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 m
10/14/08
F(t)
=
Dynamic Analysis
m
ma = mx
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 = m Then : 2
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.
xm
10/14/08
τ
Dynamic Analysis
8
One Dimensional Bar Element
10/14/08
Dynamic Analysis
9
Step 1 - Select Element Type dˆ1x fˆ1ex ( t )
1
2
dˆ 2 x
xˆ
fˆ2ex ( 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 uˆ = a1 + a 2 xˆ uˆ = N 1dˆ1x + N 2dˆ 2 x xˆ N1 = 1 − L xˆ N2 = L 10/14/08
Dynamic Analysis
11
Step 3 - Define Strain/Displacement and Stress/Strain Relationships
{}
ˆ ∂ u {εx } = ˆ = [ B] dˆ ∂x [ B] = − 1 1 L L ˆ d dˆ = 1x ˆ d 2x
{}
{σ} = [ D]{εx } = [ D][ B] {dˆ} 10/14/08
Dynamic Analysis
12
Step 4 - Derive Element Stiffness and Mass Matrices and Equations With time dependent loading
ˆf ≠ fˆ 1x 2x 10/14/08
Dynamic Analysis
13
Newton’s Second Law
f = ma
10/14/08
Dynamic Analysis
14
NODAL EQUILIBRIUM EQUATIONS
ˆ ∂ d 1x ˆf e = fˆ + m 1x 1x 1 2 ∂t 2ˆ ∂ d e 2x ˆf = fˆ + m 2x 2x 2 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 dˆ1x fˆ1ex ( t )
1
m1
2
m2
xˆ
dˆ 2 x fˆ2ex ( t )
L
10/14/08
Dynamic Analysis
17
Equilibrium in Matrix Form fˆ1ex fˆ1x m1 ˆ e = ˆ + f 2 x f 2 x 0
10/14/08
Dynamic Analysis
∂ dˆ1x 2 0 ∂ t ˆ m 2 ∂ d 2x 2 ∂ t
18
Equilibrium in Matrix Form
{
} [ ]{ }
{}
ˆf ( t ) = kˆ dˆ + [ m ˆ ] dˆ e
10/14/08
Dynamic Analysis
19
Defining Terms 1 − 1 AE [k ] = L − 1 1 ρAL 1 0 [m] = 2 0 1 dˆ 2 ˆ ˆ ∂ d d = ∂ t2
{}
{}
10/14/08
{}
Element Stiffness Matrix Element Lumped Mass Matrix Nodal Displacements Nodal Accelerations
Dynamic Analysis
20
Consistent Mass Matrix
{}
{ }
X = −ρ uˆ e
{fb } = ∫∫∫[ N] { X} dV T
V
{fb } = −∫∫∫ ρ[ N ]
T
{}
uˆ dV
V
10/14/08
Dynamic Analysis
21
Consistent Mass Matrix
{}
ˆ { uˆ} = [ N ] d uˆ = [ N ] dˆ uˆ = [ N ] dˆ
{} {}
10/14/08
{} {}
Dynamic Analysis
22
Consistent Mass Matrix
{}
ˆ {fb } = −∫∫∫ ρ[ N] [ N] d dV T
V
{}
ˆ {fb } = −[mˆ] d
[mˆ] = ∫∫∫ ρ[ N] [ N] dV T
V
10/14/08
Dynamic Analysis
23
Consistent Mass Matrix Bar Element [mˆ] = ∫∫∫ ρ[ N] [ N ] dV T
V
1 − xˆ L [mˆ] = ∫∫∫ ρ xˆ 1 − xˆ V L L xˆ L 1 − ˆ x L ˆ [m] = ρ A ∫ xˆ 1 − 0 L L 10/14/08
Dynamic Analysis
xˆ dV L xˆ dxˆ L 24
Consistent Mass Matrix Bar Element ˆ ˆ x x L 1 − 1 − L L [mˆ] = ρ A ∫ ˆ ˆ x x 0 1 − L L ρ A L 2 [mˆ] = 6 1 10/14/08
Dynamic Analysis
1 − xˆ xˆ L L dxˆ ˆ ˆ x x L L 1 2 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
V
10/14/08
Dynamic Analysis
28
Beam Element yˆ , vˆ ˆ ,m φ 1 ˆ1
1
xˆ
fˆ1y , dˆ 1y
10/14/08
2 φˆ 2 , m ˆ2 L
Dynamic Analysis
fˆ2y , dˆ 2y
29
Shape Functions
( 2xˆ 3 − 3xˆ 2 L + L3 ) L3 1 3 ( N2 = xˆ L − 2xˆ 2 L2 + xˆL3 ) L3 1 ( N3 = − 2xˆ 3 + 3xˆ 2 L ) L3 1 3 ( N4 = xˆ L − xˆ 2 L2 ) 3 L N1 =
