POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

FREE VIBRATION FREE VIBRATION

System of n II order ODEs   kq  0 ; q(0), q (0)

mq   kq  0 ; q(0), q (0)
mq

q( t )  φ  ( t ) qt   φ  t 
tentative integral=amplitude vector x time variation
m φ  t   k φ  t   0
taking the second derivative and replacing into
the system of equations:
rhs is constant over time, so must be lhs
m φ t   k φ  t   0
t 
  t   C ; t   C  t   0
mφ  k φ  t 
 t 

1 2

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

FREE VIBRATION FREE VIBRATION
  kq  0 ; q(0), q (0)
mq qt   φ t    kq  0 ; q(0), q (0)
mq
t 
qt   φ  t   C ; t   C  t   0
 t 
  t  C must be negative due to the
mφ  k φ
 t  stability of the system

rhs is constant over time, so must be lhs   t    2   t   0 ; with C   2

t 
 C ; t   C  t   0
 t   (t )  A sin t  B cos t

3 4

1
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

FREE VIBRATION FREE VIBRATION

qt   φ t    kq  0 ; q(0), q (0)

mq qt   φ t    kq  0 ; q(0), q (0)
mq

 (t )  A sin t  B cos t  (t )  A sin t  B cos t

q  t   φ  A sin t   B cos t   q  t   φ  A sin t   B cos t  

is a particular integral (in fact they is a particular integral (in fact they
are two lin independent integrals) are two lin independent integrals)

if C   2   t  if C   2  m φ   2    k φ
mφ  k φ
 t 

5 6

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

FREE VIBRATION FREE VIBRATION

qt   φ t    kq  0 ; q(0), q (0)

mq qt   φ t    kq  0 ; q(0), q (0)
mq

q  t   φ  A sin t   B cos  t   q  t   φ  A sin t   B cos  t  

m φ   2    k φ m φ   2    k φ
 k   2 m  φ  0  k   2 m  φ  0
algebraic eigenvalue problem
(non standard form)
standard form  A  I x  0
7 8

2
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

FREE VIBRATION FREE VIBRATION

qt   φ t    kq  0 ; q(0), q (0)

mq k   2m  0
n solutions 12 ,22 ,...,n2 ( eigenvalues )
 k   m  φ  0
2
 (t )  A sin t  B cos t solutions are real and positive if both matrices are symmetric
and positive definite; we assume, for now, that solns are also
linear homogeneous system: non-trivial solutions only for distinct. Replacing each soln into system:

k   2m  0  k   2j m  φ j  0
k   2 m   n  2 n   n 1  2 n 1  ...   0  0 n corresponding solutions φ1 ,φ2 ,..., φn
the system is singular, so solutions/eigenvectors are not unique
n solutions 12 ,22 ,...,n2 but defined up to an arbitrary factor: vectors are linearly
independent
9 10

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

FREE VIBRATION NORMAL MODES OF VIBRATION

 k   2j m  φ j  0 q ( j )  t   φ j  A j sin  j t   B j cos  j t  

n corresponding solutions φ1 ,φ2 ,..., φn are named “normal modes of

vibration” of the system. They are
oscillations in which all points
the system is singular, so solutions/eigenvectors are not
vibrate with the same amplitude and
unique but defined up to an arbitrary factor
phase; in addition the ratio between
the amplitude at two points
This means that eigenvectors can be normalized according
(i.e. the shape) remains constant
to an arbitrary, or better «efficient», criterion
during the motion

11 12

3
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

NORMAL MODES OF VIBRATION NORMAL MODES OF VIBRATION

q ( j )  t   φ j  A j sin  j t   B j cos  j t  
n
q( t )   φ j ( Aj sen j t  B j cos  j t )
j 1
are named “normal modes of vibration” of the system. They are free vibration solution: the Aj e Bj constants are computed according
oscillations in which all points vibrate with the same amplitude and to the initial conditions
n
q  t    φ j  j  Aj cos  j t   B j sin  j t  
phase; in addition the ratio between the amplitude at two points
(i.e. the shape) remains constant during the motion
j 1
n n
q( t )   φ j ( A j sen j t  B j cos  j t ) q0    φ j B j Φ B  q 0   B
j 1 j 1
n
free vibration solution: the Aj e Bj constants are computed according q 0     j φ j A j Φ Ω A  q  0   A
to the initial conditions j 1

matrices Φ and ΦΩ are nonsingular

13 14

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

EIGENPROPERTIES (EXAMPLE) EIGENPROPERTIES (EXAMPLE)
 96 30 
 1 0 
EJ  7 7 
k 3   m  m 
l  30 12  0 1 
  2 
 7 7 

Switch to non
dimensional
form

15 16

4
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

EIGENPROPERTIES (EXAMPLE) EIGENPROPERTIES (EXAMPLE)

 1   1  non standard
1    ; 2   
3.055  0.6547  normalization !

17 18

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

EIGENPROPERTIES (EXAMPLE) EIGENPROPERTIES (EXAMPLE)
 96 30 
 1 0 
EJ  7 7 
k 3   m  m 
l  30 12  0 1 
  2 
 7 7 
by solving
k   m   0
2

EJ EJ
12  0.62261 ;  22  16.52 3
l 3m l m
 1   1 
 1   1  1    ; 2   
1    ; 2    3.055   0.6547 
3.055  0.6547 
19 20

5
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

FREE VIBRATION (EXAMPLE) FREE VIBRATION (EXAMPLE)
EJ EJ
assume  100 ; qT ( 0 )  0 ; q T ( 0 )  1 0  100 ; qT ( 0 )  0 ; q T ( 0 )  1 0
l 3m l 3m
1  7.89058 rad / s ; f1  1.25583 Hz ; T1  0.79629 s 1  7.89058 rad / s ; f1  1.25583 Hz ; T1  0.79629 s
 2  40.6451rad / s ; f 2  6.46887 Hz ; T2  0.15459 s  2  40.6451rad / s ; f 2  6.46887 Hz ; T2  0.15459 s
n
0.42012  q( t )   φ j ( Aj sen j t  B j cos  j t )
φ1    j 1
 1.2834 
 0.90747 
φ2   
0.59414 
NB: φTj mφ j  1
(standard normalization)
21 22

21 22

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

orthogonality of eigenvectors orthogonality of eigenvectors

 k  i2 m  φi  0 Let’s consider two

φTj  k  i2 m  φi  0 (a)
eigenpairs, i and j

φTj  k  i2 m  φi  0 (a) φ Tj  k   2j m  φ i  0 (b )

take the difference (a) – (b)
 k   m  φ j  0
2
j

φTj m φi i2   2j  0 
φ Tj  k   2j m  φ i  0 (b ) if the eigenvalues are distinct, the eigenvectors
are orthogonal wrt to m

23 24

6
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

orthogonality of eigenvectors orthogonality of eigenvectors

φTj  k  i2 m  φi  φTj  k  i2 m  φi 

 φTj kφi  i2 φTj mφi  0 (a)  φTj kφi  i2 φTj mφi  0 (a)
φTj m φi  0 for i  j φTj m φi  0 for i  j
φTj k φi  0 from ( a ) φTj k φi  0 from ( a )
if the eigenvalues are distinct, the eigenvectors are
from orthogonality, linear independence among
orthogonal wrt to m and k; it is different from
eigenvectors can be easily demonstrated
geometrical orthogonality (uTv = 0) which occurs in the
standard form

25 26

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

orthogonality of eigenvectors orthogonality of eigenvectors

φTj kφi  i2 φTj mφi  0 (a) K  ΦT k Φ

T K ij  eiT Ke j  eiT ΦT k Φe j  φiT k φ j
φ k φj
 2j  T
j
for i  j M  ΦT m Φ  ...  φiT m φ j
φ m φj
j

M and K are diagonal matrices (with positive

if you can find the eigenvector, the eigenvalue is readily coefficients if m and k are positive definite)
obtained

27 28

7
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

orthogonality of eigenvectors FORCED VIBRATION
I must integrate the following system of ODEs
K Φ kΦ T
M  Φ mΦ T
mq  kq  Q
M and K are diagonal matrices (with positive n

coefficients if m and k are positive definite) F j (t )   k ji qi  m j q j  0

i 1
T the soln can be again obtained as combination of eigenvectors
φ k φj Kj
 2j  j
 n (linearly independent):
T
φ m φj Mj q( t )   φ j y j ( t )
j
j 1

  K j if φ m φ j  1
2
j
T
j
yj(t) are new coordinates, named “normal”; each of them represents
the contribution, at each time, of the j-th normal mode

29 30

n DoF LINEAR UNDAMPED SYSTEM n DoF LINEAR UNDAMPED SYSTEM

SOLUTION BY MODE SUPERPOSITION SOLUTION BY MODE SUPERPOSITION
n n
q(t )  φj y j (t ) q(t )  φj y j (t )
j 1 K  ΦT k Φ j 1 properties of
eigenvectors are exploited
mq  kq  Q M  Φ mΦ
T

m q  kq  Q
  kΦy  Q
mΦy ortogonality φTj mφk  0

ΦT mΦy  ΦT kΦy
  ΦT Q My  Ky   T Q M and K are diagonal

y j  K j y j  φTj Q
M j 
My  Ky   Q T
equations are uncoupled!!
31 32

8
POLITECNICO DI MILANO

N DoF LINEAR UNDAMPED SYSTEM N DoF LINEAR UNDAMPED SYSTEM

SOLUTION BY MODE SUPERPOSITION 2 SOLUTION BY MODE SUPERPOSITION

y j  K j y j  φTj Q
M j  y j   2j y j  φTj Q / M j

y j   2j y j  φTj Q / M j
  ICs y j ( 0 ), y j ( 0 )
Pros
is the eqn of 1 DoF system; I can easily solve it • formation of eqns simple and economical (after
n
eigenvalue extraction)
Once yj(t) are known q(t )  φj y j (t ) • are the eqns of 1 DoF systems; I can easily solve them
j 1 • for typical CE loadings, superposition can be
extended to rather limited number of normal modes
NB: if standard normalization p<<n
of eigenvectors is used M j  1 ; K j   2j

33 34

N DoF LINEAR UNDAMPED SYSTEM N DoF LINEAR UNDAMPED SYSTEM

SOLUTION BY MODE SUPERPOSITION SOLUTION BY MODE SUPERPOSITION

y j   2j y j  φTj Q / M j
 y j   2j y j  φTj Q / M j

 ICs y j ( 0 ), y j ( 0 )
 ICs y j ( 0 ), y j ( 0 ) n
q(0 )  φj y j (0 )  Φy(0 )
j 1
Cons
• eigenvalue extraction can be a heavy computational Φ1q(0 )  y(0 ) ???
task
• there are problems for treating non standard damped • matrices are never inverted in structural analyses
systems • all eigenvectors are needed (never happens)

35 36

9
POLITECNICO DI MILANO

N DoF LINEAR UNDAMPED SYSTEM N DoF LINEAR UNDAMPED SYSTEM

SOLUTION BY MODE SUPERPOSITION: SOLUTION BY MODE SUPERPOSITION:
initial conditions initial conditions
n
q(0 )  φj y j (0 )  Φy(0 ) n
j 1 q(0 )  φj y j (0 )  Φy(0 )
j 1
ΦT mq(0 )  ΦT mΦy(0 )  My(0 )
1
ΦT mq(0 )  ΦT mΦy(0 )  My(0 )
y(0 )  M Φ mq(0 ) T

y(0 )  M 1ΦT mq(0 )

NB ( p, p )( p,n )( n,n )
1 T T 1 T
• it is a form of approximate inversion yi (0 )  ei Φ mq(0 )  φi mq(0 )
• works if a limited number of modes is available Mi Mi
• same for velocity

