Structural

Dynamics

10/14/08 Dynamic Analysis 1

subjected to a time
dependent load.

k
F(t)
m

10/14/08 Dynamic Analysis 2

 Free-body diagram of the
mass.

T = kx F(t) ma = mx
=
m m

10/14/08 Dynamic Analysis 3

 F( t ) − k x = m x

m x + k x = F( t )

10/14/08 Dynamic Analysis 4

Solution of D.E. is sum
of homogeneous and
particular solutions:
Homogeneous :

F( t ) = 0

m x + k x = 0
10/14/08 Dynamic Analysis 5
 Let :
k
ω =2

m
Then :
x + ω x = 0
2

ω 2 is the natural circular frequency

10/14/08 Dynamic Analysis 6
 2π
τ=
ω
τ is the period (measured in seconds)

10/14/08 Dynamic Analysis 7

 Displacement due to simple harmonic motion.

τ
xm

10/14/08 Dynamic Analysis 8

 One
Dimensional
Bar Element

10/14/08 Dynamic Analysis 9

 Step 1 - Select Element Type

d̂1x 1 2 d̂ 2 x
x̂
f̂1ex ( t ) f̂ 2ex ( t )
L
E - modulus of elasticity
A - cross-sectional area
ρ - mass density

10/14/08 Dynamic Analysis 10

 Step 2 - Select a
Displacement Function
û = a1 + a 2 x̂
û = N 1d̂1x + N 2d̂ 2 x
x̂
N1 = 1 −
L
x̂
N2 =
L
10/14/08 Dynamic Analysis 11
 Step 3 - Define
Strain/Displacement and
Stress/Strain Relationships
∂
{εx } = = [ B] d̂
û
∂x̂
{}
[ B] = − 1 1 
 L L

{} d̂1x 
d̂ =  
d̂ 2 x 
{σ} = [ D]{εx } = [ D][ B] {d̂}
10/14/08 Dynamic Analysis 12
 Step 4 - Derive Element
Stiffness and Mass
Matrices and Equations

With time dependent loading

f̂1x ≠ f̂ 2 x

10/14/08 Dynamic Analysis 13

Newton’s Second Law

 
f = ma
10/14/08 Dynamic Analysis 14
 NODAL EQUILIBRIUM
EQUATIONS

∂ d̂1x 2
f̂ = f̂1x + m1
e
1x
∂t 2

∂ d̂ 2 x 2
f̂ e
2x = f̂ 2 x + m 2
∂t 2

10/14/08 Dynamic Analysis 15

 m1 and m2 are obtained by
lumping the total mass of the
bar equally at the two nodes

ρAL
m1 =
2
ρAL
m2 =
2
10/14/08 Dynamic Analysis 16
 Lumped Mass Model

d̂1x 1 2 d̂ 2 x
m1 x̂ m2
f̂1ex ( t ) f̂ 2ex ( t )
L

10/14/08 Dynamic Analysis 17

 Equilibrium in Matrix
Form
 ∂ d̂1x 
f̂1ex  f̂1x  m1 
0  ∂ t  2 
 e  =  +    
f̂ 2 x  f̂ 2 x   0 m 2   ∂ d̂ 2 x 
 ∂ t 
2

10/14/08 Dynamic Analysis 18

 Equilibrium in Matrix
Form

{ } [ ]{ }
f̂ ( t ) = k̂ d̂ + [ m̂ ] d̂
e 
{}
10/14/08 Dynamic Analysis 19
 Defining Terms
 1 − 1
[k ] = 
AE Element Stiffness Matrix
L − 1 1 
ρAL 1 0
[m] = Element Lumped Mass Matrix
2 0 1
{}d̂ Nodal Displacements
{} ∂ d̂
d̂ =
2
{} Nodal Accelerations
∂ t2

