̈ + K q = p(t) (2.51)
The eigenmodes xr of the undamped system can be used as a basis to represent the response of the
structure since the normal vibration modes provide a complete set for the expansion of an arbitrary
vector:
n
X
q(t) = ηs xs . (2.53)
s=1
Substituting the equation (2.53) in equation (2.51) and premultiplying by each eigenmode xr we obtain:
By exploiting the orthogonality property of the modes, we get n independent normal equations:
with the modal participation factor of mode r (µr is the modal mass of the rth mode):
xTr p(t)
φr (t) = , (2.56)
µr
So our response, thanks to the projection into the modal basis, is reduced to the solution of n uncoupled
single-degree-of-freedom systems.
The projection of the initial conditions in modal representation is obtained by multiplying the two
n
X n
X
q0 = ηs (0)xs and q̇0 = η̇s (0)xs . (2.57)
s=1 s=1
So we obtain:
xTr Mq0 xTr Mq̇ 0
ηs (0) = and η̇s (0) = . (2.58)
µr µr
Each of the normal equations η¨r (t) + ωr2 ηr (t) = φr (t) may be integrated in the form of a convolution
product. Indeed, using the impulsive response approach, the general solution of a normal equation
The last term of the previous equation is a convolution product known as Duhamel’s integral and
[In details] The convolution product can be integrated in closed form assuming a piecewise linear
variation of the excitation for undamped oscillator submitted to initial conditions. We can perform the
time integration in a recurrent manner. In this iteration manner in the time interval [tn , tn+1 ], we
consider ηr,n and η̇r,n as initial conditions at time tn , while ηr,n+1 and η̇r,n+1 as initial conditions at
time tn+1 = tn + ∆t. So the solution at step tn+1 can be expressed as:
Z tn+1
η̇r,n 1
ηr,n+1 = ηr,n cos ωr ∆t + sin ωr ∆t + φr (τ ) sin ωr (tn+1 − τ )dτ (2.60)
ωr ωr tn
The velocity is instead obtained computing the derivative of the previous equation.
Z tn+1
η̇r,n+1 = −ωr ηr,n sin ωr ∆t + η̇r,n cos ωr ∆t + φr (τ ) cos ωr (tn+1 − τ )dτ (2.61)
tn
Now let us introduce the linearity assumption for the excitation on the time interval and the frequency
parameter α (= ωr ∆t):
sin α sin α
1 − sinα α
ηr,n+1 cos α α ηr,n − cos α
α φr,n 1
= +
∆t η̇r,n+1 −α sin α cos α ∆t η̇r,n α sin α + cos α − 1 1 − cos α φr,n+1 ωr2
(2.62)
Generally, when we express the system response to external loading through mode superposition, it
occurs that the number of degrees of freedom is so large that the modal expansion has to be restricted
to a subset of k modes:
k
X
q(t) = ηs (t) xs . (2.63)
s=1
where g indicates a static load distribution applied with the time-variation law φ(t). For a system
initially at rest, the solution of the normal equation number r is given by:
Z t
ηr (t) = φr (τ )h(t − τ )dτ, (2.65)
0
with:
xTr p(t)
φr (t) = (2.66)
µr
t
xTs g
Z
ηs (t) = sin ωs (t − τ )φ(τ )dτ (2.67)
ωs µs 0
The global convergence of the response can be measured using the M-norm:
k
X
||q||2M = qT Mq = µs ηs2 (t) (2.68)
s=1
k Z t 2
X (xTs g)2 1
||q||2M = sin ωs (t − τ )φ(τ )dτ (2.69)
µs ω
s=1 | {z } | s 0 {z }
Spatialf actor T emporalf actor
Let us notice that the convergence is governed by two terms, the spatial factor and the temporal
factor.
Spatial factor gives a convergence of quasi-static type. This kind of converges occurs if the applied
load g admits a sufficiently accurate spatial representation on the basis of the k < n retained
eigenmodes. This is equivalent to assuming that the load distribution g must be nearly orthogonal to
the n − k omitted eigenmodes which therefore will not be excited or will be only slightly excited.
