Professional Documents
Culture Documents
2 CompA Discrete Systems
2 CompA Discrete Systems
m2
k2 ky2 ky4
k y2
kx1 kx 2 kx1 kx2 kx3
m1 m
m m1 m2
1. Two-body Systems
• System of equations: time and frequency domain
• Direct solution
• Modal analysis and modal transform
2. Multibody Systems
• System of equations and direct solution
• Modal transform
• Modal reduction
F2 ( t ) U2 ( t )
m2
Time frequency ω F1 ( t ) U1 ( t )
m1
domain
Frequency k1
domain
Example:
Direct
solution
HA ggph
www.audi.co.uk
Undamped: ɺɺ = F
KU + MU
Two DOF
Time
Damped: ɺ + MU
KU + CU ɺɺ = F
domain
Frequency
domain Time-harmonic:
F ( t ) = 3cos 2t
m2
m1 = 9kg, m2 = 1kg
k2 r2
k1 = 24Nm −1 , k 2 = 3Nm −1
Two DOF
m1
Time
domain k1 r1 Rayleigh Damping: α = 0, β = 0.1
Frequency
domain
27 −3
Direct
9 0 0
K= , M= , C = β K, F =
solution
− ω
Modal
Transform
3 3 0 1 cos t
Excitation „in a
System of equations in time domain: frequency band“
ɺ + MU
KU + CU ɺɺ = F
ɺ + MU
KU + CU ɺɺ = F
Two DOF
K + iωC − ω M u = f
Modal
Transform 2
K + iωC − ω 2 M u = f
Two DOF
Time
domain
Frequency
−1
u = K + iωC − ω M f
domain 2
Direct
solution
Modal
Transform Dynamical stiffness matrix
0
clc, close all, clear all
f =
k1=24;k2=3;
m1=9;m2=1;
Two DOF beta=0.1;
% stiffness, mass matrix
1
Time k = [[k1+k2, -k2] ; [-k2, k2]];
m = diag([m1,m2]); Excitation at mass #2
domain
% damping matrix, force vector
Frequency c = beta*k;
domain f=[0;1]
% direct solution for damped response
Direct omega=0.2:0.02:10; au=[];
solution for o=omega
k_d=-o^2*m+(1+i*o*beta)*k; % dynamic stiffness
Modal u=inv(k_d)*f;
Transform au(end+1)=norm(abs(u));
end
% graphics
lw=3; fs=22;
plot(omega, au, 'linewidth', lw)
1.5
|u(ω)|
1
|u(
0.5
0
0 1 2 3 4 5 6 7 8 9 10
Driving Frequency ω [rad/s]
ω1 ω 2
HAW/M+P, Ihlenburg, CompA Vibrations: Discrete Systems 9
Frequency response with different damping values
2
|u(ω)|
1.5
|u(
0.5
0
0 1 2 3 4 5 6 7 8 9 10
Driving Frequency ω [rad/s]
K − ω 2M u = f
Two DOF
Direct det K − ω 2M ≠ 0
solution
Modal
Transform
Kx = λ Mx 0 ≤ λ1 ≤ λ2
λ1 = ω12 , λ2 = ω22
Angular eigenfrequencies.
x11 x21
X MX=I
Two DOF
X = x1 x 2 =
T
x22
with
Time
domain
x12
Frequency
ω 2
0
X KX= =
domain T 1 2
2
ω
Direct
solution 0 ω2
Modal transform of
Modal steiffness matrix.
Transform Diagonal = uncoupled!
Assume: u = Xq XT K − ω 2 M Xq = XT f
Two DOF
(ω 2
)
− ω 2 q1 = x11 f1 + x12 f 2
ω 2 − ω 2 I q = XT f
Time 1
(ω )q
domain 2
2 −ω2 2 = x21 f1 + x22 f 2
Frequency ^
domain Diagonal matrix! f
Direct
solution
Modal Solution: Response is computed in two steps:
Modal −1
1. q = ω − ω I f ,
^
Transform
2 2
2. u = Xq
(ω )y
domain
( 1 ) 1 f1
^
Direct
2
1 −ω 2
1 = f1 ω 2
+ 2i ωζ ω
1 1 − ω 2
q
^
=
(ω )y
^
( 2 ) q2 = f 2
solution
2
2 − ω2 2 = f2 ω 2
+ 2i ωζ ω
2 2 − ω 2 ^
Modal
Transform
Modal Damping
= adding a viscous damping term to each modal equation
= assigning a damping ratio to each mode.