37 38

n DoF SYSTEM UNDER PERIODIC FORCES n DoF SYSTEM UNDER PERIODIC FORCES
Application of mode superposition analysis
Application of modal superposition analysis
1) write equations of motion
 3) write equations in normal coordinates
mq+ cq + kq = Q(t )  F  Rk sen  kt  k  
k 1 y j   2j y j  φTj F  Rk sin  kt  k  / M j

2) compute eigenvalues and eigenvectors k 1
4) solve equations of motion in normal coordinates
 k   m  φi  0
i
2 12 ,22 ,..., p2 ; φ1 , φ2 ,..., φ p
φTj F 
3) write equations in normal coordinates y j t   R N k
( j)
k sen  k t  k   k , j 

Kj k 1

y j   2j y j  φTj F  Rk sin  kt   k  / M j

 N k( j )  N ( k /  j ) ;  k( j )   ( k  /  j )
k 1

1  
y j   2j y j  φTj Q / M j
 v t  
F
N sen t    v t    Rk N k sen  kt  o k   k 
k k  k 1

39 40

10
POLITECNICO DI MILANO

n DoF SYSTEM UNDER PERIODIC FORCES n DoF SYSTEM UNDER PERIODIC FORCES
Application of modal superposition analysis
Application of modal superposition analysis
4) solve equations of motion in normal coordinates
φTj F  4) solve equations of motion in normal coordinates
y j t   R N ( j)
sen  k t  k   k , j  φTj F 
sen  k  t  k   k , j 
k k
Kj k 1 y j t  
Kj
R N
k 1
k
( j)
k

N ( j)
k  N ( k /  j ) ;  ( j)
k   (k /  j )
5) superimpose modal responses
  00
n φTj F 
  00.05
.0 5
q (t )   φ j R N k
( j)
k sen  kt   k   k , j 
  0   00.1
.1 Kj
j 1 k 1
  0 .0 5   00.2
.2
  0 .1   00.5
.5
  0 .2
  0 .5

41 42

n DoF SYSTEM UNDER PERIODIC FORCES n DoF SYSTEM UNDER PERIODIC FORCES:
summary of modal superposition example

mq + cq + kq = F  Rk sen  kt  k  F = 10 kN
k 1 EJ
  50 kN / m
y j   y j   F  Rk sen  kt  k  / M j
 2
j
T
j
l3 f  0.5 Hz Force:
forcing frequency
frequency
k 1
in normal coordinates m 1 t 80
Rk  harmonic
Force: amplitudes
harmonic
 F T   k amplitude coefficients
y j t   j
R N k
( j)
k sen  k t   k   k , j 
Kj k 1 1  5.579rad / s ; f1  0.888Hz ; T1  1.126s
response in terms of
N ( j)
k  N (k /  j ) ;  ( j)
k   (k /  j ) normal coordinates
0.42012  2  28.74 rad / s ; f2  4.574Hz ; T2  0.2186s
1   
n  F
T 
 1.2834  φTj F
 Rk N k( j ) sen  kt  k   k , j   

q (t )    j y j t   R N sin kt  k   k j 
j j
k k
Kj Kj k 1
j 1 k 1
 0.90747 
modal superposition 2   
0.59414  
N k j   N k  j  ;  k j    k  j 

43 44

11
POLITECNICO DI MILANO

n DoF SYSTEM UNDER PERIODIC FORCES: n DoF SYSTEM UNDER PERIODIC FORCES:
example example
force 1 3 5 7 9 11 13
      
fk [Hz] 0.5 1.5 2.5 3.5 4.5 5.5 6.5

 Rk 25.47 8.488 5.093 3.638 2.829 2.315 1.959

1.464 0.54 0.144 0.0688

N k(1)
 F
T 1.012 1.121 1.426 2.412 31.08 2.244 0.981
2
N k( 2)
0.21 -0.0256 -0.0041 -0.00141
q(1)
k ,1
1T F q (2)
0.0258 0.0095 0.00725 0.00876 0.0878 -0.00519 -0.00192
system k ,1
q k ,1 0.2358 -0.0161 0.00315 0.00735 0.0878 -0.00519 -0.00192

 
q
k ,1 -2.327 1.430 -0.7722 -3.555 -70.19 6.198 3.2025

45 46

N DoF LINEAR UNDAMPED SYSTEM n DoF LINEAR SYSTEM (UNDAMPED):

SOLUTION BY MODE SUPERPOSITION properties of eigenvalues

 k  i2 m  φi  0 12 ,22 ,...,n2 ; φ1 , φ2 ,...,φn

n
q  t    φ j  Aj sin  j t   B j cos  j t  
https://www.youtube.com/watch?v=uWoiMMLIvco j 1
free vibration

on physical grounds:
the eigenvalues are the squares of the (circular)
natural frequencies of the system, so they must be
(strictly?) positive and real

47 48

12
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

meaning of null eigenvalues properties of eigenvalues

 k   2 m  φ  0 the case of singular stiffness  k   2 m  φ  0 the case of singular stiffness

k   2 m   n  2n   n 1  2 n 1  ...   0  0 if the stiffnes is singular are the eigenvalues still real?
We can look at the properties of the standard form
note: k   0  there are null eigenvalues if
the stiffness matrix is singular A  I x  0
T (E)
φ k φj 2V
 2j  T
j
 T
j I) The eigenvalues, λ1, ..., λn, of a symmetric matrix A(n,n) are real
II) If the matrix A(n,n) is symmetric and positive definite, its
φ m φj
j φ m φj
j eigenvalues, λ1, ..., λn, are real and strictly positive, while if the
if  j is null the modal shape is either a rigid body matrix is only positive semi-definite its eigenvalues are real and
2
non-negative
mode or the critical shape associated to the critical load

49 50

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

properties of eigenvalues properties of eigenvalues m  L LT
 k   2 m  φ  0 the case of singular stiffness if the stiffnes is singular are the eigenvalues still real?
assuming positive definite mass we can factorize it and:
if the stiffnes is singular are the eigenvalues still real? define x = LT φ  kL-T x   2 L x
assuming positive definite mass we can factorize it as: L-1kL-T x   2 x
m  L LT  kφ   2 mφ   2 L LT φ compare to  A  I x  0
define x = LT φ  kL-T x   2 L x we transformed the problem into standard form, with

L-1kL-T x   2 x A = L-1 kL-T ( symmetric )   j   2j real !

51 52

13
POLITECNICO DI MILANO

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

properties of eigenvalues properties of eigenvalues
 1  the case of singular mass  k   2 m  φ  0  1 
m   2 m   2 k φ  0
k φ  0
   
n  n 1 
1  1   1  the case of singular mass
m k   n  2    n 1  2   ...  0  0
2     if the mass is singular are the eigenvalues still real?
note: m  0  there are un lim ited eigenvalues if We can look at the properties of the standard form
the mass matrix is singular
 T 1 1  T
k  NN  mφ  kφ  NN φ
T
φ k φj 2V (E)
 2
 2

 2j  T
j
 j positive definite
1 
φ m φj
j φTj m φ j define r = N T φ  mN T r  2 Nr
if  j is unlimited the modal shape is such that no
2 
kinetic energy is developed N mN r  1 /  2 r
1 T

53 54

n DoF LINEAR SYSTEM (UNDAMPED): n DoF LINEAR SYSTEM (UNDAMPED):

properties of eigenvalues properties of eigenvalues

if the mass is singular are the eigenvalues still real? • Both the mass and the stiffness are positive definite
assuming positive definite stiffness we can factorize it and eigenvalues are all real positive and limited
1 
define r = N T φ  mN T r  2 Nr • The mass is p.d. and the stiffness is singular
 there are null eigenvalues corresponding to rigid-
1 body modes or critical shapes
N 1mN T r  2 r
 • The stiffness is p.d. and the mass is singular
compare to  B  I r  0 there are unlimited eigenvalues corresponding to
we transformed the problem into standard form, with modal shapes NOT activating any inertia

• Both stiffness and mass are singular

B = N 1mN T ( symmetric )   j  1 /  2j real ! unacceptable, since equilibrium is not guaranteed

55 56

14
POLITECNICO DI MILANO

The Rayleigh quotient

φiT k φi K
i2  T  i
φi m φi M i

uT k u
 u   T Rayleigh quotient
u mu

  φ j    2j

57 58

The Rayleigh quotient: properties Effect of additional constraints

uT k u
1. 12   u   n2  u  Take a n DoF system and add an additional constraint
uT m u (holonomic); the system has now (n-1) DoFs

2. The Rayleigh’s quotient is stationary in

correspondence of the eigenvectors of the 12 ,22 ,...,n2  12 ,22 ,...,n21
system (i.e. for u  φ j )
the following property holds:
3. If u  φi   x (approx of an eigenvector),
then  u   i
2
 O  2
, which means that the 12  12  22  22  32  ...  n21  n21  n2
eigenvalue is better approx than the eigenvector
«inclusion» principle

59 60

15
POLITECNICO DI MILANO

Multiple eigenvalues Multiple eigenvalues

φs ,1 ,..., φs , r
assume that s2 has a multiplicity equal to r
k φs ,i  m φs ,is2
there are r eigenvectors associated to s2
k φs , j  m φs , js2
φs ,1 ,..., φs , r

• they are linearly independent

  
k  φs , j   φs ,i  m  φs , j   φs ,i s2 
• each of them is uniquely defined except for
a scale factor any linear combination of these vectors is an
• orthogonality is generally not satisfied eigenvector associated to the same multiple
among these vectors eigenvalue

61 62

Multiple eigenvalues: how to orthogonalize Multiple eigenvalues: how to orthogonalize

vectors? vectors? Gram-Schmidt procedure
φs ,1 , φs ,2 non orthogonal
φs ,1 , φs ,2 ,...φs ,r non orthogonal
take φ s ,1  φs ,1 and φ s ,2  φs ,2   φ s ,1
take φ s ,1  φs ,1 and assume that φ s ,2 , φ s ,3 ,..., φ s , j 1
φ sT,1mφ s ,2  φ sT,1mφs ,2   φ sT,1mφ s ,1  0
has been already orthogonalized
φ sT,1mφs ,2
 T new vectors are orthogonal!! j 1
φ s ,1mφ s ,1 define φ s , j  φs , j    k φ s ,k
k 1

can do this for many vectors; once j-1 have T

φ mφs , j
been orthogonalized, can orthogonalize the k  s ,k
T
new vector is orthogonal
next φ mφ s ,k
s ,k

63 64

16
POLITECNICO DI MILANO

The inverse iteration technique for extracting The inverse iteration technique for extracting
eigenpairs: basic ideas eigenpairs: basic ideas

k φ   2m φ eigenpairs satisfy this system k φ   2m φ eigenpairs satisfy this system

note that if we form this vector (starting from an exact note that if we form this vector (starting from an exact
eigenvector) eigenvector)
Note: for standard eigenvalue problems
R  m φj R  m φj it is like applying a loading vector and
computing a deflection which has the
and solve the system and solve the system same shape as the load!!!
ku  R ku  R
1 1
we obtain  u φj i.e. a vector having the we obtain  u φj i.e. a vector having the
 2
j same shape as the original  2j same shape as the original

65 66

The inverse iteration technique for extracting The inverse iteration technique for extracting
eigenpairs: basic ideas eigenpairs: the iteration

k φ   2m φ eigenpairs satisfy this system

R ( s 1 )  m v ( s 1 )
now, take a tentative vector and pre-multiply by mass
kvˆ ( s )  R( s 1 )
R (0 )
 mv (0 )

then solve the system vˆ ( s ) converges to φ1 ( first eigenvector )

kvˆ ( 1 )  R( 0 )
we obtain a soln v̂ ( 1 ) having a shape which closer to φ1
than the one of the starting vector v ( 0 )
Note: reason for the «hat» will be clear