20 Ch2, part2, slide 21-27 and Ref. book p.95-98
Temporal factor gives a convergence of spectral type, conditioned by the convergence to zero of the
convolution products in the temporal term when progressing in the eigenspectrum of the system. It
thus depends both on the frequency content of the excitation and on the system eigenspectrum.
It is possible to increase the accuracy of the solution by the modal acceleration method. In this method
indeed only the inertial forces are expressed by modal superposition such as, starting from the equation
and then, by applying the truncated modal representation to only the inertia forces:
k
X
K q = p(t) − η̈s (t) Mxs , (2.71)
s=1
xTs p(t)
η¨s (t) + ωs2 ηs (t) = (2.73)
µs
Furthermore, taking into account that Kxs = ωs2 Mxs , the solution can thus be rewritten as:
k k
!
X −1
X xs xTs
q(t) = ηs (t)xs + K − p(t) (2.74)
s=1
ω2 µ
s=1 s s
| {z } | {z }
mode displacement solution correction term
Pn x(s) xT
Since K−1 can be rewritten as K−1 = s=1
(s)
ωs2 µs thanks to the spectral expansion, then:
k n
!
X X xs xTs
q(t) = ηs (t)xs + p(t) (2.75)
ω 2 µs
s=1 s=k+1 s
| {z } | {z }
mode displacement solution correction term
In conclusion, we can see that in the modal acceleration method, the mode displacement solution is
complemented with the missing terms from the modal expansion of the static response.
2.7 Show how to use the method of additional masses to approximate the
response of a system submitted to ground acceleration excitation.
TO DO
2.8 Explain why the concept of effective modal masses is useful when the
support of a structure undergoes a rigid body motion. Define what the
different terms represented in equation (2.76).
n1
X (uTs(i) γ(r) )2
mT = mS + (2.76)
r=1
µr
TO DO
M q̈ + C q̇ + K q = p(t) (3.1)
Let us consider the associated conservative system associated with the real system. It is governed by
the equations:
M q̈ + K q = 0 (3.2)
2
for which the eigenmodes and eigenfrequencies are noted respectively x(r) and ω0r with r = 1, ..., n.
Since the modes x(r) of the associated conservative system form a complete basis one can, as in the
undamped case, solve the system of equations ( 3.1) through the modal expansion:
n
X
q= ηs (t) x(s) (3.3)
s=1
n
X βrs 2
η̈r + η̇s + ω0r ηr = φr (t) r = 1, ..., n (3.4)
s=1
µr
with the modal damping coefficients: βrs = xT(r) Cx(s) . As the modal damping coefficients βrs don’t
vanish in general, the modal damping matrix [βrs ] is not diagonal so the normal equations remain
Let us analyze the free vibrations of a damped system starting from the homogeneous system of
equations:
M q̈ + C q̇ + K q = 0 (3.5)
and let us introduce the assumption that the damping terms are of an order of magnitude lower than
q = z eλt (3.6)
Without damping, we would have λk = ±iω0k and z(k) = x(k) , ω0k and x(k) being the eigenmodes and
2
(K − ω0k M) x(k) = 0 (3.8)
If the system is assumed to be lightly damped, it can be supposed that the λk and z(k) differ only
λk = iω0k + ∆λ
(3.9)
z(k) = x(k) + ∆z
2
{(−ω0k + 2iω0k ∆λ + (∆λ)2 )M + (iω0k + ∆λ) C + K}(x(k) + ∆z) = 0 (3.10)
2
(K − ω0k M) ∆z + (2iω0k M + C)x(k) ∆λ + iω0k C(x(k) + ∆z) ' 0 (3.11)
The light-damping assumption allows to neglect the terms in C∆λ and C∆z. We then obtain the
first-order relationship:
2
(K − ω0k M) ∆z + iω0k (C + 2∆λM)x(k) ' 0 (3.12)
from which, after multiplication by xT(k) , the correction to the eigenvalue may be extracted:
βkk (3.13)
∆λ ' −
2µk
providing the approximate expression:
βkk
λk ' iω0k − (3.14)
2µk
• Each eigenvalue correction takes the form of a real negative part and thus transforms each term of
• This first-order correction involves only the diagonal damping terms βkk = xT(k) Cx(k) and thus the
Therefore one is authorized to neglect the nondiagonal damping terms without significant alteration of
It is also possible to obtain from equation ( 3.12) the eigenmode correction through a modal expansion
n
X
∆z = αs x(s) (3.15)
s=1,s6=k