ii ii
KU + MU = F
i
KU + CU + MU = F
C = αM + β K
Two DOF
with
Time
domain
Frequency ω 2
0 1 0
domain X KX =
T 1
2
, X MX =
T
0 ω2 0 1
Direct
solution
Modal
Transform
α 0 βω 2
0 α + βω12 0
X T CX = +
0
1
2
=
0 α βω2 0 α + βω22
Ku + Cuɺ + Mu
ɺɺ = f K + iωC − ω 2 M u = f
Two DOF
Time u = Xq
domain
Frequency
domain
ω12 0 1 0
X KX = ω =
T 2
2
, X MX = I =
T
Direct
solution 0 ω 2 0 1
Modal
Transform
ω 2 + iω XT CX − ω 2 I q = XT f
1 1 −1
Two DOF 3 x1 3 X=
18 3 3
x2
Time
domain 1 −1
2 0 1 0
Frequency XT KX = , X T
MX = 0 1
domain 0 4
ω1 = 2 ω2 = 2
Direct
solution
2 0
Modal Rayleigh Damping: X CX = β
T
Transform 0 4
−ω q + (1 + iωβ ) ω q = X f
2 2 T
−ω q + (1 + iωβ ) ω q = X f
2 2 T
Two DOF
Time
domain
Frequency
domain fɶi
qi (ω ) = 2 Solution in „modal space“
Direct ω i − ω 2 + iωβω i 2
solution
Modal
Transform Cf. Solution for simple oscillator
u = Xq
Back transform
2
|u(ω)|
1.5
|u(
0.5
0
0 1 2 3 4 5 6 7 8 9 10
Driving Frequency ω [rad/s]
-0.5
-1
0 1 2 3 4 5 6 7 8
Time t [s]
4
2
Re(u (t))
2
-2
-4
0 1 2 3 4 5 6 7 8
HAW/M+P, Ihlenburg, CompA Time tDiscrete
Vibrations: [s] Systems 23
Forced Damped Vibrations of 2-Mass-System
3
direct
modal
2.5
Amplitude at ω=2: |u|=2.23
2
|u(ω)|
1.5
0
0 1 2 3 4 5 6 7 8 9 10
Driving Frequency ω [rad/s]
0.02
Re(u (t))
1
-0.02
-0.04
0 1 2 3 4 5 6 7 8
Time t [s]
1
0.5
Re(u (t))
2
-0.5
-1
0 1 2 3 4 5 6 7 8
Time t [s]
T
u Ku
R ( u) = T
Potential Energy
Definition:
uɺ Muɺ Kinetic Energy
Properties:
xiT Kxi
1. R ( xi ) = ωi
2
ωi = T
xi Mxi
Ku + Cuɺ + Mu
ɺɺ = f
System of
equations (1)
are then n or n x n, resp. Modal analysis of the undamped system
Modal
analysis
yields n real eigenvalues. The corresponding n eigenvectors can be
written as a matrix X of size n x n.
Modal
Reduction Using the matrix X, the system then can be transformed into n
decoupled equations, representing n independent simple oscillators,
with modal amplitudes qi(t) . The back-transform u=Xq yields the
physical solution in the time domain.
det K − ω 2M = 0 4-DOF
Multibody
System of
n×n
with symmetric, positive-definite n x n Matrices has exactly n non-negative
eigenvalues
equations
ω , ω2 , … , ωn ≥ 0
1
Modal
analysis and corresponding real-valued n × 1 eigenvectors (modes) x1 , x 2 ,… , x n .