10/14/08 Dynamic Analysis 20

 Consistent Mass Matrix

{ } 
X = −ρ û
e
{}
{fb } = ∫∫∫[ N] { X} dV
T

{fb } = −∫∫∫ ρ[ N ] T
{}
û dV
V

10/14/08 Dynamic Analysis 21

 Consistent Mass Matrix

{ û} = [ N ] { d̂}
{û} = [ N] d̂
 
{}
{û} = [ N] d̂
 
{}
10/14/08 Dynamic Analysis 22
 Consistent Mass Matrix

{fb } = −∫∫∫ ρ[ N] [ N] d̂ dV
 T
{}
{}
V

{fb } = −[m̂] d̂


[m̂] = ∫∫∫ ρ[ N] [ N] dV T

10/14/08 Dynamic Analysis 23

 Consistent Mass Matrix
Bar Element
[m̂] = ∫∫∫ ρ[ N] [ N ] dV
T

1 − x̂ 
 L   x̂
[m̂] = ∫∫∫ ρ x̂  1 − x̂  dV
V   L L 
 L 
L 1 −
x̂ 
 L   x̂
[m̂] = ρ A ∫  x̂  1 − x̂  dx̂
0  L L 
 L 
10/14/08 Dynamic Analysis 24
 Consistent Mass Matrix
Bar Element

L 1 −
x̂  − x̂  1 − x̂  x̂ 
 L  L 1 
[m̂] = ρ A ∫   L  L  dx̂
 x̂  x̂   x̂  x̂  
0  1 −   
  L  L   L  L  
ρ A L 2 1
[m̂] =  
6 1 2

10/14/08 Dynamic Analysis 25

STEP 5 - Assemble the Global
Equations and Apply B.C.’s

{ F( t )} = [ K ] { d} + [ M ] {d}


10/14/08 Dynamic Analysis 26

{ F( t )} = [ K ] { d} + [ M ] {d}


Now must solve coupled set

of ODE’s instead of set of
linear algebraic equations!

10/14/08 Dynamic Analysis 27

 Consistent Mass Matrix

[m] = ∫∫∫ ρ[ N] [ N] dV T

10/14/08 Dynamic Analysis 28

 Beam Element

ŷ , v̂

φ̂1 , m̂ 1
1 x̂ 2 φ̂ 2 , m̂ 2

L
f̂1y , d̂ 1y f̂ 2y , d̂ 2y

10/14/08 Dynamic Analysis 29

 Shape Functions

N1 =
L3
1
( 2x̂ 3 − 3x̂ 2 L + L3 )

N2 =
1 3
L3
( x̂ L − 2x̂ 2 L2 + x̂L3 )

N3 =
1
L3
( − 2x̂ 3 + 3x̂ 2 L )

N4 =
L
1 3
3
( x̂ L − x̂ 2 L2 )

10/14/08 Dynamic Analysis 30

 Shape Functions
1.000
N1 N3
0.500
N2
L
0.000
0 N4

-0.500

10/14/08 Dynamic Analysis 31

 Consistent Mass Matrix
[ m] = ∫∫∫ ρ[ N] [ N] dV
T

 156 22L 54 − 13L 

 22L 4L2
13L 2 
− 3L 
[m] = 
m
420  54 13L 156 − 22L 
 2 
− 13L − 3L − 22L 4L 
2

10/14/08 Dynamic Analysis 32

 Lumped Mass Matrix

1 0 0 0 
 αL 2

m  0 0 0 
[ m ] = 0 210
0 1 0 

2
0 αL 
2
0 0
 210 

10/14/08 Dynamic Analysis 33

 Lumped Mass Matrix

2nd and 4th terms account for rotary inertia.

α = 0 if this is ignored.
α = 17.5 if mass moment of inertia of bar
spinning about one end is selected
2
 m  L 
  
 2  2 
I=
3
10/14/08 Dynamic Analysis 34
 Consistent Mass Matrix -
CST
Q 0 
[ m] =  
 0 Q
 2 1 1 u1
m
[ Q] = 1 2 1 u 2
12
1 1 2 u 3