After substitution and premultiplication by xT(l) for l 6= k and taking into account the orthogonality
relationships,
i ω0k βks
αs = 2 − ω2 ) (3.16)
µs (ω0k 0s
n
X βks
z(k) = x(k) + i ω0k 2 2 ) x(s) (3.17)
µs (ω0k − ω0s
s=1,s6=k
2 2
This formula is valid only if the βkl are first-order quantities and if (ω0k − ω0s ) remains finite. It shows
that, as long as the eigenfrequencies of the associated conservative system are well-seperate, it may be
considered that the influence on the eigenmodes of the coupling through damping is of the same order
of magnitude as the damping coefficients βks . It also shows that the eigenmode correction is an
imaginary term, having as its consequence the fact that a free vivbration of the damped sustem is no
and degree of freedom qi is no longer in phase with another degree of freedom qj . For lightly damped
systems with well-seperate eigenvalues, the eigenmode expression ( 3.17) can be approximated by
z(k) ' x(k) and the general solution of the free vibration system takes the form
The discussion above shows that the modal damping assumption is in most cases physically consistent
with the low-damping assumption: when a system is weakly damped and when its eigenfrequencies are
well-seperate, the effect of the cross-damping terms βks , k 6= s on the eigenspectrum can be neglected
M q̈ + C q̇ + K q = f eiωt (3.21)
and let us assume that the response is limited to the forced term q = z eiωt , so that the excitation and
(K − ω 2 M + i ω C) z = f (3.22)
or
z = H(ω)f (3.23)
The modal damping assumption allows us to construct the modal expansion of the dynamic influence
coefficient matrix in terms of undamped eigenmodes. Indeed, if the response amplitude is developed in
terms of eigenmodes:
n
X
z= αs x(s) (3.24)
s=1
the orthogonality relationships together with the modal damping assumption provide the coefficients:
xT(r) f
αr = 2 − ω2 + 2 i ω ω ) (3.25)
µr (ω0s s 0s
Obviously, when the excitation frequency ω tends towards 0, the dynamic influence coefficients converge
to the static influence coefficients gkl , as long as the system does not contain rigid-body modes.
The most frequent objective of a vibration test is to determine the modal characteristics (eigenmodes
and eigenfrequencies) of the associated conservative system. A natural procedure consists of forcing the
22 slide 33-34 and reference book p.160-162
23 slide 37-39 and reference book p.164-170
vibration of the structure in each of its eigenmodes successively, requiring proper tuning of the
frequency and force amplitudes in order to reach the appropriate excitation for each mode.
When a harmonic vibration test is performed on a damped system, the amplitudes of applied forces and
the response amplitudes at the different points verify the complex relationship:
(K − ω 2 M + i ω C) z = f (3.28)
Extracting a given eigenmode of the associated conservative system through appropriate excitation is
2
(K − ω0k M + i ω0k C) x(k) = f(k) (3.29)
2
where f(k) is the excitation mode which allows one to achieve the appropriate excitation. Because ω0k
and x(k) are eigensolutions of the associated conservative system, one has:
2
(K − ω0k M) x(k) = 0 (3.30)
and the expression of the excitation which makes it possible to excite eigenmode x(k) at its resonance
This shows that the excitation, when appropriate, is in phase with the dissipation forces and thus has a
In order to understand the concept of phase quadrature, let us consider the representation of the
equation of motion in the complex plane. When the excitation is approriate, all degrees of freedom are
2
(K − ω0k M + i ω0k C) x(k) − f(k) = 0 (3.32)
This shows that when the excitation is appropriate, the elastic forces Kx(k) and the inertia forces
2
ω0k Mx(k) cancel each other as if the system were vibrating in an undamped fashion, while the
excitation force f(k) equilibrates the damping forces which are proportional to the velocity and thus
Figure 3.1: Representation of the phase quadrature in the complex plane under appropriate excitation.