67 68

17
POLITECNICO DI MILANO

The inverse iteration technique for extracting The inverse iteration technique for extracting
eigenpairs: proof of convergence eigenpairs: proof of convergence
n n φTj mv 0 
expand v 0 
  j φj with expand v 0 
  j φj with  j 
j 1 j 1 φTj mφ j
n n
φkT mv 0     j φkT mφ j   k φkT mφk R( 1 )  m v ( 0 )  v ( 1 )  k 1mv 0   k 1m   j φ j
j 1 j 1

φkT mv  
n n
0 1
k  v ( 1 )    j k 1 m φ j    j φ j k φ j   2j m φ j
φkT mφk j 1 j 1  2
j

1
at each iteration the factor 2 multiplies
each term in the sum j

69 70

The inverse iteration technique for extracting The inverse iteration technique for extracting
eigenpairs: proof of convergence eigenpairs: second eigenvector
n n
1 n
v ( 1 )    j k 1 m φ j    j φ j 2 expand v      j φ j
0
with  1  0
j 1 j 1 j j2

i.e. v 0  orthogonal to φ1
s s s
n  1   1  n  12 
v (s)
  j φ j  2    2    j φj  2  s s s
j j  1   1  n  22 
  1 
n
j 1  j 1   v( s )   j φ j  2    2    j φj  2 
   
 1 
s j2  j   2  j  2  j
v (s)
  2   1 φ1 s
 1   1 
v   2   2 φ2
(s)
12  2  k2
the speed of convergence depends on
22 the speed of convergence depends on
k21
71 72

18
POLITECNICO DI MILANO

The inverse iteration technique for extracting The inverse iteration technique for extracting
eigenpairs: k-th eigenvector eigenpairs: k-th eigenvector
n
expand v 0     j φ j with  1  ...   k 1  0 n  1
s
  1 
s
n  k2 
s
(s)
  j φ j  2    2    j 
  2 
j k v φ

  k 
j

i.e. v 0  orthogonal to φ1 ,...φk 1

j k  j j k  j
s s s s
n  1   1  n   2   1 
v ( s )    j φ j  2    2    j φ j  k2    2   k φk
(s)
v
     k 
j k  j   k  j  k  j
k2
 1 
s the speed of convergence depends on
v ( s )   2   k φk k21
 k  k2 There are methods for making
the speed of convergence depends on convergence faster
k21
73 74

The inverse iteration technique for extracting The inverse iteration technique for extracting
eigenpairs: k-th eigenvector eigenpairs: k-th eigenvector
s s s
n n  1   1  n  k2 
expand v 0 
  j φj with  1  ...   k 1  0 (s)
  j φ j  2    2    j 
  2 
v φ

  k 
j
j k j k  j j k  j
i.e. v 0  orthogonal to φ1 ,...φk 1  1 
s

v (s)
  2   k φk
 k 
Note: orthogonalization cannot be applied
only at the beginning of the iteration (we find Note: the factor can grow very large or very
always the first eigenvector) but at each small, must normalize the vector at each
iteration iteration
vˆ ( s )  v ( s ) upon normalization and
orthogonalization
75 76

19
POLITECNICO DI MILANO

The inverse iteration technique for extracting The inverse iteration technique for extracting
eigenpairs: k-th eigenvector eigenpairs: the iteration
s
 1  R ( s 1 )  m v ( s 1 )
v   2   k φk
(s)

 k  vˆ   R s 1
sT
2 s 
kvˆ (s)
R ( s 1 )
   s T
vˆ m vˆ  s 
k
How to control convergence?

vˆ   k vˆ   vˆ   R s 1 vˆ ( s )  v ( s )
sT s s T
2 s  normalization and orthogonalization
   s T  (Gram Schmidt ok for small systems)
vˆ m vˆ  s  vˆ  s T m vˆ  s 
k

k2 s   k2 s 1

k2 s   k2  s 1 
 k2  s 
k2 s 

77 78

The inverse iteration technique for extracting The inverse iteration technique for extracting
eigenpairs: multiple eigenvalue eigenpairs: pros and cons
s s s
n  1   1  n  k2  Pros:
v (s)
  j φj  2    2    j φj  2 
 j  j  • simple and, generally, computationally
j k    k  j k   efficient
s
 1  r  k2  • can choose how many eigenpairs to be
v (s)
 2    j φ j  2  r is multiplicity
 j  extracted, frequency limits etc…
 k  j k   • no problems with multiple eigenvalues
Note: converges to a linear combination of Cons
eigenvectors associated to the same eigenvalue, • can slow down (groups of close eigenvalues)
still eigenvectors. I can find the entire set upon • orthogonalization is critical
orthogonaliation.

79 80

20
POLITECNICO DI MILANO

Large eigenvalue problems: the Rayleigh-Ritz Large eigenvalue problems: the Rayleigh-Ritz
method method
r n
«system size reduction»
q( t )   ψ k xk  t   Ψ x( t )
k 1
• to reduce from continuous to discrete
n
v  x,t     k  x  qk  t  Ritz vectors
k 1
• To reduce from large (discrete) to smaller Note: additional constraints are added
r n to the system
q( t )   ψ k xk  t   Ψ x( t )
k 1

81 82

Large eigenvalue problems: the Rayleigh-Ritz Large eigenvalue problems: the Rayleigh-Ritz
method for approximating eigenpairs method for approximating eigenpairs
r n
q( t )   ψ k xk  t   Ψ x( t ) φ Ψz
k 1

φ Ψz Problem: how to estimate the z vector leading

to the «best» approx of an eigenvector?

Typical situation in CE dynamics: system is Solution: express Rayleigh quotient for the
very large but a smaller number of eigenpairs approx eigenvector and impose stationarity
must be extracted with sufficient approx.
Orthogonality is important

83 84

21
POLITECNICO DI MILANO

Large eigenvalue problems: the Rayleigh-Ritz Large eigenvalue problems: the Rayleigh-Ritz
method for approximating eigenpairs method for approximating eigenpairs

φ Ψz Solution: express Rayleigh quotient for the

approx eigenvector and impose stationarity
~
φT k φ zT Ψ T k Ψ z k
Solution: express Rayleigh quotient for the  φ   T   ~
approx eigenvector and impose stationarity φ m φ zTΨ T m Ψ z m
~ ~ ~ ~
T T ~T ~ k  k~ m k   m
m
φ kφ z Ψ kΨ z k
 φ   T   ~ 

z j z j z j

z j
0 j  1,..., r
φ m φ zTΨ T m Ψ z m z j ~
m 2 ~
m

85 86

Large eigenvalue problems: the Rayleigh-Ritz Large eigenvalue problems: the Rayleigh-Ritz
method for approximating eigenpairs method for approximating eigenpairs
~ ~
φT k φ zTΨ T k Ψ z k φT k φ zT Ψ T k Ψ z k
 φ   T  T T  ~  φ   T   ~
φ m φ z Ψ mΨ z m φ m φ zT Ψ T m Ψ z m
~ ~ ~ ~
~ k  k~ m k   m
m defining k  Ψ T k Ψ ; m  Ψ T mΨ
 z j z j z j z j
 ~  ~ 0 j  1,..., r the above stationarity condition leads to
z j m 2
m
a new " small" ( r ,r ) eigenvalue problem
defining k  Ψ k Ψ T
; m  Ψ mΨ
T

the above stationarity condition leads to

a new " small" eigenvalue problem
k   m z  0
87 88

22
POLITECNICO DI MILANO

Large eigenvalue problems: the Rayleigh-Ritz The Rayleigh-Ritz method for approximating
method for approximating eigenpairs eigenpairs: orthogonality

k ΨT kΨ ; m  Ψ T mΨ k ΨT kΨ ; m  Ψ T mΨ

 k   m  z  0  1 ,...,  r , z1 ,..., zr φ j  Ψ z j approx to eigenvector

φ j  Ψ z j approx to eigenvector φiT m φ j  ziT Ψ T mΨ z j  ziT m z j  0 i j

z Tj k z j φ Tj k φ j the φ j vectors are approx to the exact
 j   φj   
z Tj m z j φ Tj m φ j eigenvectors but are exactly orthogonal
approx to the eigenvalue
89 90

The Rayleigh-Ritz method for approximating The Rayleigh-Ritz method for approximating
eigenpairs: accuracy eigenpairs: accuracy

φ j  Ψ z j approx to eigenvector φ j  Ψ z j approx to eigenvector

z Tj k z j φ Tj k φ j z Tj k z j φ Tj k φ j
 j   φj     j   φj   
z Tj m z j φ Tj m φ j z Tj m z j φ Tj m φ j

• to get the best approx, the starting vectors must be as • to get the best approx, the starting vectors must be as
close as possible to the exact eigenvectors close as possible to the exact eigenvectors
• for having a satisfactory approx of p eigenvectors th • the inverse iteration procedure can be adopted to
size of the «subspace» r must be at least 2p improve the «similitude» (Subspace Iteration)
• eigenvalues approximate the exact values «from
above», since the system is more constained

91 92

23
POLITECNICO DI MILANO

The Rayleigh-Ritz method as an The Rayleigh-Ritz method as an

orthogonalization procedure orthogonalization procedure
we start from ψ1 ,...ψ r linearly independent
we start from ψ1 ,...ψ r linearly independent form k  Ψ T k Ψ ; m  Ψ T mΨ
form k  Ψ T k Ψ ; m  Ψ T mΨ solve  k   m  z  0  z1 ,..., zr

solve  k   m  z  0  z1 ,..., zr obtain φ j  Ψ z j j  1,...,r

linearly independent and orthogonal
obtain φ j  Ψ z j j  1,...,r
linearly independent and orthogonal The RR method is a powerful orthogonalization, can be
used within the inverse iteration algorithm (Subspace
Iteration)

93 94

n DoF LINEAR SYSTEM (DAMPED): n DoF LINEAR UNDAMPED SYSTEM

FORCED VIBRATION SOLUTION BY MODE SUPERPOSITION
n
q(t )  φj y j (t )
I must integrate the following system of ODEs
K  ΦT k Φ
mq  cq  kq  Q j 1
M  ΦT m Φ
mq  cq  kq  Q
the soln can be again obtained as combination of
n eigenvectors mΦy  cΦy  kΦy  Q
q( t )   φ j y j ( t ) (linearly independent):
ΦT mΦy  ΦT cΦy  ΦT kΦy  ΦT Q
j 1
assume that we use the (classical) eigenvectors of the
undamped system My  Cy  Ky  Φ T Q
95 96

24
POLITECNICO DI MILANO

n DoF LINEAR UNDAMPED SYSTEM n DoF LINEAR UNDAMPED SYSTEM

SOLUTION BY MODE SUPERPOSITION SOLUTION BY MODE SUPERPOSITION
n
q(t )  φj y j (t ) K  ΦT k Φ My  Cy  Ky  Φ T Q
j 1 M  ΦT m Φ equations are no longer uncoupled, how to solve?
Rem: we have still the problem of defining c
My  Cy  Ky  Φ T Q C  Φ T cΦ
1) solve the equations with damping coupling
generally non diagonal uneffcient, must still define c
2) compute eigenpairs for the damped system and solve
equations are no longer uncoupled, in general with exact uncoupling of damped equations
can be done; problem becomes 2n,2n and eigenpairs
are complex. Still c must be defined
3) introduce «proportional» damping matrix