For each degree of freedom

10/14/08 Dynamic Analysis 35

 Consistent Mass Matrix -
CST
2 0 1 0 1 0
0 2 0 1 0 1

ρAt 1 0 2 0 1 0
[ m] =  
12  0 1 0 2 0 1
1 0 1 0 2 0
 
 0 1 0 1 0 2
10/14/08 Dynamic Analysis 36
 Lumped Mass Matrix -
CST

Q 0 
[ m] =  
 0 Q
1 0 0 u 1
m
[ Q] = 0 0 1 u 2
3
0 0 1 u 3

10/14/08 Dynamic Analysis 37

 Lumped Mass Matrix -
CST
1 0 0 0 0 0
0 1 0 0 0 0

ρAt 0 0 1 0 0 0
[m] =  
3 0 0 0 1 0 0
0 0 0 0 1 0
 
0 0 0 0 0 1
10/14/08 Dynamic Analysis 38
 Consistent Mass Matrix -
Quad

Q 0 
[m] =  0 Q 
 
4 2 1 2
2 4 2 1 
[ Q] = m 
36 1 2 4 2
 
2 1 2 4
m = ρAt

10/14/08 Dynamic Analysis 39

 Consistent Mass Matrix -
Quad
4 0 2 0 1 0 2 0
0 4 0 2 0 1 0 2

2 0 4 0 2 0 1 0
 
m 0 2 0 4 0 2 0 1
[ m] = 
36 1 0 2 0 4 0 2 0
 
0 1 0 2 0 4 0 2
2 0 1 0 2 0 4 0
 
0 2 0 1 0 2 0 4

10/14/08 Dynamic Analysis 40

 Hybrid Methods

Attempts have been made to combine

consistent and lumped mass approaches
to achieve some of the benefits of each!

10/14/08 Dynamic Analysis 41

 HRZ Lumping
 Hinton, Rock, and Zienkiewicz
 Compute the diagonal terms of
consistent mass matrix.
 Compute total mass of element, m
 Compute s by adding diagonal
coefficients associated with
translational D-O-F that are in same
direction.
 Scale all diagonal coefficients by
multiplying by m/s
10/14/08 Dynamic Analysis 42
 HRZ - Bar Element
ρ A L 2 1 
[ m̂] =  
6 1 2 
m = ρAL
ρAL
s = 4×
6
m 3
=
s 2
ρ A L 3 0
[ m̂] =
6 0 3
10/14/08 Dynamic Analysis 43
 HRZ - Beam Element
 156 22L 54 − 13L 
 22L 4 L 2
13L − 3L 2 
[m] = m  
420  54 13L 156 − 22L 
 2 
− 13L − 3L − 22L 4L 
2

m = ρAL
ρAL
s = 312 ×
420
m = 420
s 312

10/14/08 Dynamic Analysis 44

 HRZ - Beam Element

 420 
 312 × 156 0 0 0   39 
 420   
 0 × 4 L2 0 0   L2

m  m
[ m̂ ] = 312 =  
420  420  78  39 
 0 0 × 156 0 
312  
   L2

 420 2  
0 0 0 × 4L
 312 

10/14/08 Dynamic Analysis 45

 HRZ – Quadratic Serendipity

3 1
76 36

16 8
76 36
3x3 2x2
Gauss Rule Gauss Rule

10/14/08 Dynamic Analysis 46

 HRZ – Quadratic Lagrangian

1 1
36 36

4 16 16 4
36 36 36 36
3x3 2x2
Gauss Rule Gauss Rule

10/14/08 Dynamic Analysis 47

 % error in natural frequencies of a thick
simply-supported plate.
Half of the plate modeled with 8-noded
24 d-o-f elements
Mode Type of Mass Matrix
m n Consistent (%) HRz Lumping (%) Ad Hoc Lumping (%)
1 1 0.11 0.32 0.32
2 1 0.4 0.45 0.45
2 2 0.35 2.75 4.12
3 1 5.18 0.05 5.75
3 2 4.68 2.96 10.15
3 3 13.78 5.18 19.42
4 2 16.88 1.53 31.7

10/14/08 Dynamic Analysis 48

Optimal Lumping
 Only translational d-o-f
 Based on consistent mass matrix
 Use appropriate quadrature rule
 Chose integration points to
coincide with nodal locations
 [m] will be diagonal

10/14/08 Dynamic Analysis 49

  Let p be the highest order
complete polynomial in shape
function N
 let m be the highest order
derivative in strain energy (m =
1 elasticity, m = 2 bending)
 Chose quadrature rule with
degree of precision 2(p-m)