”The structure vibrates according to one of the eigenmodes of the associated conservative system if and
only if all degrees of freedom vibrate synchronously and have a phase lag of π/2 with respect to the
excitation.”
Once the phase quadrature criterion is verified during the experimental test, one may take note of ω0k
3.4 Explain what are the different terms of the FRF matrix written in the
form (see below) equation ( 3.34). How can this equation be used for
modal identification? What is the information that can be extracted
from the identified variables? What is the difference between a complex
mode and a real mode? 24
where the bar denotes the complex conjugate and ρj is the squared norm of the eigenvector z(j) .
n
!
X Ars(k) Ārs(k)
Hrs (ω) = + (3.35)
i ω − λk i ω − λ̄k
k=1
with
zr(k) zs(k)
Ars(k) = (3.36)
ρk
For the complex mode, the components vary with the lag φ between the components.
24 s.40
4 Continuous systems
4.1 Define the Green’s strain tensor. Apply Hamilton’s principle to a
continuous system and derive the equations of motion.25
Let us consider an elastic body undergoing in time a certain motion measured from the undeformed
configuration. The latter is supposed to be time-invariant, and corresponds to the equilibrium position
Starting by calculing the deformation at every point of the volume with respect to the reference
configuration, i.e. to express the deformation at a point A of coordinates xi before deformation. Let B
be a second point in the neighbourhood of A and occupying the position xi + dxi in the undeformed
state. After deformation, the points A and B occupy the new positions xi + ui and xi + ui + d(xi + ui ),
Let us denote ds0 and ds as the lengths of segment AB before and after deformation:
ds20 = dxi dxi = dx21 + dx22 + dx23
(4.1)
ds2 = d(xi + ui ) d(xi + ui )
By taking account of:
∂ui
dui = dxj i = 1, 2, 3 (4.2)
∂xj
Let us next define the components of Green’s symmetric strain tensor ij in such a way that the
Hence, according to the previous expressions, the strain components may be written:
1 ∂ui ∂uj ∂um ∂um
ij = 2 ∂xj + ∂xi + ∂xi ∂xj (4.5)
The displacement variational principle is Hamilton’s principle expressed for a continuous system. Let us
Among the feasible trajectories of the system subjected to the restrictive conditions:
at the end of the considered time interval [t1 , t2 ], the real trajectory of the system is the stationary point
Z t2 Z t2
δ Lg [u] dt = δ (T − V) dt = 0 (4.8)
t1 t1
The kinetic energy of the continuous system of figure 4.1 may be evaluated from an integration over
the reference volume (by making use of Einstein’s notation), it takes the form:
Z
1
T(u) = ρ0 u̇i u̇i dV (4.9)
2 V0
The total potentiel energy V results from the summation of the strain energy of the body and the
Z
Vint = W (ij ) dV (4.11)
V0
and the external potential energy is computed by assuming the existence of body forces: X̄i and
We apply Hamilton’s principle to the continuous system with the assumption that the forces applied to
expressing the dynamic equilibrium of the body in the volume and on the surface:
∂ ∂uj
σij + σim − ρ0 üj + X̄j = 0 in V0
∂xi ∂xm
(4.15)
∂uj
tj = ni σij + σim = t̄j on Sσ
∂xm
Linear case
When both rotations and displacements are small, the quadratic term can be neglected:
1 ∂ui ∂uj
ij = + (4.16)
2 ∂xj ∂xi
displacements and rotations. They express equilibrium in the undeformed state V0 ' V .
4.2 Write the Hamilton’s principle in the case of prestressed structures and
discuss the case of structural stability analysis.26
0
A structure is said to be prestressed when it is submitted to a prescribed fiels of initial stresses σij and
to the associated field of initial strains 0ij , both being independent in time. The analysis of such a
∗
structure consists thus in determining the fields σij and u∗i which are added to this initial state.
26 slide 16-17 and reference book p.227-230
The notations adopted are summarized by Figure 4.3, and the relationships
0 ∗
σij = σij + σij (4.19)
express respectively the strain state and the stress state measured in the prestressed configuration V ∗
adopted as reference.
where the variation operates on the displacement from the initial state, δui = δu∗i , T and V are the
kinetic and potential energies accumulated by the structure when passing from the initial to the
deformed state.