97 98

n DoF LINEAR UNDAMPED SYSTEM n DoF LINEAR UNDAMPED SYSTEM

SOLUTION BY MODE SUPERPOSITION SOLUTION BY MODE SUPERPOSITION

My  Cy  Ky  Φ T Q My  Cy  Ky  Φ T Q

2) compute eigenpairs for the damped system and solve equations are no longer uncoupled, how to solve?
with exact uncoupling of damped equations …
can be done; problem becomes 2n,2n and eigenpairs 3) introduce «proportional» damping matrix
are complex. Still c must be defined
c  m   k Rayleigh damping
(C becomes exactly diagonal)
note: damping model depends only on two parametrs

4) assume that out-of-diagonal terms in C can be neglected

without assuming a form for c

99 100

25
POLITECNICO DI MILANO

N DoF LINEAR UNDAMPED SYSTEM N DoF LINEAR UNDAMPED SYSTEM

SOLUTION BY MODE SUPERPOSITION SOLUTION BY MODE SUPERPOSITION

My  Cy  Ky  Φ T Q y j  2 j j y j   2j y j   Tj Q / M j

4) assume that out-of-diagonal terms in C can be neglected  j   j  1,...p
without assuming a form for c
in standard cases, i.e. material and structural
y j  C j y j  K j y j  φTj Q
M j  system is standard and homogeneous in terms of
dissipation
y j  2 j j y j   2j y j   Tj Q / M j

one parameter describes the damping model!!
Cj
j  directly introduced on the basis in other situations modal damping factors can be
2 K jM j of material/structural properties estimated, always under the proportionality hp

101 102

103 104

103 104

26
POLITECNICO DI MILANO

LUMPED MASS METHOD: WRITING

THE EQUATIONS OF MOTION

The equations of motion express

equilibrium of rigid bodies under
perturbative (external) forces,
elastic restoring forces, inertia
forces (and damping forces).

k ji stiffness coefficients
Matrix form
n
F j (t )   k ji qi  m j q j  0
i 1 m q  kq  Q
105

105 106

LUMPED MASS METHOD: internal forces analysis FROM ELASTIC RESTORING FORCES TO
via elastic restoring forces EQUIVALENT STATIC FORCES

Once the response in terms of Lagrangian q (t )   j y j (t ) assume single-mode response

coordinates is known, restoring forces can be
evaluate at each time and applied, with minus
sign, to the «elastic support», this being the
f E( j ) (t )  kq( j ) (t )  k j y j (t )   2j m j y j (t )
structure to be designed. elastic forces to be applied to the structural system

f E( j ) (t )  kq( j ) (t ) f E( ,jMAX
)
(t )  kq(MAX
j)
 k j y j ,MAX   2j m j y j , MAX
This is not a standard procedure in dynamic forces leading to peak values of modal effect
structural analysis
If the contribution of several modes must be accounted for a
combinations rule must be adopted (SRSS, CQC, …)

108
108

107 108

27
POLITECNICO DI MILANO

LUMPED MASS METHOD: direct internal forces LUMPED MASS METHOD: direct internal forces
analysis via displacements analysis via dispacements
n
The standard, and by far more efficient,
procedure relies on the direct computation of
S( t )   si qi ( t ) sT q( t )
i 1
internal forces (or any linear effect) from
displacements Note : si is the effect of
n
qi  1 and q j  0 for j  i
S( t )   si qi ( t ) sT q( t )
i 1
The computation of such coefficients
Note : si is the effect of is totally unexpensive, since the
qi  1 and q j  0 for j  i necessary structural analysis is
already performed during the
stiffness evaluation

109 110

LUMPED MASS METHOD: direct internal forces System with moving reference frame
analysis via dispacements
n
S( t )   si qi ( t ) sT q( t )
i 1
n n n
S( t )   sT  j y j ( t )   s( j ) y j ( t )  S ( j ) ( t )
j 1 j 1 j 1

in modal superposition analysis

«modal effects» are usually stored

112

111 112

28
POLITECNICO DI MILANO

System with moving reference frame: apparent forces 1 DoF LINEAR SYSTEM WITH MOVING
REFERENCE FRAME

v(t) relative displacement

(wrt moving ref)
u(t) displacement of the
moving ref

m(v  u)  cv  kv  0

Equation of motion
v  21v  12 v  u(t )
m v  cv  kv   mr u 113
NB: depends on 2 parameters only !! (system is def by 3)

113 114

1 DoF LINEAR SYSTEM UNDER SEISMIC 1 DoF LINEAR SYSTEM UNDER SEISMIC
GROUND MOTION MOTION: RESPONSE SPECTRUM DEFINITION
Displacement response
spectrum:
The previous derivation can
S d T1 ,    max v  t 
be applied to the seismic case
if the following two Remind:
assumptions can be justified:
2 m c
T1   2 ; 
1 k 2 km

1) the seismic motion is equal and simultaneous in all “contact”

S a T1 ,    12 S d T1 ,   Pseudo- acceleration response spectrum
soil-structure zones (i.e. plan dimensions << seismic wavelenght); Note: 21v  12 v  (u  
v) (absolute acceleration)
2) the seismic motion u(t) is not significantly affected by the
structure (i.e.structure is not too heavy and/or rigid). For small damping S a T1 ,    max u(t )  v(t )

115 116

29
POLITECNICO DI MILANO

1 DoF LINEAR SYSTEM UNDER SEISMIC Recorded Ground Motions (horizontal component)
MOTION: RESPONSE SPECTRUM DEFINITION
Displacement response To define earthquakes – ground
spectrum: shaking: time variation of
ground acceleration
S d T1 ,    max v  t  3 components: 2 horizontal, 1
vertical
Remind:
Strong-motion accelerometers
2 m c
T1   2 ;  Frequency range of recording:
1 k 2 km 0-30 Hz (up to 50 Hz for digital

S a T1 ,     S d T1 ,   Pseudo- acceleration response spectrum

2 instruments)
1

118

117 118

Horizontal ground acceleration (Parkfield Station) Displacement, pseudo-velocity and pseudo-

acceleration response spectra ( 2%)

Displacement

Pseudo-velocity:

2
Sv   1 Sd  Sd
T1
Pseudo-acceleration:
2
S a  1S v  Sv
T1
119 Sa /g base shear coefficient 120

119 120

30
POLITECNICO DI MILANO

RESPONSE SPECTRUM PROPERTIES n DoF LINEAR SYSTEM UNDER SEISMIC

MOTION: RESPONSE SPECTRUM METHOD
Displacement response equations of motion (system subjected
spectrum: to “apparent” forces ):

S d T1 ,    max v  t  mq  cq  kq  m r u(

 t )
rk= 1 for coords. parallel to ref motion
Sa T1 ,    max u(t )  v(t ) rk= 0 otherwise
transform into normal coordinates
n
S a T1  0,    max u(t )  PGA q( t )    j y j ( t )
(very stiff structure) j 1

S a T1   ,   0  Tj mr
y j  2 j j y j   y j  
 2
j u(t )   j u(t )
Mj
S d T1   ,   max u( t ) (equation of 1DoF system under participation factor
(very deformable/isolated structure) seismic motion   j u(t ) )

121 122

THE RESPONSE SPECTRUM METHOD FOR RESPONSE SPECTRUM METHOD:

APPLIED TO n DoF LINEAR SYSTEMS COMBINATION OF MODAL CONTRIBUTIONS
 Tj mr n n
y j  2 j j y j   2j y j  
 u(t )   j u(t ) qk (t )    j ,k y j (t )  qk( j ) (t )
Mj j 1 j 1

I can exploit the response spectrum for

computing the extreme value: max absolute value of the modal contribution:

max y j (t )   j Sd (T j ,  j ) 2 max qk( j ) (t )  max  j ,k y j (t )   j ,k  j S d (T j ,  j )

Tj 
j
n n
q (t )    j y j (t )  q ( j ) (t ) modal contributions to response
The question is how to combine modal maxima to obtain extreme values
j 1 j 1 of the total response, This being functions for which only the extreme
value is known. The standard combination rule is the following:
n n
qk (t )    j ,k y j (t )  qk( j ) (t )
 max q 
n
for the k-th coordinate max qk (t )  ( j)
(t )
2
SRSS combination
k
j 1 j 1
j 1

123 124

31
POLITECNICO DI MILANO

RESPONSE SPECTRUM METHOD: RESPONSE SPECTRUM METHOD:

EXAMPLE OF APPLICATION EXAMPLE OF APPLICATION - 2 0.42012 
1   
 1.2834 
EJ
 100 kN / m
l3  0.90747 
2   
m 1 t 0.59414 
1  7.89058 rad / s ; f1  1.25583 Hz ; T1  0.79629 s
1 1
r 
r  1  2  40.6451rad / s ; f 2  6.46887 Hz ; T2  0.15459 s
1
  7.89058 rad / s ; f1  1.25583 Hz ; T1  0.79629 s  j   Tj mr / M j   1  1.06182 ;  2  0.6104
0.42012  1
1      40.6451rad / s ; f  6.46887 Hz ; T  0.15459 s
 1.2834  2 2 2 Sa (T1 ,0.05 )  4.8123 m / s 2 ; Sd ( T1 ,0.05 )  S a / 12  0.07729 m
max q k( j ) (t )  max  j , k y j (t )   j , k  j S d (T j ,  j )
 0.90747  Sa (T2 ,0.05 )  7.6641m / s 2 max q k( j ) (t )  max  j , k y j (t )   j , k  j S d (T j ,  j )
2   
 max q 
n

0.59414   max q 
2 n
max q k (t )  ( j)
(t ) max q k ( t )  ( j)
(t )
2

Sd (T2 ,0.05 )  0.00464 m

k k
j 1 j 1

125 126

RESPONSE SPECTRUM METHOD: RESPONSE SPECTRUM METHOD:

EXAMPLE OF APPLICATION - 3 EXAMPLE OF APPLICATION - 4
 1  1.06182 ;  2  0.6104 internal force or stress component S(t)
n
Sd (T1 ,0.05 )  0.07729 m ; S d ( T2 ,0.05 )  0.00464 m S (t )   si qi (t ) s T q (t )
i 1

0.42012   0.90747  n n n
1    ; 2    S( t )  s T q( t )   sT  j y j ( t )   s ( j ) y j ( t )  S ( j ) ( t )
 1.2834  0.59414  j 1 j 1 j 1

max q ( t )  0.03448 m ; max q

(1)
1
(2)
1 ( t )  0.00257 m modal superposition

max q1 ( t )  0.03448 2  0.00257 2  0.03458 m max S ( j ) (t )  max s ( j ) y j (t )  s ( j )  j S d (T j ,  j )

extreme value of the modal contribution
max q2( 1 ) ( t )  0.10533 m ; max q2( 2 ) ( t )  0.001682 m
 max S 
n 2
max q k( j ) (t )  max  j , k y j (t )   j , k  j S d (T j ,  j ) max S (t )  ( j)
(t ) SRSS combination
max q2 ( t )  0.10534 m j 1
 max q 
n 2
max q k ( t )  ( j)
k (t )
j 1

127 128

32
POLITECNICO DI MILANO

RESPONSE SPECTRUM METHOD: RESPONSE SPECTRUM METHOD:

EXAMPLE OF APPLICATION - 5 EXAMPLE OF APPLICATION - 6
EJ
 1  1.06182 ;  2  0.6104  1  1.06182 ;  2  0.6104  500 kN
l2
Sd (T1 ,0.05 ) ; Sd ( T2 ,0.05 )  0.00464 m Sd (T1 ,0.05 ) ; Sd ( T2 ,0.05 )  0.00464 m
B
0.42012   0.90747  extreme values of M A,B ( t )  sTA,B q( t )
1    ; 2    bending moments
 1.2834  0.59414  sTA 
EJ  36 6
 
extreme value of shear
n
A l2  7 7
S (t )   si qi (t ) s T q (t )
in upper column i 1
EJ  30 12 
s( 1 )  sT 1  39.96 kN / m s(A1 )  sTA1  530.3 kN sTB    
EJ  30 12  l2  7 7 
s  3 
T

s( 2 )  sT 2  490.77 kN / m l  7 7  s(A2 )  sTA2  2588.2 kN
0.42012   0.90747 
1    ; 2   
max S ( 1 ) ( t )  3.28 kN ; max S ( 2 ) ( t )  1.39 kN s (1)
 s   199.8 kN
T
 1.2834  0.59414 
B B 1
max S( t )  3.56 kN
(SRSS) max S ( j ) (t )  max s ( j ) y j (t )  s ( j )  j S d (T j ,  j ) s(B2 )  sTB2  2453.9 kN max S ( j ) (t )  max s ( j ) y j (t )  s ( j )  j S d (T j ,  j )

129 130

RESPONSE SPECTRUM METHOD: Linear correlation and the “sum problem”

EXAMPLE OF APPLICATION - 7 4

extremes of modal internal forces

2

I want to define the

x(1)(t)
0
properties of the sum
max M (t )  43.52 kNm ()
(1)
A -2
function z(t)
max M ( 2)
A (t )  7.33 kNm () -4

0 0.4 0.8 1.2 1.6 2

time [s]

max M (t )  16.40 kNm ()

(1)
B
4

max M B( 2 ) (t )  6.95 kNm () assume zero-mean

x(2)(t)
0

functions -2

extremes of total response -4

0 0.4 0.8 1.2 1.6 2

max M A (t )  44.13 kNm time [s]

z  t   x1  t   x 2  t 
(SRSS)
max M B (t )  17.81 kNm
max S (t )  max s y j (t )  s  j S d (T j ,  j )
( j) ( j) ( j)

131 132

33
POLITECNICO DI MILANO

Linear correlation and the “sum problem” Linear correlation and the “sum problem”

E  z 2   E  x (1)   2 E  x (1) x (2)   E  x (2)  E  z 2   E  x (1)   2 E  x (1) x (2)   E  x (2) 

2 2 2 2

       
variance = mean square value define the linear correlation coefficient
1 T
E  x (1) x (2)  12
 12  E  x (1)    x (1) dt
2 2
variance=RMS2
  12  
T 0
 1 2
E  x (1)  E  x (2) 
2 2

1 T    
12  E  x (1) x (2)  
T  0
x (1) x (2) dt covariance
E  z 2   E  x (1)   2 12 E  x (1)  E  x (2) 
2 2 2

     
the SRSS rule holds for the RMS value of the sum of
 E  x (2) 
2
uncorrelated functions (holds also for n functions)
 

133 134

EXTREME VALUE IN A TIME INTERVAL Linear correlation and the “sum problem”
3

E  z 2   E  x (1)   2 E  x (1) x (2)   E  x (2) 

2 2

   
2

1
x(j)(t)

-1 2
 max z (t )   max x 
2 (1)

E  z       
-2
2

 pz   p1 
-3

0 2 4 6 8 10

time [s] 2
 max x (1)  max x (2)   max x (2) 
RMS value (=1 in this example) 2 12    
extreme value = p x RMS with p «peak factor»  p1  p2   p2 
    

135 136

34
POLITECNICO DI MILANO

Linear correlation and the “sum problem” The CQC formula for combining modal effects
2
 max z (t )   max x 
2 (1) n n n

E  z   
2
    S( t )  sT q( t )   sT  j y j ( t )   s( j ) y j ( t )  S ( j ) ( t )
 pz   p1 

j 1 j 1 j 1

2
response parameter
 max x (1)  max x (2)   max x (2) 
2 12     max S ( j ) (t )  max s ( j ) y j (t )  s ( j )  j S d (T j ,  j )
 p1  p2   p2 
    
extreme of modal effect
if all peak factors are about the same: 1/2
 n n 
max z (t )  max S  t    s ( j ) s ( k )  j  k  jk Sd T j ,  j  Sd Tk ,  k  
2 2
 j 1 k 1 
 max x (1)  2 12 max x (1) max x (2)  max x (2) CQC combination

138

137 138

CASE OF TWO COMPONENTS OF VIBRATION SENSORS

GROUND MOTION
 n n x(t) measured
max S  t    s( j ) s ( k )  jk  (jx ) (kx ) Sd( x ) T j ,  j  Sd( x ) Tk ,  k  u(t) to be estimated
 j 1 k 1
  (jy ) (ky ) Sd( y ) T j ,  j  Sd( y ) Tk ,  k  

 2  XY  (jx )  (ky ) Sd( x ) T j ,  j  Sd( y ) Tk ,  k  

1/2

 XY
 XY  “extended” CQC
 XY
correlation coefficients between the two ground
acceleration components

139 140

35
POLITECNICO DI MILANO

VIBRATION SENSORS VIBRATION SENSORS: accelerometer

x(t) measured x(t) measured
u(t) to be estimated, u(t) to be estimated,
assume first: u t   A sin 2 ft
u t   A sin 2 ft steady state response
 0
  0.05
mA
mx  cx  kx  mu  mu(t )  mA sin 2 ft x t    N sin  2 ft      0.1
k   0.2
  0.5
mA assume very stiff sensor
x t    N sin  2 ft    steady state response
k f
 1   1
f1
141 142

VIBRATION SENSORS: accelerometer VIBRATION SENSORS: accelerometer

x(t) measured x(t) measured

 0
u(t) to be estimated, u(t) to be estimated, u t   A sin 2 ft
  0.05
  0.1
u t   A sin 2 ft
  0.2
mA   1   0.6
  0.5
x t   
mA
N sin  2 ft    x t    N sin  2 ft   
k k N 1   0

A 1
  0
x t    sin  2 ft    u(t )
  0 .0 5 2
1 12
  1   0.6 



0 .1
0 .2
N 1   0   0 .5 the measured relative displacement is
proportional to input acceleration!

143 144

36
POLITECNICO DI MILANO

VIBRATION SENSORS: accelerometer VIBRATION SENSORS: accelerometer

x(t) measured M 1 M
Ak
u t    Ak sin  2 k ft  k  x t     N sin  2 k ft  k   k 
u(t) to be estimated
k 1

2
1 k 1 k k
1 M
Ak now f1  M f (usually  2 M f )
x t     N sin  2 k ft  k   k 
 2
1 k 1 k k
Ok for N, but if natl frequency is not large enough,
harmonic components take a different (nonzero) phase
now f1  M f (usually  2 M f ) and the record is distorted

Ok for N, but if natl frequency is not large enough,   0

  0 .0 5
harmonic components take a different (nonzero) phase
  0 .1
and the record is distorted   0 .2
  0 .5

145 146

VIBRATION SENSORS: accelerometer VIBRATION SENSORS: frequency content of

acceleration
1 M
Ak 
x t     N sin  2 k ft  k   k 
2
1 k 1 k k u  f    u t  e
 i 2 ft
dt

now f1  M f (usually  2 M f ) 
u t    u  f  e
i 2 ft
df
must we try to make f1 as large as possible???


NO Otherwise we do not read the relative 

displacement x(t) with the adequate resolution! u  t    u  f  i2 f e

i 2 ft
df
(see square of natl frequency at denominator) 

We need different sensors for different measurments! u  f   u  f  i2 f ; u  f   u  f  ( 2 f )2

147 148

37
POLITECNICO DI MILANO

VIBRATION SENSORS: frequency content of VIBRATION SENSORS: seismometer

acceleration
u  t   B sin 2 ft  u t    2 B sin 2 ft
u  f   u  f  i2 f ; u  f   u  f  ( 2 f )2
mB 2
x t   N sin  2 ft    
if you have significant harmonic components at high k
frequencies, you have large aceleration values B 2
 N sin  2 ft    
1
2
physics is helping us!! in fact:   1   0.6
• if you expect large components at high frequencies you   B sin 2 f  u (t )
2
must/can take a stiff accelerometer (rem f1  M f ) N 1  
12
• if you expect large components only at low frequencies used in seismological studies;
you must/can take a softer sensor, to gain better large masses are usually required
sensitivity and measure low acceleration levels

149 150

VIBRATION SENSORS: accelerometer SEISMOMETER

x(t) measured
u(t) to be estimated, u t   A sin 2 ft

mA   1   0.6
x t    N sin  2 ft   
k N 1   0

A 1
x t    sin  2 ft    u(t )
2
1 12

the measured relative displacement is Seismometers:

proportional to input acceleration! •The paper roll moves with the ground
•The pen remains stationary, because of the spring, hinge and weight

151 152

38
POLITECNICO DI MILANO

TRADITIONAL ACCELEROMETERS CAPACITIVE ACCELEROMETERS

uses an inductive position sensor to measure relative

displacement the motion of the mass affects
the electical capacitance of the
circuit

damping of the sensor is often

significantly dependent on
can measure at f = 0
temperature
can measure at f = 0

153 154

PIEZO-ELECTRIC ACCELEROMETER SERVO-ACCELEROMETER (closed loop)

force-balance sensor

stiffness and damping

are supplied by
electromagnetic circuits

the mass is kept in

position by the
electromagnetic force,
which can be accurately
measured

measures stress variations; has problems at low frequencies

155 156

39
POLITECNICO DI MILANO

MEASUREMENT CHAIN ANALOG TO DIGITAL CONVERSION

signal “conditioning” means generally the amplification of a

voltage output from mV to V; when amplified the signal can be
transmitted at longer distances (10-50 m) without an excessive
decay of the signal-to-noise ratio
number of intervals = 2 Nbit
157 158

Frequency analysis: Fourier Series and

Fourier Transform
ON THE
ADVANTAGES OF Periodic function: Fourier Series
DIGITAL SIGNAL
1 T /2 1

ACQUISITION
Xn  x(t ) e  i 2 nft dt f 
T
T T / 2
 T period
x(t )  X
n 
n e i 2 nft

Non-periodic function: Fourier Transform


x ( f )   x(t ) e i 2 ft dt the FT exixts if:

 
x(t )   x ( f ) ei 2 ft df

0
x(t ) dt  

159 160

40
POLITECNICO DI MILANO

PRACTICAL HARMONIC ANALYSIS EXAMPLE OF RECORDED VIBRATION AND SPD

USEFUL FORMULAS 1

0.8

N /2
x  k t    n Rn sin  2 nfk t  n 
0.6

accelerazione [m/s2]
0.4

0.2

1 Fourier Series
0

-0.2

N 1 M 2
1 R
 x  k t    2
-0.4

E  x 2 (t )   2 n -0.6

N
-0.8

k 0 n 1 -1

6 6.4 6.8 7.2 7.6 8

tempo [s] 0.1

mean square = variance Piede b - 450 rpm

vertical vibration recorded on 0.08

Supporto rigido

the first deck of a a weaving

 Rn( j )2 

spettro di potenza
GX  nf   lim T E 
0.06
looms plant
T 
 2  0.04

0.02

one-sided spectral power density (averaged over realizations) 0

0 10 20 30 40 50 60 70 80
frequenza [Hz]