10/14/08
1
Dynamic Analysis
30
Shape Functions 1.000
N1 0.500
N3
N2 L
0.000
N4
0 -0.500 10/14/08
Dynamic Analysis
31
Consistent Mass Matrix
[ m] = ∫∫∫ ρ[ N] [ N] dV T
V
22L 156 22L 2 4 L m [m] = 420 54 13L 2 − 13L − 3L 10/14/08
Dynamic Analysis
− 13L 2 13L − 3L 156 − 22L 2 − 22L 4L 54
32
Lumped Mass Matrix 1 0 m [ m ] = 0 2 0 10/14/08
0 2 αL 210 0
1
0
0
Dynamic Analysis
0 0
0 0 0 2 αL 210 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
m L 2 2 I= 3 10/14/08
Dynamic Analysis
2
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 ρAt 1 [ m] = 12 0 1 0 10/14/08
0 2 0 1 0 1
1 0 2 0 1 0
Dynamic Analysis
0 1 0 2 0 1
1 0 1 0 2 0
0 1 0 1 0 2 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 ρAt 0 [m] = 3 0 0 0 10/14/08
0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1
Dynamic Analysis
38
Consistent Mass Matrix Quad Q 0 [m] = 0 Q 4 2 2 4 [ Q] = m 36 1 2 2 1
1 2 4 2
2 1 2 4
m = ρAt
10/14/08
Dynamic Analysis
39
Consistent Mass Matrix Quad 4 0 2 m 0 [ m] = 36 1 0 2 0 10/14/08
0 2 0 1 0 2 0 4 0 2 0 1 0 2 0 4 0 2 0 1 0 2 0 4 0 2 0 1 0 2 0 4 0 2 0 1 0 2 0 4 0 2 0 1 0 2 0 4 0 2 0 1 0 2 0 4 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
10/14/08
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 Dynamic Analysis
42
HRZ - Bar Element ρ A L 2 1 [ mˆ] = 6 1 2 m = ρAL ρAL s = 4× 6 m 3 = s 2 ρ A L 3 0 [ mˆ] = 6 0 3 10/14/08
Dynamic Analysis
43
HRZ - Beam Element 22L 54 − 13L 156 22L 2 2 4 L 13 L − 3 L [m] = m 420 54 13L 156 − 22L 2 2 − 13L − 3L − 22L 4L m = ρAL ρAL s = 312 × 420 m = 420 s 312 10/14/08
Dynamic Analysis
44
HRZ - Beam Element 420 312 × 156 0 m [ mˆ ] = 420 0 0
10/14/08
0
0
420 × 4 L2 312
0
0
420 × 156 312
0
0
Dynamic Analysis
39 2 0 L m = 78 39 0 2 L 420 2 × 4L 312 0
45
HRZ – Quadratic Serendipity 1 36
3 76
8 36
16 76 3x3 Gauss Rule
10/14/08
2x2 Gauss Rule
Dynamic Analysis
46
HRZ – Quadratic Lagrangian 1 36
1 36 4 36
16 36
16 36
3x3 Gauss Rule
10/14/08
4 36 2x2 Gauss Rule
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 m n 1 1 2 1 2 2 3 1 3 2 3 3 4 2
10/14/08
Type of Mass Matrix Consistent (%) HRz Lumping (%) Ad Hoc Lumping (%) 0.11 0.32 0.32 0.4 0.45 0.45 0.35 2.75 4.12 5.18 0.05 5.75 4.68 2.96 10.15 13.78 5.18 19.42 16.88 1.53 31.7
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) 1 4 b+a 1 ∫a f (x)dx = ( b − a ) 6 f ( a ) + 6 f 2 + 6 f ( b ) 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 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
m ij = ρA
1 0 0 ρAL [ m] = 0 1 0 6 0 0 4 10/14/08
1
Dynamic Analysis
3
2
52
Serendipity 1 − 12 1 3
10/14/08
Dynamic Analysis
53
Lagrangian 1 36
4 9
10/14/08
Dynamic Analysis
1 9
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 2. 3. 4. 5.
frequencies less than exact values. Simpler to form. Occupy less storage. Require less computational effort. 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 3. 4. 5. 6.
and slip in connections. Mechanisms not well understood. Awkward to incorporate into structural dynamic equations. Makes equations computationally difficult. 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 ➲ ➲ 10/14/08
Introduce Viscous Damping Relies on Fraction of Critical Damping 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% 2% ≤ ξ ≤ 15% 2% ≤ ξ ≤ 15%
Steel Piping Bolted or riveted steel structures 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: β 1 ξ = αω− 2 ω ( ξ 2 ω 2 − ξ 1 ω1 ) α=2 2 2 ω 2 − ω1
(
β = 2 ω1 ω 2 10/14/08
)
( ξ 1 ω 2 − ξ 2 ω1 )
(ω
2 2
Dynamic Analysis
−ω
2 1
)
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 = 2π 10/14/08
( Hz ) Dynamic Analysis
66
Results in generalized eigenproblem
( [ K ] − λ [ M ] ) { D } = { 0} λ=ω
10/14/08
2
Dynamic Analysis
67
Trivial Solution:
[ K ] − λ[ M]
{ D} = 0
10/14/08
Dynamic Analysis
≠0
68
Nontrivial Solution:
[ K ] − λ[ M ]
{ D} ≠ 0
10/14/08
=0
Dynamic Analysis
69
λi
{ D} 10/14/08
≡ Roots of Characteristic Polynomial (eigenvalues)
i
≡ Associated
Dynamic Analysis
Eigenvectors
70
ω i Natural Frequencies
{D}
10/14/08
i
Normal Modes
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
{ D} [ K ]{ D} λ= { D} [ M ]{ D} T
T
{ D}
[ K ] symmetric [ M ] positive definite th
approximat ion to i eigenvector
λ approximat ion to i th eigenvalue 10/14/08
Dynamic Analysis
73
Rayleigh Quotient
λ min
{ v} [ K ]{ v} ≤ ≤ λ max T { v} [ M ]{ v}
λ min
smallest eigenvalue
λ max
l arg est eigenvalue
T
10/14/08
Dynamic Analysis
74
Modal Methods When [K], [C], [M] are known and time independent the problem is linear.
{
ext [ M ] { D} + [ C ] { D} + [ K ] { D} = R
{ D ( 0 ) } , { D ( 0 ) } 10/14/08
}
given as initial conditions Dynamic Analysis
75
Modal Methods Assume orthogonal damping, such as Rayleigh Damping. Modes can be uncoupled:
{ D} [ M ] { D} { D} [ K ] { D} { D} [ C] { D} T i
j
=0
T i
j
=0
T i
j
=0
i≠j 10/14/08
Dynamic Analysis
76
{ D} [ M ] { D} = 1 { D} [ K ] { D} = ω { D} [ C] { D} = 2ξ ω T i
i
T i
T i
10/14/08
2 i
i
i
Dynamic Analysis
i
i
77
Eigenvectors are linearly independent
[ φ] = matrix of
eigenvectors (mode shapes)
{ Z} 10/14/08
{ D} = [ φ]{ Z}
modal amplitudes Dynamic Analysis
78
Substitute into:
{
ext [ M ] { D } + [ C ] { D } + [ K ] { D} = R { D ( 0 ) } , { D ( 0 ) } given
10/14/08
Dynamic Analysis
}
79
{
ext [ M ] [ φ] { Z} + [ C ] [ φ] { Z} + [ K ] [ φ] { Z} = R [ φ ] { Z ( 0 ) } = { D ( 0 ) }
[ φ ] { Z ( 0 )} = { D ( 0 ) }
10/14/08
Dynamic Analysis
given
80
}
Mode Displacement Method
[ ]
2 [ I ] { Z } + [ ξ ] { Z } + ω { Z} = { p}
10/14/08
Dynamic Analysis
81
Mode Displacement Method Pre-multiply by [φ]T
[ φ]
T
T [ M ] [ φ] { Z } + [ φ] [ C ] [ φ] { Z }
[ K ] [ φ] { Z} = [ φ] { R } [ φ] { Z ( 0 ) } = { D ( 0 ) } [ φ] { Z ( 0 ) } = { D ( 0 ) } given
+ [ φ]
10/14/08
T
T
Dynamic Analysis
ext
82
Mode Displacement Method
[ φ] [ M ] [ φ] = [ I ] T [ φ] [ C] [ φ] = [ ξ ] T 2 [ φ] [ K ] [ φ] = [ ω ] T
10/14/08
Dynamic Analysis
83
Mode Displacement Method
[ ]
2 [ I ] { Z } + [ ξ ] { Z } + ω { Z} = { p}
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
10/14/08
i
i
i
i
Dynamic Analysis
i
85
[ φ ] { Z ( 0 ) } = { D ( 0 ) } [ φ ] { Z ( 0 )} = { D ( 0 ) } T T [ φ ] [ M ] [ φ ] { Z ( 0 ) } = [ φ ] [ M ] { D ( 0 ) } T [ I ] { Z ( 0 ) } = [ φ ] [ M ] { D ( 0 ) } T { Z ( 0 ) } = [ φ ] [ M ] { D ( 0 ) } T { Z ( 0 )} = [ φ ] [ M ] { D ( 0 ) } 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} {R } ext
e( t ) =
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
[ ]
2 [ I ] { Z } + [ ξ ] { Z } + ω { Z} = { p}
10/14/08
Dynamic Analysis
90
Modal transformation only on inertial and damping terms
{
ext [ M ] [ φ ] { Z } + [ C ] [ φ ] { Z } + [ K ] { D} = R [ φ ] { Z ( 0 ) } = { D ( 0 ) }
[ φ ] { Z ( 0 )} = { D ( 0 ) }
10/14/08
Dynamic Analysis
given
91
}
} − [C][ φ]{Z } [K ] {D} = {R ext } − [M ][φ]{Z } − [C][ φ]{Z }) {D} = [K ] {R } − [K ] ([M ][φ]{Z −1
10/14/08
ext
−1
Dynamic Analysis
92
[ φ] [ M ] [ φ] = [ I ] −T [ M ] [ φ] = [ φ] T
10/14/08
Dynamic Analysis
93
{ D} = [ K ] { R } − [ K ] −1
10/14/08
ext
−1
( [ φ ] { Z } − [ C ] [ φ ] { Z } )
Dynamic Analysis
−T
94
[ φ] [ K ] [ φ] = [ ω ] T 2 −1 [ φ] [ K ] [ φ] [ ω ] = [ I ] −T −T 2 −1 [ K ] [ φ] [ ω ] = [ φ] [ I ] = [ φ] −T 2 −1 −1 [ φ] [ ω ] = [ K ] [ φ] T
10/14/08
2
Dynamic Analysis
95
{ D} = [ K ] {R } − [ K ] [ φ] −1 ext 2 −1 −1 { D} = [ K ] {R } − [ φ] [ω ] {Z} + [ K ] [ C][ φ]{Z } −1
10/14/08
ext
−1
−T
−1 {Z} + [ K ] [C][ φ]{Z }
Dynamic Analysis
96
{ D} = [ K ] {R } − [ K ] [ φ] {Z } + [ K ] −1[ C][ φ]{Z } −T T −1 ext 2 −1 −1 { D} = [ K ] {R } − [ φ] [ω ] {Z} + [ K ] [ φ] [ φ] [ C][ φ]{Z } −1 ext 2 −1 2 −1 { D} = [ K ] {R } − [ φ] [ω ] {Z} + [ φ] [ω ] [ ξ]{Z } −1
10/14/08
ext
−1
−T
Dynamic Analysis
97
{ D} = [ K ] { R } − [ φ ] [ ω −1
ext
{ D} = [ K ] { R } −1
ext
] ( { Z } + [ ξ ] { Z } )
2 −1
1 2ξ i − ∑ { φ} i 2 Z i + Z i ωi i =1 ωi m
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 i i i i i
10/14/08
Dynamic Analysis
i = 1, n
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 KT ms
K ms M mm − λ K ss M Tms
M ms D m 0 = 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 KT ms
0 D m 0 = 0 D s 0
K ms M mm − λ K ss 0
{ D } = −[ K ] [ K ] { D } −1
s
10/14/08
ss
T
ms
Dynamic Analysis
m
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} T [ K r ] = [ T] [ K ] [ T] T [ Mr ] = [ T] [ M] [ 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 ext ext {R r } = [ T ] {R } ext [ M r ]{D m } + [ Cr ]{D m } + [ K r ]{ D m } = {R r } T
10/14/08
Dynamic Analysis
106
Compute Slaves
{D } s
i
= − [ K ss − λ i M ss ]
10/14/08
−1
[K
Dynamic Analysis
T ms
− λiM
T ms
]{ D } m
107
i
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-OF 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 10/14/08 109 optimal way. Dynamic Analysis
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