This result shows that the additional displacement field u∗i of a prestressed structure is obtained by a
linear analysis from the prestressed state to the deformed state, with the geometric strain energy due to
The theory of the prestressed case forms the basis of structural stability analysis: the latter consists in
computing the prestressing forces applied to the structure which render possible the existence of an
equilibrium under only geometrically linear and nonlinear elastic forces. Hamilton’s principle can then
As shown by equation ( 4.21), prestressing modifies the vibration eigenfrequencies, and the limit case of
Principle: Subdivision of the structure into a finite number of elements of simple geometry with
well-identified structural behaviour (bar, beam, membrane, plate, shell, 3-D solid, ...).
• The intensity parameters of these functions (i.e. the generalized coordinates) are local values of
⇒ If both conditions are strictly satisfied, the approximation obtained is kinematically admissible F.E.
Bar in extension:
Let us consider the case of a bar in extension possibly subjected to distributed loads X̄(t). The bar is
first divided into N elements of length l as shown in figure 5.1. We call P 1 and P 2 the loads at the
ends of an element, representing the sum of applied external forces on the nodes and of inner normal
forces originating from interaction with neighbouring elements. Next, the displacement field in the
where u1 (t), u2 (t) are the connector degrees of freedom, namely the axial displacements at both element
φ1 (x), φ2 (x) are the shape functions of the element, chosen in such a way that:
x x
φ1 (x) = 1 − and φ2 (x) = (5.3)
l l
where
Z l
P1 (t)
pe (t) = ΦTe X̄(x, t) dx + (5.10)
0 P2 (t)
We verify that:
• the mass associated to the rigid body mode, obtained by projecting the mass matrix on it, yields
If we apply the virtual variation to the generalized coordinates q, we arrive at the equations of motion
Me q̈e + Ke qe = pe (5.13)
where pe contains the a priori unknown reaction forces between element and its neighbours.
Assembly process
In order to express dynamic equilibrium for the global system of figure 5.1, let us construct the matrix
qT = u0
u1 u2 ... uN (5.14)
qe = Le q (5.15)
where the localization operator Le is a Boolean matrix with dimension 2x(N + 1). It contains only 1
and 0 terms and can be seen as a filter picking the nodal displacements qe of an element out of the
By summing all the elements of the system, the structural variational equation becomes:
Z N
t2 X Z N
t2 X
1 T 1
δ q̇e Me q̇e − qTe Ke qe dt + δ qTe pe dt = 0 (5.16)
t1 e=1
2 2 t1 e=1
We define the structural mass matrix M, the structural stiffness matrix K and the structural load
vector p.
Z t2 Z t2
1 T 1
δ q̇ M q̇ − qT K q dt + δ qT p dt = 0 (5.18)
t1 2 2 t1
By taking the variation of this final discretized expression, one obtains the discretized structural
M q̈ + K q = p(t) (5.19)
5.2 Apply the finite element method to beam in bending without shear
deflection. Extend the method to the case of a three-dimensional beam
element.28
We consider the case of a beam in bending (figure 5.2), possibly excited by a distributed load p̄(x, t).
In the context of the kinematic Bernoulli assumptions for beams with no shear deflection, let us express
the strain energy of the beam element (with no prestress) in the form:
L 2
∂2w
Z
1
Vint,e = EI dx (5.20)
2 0 ∂x2
x
In terms of nondimensional variable ξ = l over the element domain, the approximation of the
where the shape functions φi (ξ), (i = 1 ... 4) are the third-order Hermitian polynomials, matching the
kinematic conditions:
dφi
φ1 (0) = 1 φ˙1 (0) = 0 φ1 (1) = φ̇1 (1) = 0 where φ̇i =
dξ
φ2 (0) = 0 φ˙2 (0) = l φ2 (1) = φ̇1 (1) = 0 (5.23)
φ3 (ξ) = φ(1 − ξ) φ4 (ξ) = −φ2 (1 − ξ) = 0
qTe = w1
ψ1 w2 ψ2 (5.25)
For an element of uniforme characteristics excited by a constant distributed load p̄0 , we explicitly
obtain:
12 6l −12 6l
EI 6l 4l2 −6l 2l2
Ke = 3 (5.30)
l −12 −6l 12 −6l
6l 2l2 −6l 4l2
156 22l 54 −13l
ml 22l 4l2 13l −3l2
Me = (5.31)
420 54 13l 156 −22l
−13l −3l2 −22l 4l2
p̄0 l
pTe = l
− 6l
1 6 1 (5.32)
2
It is easily verified that the stiffness matrix has one translational rigid body mode:
uT(1) = 1
0 1 0 (5.33)
uT(2) = 1 − 2l l
1 2
(5.34)
The quadratic forms uT(1) Mu(1) and uT(2) Mu(2) are then equal to the translation and rotatory inertias
ml3
ml and 12 .