Modal The modal matrix
Reduction
X = x1 x 2 … x n
is an orthogonal n × n matrix which can be scaled such that it is mass-
orthogonal,
X T M X = X M X T = I n ×n
It follows that X KX = ω
T 2
ω1
ω
ω= 1 : Matrix of eigenfrequencies
⋱
ωn
• Assuming time-harmonic excitation in the form F = feiωt the response can be written as
u1
U = ⋮ eiωt = ueiωt
un
and the time-derivatives are replaced by the factor ( iω ) „in the frequency range“.
K + iωC − ω 2 M u = f (**)
• The matrix on the left of (**) is called the dynamic stiffness matrix. It depends on the
driving frequency ω.
• The linear n × n system (**) can be solved (separately for each ω of interest) using
either a direct approach (SOL108) or the method of modal superposition (SOL111).
• The modal equations are diagonal n × 1 if modal or proportional damping is assumed.
Single spring:
c c −c
fa fb Kc =
− c c
ua ub
Assembly
c1 c2 c1 −c1 0
K c1 ,c2 = −c1 c1 + c2 −c2
0 −c2 c2
ub1= u a2
Equal displacements!
HAW/M+P, Ihlenburg, CompA Vibrations: Discrete Systems 33
Systematischer Aufbau der Steifigkeitsmatrix (2/2)
Constraint
c1 c2 c1 −c1 0
K c1 ,c2 = −c1 c1 + c2 −c2
0 −c2 c2
la =
1.0e+004 *
0.1471 0 0 0 0 0 0 0 0 0
0 0.7020 0 0 0 0 0 0 0 0
0 0 1.0000 0 0 0 0 0 0 0
0 0 0 1.2857 0 0 0 0 0 0
0 0 0 0 1.3081 0 0 0 0 0
0 0 0 0 0 2.0403 0 0 0 0
0 0 0 0 0 0 2.2500 0 0 0
0 0 0 0 0 0 0 3.7618 0 0
0 0 0 0 0 0 0 0 4.5000 0
0 0 0 0 0 0 0 0 0 4.5499
xx =
-2
10 Eigenfrequencies
-3
10
-4
10
-5
10
-6
10
0 50 100 150 200 250 300 350 400
Driving Frequency [rad/s]
-4
10
-5
10
-6
10
0 50 100 150 200 250 300 350 400
-2
Eigenfrequencies
10
-3
10
-4
10
-5
10
0 50 100 150 200 250 300 350 400
Driving Frequency [rad/s]
In many practical applications one wants to determine the response of a system within a
certain frequency band. It is then sufficient to compute the eigenvalues and –vectors only up
to the maximal frequency of this band (plus some margin value for reliability). Usually this
yields a limited number of N<<n eigenvectors which are assembled into a reduced modal
matrix.
Using the reduced modal matrix, the system is transformed into a reduced modal space. This
method is also called modal projection. The solution in the modal space and the back
transform are performed in the same way as for modal transformation. In fact, modal
reduction and modal transformation are equivalent if N=n.
Projection: ℝ3 → ℝ 2
Ku + Cuɺ + Mu
ɺɺ = f dim K = n
Step 1: Determine (numerically and approximately) N << n = dim K,M eigenvalues and –
vectors of the undamped eigenproblem
K − ω 2 M x = 0
and build a rectangular matrix X=XN.
Step 2: Assume
X = x1 ⋯ x N n≫N u ≈ u N = Xq
with N „modal coordinates“
N q1 ,⋯ qN
ω12 1
⋱
XT KX = ω 2 = ⋱ , X T
MX = I =
ω N N × N
2
1 N × N
K + iωC − ω 2 M u = f
n× n
ω 2 + iω XT CX − ω 2 I q = XT f
N ×N
The modal system decouples for suitable damping assumptions. The principal advantage
of the method is the dimensional reduction to N<<n equations!
−1
q = ω + iω X CX − ω I X f
2
T T 2
u N = Xq
As a rule, the approximate solution agrees well with the exact solution in the frequency
band of interest.
Rule of thumb: perform the modal analysis up to 1.2..1.5fmax (highest driving frequency).