161 162

FREQUENCY DOMAIN INTEGRATION FREQUENCY DOMAIN INTEGRATION

n n
    

a  t    cn  2 nf  sin  2 nft   d n  2 nf  cos  2 nft 
v  t   v0   n cn  cos  2 nft   d n sin  2 nft  n 1
1 

velocity    n sin  2 nft    n cos  2 nft 

n 1 acceleration
n n
    
a  t    cn  2 nf  sin  2 nft   d n  2 nf  cos  2 nft 
n 1 viceversa (integration)

   n sin  2 nft    n cos  2 nft 
n 1
n n
acceleration
cn   dn 
2 nf 2 nf

163 164

41
POLITECNICO DI MILANO

FREQUENCY DOMAIN INTEGRATION EXAMPLE OF VIBRATION RECORD INTEGRATION

 2

v  t   v0   n cn  cos  2 nft   d n sin  2 nft  1

accelerazione [m/s2 ]
1 0

velocity
n n -1

acceleration
cn   dn  ; n0 -2

2 nf 2 nf 0.15

0 20
tempo [s]
40 60

0.1

and, by introducing t=0 0.05

velocità [m/s]
 0

v0  v (0)   cn -0.05

-0.1
velocity
n 1 -0.15

0 20 40 60
tempo [s]

165 166

FROM FOURIER SERIES TO DISCRETE FROM FOURIER SERIES TO DISCRETE

FOURIER TRANSORM FOURIER TRANSORM
1 T /2
1 T /2
X n   x(t ) e i 2 nft dt f 
1
T
Xn 
T  T / 2
x(t ) e  i 2 nft dt f 
1
T
T T / 2
T period  T period

x(t )  xˆ (k t ) k  1,...N
x(t )  Xe
n 
n
i 2 nft

xˆ ( N t )  xˆ (0t ) ; T  N t N 1
Xˆ (nf )   xˆ (k t )e i 2 kn / N
N 1 k 0
X n  Xˆ (nf )   xˆ (k t )e i 2 kn / N
DFT – Discrete Fourier Transform
N 1
k 0 xˆ (k t )   Xˆ ( nf )ei 2 kn / N
k 0 IDFT – Inverse
DFT – Discrete Fourier Transform Discrete Fourier Transform
167 168

42
POLITECNICO DI MILANO

FROM FOURIER SERIES TO DISCRETE FROM FOURIER SERIES TO DISCRETE

FOURIER TRANSORM FOURIER TRANSORM
N 1
x(t )  xˆ ( k t ) k  1,...N Xˆ (nf )   xˆ (k t )e i 2 kn / N
xˆ ( N t )  xˆ (0t ) ; T  N t k 0
N 1 Criteria for time step?
Xˆ (nf )   xˆ (k t )e  i 2 kn / N To capture a pure waveform you must have at least 2
k 0 samples per period!
Criteria for time step?
To capture a pure waveform you must have at least 2 T 1
t   for purely harmonic
samples per period! 2 2f
T 1
t  min  for multi harmonic
2 2 f max
169 170

ALIASING – time step larger than period/2 FREE VIBRATION OF DAMPED n DoF SYSTEM
equations of motion

mq + cq + kq = 0 ; q(0) q (0) initial conditions

1.5

1
in normal coordinates

0.5 y j  2 j j y j   2j y j  0 ; y j (0)  φTj mq (0) / M j


y j (0)  φTj mq (0) / M j
v(t)

-0.5 response in terms of normal coordinates

 y j  0    j j y j  0  
y j  t   exp   j j t   y j  0  cos Dj t 
-1
sen Dj t 
-1.5  Dj 
0 2 4 6 8 10
modal superposition n
t [s]
q(t )  φ
j 1
j y j (t )

171 172

43
POLITECNICO DI MILANO

FREE VIBRATION OF DAMPED n DoF SYSTEM FREE VIBRATION OF DAMPED n DoF SYSTEM
initial displacement only initial velocity only (obtained by impulse excitation)

y j  2 j j y j   2j y j  0 ; y j (0)  φTj mq(0) / M j

 y j  2 j j y j   2j y j  0 ; y j (0)  0

y j (0)  0 y j (0)  φTj m 0 q2 (0) / M j

response in terms of normal coordinates response in terms of normal coordinates

  j j y j  0  
 y j  0 
y j  t   exp   j j t   y j  0  cos Dj t  sen Dj t  
 y j  t   exp   j j t   sen Dj t 
 
 Dj
Dj


initial condition is similar to first modal shape initial condition excites both modes

n
modal superposition q(t )  φ1 y1 (t ) modal superposition q (t )  φ
j 1
j y j (t )

173 174

FREE VIBRATION OF 1 DoF DAMPED SYSTEM FREE VIBRATION OF 1 DoF DAMPED SYSTEM

damping factor

alternative form of the

soln; amplitude and phase

TD

 v  0   1v  0  
v  t   exp  1t  v  0  cos D t  sen D t 
  D 
 v t  
free-amplitude decay method: ln    2
 v  t  TD  

175 176

44
POLITECNICO DI MILANO

FREE VIBRATION OF 1 DoF DAMPED SYSTEM DIRECT METHODS

• Time domain integration (step-by-

take a maximum/minimum at a given time
step)

• Frequency domain integration (via

Fourier analysis)

Both can be seen as extension of 1 DoF

analysis

177 178

DIRECT METHODS DIRECT METHODS

Frequency analysis; harmonic excitation Frequency analysis; harmonic excitation

mq  cq  kq  Fsen2 ft (a) The solution can be cast into the following matrix form:

z (t )  H ( f ) Fe i 2 ft  X ( f )e i 2 ft X ( f )  H ( f ) F
mr  cr  kr  F cos 2 ft (b)
1
H ( f )     2 f  m  i 2 f c  k 
2
complex form steady state solution (b+ia)  
mz  cz  kz  Fe i 2 ft z (t )  r (t )  iq (t ) H(f) is the frequency response function matrix; its inverse E(f)
is denoted as “mechanical impedance” matrix for the system.
ei 2 ft  cos2 ft +i sen2 ft ; i = -1 To compute the response amplitudes I solve the linear system:
The real and imaginary part of the complex solution z(t)
E ( f ) X ( f )     2 f  m  i 2 f c  k  X ( f )  F
2
deliver, respectively, the response to the forces Fcos2ft  
and Fsin2ft

179 180

45
POLITECNICO DI MILANO

DIRECT METHODS DIRECT METHODS

Frequency analysis; harmonic excitation Frequency analysis; harmonic excitation
The solution can be cast into the following matrix form:

E ( f ) X ( f )     2 f  m  i 2 f c  k  X ( f )  F
2
 
z (t )  H ( f ) Fe i 2 ft  X ( f )e i 2 ft

q (t )  Im  z (t )  Re  X ( ) sin t  Im  X ( ) cos t
The frequency response functions are “dynamic flexibility” functions.
r (t )  Re  z (t )   Im  X ( ) sin t  Re  X ( ) cos t The impedance functions are “dynamic stiffness” functions

The real part of a FRF represents the part of the response which is in
phase with the applied force; the imaginary part is in quadrature

181 182

DIRECT METHODS DIRECT METHODS

frequency analysis; frequency response functions Frequency analysis; harmonic excitation
•complex system excited by a limited number of
dynamic force components 2 1
• a limited number of response parameters are of 3 6  5 2 
interest (eg points of application of forces) m  l   ; k k 
1 1  2 1 
Given a set of suitable frequency response  6 3 
functions I can write the linear relation:  2 1
D
1
2
 
cq12  c  q2  q1   c  c 
2

 1 1 
 v1 ( f )   H11 ( f ) H12 ( f )   F1 ( f ) 
v ( f )     
 2   H12 ( f ) H 22 ( f )   F2 ( f ) 
E( f )X ( f ) 
NOTE: it can be very useful for substructuring
purposes    2 f 2 m  i 2 f c  k  X ( f )  F
 
183 184

46
POLITECNICO DI MILANO

DIRECT METHODS DIRECT METHODS

Frequency analysis; harmonic excitation Frequency analysis; harmonic excitation
2 1
3 6  5 2   2 1  2m c  k 2
m  l   k k  c  c    m  i c  k  kX ( )  F ;  
 2
;  2

1 1  2 1   1 1  k k m  2
 6 3   2 c  c  2 km
m l
  m  i k c  k  X ( )  F ;  k  k
 2
  m  i c  k  X ( )  F
2

  2 mm  i cc  kk  X ( )  F   2 m  i 2 c  k  X ( )  F ; X ( )  kX ( )
 2m c  k 2
          q(t )  Im  Η ( ) Feit   Im  X ( )eit  
2 2
 m i c k  kX ( ) F ;  ;
k k m  2
 2 c  c  2 km  Re  X ( ) sin t  Im  X ( ) cos t
  m  i k c  k  X ( )  F ;  k  k
 2

185 186

FREQUENCY RESPONSE FUNCTIONS FREQUENCY RESPONSE FUNCTIONS

example example

X1 X2

non dimensional frequency non dimensional frequency

187 188

47
POLITECNICO DI MILANO

DIRECT METHODS DIRECT METHODS

Frequency analysis; harmonic excitation Frequency analysis; harmonic excitation

k T k
  2 m  i 2 c  k  X ( )  F MB  3
s q  3 s T Im  X ( )eit  
l l
X ( )  kX ( ) k 1 
 Im  3 s T X ( )eit  
k  q2  q1 q1  l k 
MB    
l2  l l  1 
 Im  3 s T X ( )eit  
k k l 
 3 2 1 q  3 s T q
l l 1  1 
 Re  3 s T X ( )  sin t  Im  3 s T X ( )  cos t
l  l 

189 190

FREQUENCY RESPONSE FUNCTIONS Frequency response functions: example – mast

example structure
Frequency response functions can be
defined for any pair of input-output
functions.

Example:
T
s X •mast with lumped masses
•only flexural deformation
l4 •hysteretic damping

(M B / l) FRFs are given for a unit harmonic force

acting along q10(t)

non dimensional frequency 192

191 192

48
POLITECNICO DI MILANO

Frequency response functions: example – mast Frequency response functions: example – mast
structure structure

FRF for acceleration 

H10 10 ( f )   2 f  H10 10 ( f )
2

193 194

193 194

Frequency response functions: example – mast DIRECT METHODS

structure Frequency analysis: Fourier Series and Fourier Transform
Periodic function: Fourier Series
Top shear
1 T /2 1
Xn  
T T / 2
x(t ) e  i 2 nft dt f 
T
 T period
x(t )  X
n 
n e i 2 nft

Non-periodic function: Fourier Transform


x ( f )   x(t ) e i 2 ft dt the FT exixts if:
Base bending 
 
moment
195
x(t )   x ( f ) ei 2 ft df

0
x(t ) dt  

195 196

49
POLITECNICO DI MILANO

DIRECT METHODS DIRECT METHODS

Frequency analysis: response of n DoF systems Frequency analysis: response of n DoF systems
n t
hkj  t    response qk ( t ) when unit impulse
qk  t     Q j   hkj  t    d
j 1 t0 is applied along q j at time 
hkj  t    response qk ( t ) when unit impulse
is applied along q j at time 
t
rem Duhamel integral v  t    g   h  t    d
0

197 198

DIRECT METHODS DIRECT METHODS

Frequency analysis: response of n DoF systems Frequency analysis: response of n DoF systems
n t
qk  t     Q j   hkj  t    d   
n t

j 1 t0
qk  t    t   Q j f  e
i 2 f 
df  hkj  t    d

j 1 0 
assume that Q j  f    Q j t  e  i 2 ft
dt applying exactly the same reasoning as for the

SDoF systems, we reach the following result:

exists for j  1,...n n
qk  t     Q  f  H  f  e
j kj
i 2  ft
df
j 1 
n t
  
qk  t     Q j f  e
i 2 f 
df  hkj  t    d where H kj  f  is the FT of hkj  t 
j 1 t0   
199 200

50
POLITECNICO DI MILANO

DIRECT METHODS DIRECT METHODS

Frequency analysis: response of n DoF systems Frequency analysis: response of n DoF systems
n  
qk  t     Q j  f  H kj  f  ei 2 ft df q t    H  f  Q  f  ei 2 ft df
j 1  
~
where H kj  f  is the FT of hkj  t  applying Q  t   e j ei 2 f0t  Q  f   e j   f  f0 

q  t   H  f 0  e j ei 2 f0 t

q t    H  f  Q  f  ei 2 ft df

so that H  f 0   H  f 0 
~
q~  f   H  f  Q  f  1
H ( f )     2 f  m  i 2 f c  k 
2

 
201 202

DIRECT METHODS DIRECT METHOD

Frequency analysis; frequency response functions Response to arbitrary loading

mq  cq  kq  Q m q  cq  kq  Q


Q ( f )   Q (t ) e  i 2 ft dt FT of force components time domain frequency domain
 t
q(t )   h(t   )Q ( )d q ( f )  H ( f )Q ( f )
   2 f  m  i 2 f c  k  q ( f )  Q ( f )
2 0

  

Unit impulse
q ( f )   q(t ) e  i 2 ft dt Frequency
Linear system solution  obtain FT of the response 
response  response
 function Q ( f )   Q (t ) e i 2 ft dt function
q (t )   q ( f ) e i 2 ft
df 

 
obtain time response by inverse FT H ( f )   h(t ) e i 2 ft dt


203 204

51
POLITECNICO DI MILANO

FROM FOURIER SERIES TO DISCRETE

FOURIER TRANSORM
1 T /2
Xn 
T  T / 2
x(t ) e  i 2 nft dt f 
1
T
 T period
x(t )  
n 
X n ei 2 nft

FS: totally equivalent to FT for finite duration

transient function

1
Gn  G  n f 
T
206

205 206

FROM FOURIER SERIES TO DISCRETE FROM FOURIER SERIES TO DISCRETE

FOURIER TRANSORM FOURIER TRANSORM

1 T /2 1 1 T /2 1
Xn  
T T / 2
x (t ) e i 2 nft dt f 
T
Xn  
T T / 2
x (t ) e i 2 nft dt f 
T
x(t )  xˆ ( k t ) k  1,...N T period
x(t )  xˆ ( k t ) k  1,...N T period

xˆ ( N t )  xˆ (0t ) ; T  N t xˆ ( N t )  xˆ (0t ) ; T  N t
we apply the trapezoidal rule to estimate the FS
coefficient:
Nt
1 1 
 at  dt  t  2 a0  a1  a2  ...  aN 1  2 aN 
0

207 208

52
POLITECNICO DI MILANO

FROM FOURIER SERIES TO DISCRETE FROM FOURIER SERIES TO DISCRETE

FOURIER TRANSORM FOURIER TRANSORM
1 T /2 1 T /2
Xn  
T T / 2
x(t ) e i 2 nft dt Gn  
T T / 2
g (t ) e i 2 nft dt
Nt
1 1  1 N 1
1
 at  dt  t  2 a0  a1  a2  ...  a N 1  2 aN  Gn   gˆ  k t  e  i 2 n fk t
; n fk t  n k t
0 N k 0 N t
 
N 1
1 1
Gn  gˆ 0t   gˆ  Nt e  i 2nfNt   gˆ kt e  i 2nfkt
1 N 1
1 ˆ
2N N k 1 Gn 
N
 gˆ  k t  e
k 0
 i 2  nk / N

N
G( n f )
ˆ N t )  g( 0 t )
but e i2 n fN t  ei 2 n  1 and g( N 1

1 N 1 ˆ n f ) 
G(  gˆ  k t  ei 2 nk / N
Gn 
N
 gˆ kt ei 2nfkt k 0 Discrete Fourier Transform
k 0 DFT

209 210

DISCRETE FOURIER TRANSORM: properties DISCRETE FOURIER TRANSORM: properties

N 1 the coefficients of the DFT are periodic and conjugate

ˆ n f ) 
G(  gˆ  k t  ei 2 nk / N
k 0
symmetric
Modulo DFT
DFT absolute value
5

N 1 N N 1
f 
Gˆ n  N f  

 gˆ kt ei 2 n  N k / N
4.5

2 2 N t 4

k 0
3.5
1
  f MAX 3

N 1 2 t
 gˆ kt ei 2nk / N ei 2k
2.5

ei 2k  1  2

1.5
k 0 1

the coefficients of the DFT are periodic, this being 0.5

0
physically not justified -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9

Note they have also conjugate symmetry (from Gn) frequency component
indice frequenza

211 212

53
POLITECNICO DI MILANO

FROM FOURIER SERIES TO DISCRETE DISCRETE FOURIER TRANSORM: properties

FOURIER TRANSORM
N 1 N 1
l
Xˆ (nf )   xˆ (k t )e i 2 kn / N gˆ kt    Gˆ nf ei 2nk / N
k 0
N n 0
Inverse Discrete Fourier
Criteria for time step? Transform IDFT
To capture a pure waveform you must have at least 2 N 1
samples per period! ˆ n f ) 
G(  gˆ  k t  ei 2 nk / N
k 0
T 1 Discrete Fourier Transform
t   for purely harmonic DFT
2 2f
are two periodic sequencies; only the first period is
T 1 physically significant (half period for frequency
t  min  for multi harmonic coefficients)
2 2 f max
213 214

DISCRETE FOURIER TRANSORM: properties

1
Gn  G  n f 
T
• the FS is a perfect estimator of the FT
• what about the performance of the DFT in estimating
the FT?
The theory of distributions allows for deriving this
reltionship in the following form
 
1 1  j 
Gˆ nf  
t
 Gn  jN f   t  G nf  
t 
j   j   

aliasing relationship
215

215 216

54
POLITECNICO DI MILANO

DISCRETE FOURIER TRANSORM: aliasing DISCRETE FOURIER TRANSORM: aliasing

   
1 1  j  1 1  j 
Gˆ nf  
t
 Gn  jN f   t  G nf  
t 
Gˆ nf  
t
 Gn  jN f   t  G nf  
t 
j   j    j   j   

the DFT is a correct estimator if the only significant term

in the previous summation is j=0, so that To illustrate the relationship an example is shown,
1 regarding the transform of the h(t) function, which is
Ĝ  n f   G  n f  known and equal to the FRF H(f). An oscillator with unit
t stiffness, natural period equal to 0.3 s and damping ratio
To illustrate the relationship an example is shown, equal to 0.05, was considered.
regarding the transform of the h(t) function, which is The exact H(f) is compared to the DFT of the h(t), first
known and equal to the FRF H(f). An oscillator with unit sampled at a quite large time step, equal to 0.1s
stiffness, natural period equal to 0.3s and damping ratio
equal to 0.05, was considered.

217 218

DISCRETE FOURIER TRANSORM: properties DISCRETE FOURIER TRANSORM: properties

Re{D.F.T.}*Δt Re{H(f)} Re{D.F.T.}*Δt Re{H(f)}
Re{H(f-1/Δt)} Re{H(f+1/Δt)} 6 Re{H(f-1/Δt)} Re{H(f+1/Δt)}
6
t  0.1s 4 t  0.05s
4
1 1
2  f MAX  5Hz 2  f MAX  10Hz
2t 2t
0 0

-2 -2

-4 -4

-6 -6
-10 -5 0 5 10 -20 -10 0 10 20
 f [Hz] 
1  j  1  j 
f [Hz] Ĝ  n f  
t
 G  n f  t 
j 
Ĝ  n f  
t
 G  n f  t 
j 

219 220

55
POLITECNICO DI MILANO

DISCRETE FOURIER TRANSORM: properties Frequency domain analysis of the response of a SDoF
system to arbitrary excitation
1.5

vt    H  f G  f  e
i 2ft
1
df
0.5


v(t)

-0.5
v t    V  f ei 2ft df

6
-1
4 from definition of inverse FT, so that, upon comparison
-1.5 2
0 2 4 6 8 10

V  f   G f  H  f 
0
t [s] -2

-4

-6
-10 -5 0 5 10 frequency domain convolution (just a product of fcns…)
f [Hz] 222

221 222

DIRECT METHODS DIRECT METHODS

Frequency analysis; response computation Frequency analysis; numerical performance
N
assume that the time step is adequate (no aliasing), so that Vˆ nf   Vˆ * N  n f  nN
2
1
X̂  n f   X  n f  impose symmetry
t DFT absolute value
Modulo DFT

V  f   G f  H  f 
5

4.5

3.5

1 1
Vˆ nf   V nf   t Gˆ nf  H nf 
2.5

t t 1.5

N
 Gˆ nf  H nf 
0.5

0n 0

2 -5 -4 -3 -2 -1 0 1 2 3 4 5 6

frequency
7 8 9

component
indice frequenza

223 224

56
POLITECNICO DI MILANO

DIRECT METHODS DIRECT METHODS

Frequency analysis; response computation Frequency analysis; response computation
The period must be large enough to «contain» the entire
V̂  n f   G
ˆ  n f  H  n f  response

Note that response is periodic! The period must be large

enough to «contain» the entire response, which lasts
longer than the input, since free vibration occur at the end
of the excitation.

225 226

DIRECT METHODS DIRECT METHODS

Frequency analysis; response computation Frequency analysis; numerical performance
Qˆ j lt  l  0,..., N  1 ; j  1,..., n
Note that response is periodic! The period must be large input time series
enough to «contain» the entire response, which lasts ˆ  k f  , k  0,..., N
1 – obtain DFT of force components Q
longer than the input, since free vibration occur at the end 2
of the excitation. To do this we can add D seconds to the 2 - solve linear complex system
excitation, such that
   2 k f 2 m  i 2 k f c  k  qˆ  k f   Q
ˆ  k f 
 
e1D   k  0,...,
N
2
ln  3 – impose symmetry

D qˆ  k f   qˆ *  N  k  f 
N
k
 1,..., N  1
1 2
4 - obtain time series of response qˆ  l t  l  0,..., N  1 by IDFT

227 228

57
POLITECNICO DI MILANO

DIRECT METHODS DIRECT METHODS

Frequency analysis; numerical performance Frequency analysis: example of application
As an example, the procedure was applied to study the response
   2 k f 2 m  i 2 k f c  k  qˆ  k f   Q
ˆ  k f 
  of an harmonic oscillator, with period 1s and damping ratio 0.05,
N subjected to the external force depicted in figure 2.4.9-a; as it can
k  0,..., be inferred from the figure, the external force is not zero only
2 within the interval (0,5 s)
N
qˆ  k f   qˆ *  N  k  f  k  1,..., N  1
2 60

40

to deal with periodicity the natural period and the damping factor for 20

g(t) [kN]
the first mode must be estimated 0

ln  -20

D -40

11
-60
0 1 2 3 4 5 6
t [s]
230

229 230

DIRECT METHODS DIRECT METHODS

Frequency analysis: example of application Frequency analysis: example of application
60

…as it can be inferred from the figure, the external force is not 40

zero only within the interval (0,5 s) A sequence of zeros with 20

g(t) [kN]
time length D = 7.8 s was first added to the external force before
0

-20

evaluating the discrete Fourier transform. -40

-60
0 1 2 3 4 5 6
t [s]
60 80
Re{G(f)}
60 Im{G(f)}
40

40
20
g(t) [kN]

20

G(f) [kNs]
0
0
-20
0 1 2 3 4 5
-20
-40
-40
-60
0 1 2 3 4 5 6 -60
t [s]
-80
231 f [Hz] 232

231 232

58
POLITECNICO DI MILANO

DIRECT METHODS DIRECT METHODS

Frequency analysis: example of application
80
Frequency analysis: example of application
Re{G(f)}
60 Im{G(f)}

40

G(f) [kNs] 20 0.08

0
N=128
-20
0 1 2 3 4 5 0.06 N=256
-40 acc.lin. Dt=0.1s
0.04 acc. lin. Dt=0.01 s
-60
0.0025
-80 Re{H(f)} 0.02
f 0.002
[Hz]
Im{H(f)}

v(t)
0.0015 0
0.001 -0.02
0.0005
-0.04
0
-0.0005 -5 -4 -3 -2 -1 0 1 2 3 4 5 -0.06

-0.001 -0.08
-0.0015 0 1 2 3 4 5 6 7 8 9 10 11 12 13
-0.002
t [s]
-0.0025
f [Hz] 233 234

233 234

DIRECT METHODS DAMPING MODELLING

Frequency analysis: example of application
0.08
N=128 TOPICS
0.06 N=256
acc.lin. Dt=0.1s • the linear hysteretic model
0.04 acc. lin. Dt=0.01 s
0.02 • damping factors for modal analysis – the case of
non homogeneous systems
v(t)

0 0.08
N=128
0.06 N=256
-0.02
0.04
acc.lin. Dt=0.1s
acc. lin. Dt=0.01 s
• building a viscous matrix from damping factors
-0.04

-0.06
0.02
• Rayleigh damping
v(t)

-0.08 -0.02 • hysteretic damping and direct frequency analysis

0 1 2 3 4 -0.045 6 7 8 9 10 11 12 13
-0.06 t [s]
-0.08
0 0.5 1 1.5 2
t [s]
235

235 236

59
POLITECNICO DI MILANO

DAMPING MODELLING DAMPING MODELLING

comments on viscous model from viscous to linear hysteretic damping model
SDOF system under harmonic excitation – steady state response modify viscous model to cancel frequency dependence
F cH
v t   N sen t     V sen t    let c WD, H   cH V 2
k 
we are interested in dissipated energy (per cycle)
cH   k has same dimensions as stiffness
t T t T
T
WD,V   cv dt  c V 22  cos t    dt  c V 
2 2 2 2

t t
2 cH hysteretic damping coefficient
WD,V   c V  2
 hysteretic damping factor (non-dimensional)

experimentally: energy dissipation in structural elements under

cyclic loading is largely independent of frequency

237 238

DAMPING MODELLING DAMPING MODELLING

dissipation in hysteretic and viscous damping models equivalent viscous damping

make dissipated energy non dimensional wrt maximum elastic energy by equating the dissipated energy expressions we get the
equivalent viscous damping factor
for viscous
c k  1  1
WD ,V  cV 2       (*)
  4  2 km 2  km 2  2 
VE , MAX 1 2
kV 1
2 
and hysteretic model in practical applications  (**)
2
WD, H  cH V 2 can be seen as an alternative
  2 definition of the hysteretic this approximation amounts to calibrating the viscous model
VE , MAX 1 2 with respect to resonant components of response, these being
kV damping factor, also on
2 experimental grounds those most affected by damping effects

239 240

60
POLITECNICO DI MILANO

DAMPING MODELLING DAMPING MODELLING

frequency response of the hysteretic oscillator frequency response of the hysteretic oscillator

By substituting the equivalent viscous damping factor (*) into the

expressions of the frequency response function, dynamic magnification If we compare the frequency response functions for the hysteretic and
factor and phase we get the following: viscous oscillator, under the assumption (**) we get very close results
for typical damping values.
1 1
HH  f   2 Question: can we use the hysteretic model in frequency domain
k f
1 i analysis with arbitrary loading ? i.e. for a n-DOF system according to:
f12  
     2 f 2 m  i 2 f c  ic H  k  q ( f )  Q ( f )
1  
NH  f    H  arctg   
2  f 
2
 f 
2
 1  
 1  2 
 2  f12 
 f1 

241 242

DAMPING MODELLING DAMPING MODELLING

hysteretic damping and transient analysis hysteretic damping – physical consistency
Question: can we use the hysteretic model in frequency domain Secondly, we can check the physical consistency of the linear hysteretic
analysis with arbitrary loading ? model by Fourier transforming the FRF back into the time domain and
inspect the obtained unit impulse response function. We get:
We have first to make the FRF conjugate symmetric in order to deliver 15
isteretico
a real time function for the system response. For a SDOF systems this viscoso
amounts to setting: 10

1 1 5
Hf  2
k f

h(t)
0
1  i sgn( f )  -5 -4 -3 -2 -1 0 1 2 3 4 5
f12 -5

This is also useful to avoid the introduction of fictitious forces in -10

static analysis (f=0).
-15
t [s]

243 244

61
POLITECNICO DI MILANO

DAMPING MODELLING DAMPING MODELLING for modal analysis

hysteretic damping – physical consistency non homogeneous systems - definitions
15
isteretico
viscoso Assume that our system is composed of subsystems of two types:
10
1.Spring + dashpot elements
5 2.Hysteretic subsystems

h(t) 0 We treat the problem to derive consistent values of the modal damping
-5 -4 -3 -2 -1 0 1 2 3 4 5
factors with a twofold purpose
-5
• Perform classical modal superposition
-10 • Compare with the damping factors which are implicit in other
procedures (eg based on Rayleigh damping)
-15
t [s]
Note: we’ll always assume, in the following, the system under steady
Note: the causality principle is violated ! state harmonic motion with circular frequency equal to,  i.e.;
Comment: the violation is limited for normally damped structural
systems (the example in the figure had a 20% damping factor) qt   q sent   

245 246

DAMPING MODELLING for modal analysis DAMPING MODELLING for modal analysis
non homogeneous systems - viscoelastic component non homogeneous systems - hysteretic subsystem

cHr   r  r  contribution to the global damping matrix

z   (t )  Z   sen  t   
l l c r    k
Assume   (i.e. c   r c ( r ) )
(l ) where
z variation of the distance between the nodes connected by the
component
 (r ) hysteretic damping factor for the subsystem

(l ) (l )
cH( r ) , k ( r ) contribution of the subsystem to the global hysteretic
c ,k damping and stiffness coefficients of the component damping matrix and stiffness matrix

WD  c l  Z ( l )2 c l  WDr 

 2  r 
l
  2 r 
VE l, MAX
 1 (l ) (l )2
k Z k l  VE , MAX
2
non dimensional dissipated energy non dimensional dissipated energy, for qt   q sent   

247 248

62
POLITECNICO DI MILANO

DAMPING MODELLING for modal analysis DAMPING MODELLING for modal analysis
non homogeneous systems - dissipated energy non homogeneous systems - total dissipated energy
Assume single-mode resonant harmonic motion nH nV
c l   l 
q  t   φ j sin  j t    WD , j  2 ( r ) VE ,rMAX

, j   2 j
k l 
VE , MAX , j
 r 1
nH
l 1
nV  
VE , MAX , j
compute total dissipated energy V
r 1
r
E , MAX , j   VE l, MAX
l 1

,j

V 
nV
c l  l 
WD , j   2 j l  VE , MAX , j in viscoelastic components total non-dimensional dissipated energy
l 1 k
nH Criterion
WD( H, j )   2 ( r )VE( ,rMAX
)
,j
in hysteretic components the modal damping factor (viscous) for mode j must lead to the
r 1
same amount of dissipated energy
1 T i 
where VE i, MAX

,j  φj k φj
2

249 250

DAMPING MODELLING for modal analysis DAMPING MODELLING for modal analysis
non homogeneous systems - equivalent modal damping non homogeneous systems - equivalent modal damping
Criterion nH nV
cl   l 
the modal damping factor (viscous) for mode j must lead to the
same amount of dissipated energy 1   ( r )VE r,MAX

, j   j
k l 
VE , MAX , j
j  r 1
nH
l 1
nV
j 2
WD ,V , j
 4  j  4  j
for a viscous oscillator
having same properties
V   r
E , MAX , j   VEl, MAX

,j
VE , MAX , j j as mode j
r 1 l 1

“weigthed damping” formula

equating to    :

c  l 
nH nV l nH nV
1 T T 1 2
1
 (r ) r 
VE , MAX , j    j  l  VE , MAX
k
,j
Note: V r 
E , MAX , j   VEl, MAX

,j 
2
 j k j   j
2
j  r 1
nH
l 1
nV
r 1 l 1

2  Tj m j  1
V  
r 1
r
E , MAX , j   VE l, MAX
l 1

,j
if, as usual

251 252

63
POLITECNICO DI MILANO

DAMPING MODELLING DAMPING MODELLING

how to form a viscous matrix- from normal modes how to form a viscous matrix- from normal modes

definition of modal damping factors transformation of the viscous matrix into the modal space

j 
Cj

1 φTj c φ j
 
C  diag 2  j j M j  ΦT c Φ
2 j M j 2 j φ m φ jT
j

transformation of the viscous matrix into the modal space 

c  Φ T diag 2  j j M j Φ 1 

C  diag 2  j j M j  ΦT c Φ  problem: cannot be done if, as usually, a limited number of modes
is extracted
solution: approximate inversion of the modal matrix can be
perfomed by exploiting orthogonality
note: it can be inverted !

253 254

DAMPING MODELLING DAMPING MODELLING

how to form a viscous matrix- approximate inversion how to form a viscous matrix – orthonormal damping


c  Φ T diag 2  j j M j Φ 1  
c  Φ T diag 2  j j M j Φ 1 
solution: approximate inversion of the modal matrix where: Φ  T  m Φ M 1 Φ 1  M 1Φ T m
 
M  diag M j  ΦT m Φ upon substitution

Φ T M  m Φ 
c  m Φ M 1 diag 2  j j M j M 1ΦT m 
Φ 1  M 1Φ T m
and, by
Φ  T  m Φ M 1 transposing: p 2  j j 
or c  m   φ j φTj m
note: inversion is trivial since M is diagonal  j 1 M j  
255 256

64
POLITECNICO DI MILANO

DAMPING MODELLING DAMPING MODELLING

how to form a viscous matrix - comments how to form a viscous matrix - Rayleigh damping
  j
c  m  k j  
 2  j j 
p 2 j 2
c  m   φ j φTj m  and  can be computed by imposing the damping factors
 j 1 M j  of two modes

Notes: the matrix is full in principle

the matrix is not positive definite if p < n

modal damping
total

higher mode oscillations are undamped

stiffness proportional

mass proportional

modal frequency

257 258

DAMPING MODELLING DAMPING MODELLING

non homogeneous Rayleigh damping in direct frequency domain analysis
nm nm use combined (viscous+hysteretic) damping
c =   ( m )m ( m )    ( m )k ( m )
cH
m 1 m 1 c  cV 
nm 2 f cH hysteretic damping matrix
1
nm  j
( m)
VE m, MAX

,j
1
j   (m)
M (j m )  m 1
   2 f 2 m  i 2 f cV  k  icH  q ( f )  Q ( f )
2 j M j 2 nm
 
m 1
V  m
E , MAX , j
modal damping factor m 1
compare to weighted damping
cH  k for a system with homogeneous damping
nH
1 ( r )  E , r  nV c l 
c H    ( j )k ( j )
nH


r 1 2
 VMAX , j    j  l  VE l,MAX
2 k

,j System is made of nH hysteretic subsystems
j  nH
l 1
nV j 1
VE r,MAX

r 1
, j   VE , MAX , j
l 
l 1 Note: the hysteretic matrix can be easily assembled in parallel to stiffness

259 260

65