Kx = ω 2 M x (6.1)
To reduce the size of K and M, we eliminate a subset of DOFs, which gives the nR remaining and nC
condensed coordinates respectively written xR and xC (nR n). Eq. 6.1 is therefore partitionned as
follows:
KRR KRC xR 2 MRR MRC xR
=ω (6.2)
KCR KCC xC MCR MCC xC
The inertia forces FC are negligible if the masses affected to the condensed DOF are negligible or equal
Finally, the general stiffness and mass matrices are given by:
(
−1
K = RT KR = KRR − KRC KCC KCR
T −1 −1 −1 −1
M = R M R = MRR − MRC KCC KCR − KRC KCC MCR + KRC KCC MCC KCC KCR
(K − ω 2 M )xR = 0 (6.6)
Notice that the Guyan’s reduction is only valid if the inertial forces FC are indeed negligible, and that
it can be shown that this method always lead to an excess approximation of the eigenvalue spectrum
anyway.
is computed by solving the static problem (with nR second members): KCC RCR = −KCR .
29 Ref. book p.481-482 & slides 6.10-13
6.2 Describe the concept of mechanical impedance and how it can be used
for substructuring a finite element model.30
Large number of DOFs ⇒ dynamic analysis unfeasible. Hence we break down the system in several
substructures (Fig. 6.1) and we study their response to forces applied on their interface boundaries.
The result is that we have a reduced set of variable which approximate each substructure and by doing
Z(ω 2 ) = K − ω 2 M = H −1 (ω 2 ) (6.8)
and is the matrix which relates the applied force amplitudes s to the displacement amplitudes q as
follows:
s = Z(ω 2 )q (6.9)
Since the internal degrees of freedom q1 (cf. Fig. 6.1) are not externally loaded, we may write:
Z11 (ω 2 ) Z12 (ω 2 ) q1
0
= (6.10)
Z21 (ω 2 ) Z22 (ω 2 ) q2 s2
Taking the first equation of this system to eliminate the internal degrees of freedom gives:
−1
q1 = −Z11 Z12 q2 , (6.11)
∗
Z22 (ω 2 )q2 = s2 (6.12)
where
∗ −1
Z22 = Z22 − Z21 Z11 Z12 . (6.13)
∗
Notice that Z22 admits as poles the zeros of Z11 which are the eigenfrequencies of the subsystem with
Considering the subsystem clamped on its boundary, we can compute the eigensolutions of the
subsystem:
−1
Based on the spectral expansion of Z11 , it can be shown that the reduced impedance matrix takes the
form:
∗ −1
Z22 = K22 − K21 K11 K12
−1 −1 −1 −1
− ω 2 (M22 − M21 K11 K12 − K21 K11 M12 + K21 K11 M11 K11 K12 )
n1
X (K21 − ω̄s2 M21 ) x̄(s) x̄T(s) (K21 − ω̄s2 M21 )T
− ω4
s=1
ω̄s4 (ω̄s2 − ω̄ 2 )
Then one can deduce that the dynamics of the substructure may be described in terms of static modes
associated to the interface (support) displacements and in terms of the internal modes (when the
interface is fixed).
This forms the main idea underlying the Craig-Bampton method (substructuring a finite element
model).