10/14/08 Dynamic Analysis 50

 Three noded bar element
p=2
m=1
2(p-m) = 2
Three point quadrature rule.
Newton -Cotes has points at the nodes.
(Simpson’ Rule)
b
1 4 b+a 1 
∫a f (x)dx = ( b − a )  6 f ( a ) + 6 f  2  + 6 f ( b ) 
10/14/08 Dynamic Analysis 51
 1
m ij = ρA ∫ N i N jdx = ∫ N i N j J dξ
−1

L
J=
2
m ij = ρA
L
( 1 − ( − 1) )  1 N i ( − 1) N j ( − 1) + 4 N i ( 0) N j ( 0) + 1 N i ( 1) N j ( 1) 
2 6 6 6 
i ≠ j m ij = 0

1 0 0 1 3 2
ρAL  
[ m] = 0 1 0
6  
0 0 4
10/14/08 Dynamic Analysis 52
 Serendipity

1
−
12

1
3

10/14/08 Dynamic Analysis 53

 Lagrangian

1
36

4 1
9 9

10/14/08 Dynamic Analysis 54

 Mass Matrices
 Product [m]{a} must yield the correct total
force on an element (F = ma) when {a}
represents a rigid-body translational
acceleration.
 Consistent mass matrices, [m] and [M] are
positive definite.
 Lumped mass matrix is positive semi-definite
when zero terms appear on main diagonal.
 Lumped mass matrix is indefinite when
negative terms appear on main diagonal.

10/14/08 Dynamic Analysis 55

 Mass Matrices
 Special treatment may be needed
to handle the last two cases.

10/14/08 Dynamic Analysis 56

 Best Type ?

1. Consistent matrices usually

more accurate for flexural
problems.
2. Consistent matrices give
upper bounds on natural
frequencies.

10/14/08 Dynamic Analysis 57

 Best Type ?
1. Lumped matrices usually give natural
frequencies less than exact values.
2. Simpler to form.
3. Occupy less storage.
4. Require less computational effort.
5. Usually more important in time-history
than in vibration problems.

10/14/08 Dynamic Analysis 58

 Damping
1. Structural damping is not viscous.
2. Due to mechanisms such as hysteresis
and slip in connections.
3. Mechanisms not well understood.
4. Awkward to incorporate into structural
dynamic equations.
5. Makes equations computationally
difficult.
6. Effects usually approximated by
viscous damping.
10/14/08 Dynamic Analysis 59
Types of Damping Models
 Phenomenological Damping
Methods (models actual dissipative
mechanisms)
➲ Elastic-Plastic Hysteresis Loss
➲ Structural Joint Friction
➲ Material Micro-cracking
 Spectral Damping Methods
➲ Introduce Viscous Damping
➲ Relies on Fraction of Critical Damping
10/14/08 Dynamic Analysis 60
 Critical Damping

ξ Fraction of Critical Damping

ξ = 1 Critical Damping

Critical Damping marks the transition between

oscillatory and non- oscillatory response of a
structure

10/14/08 Dynamic Analysis 61

Critical Damping Ratio

0.5% ≤ ξ ≤ 5% Steel Piping

2% ≤ ξ ≤ 15% Bolted or riveted steel structures
2% ≤ ξ ≤ 15% Reinforced or Prestresses Concrete

Actual value may depend on stress level.

10/14/08 Dynamic Analysis 62

 Rayleigh or Proportional
Damping

Damping matrix is a linear combination of

stiffness and mass matrices:

[ C] = α [ K ] + β [ M ]
10/14/08 Dynamic Analysis 63
 Rayleigh or Proportional
Damping
[C] is orthogonal damping matrix.
Modes may be uncoupled by eigenvectors
associated with undamped problem.

1 β
ξ = αω− 
2 ω

10/14/08 Dynamic Analysis 64

If critical damping ratio is
known at two frequencies
then:
1 β
ξ = αω− 
2 ω
( ξ 2 ω 2 − ξ 1 ω1 )
α=2
(
ω 2 − ω1
2 2
)
( ξ 1 ω 2 − ξ 2 ω1 )
β = 2 ω1 ω 2
(ω 2
2 −ω 2
1 )
10/14/08 Dynamic Analysis 65
 Natural Frequencies and Mode
Shapes
Undamped, Unforced Response

{ D} = { D } sin ωt
{ D } = ω{ D } cos ωt
{ D } = − ω { D } sin ωt
2

{ D} amplitudes of nodal d - o - f
ω circular frequency
ω
f = ( Hz )
2π
10/14/08 Dynamic Analysis 66
Results in generalized eigenproblem

( [ K ] − λ [ M ] ) { D } = { 0}
λ=ω 2

10/14/08 Dynamic Analysis 67

 Trivial Solution:

[ K ] − λ[ M] ≠0
{ D} = 0

10/14/08 Dynamic Analysis 68

 Nontrivial Solution:

[ K ] − λ[ M ] =0
{ D} ≠ 0

10/14/08 Dynamic Analysis 69

 λi ≡ Roots of Characteristic
Polynomial (eigenvalues)
{ D} i ≡ Associated Eigenvectors

10/14/08 Dynamic Analysis 70

 ω i Natural Frequencies
{D} i Normal Modes

10/14/08 Dynamic Analysis 71

 Natural Frequencies
 [K] and [M] n x n then there are n
eigenvalues and n eigenvectors
 [K] and [M] positive definite then
eigenvalues are all positive
 Mii = 0 infinite eigenvalue
 Mii < 0 negative eigenvalue - imaginary
frequency
 Use condensation to remove ith equation if
Mii = 0

10/14/08 Dynamic Analysis 72

 Rayleigh Quotient

λ=
{ D} [ K ]{ D}
T

{ D} [ M ]{ D}
T

[ K ] symmetric
[ M ] positive definite
{ D} th
approximat ion to i eigenvector
λ approximat ion to i th eigenvalue
10/14/08 Dynamic Analysis 73
 Rayleigh Quotient

{ v} [ K ]{ v}
T

λ min ≤ ≤ λ max
{ v} [ M ]{ v}
T

λ min smallest eigenvalue

λ max l arg est eigenvalue

10/14/08 Dynamic Analysis 74

 Modal Methods

When [K], [C], [M] are known and time

independent the problem is linear.

[ M ] { D} + [ C ] { D} + [ K ] { D} = R
  {
ext
}
{ D ( 0 ) } , { D ( 0 ) } given as initial conditions

10/14/08 Dynamic Analysis 75

 Modal Methods
Assume orthogonal damping, such as
Rayleigh Damping. Modes can be uncoupled:

{ D} [ M ] { D}
T
i j =0
{ D} [ K ] { D}
T
i j =0
{ D} [ C] { D}
T
i j =0
i≠j
10/14/08 Dynamic Analysis 76
 { D} [ M ] { D} = 1
T
i i

{ D} [ K ] { D} = ω
T
i i
2
i

{ D} [ C] { D} = 2ξ ω
T
i i i i

10/14/08 Dynamic Analysis 77

 Eigenvectors are linearly
independent

[ φ] = matrix of eigenvectors
(mode shapes)
{ D} = [ φ]{ Z}
{ Z} modal amplitudes

10/14/08 Dynamic Analysis 78

 Substitute into:

  {
[ M ] { D } + [ C ] { D } + [ K ] { D} = R ext
}
{ D ( 0 ) } , { D ( 0 ) } given

10/14/08 Dynamic Analysis 79

[ M ] [ φ] { Z} + [ C ] [ φ] { Z} + [ K ] [ φ] { Z} = R
  {ext
}
[ φ ] { Z ( 0 ) } = { D ( 0 ) }
[ φ ] { Z ( 0 )} = { D ( 0 ) } given

10/14/08 Dynamic Analysis 80

 Mode Displacement
Method

  [ ]
[ I ] { Z } + [ ξ ] { Z } + ω { Z} = { p}
 2

10/14/08 Dynamic Analysis 81

 Mode Displacement
Method
Pre-multiply by [φ]T

[ φ] T
[ M ] [ φ] { Z } + [ φ] [ C ] [ φ] { Z }
 T

+ [ φ]
T
[ K ] [ φ] { Z} = [ φ] { R }
T ext

[ φ] { Z ( 0 ) } = { D ( 0 ) }
[ φ] { Z ( 0 ) } = { D ( 0 ) } given

10/14/08 Dynamic Analysis 82

 Mode Displacement
Method

[ φ] [ M ] [ φ] = [ I ]
T

[ φ] [ C] [ φ] = [ ξ ]
T

[ φ] [ K ] [ φ] = [ ω ]
T 2

10/14/08 Dynamic Analysis 83

 Mode Displacement
Method

  [ ]
[ I ] { Z } + [ ξ ] { Z } + ω { Z} = { p}
 2

10/14/08 Dynamic Analysis 84

 Modes Uncouple:

  2
[ ]
[ I ] { Z} + [ ξ ] { Z} + ω { Z} = { Z}

Z + 2 ξ ω Z  + ω 2 Z = p i = 1, n
i i i i i i

10/14/08 Dynamic Analysis 85

 [ φ ] { Z ( 0 ) } = { D ( 0 ) }
[ φ ] { Z ( 0 )} = { D ( 0 ) }
[ φ ] [ M ] [ φ ] { Z ( 0 ) } = [ φ ] [ M ] { D ( 0 ) }
T  T

[ I ] { Z ( 0 ) } = [ φ ] [ M ] { D ( 0 ) }
 T

{ Z ( 0 ) } = [ φ ] [ M ] { D ( 0 ) }
 T

{ Z ( 0 )} = [ φ ] [ M ] { D ( 0 ) }
T

10/14/08 Dynamic Analysis 86

 Reduce size of problem:

m << n eq
m
{ D } = ∑ { φ} i Z i
i =1

10/14/08 Dynamic Analysis 87

 Error Estimate:

{ R } − [ M ] { D } − [ C ] { D } − [ K ] { D}
ext

e( t ) =
{R } ext

For an accurate analysis :

e ( t ) ≤ 1%

10/14/08 Dynamic Analysis 88

 In many structural dynamics
problems, more modes
participate in the quasi-static
response than in the dynamic
response. For a small m value,
the mode displacement method
may have difficulty in predicting
the quasi-static response.
10/14/08 Dynamic Analysis 89
Mode Acceleration method
Method

  [ ]
[ I ] { Z } + [ ξ ] { Z } + ω { Z} = { p}
 2

10/14/08 Dynamic Analysis 90

 Modal transformation only on
inertial and damping terms

[ M ] [ φ ] { Z } + [ C ] [ φ ] { Z } + [ K ] { D} = R
  {
ext
}
[ φ ] { Z ( 0 ) } = { D ( 0 ) }
[ φ ] { Z ( 0 )} = { D ( 0 ) } given

10/14/08 Dynamic Analysis 91

 [K ] {D} = {R ext } − [M ][φ]{Z
 } − [C][ φ]{Z
}

{D} = [K ] {R } − [K ] ([M ][φ]{Z

−1 ext −1  } − [C][ φ]{Z
 })

10/14/08 Dynamic Analysis 92

 [ φ] [ M ] [ φ] = [ I ]
T

[ M ] [ φ] = [ φ] −T

10/14/08 Dynamic Analysis 93

{ D} = [ K ] { R } − [ K ]
−1 ext −1
( [ φ ] { Z } − [ C ] [ φ ] { Z } )
−T

10/14/08 Dynamic Analysis 94

 [ φ] [ K ] [ φ] = [ ω ]
T 2

[ φ] [ K ] [ φ] [ ω ] = [ I ]
T 2 −1

[ K ] [ φ] [ ω ] = [ φ] [ I ] = [ φ]
2 −1 −T −T

[ φ] [ ω ] = [ K ] [ φ]
2 −1 −1 −T

10/14/08 Dynamic Analysis 95

{ D} = [ K ] {R } − [ K ] [ φ]
−1 ext −1 −T
{Z} + [ K ] [C][ φ]{Z }
 −1

{ D} = [ K ] {R } − [ φ] [ω ] {Z} + [ K ] [ C][ φ]{Z }

−1 ext 2 −1  −1

10/14/08 Dynamic Analysis 96

 { D} = [ K ] {R } − [ K ] [ φ] {Z } + [ K ] −1[ C][ φ]{Z }
−1 ext −1 −T

{ D} = [ K ] {R } − [ φ] [ω ] {Z} + [ K ] [ φ] [ φ] [ C][ φ]{Z }

−1 ext 2 −1  −1 −T T

{ D} = [ K ] {R } − [ φ] [ω ] {Z} + [ φ] [ω ] [ ξ]{Z }
−1 ext 2 −1  2 −1

10/14/08 Dynamic Analysis 97

 { D} = [ K ] { R } − [ φ ] [ ω
−1 ext 2 −1
] ( { Z } + [ ξ ] { Z } )
 1  2ξ i  
{ D} = [ K ] { R }
m
−1 ext
− ∑ { φ} i  2 Z i + Z i 
i =1  ωi ωi 

First term on RHS represents quasi-static

response, the second term represents
corrections for inertia and viscous effects.
10/14/08 Dynamic Analysis 98
Solve for Z terms as before:

Z + 2 ξ ω Z + ω2 Z = p i = 1, n
i i i i i i

10/14/08 Dynamic Analysis 99

 Mass Condensation
 Reduces number of d-o-f.
 Reduces expense of computing
eigenvalues.
 Detrimental to accuracy.
 Not used with optimal lumping.

10/14/08 Dynamic Analysis 100

 Guyan Reduction

  K mm K ms   M mm M ms    D m   0 
  − λ   = 
 KT 
  ms K ss   M Tms M ss    D s   0 

m - master degree of freedom

s - slave degree of freedom

10/14/08 Dynamic Analysis 101

 Guyan Reduction
m - master degree of freedom
s - slave degree of freedom

Basic Assumption:
For lowest frequency modes the inertial
forces on slave d-o-f are less important
than elastic forces transmitted by master
d-o-f. Thus we ignore all mass except Mmm

10/14/08 Dynamic Analysis 102

 Guyan Reduction

  K mm K ms   M mm 0    D m   0 
  − λ   = 
 KT 
  ms K ss   0 0    D s   0 
{ D } = −[ K ] [ K ] { D }
s ss
−1
ms
T
m

10/14/08 Dynamic Analysis 103

 Guyan Reduction

 D m 
  = [ T]{ Dm }
 D s 
 I 
[ T ] =  −1 T 
 − K ss K ms 

10/14/08 Dynamic Analysis 104

 Guyan Reduction

( [ K r ] − λ [ M r ] ) { D m } = { 0}
[ K r ] = [ T] [ K ] [ T]
T

[ Mr ] = [ T] [ M] [ T]
T

Both [Kr] and [Mr] are generally full.

[Mr] contains both mass and stiffness terms
10/14/08 Dynamic Analysis 105
 Guyan Reduction

[ Cr ] = [ T] [ C] [ T]
T

{R r } = [ T ] {R }
ext T ext

[ M r ]{D m } + [ Cr ]{D m } + [ K r ]{ D m } = {R r }
  ext

10/14/08 Dynamic Analysis 106

 Compute Slaves

{D }
s i
= − [ K ss − λ i M ss ]
−1
[K T
ms − λiM T
ms ]{ D }
m i

10/14/08 Dynamic Analysis 107

 Choosing Master D-O-F
 Choose d-o-f where inertia is most
important
 These are characterized by large mass
to stiffness ratios.
 Each d-o-f that has a time varying
applied load should be chosen.
 Master d-o-f should not be clustered.
 Process can be automated

10/14/08 Dynamic Analysis 108

Process for Choosing Master D-O-
F

 Scan diagonal coefficients of [K] and

[M].
 Choose d-o-f for which Kii/Mii is largest.
 This becomes first slave.
 Condense [K] and [M] by one order.
 Repeat process using condensed
matrices till a user specified number of
d-o-f remain.
 These are Master d-o-f chosen in near
optimal way. Dynamic Analysis
10/14/08 109
 Number of Master D-O-F
 Choose cut-off frequency ωc
 Take this to be about three times the
highest frequency of interest.
 Terminate selection of master d-o-f when
Kii/Mii < ωc2
 Can combine manual and automatic
selection (i.e. Choose each d-o-f that has a
time varying applied load and then
automatically select others.)
 Number of Master d-o-f may be 10% - 20%
of total d-o-f.
10/14/08 Dynamic Analysis 110