-2
10
-3
10
-4
10
-5
10
0 50 100 150 200 250 300 350 400
Driving Frequency [rad/s]
-2
10
-3
10
-4
10
-5
10
0 50 100 150 200 250 300 350 400
Driving Frequency [rad/s]
-2
10
-3
10
-4
10
-5
10
0 50 100 150 200 250 300 350 400
Driving Frequency [rad/s]
-2
10
-3
10
-4
10
-5
10
0 50 100 150 200 250 300 350 400
Driving Frequency [rad/s]
-2
10
-3
10
-4
10
-5
10
0 50 100 150 200 250 300 350 400
Driving Frequency [rad/s]
direct modal
-4
10
-5
10
-6
10
100 200 300 400 500 600 700 800 900 1000
Frequency [Hz]
m2
a) Berechne die Eigenfrequenzen
und Eigenmoden des skizzierten c2
m1 = 9kg, m2 = 1kg
Schwingsystems.
m1 c1 = 24Nm −1 , c2 = 3Nm −1
b) Stelle die Schwingung im
Zeitbereich für verschiedene
c1
Anfangsbedingungen dar.
X 2 (t ) ii
m2 m2
↑: m2 X 2 = −c2 ( X 2 − X 1 )
c2 Fk 2 = c2 ( X 2 − X1 )
X1 (t ) ii
m1 m1 ↑: m1 X 1 = c2 ( X 2 − X 1 ) − c1 X 1
c1 Fk 1 = c1 X 1
ii
↑: m2 X 2 = −c2 ( X 2 − X 1 )
ii
↑: m1 X 1 = c2 ( X 2 − X 1 ) − c1 X 1
ii
c1 + c2 −c2 X1 m1 0 X1 0
−c + ii
=
2 c2 X 2 0 m2 X 2 0
Steifigkeitsmatrix Massenmatrix
ɺɺ + Kx = 0
Mx
Das gekoppelte System kann im Zeitbereich als Matrizen-Differentialgleichung geschrieben
werden. Zur Lösung sind Anfangsbedingungen für die Unbekannten erforderlich.
{K − ω M}[x] = 0 λ = ω 2
2 Eigenwert
(1)
x Eigenvektor
{ K − ω M [ x] = 0 } ω Wurzel aus
2 (1) Eigenwert
x Eigenvektor
Steifigkeitsmatrix Massenmatrix
det [ K − λ M ] = 0 (2)
Steifigkeits- und Massenmatrix sind symmetrisch! Man kann zeigen dass alle
Eigenwerte reell und nichtnegativ sind.
Daher sind auch die Wurzeln (Eigenkreisfrequenzen rad/s) reell. λ = ω ,ω ∈ℝ
2
A = M −1K det [ A − λ I ] = 0
Die Berechnung der Determinante führt auf ein Polynom n. Grades (bei n
Freiheitsgraden) für die Unbekannte λ. Die Lösung dieses Gleichungssystems ist
analytisch nur bis zu fünf Freiheitsgraden möglich. Im allgemeinen werden
numerische Verfahren (z.B. Lancsosz Verfahren) verwendet.
det K − ω 2 M = 0
Dieses Eigenwertproblem hat zwei nichtnegative reelle Lösungen (Eigenwerte)
ω1 , ω2 ≥ 0
Die Eigenwerte sind die Kreisfrequenzen der Eigenschwingungen. Das Einsetzen der Eigenwerte
in (1) führt auf zwei Matrix-Vektorgleichungen zur Bestimmung der Eigenvektoren
K − ω M x1 = 0
2
1 K − ω22M x 2 = 0
Die Beträge der Eigenvektoren sind beliebig. Es ist üblich, die Vektoren so zu normieren dass gilt:
x Mx1 = 1, x Mx 2 = 1.
T
1
T
2
Die Vektoren heißen masse-normiert.
k1 xx =
0.2357 0.2357
0.7071 -0.7071
la =
2 0
0 4
3 >> xx/xx(1,1)
3
x1 1.0000 -1.0000 x2
3.0000 3.0000
1 −1
x1 , x 2
ω1 = 2 ω2 = 2
HAW/M+P, Ihlenburg, CompA Vibrations: Discrete Systems 58
Interpretation der Lösung (Frequenzbereich)
3 3
x1 x2
1 −1
ω1 = 2 ω2 = 2