With the assumption of viscous dissipation in the Lagrange equations of the system:
1 T
D= q̇ C q̇ ≥0 (7.1)
2
M q̈ + C q̇ + K q = p(t) (7.2)
Direct multistep integration methods for first-order differential systems in the form:
ẏ = f (y, t) (7.3)
where h = tn+1 − tn is the time step and yn+1 is the solution to equation (7.3) at time tn+1 calculated
from the solutions at the m preceding times, from their derivatives and from the derivative of yn + 1
itself.
For β0 6= 0, the time integration scheme is said to be implicit, since the solution at time tn+1 is a
function of its own time derivative. Therefore, the integration relationships have to be recast before
they can be solved. The solution method becomes iterative in the nonlinear case.
For β0 = 0, the time integration scheme is said to be explicit, since yn+1 can be deduced directly from
When αj = 0 and βj = 0 for j > 1, the integration scheme corresponds to a one-step method, since
the system at time tn+1 is only a function of its previous state at time tn .
7.2 Newmark
7.2.1 Applying Newmark’s integration formulas (equation(7.5)), describe the Newmark’s
integration algorithm for linear systems.32
q̇n+1 = q̇n + (1 − γ) h q̈n + γ h q̈n+1
(7.5)
2 2
qn+1 = qn + h q̇n + (0.5 − β) h q̈n + h β q̈n+1
Time-integration of the dynamic equilibrium equations yields:
M q̈ + C q̇ + Kq = p(t) (7.6)
31 Ch7,part1, slide 3-4 and Ref. book p.513-514
32 Ch7,part1, slide 17 and Ref. book p.522-523
At time tn+1 , and associated with the iteration matrix [M + γhC + βh2 K], we have the linear system
of equations:
2 1 2
[M + γhC + βh K]q̈n+1 = pn+1 − C(q̇n + (1 − γ)hq̈n ) − K qn + hq̇n + − β h q̈n (7.7)
2
7.2.2 According to Fig. 7.1 and the expressions of amplitude and periodicity error,
discuss the stability and accuracy of the method with respect to the choice of
parameters and time-step.33
1
Fig. 7.1 shows that γ must be ≥ 2 at all time for the algorithm to be stable. Unconditionnal stability is
met if β ≥ 1/4 (1/2 + γ)2 as shown in Fig. 7.1. Conditionnal stability impose a time-step:
4
(γ + 0.5)2 − 4β ≤ 2
ωmax h2
Stability limit
⇔ h≤
ωmax
You can land on the values displayed in Tab. 7.1 just by injecting the values of the free parameters into
the equation.
- The purely explicit scheme is not used in practice, since it is in any case unstable.
- The Fox and Goodwin algorithm does not generate damping and leads to a periodicity error of fourth
- The average constant acceleration algorithm is the best unconditionally stable scheme since it doesn’t
- The average constant acceleration scheme modified to introduce numerical damping is characterized
by a numerical damping ratio growing linearly with ωh. It is interesting to note that the first-order
error that results from taking γ 6= 1/2 manifests itself in the form of excessive numerical dissipation,
but not in period disrepancy since the latter remains of second-order for α 6= 0.
1 1
Finally, taking β = 6 or = 4 convey a better stability but increase the periodicity error as much as β
increase. The amplitude error will be non-existant as long as γ = 21 , which is convenient because it is
the value which gives the larger domain of unconditional stability. If γ 6= 12 , the amplitude error is given
It is used to allow the damping in the modelling without degrading the order of accuracy. The HHT-α
method is a generalisation of the Newmark method which corresponds to the case where the lag in the
damping is not equal to α and is null. Its principle consists in applying the Newmark’s formula (7.5)
whereas the time-discrete equilibrium equations are modified by averaging the internal (i.e. elastic and
damping) forces and the external forces between both time instants.
With
Mq̈n+1 + (1 − α)f (qn+1 , q̇n+1 ) + αf (qn , q̇n ) = (1 − α)g(qn+1 , tn+1 ) + αg(qn , tn ) (7.9)
The HHT-α is at least 2nd-order accurate and unconditionally stable if the following conditions are
satisfied:
1 1 1
γ= + α, β= (1 + α)2 , α∈ 0, (7.10)
2 4 3
The HHT-α method is useful in structural dynamics simulations incorporating many degrees of
freedom, and in which it is desirable to numerically attenuate (or dampen-out) the response at high
frequencies. Numerical damping (impossible in Newmark) stabilizes the numerical integration scheme
by damping out the unwanted high frequency modes and numerical noise.(36 )
7.3.2 Comparison
The Figure 7.2 shows the evolution of the numerical damping ratio through a frequency range i.e. how
each spectral component will be affected by the time integrator (=dispersion). One sees that the same
value of alpha and at a fixed frequency, a smaller timestep is needed in the Newmark algorithm for the
same numerical damping ratio. Indeed, the Newmark regular method introduces a numerical damping
ratio growing linearly with ωh while the curve computed for the HHT method appears to be quadratic.
36 http://people.duke.edu/~hpgavin/cee541/NumericalIntegration.pdf#equation.0.31
q̇n+1 = q̇n + (1 − γ)h q̈n + γh q̈n+1
(7.12)
qn+1 = qn + h q̇n + h2 ( 21 − β) q̈n + h2 β q̈n+1
First, we define the residual vector r(q) through the equilibrium equations as
∗ ∗
Since q̇n+1 = q̇n + (1 − γ)h q̈n and qn+1 = qn + h q̇n + h2 ( 21 − β) q̈n , the Newmark’s formula can be
rewritten as follows:
1
∗
q̈n+1 =
(qn+1 − qn+1 )
βh2 (7.14)
∗ γ ∗
q̇n+1 = q̇n+1
+ (qn+1 − qn+1 )
βh
k
The residual equation is expressed in terms of qn+1 only so r(qn+1 ) = 0. Let us denote qn+1 an
approximate value of qn+1 resulting from iteration k. In the neighbourhood of this value, the residual
γ t 1
S(q) = Kt + C + M (7.15)
βh βh2
The solution of r(qn+1 ) = 0 can be found iteratively as explained in Fig. 7.3. At iteration k of the time
k
step n+1, we use r(qn+1 ) = 0 to compute ∆qk and its first and second derivatives:
γ 1
S∆qk = −r(qn+1
k
), ∆q̇k = ∆qk , ∆q̈k = ∆qk . (7.16)
βh βh2
k+1
From those variables, we can compute qn+1 and continue the loop.
• Material nonliearity
Two exemples of nonlinearity in materials are the nonlinear elastic materials (Such as rubber isolators),
or the dry friciton (It is applied to brakes, rotor-stator contact, drill-string systems,...).
When we linearise, we obtain the slope at the point considered. We have to start again for each specific
point.
For small loads, we have a linear behavior. When non-linearity appear when we increase too much the
A common example of non-linearity is the simple pendulum. In case of large amplitudes, there are
If we compute for different initial conditions, we will have different results. When we increase the initial
angle, the fundamental frequency gets lower accordingly. We can also see the apparition of resonance
We can linearize for a given load only since the load varies with the variation of stiffness.
• Nonliear excitation
39
8.2 Explain what the main challenges are in nonlinear dynamics.
• No principle of superposition
• Frequency-energy dependence
• Harmonics
• Chaos
→ Interpretation and use for design of engineering structures are still a challenge.
output responses
Fig. 8.6, clearly shows the effect of the cubic non-linearity in the free response of the mass:
Figure 8.6: Single DOF Linear / Non-linear free response comparison (Left->Linear Right->Non-linear)
8.3 Apply the harmonic balance method to the example of the Duffing
oscillator to calculate an approximation of its free response. 40
Knowing that:
3
eiωt + e−iωt
3 1
cos3 (ωt) = = cos(ωt) + cos(3ωt) (8.3)
2 4 4
3
− ω 2 m + k + knl A2 A cos(ωt) = 0 (8.5)
4
3
− ω 2 m + k + knl A2 =0 (8.6)
4
Equation (8.8) shows that the natural frequency of the nonlinear system depends on the amplitude of
the oscillation. With equation (8.8), it is also possible to recover to the typical result in two specific
cases:
q
k
• When knl tends to zero: if knl → 0 then ω0 → m
q
k
• For small displacements: if A → 0 then ω0 → m
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: