Professional Documents
Culture Documents
School of Engineering
USC Viterbi
School of Engineering
Outline
General MDOF
! Forced response
Damping Models Frequency Response and Transfer Functions State-Space formulation Discrete-Time State-Space Distributed Parameter Systems
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
u3 f3 m3 m3 k3/2 k3/2 k3/2 k3/2 u2 f2 m2 m2 k2/2 k2/2 k2/2 k2/2 u1 f1 m1m1 k k /2 /2 k1 k /2 11 1/2
m3 m3 f3 u3 k3 k3 m2 m2 f2 u2 k2 k2 m1m1 f1 u1 k1 k1
f3
u3 u2 u1
m3 k3 m2 k2 m1 k1 f1 f2
! ! f !
3
u1 k1k1 m1 m1 k2 f1
u2 k2 m2 m2 k3
u3 f2 m3 k3 m3 f3
!
k1
or, compactly, + Ku = f Mu
u1 m1 f1 u2 k2 m2 f2 u3 k3 m3 f3
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
v3 v2 v1
m3 k3 m2 k2 m1 k1
vg v3 v2 v1 u3 k3/2 u2 k2/2 u1 k1/2 m1 k1/2 m2 k2/2 m3 k3/2
vg
v3 u3 v2 u2
m3 k3 m2 k2 m1 k1
! or in matrix form ! "m1 0 1 , "k1 + k2 0 %(u /k2 ! $ '* * $ ! 2 - + $ /k2 k2 + k3 $ 0 m2 0 ')u * * $ $ 0 m3 ' 0 /k3 #0 &+u 3. #
u1 vg k1
v1 m1 m1
u2 k2 m2
v2 m2
v3 u3 m3 k3 m3
vg
v1 u1
!
vg
k1
USC Viterbi
School of Engineering
Lagranges Equations
USC Viterbi
School of Engineering
! !
Using FBDs works for simple systems but difficult for complex models. Can use Lagranges Equations. Let u1, u2, , un denote independent generalized degrees of freedom and f1, f2, , fn the corresponding external forces. Let T and V denote the total Kinetic and Potential Energies 1,u 2 ,!,u n ;t ) = total Kinetic Energy T " T (u1,u2 ,!,un ;u V " V (u1,u2 ,!,un ;t ) = total Potential Energy Then, the equations of motion are given by the n equations d # "T & " T "V + = fi , i = 1,!,n % () i ' "ui "ui dt $ "u
If, instead of n independent ui, we have m > n dependent g1,g2,,gm with fj(g1,g2,,gm) = 0, j = 1,,(mn), then m )n "f j d # "T & " T "V + ) + *j = fi , i = 1,!,m % () i ' "gi "gi dt $ "g "gi j =1
See, e.g., Craig & Kurdila, Fundamentals of Structural Dynamics, Wiley, 2006
f3
!
u3 u2 u1 m3 k3 m2 k2 m1 k1 f1 f2
f3
d # "T & " T "V d ] ) [0] + [k1u1 ) k2 (u2 ) u1)] = m1u 1 + (k1 + k2 )u1 ) k2u2 = f1 + = [m u % () 1 ' "u1 "u1 dt 1 1 dt $ "u d # "T & " T "V d ] ) [0] + [k2 (u2 ) u1) ) k3 (u3 ) u2 )] = m2u 2 + (k2 + k3 )u2 ) k2u1 ) k3u3 = f2 + = [m u % () 2 ' "u2 "u2 dt 2 2 dt $ "u d # "T & " T "V d 3 ] ) [0] + [k3 (u3 ) u2 )] = m3u 3 + k3u3 ) k3u2 = f3 + = m3u [ % () 2 ' "u2 "u2 dt dt $ "u
which can be written in matrix form "m1 0 1 , "k1 + k2 0 %(u /k2 $ '* * $ 2 - + $ /k2 k2 + k3 $ 0 m2 0 ')u $0 '* $ 0 3 * 0 m3 & /k3 # +u . #
u1 k1 m1 f1 u2 k2 m2 f2 u3 k3 m3 f3
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
Example 2DOF System Unforced Case natural frequencies & mode shapes
k2!=!2
2
Lets look at a simple 2DOF system to see how to solve. First, consider an unforced (and undamped) system. "m1 0 %(u 1 + "k1 + k2 .k2 %(u1 + (0+ $ ') , + $ ') , = ) , k2 &*u2 - *0# 0 m2 &*u 2 - # .k2
"2 0%(u 1 + " 4 + 2 .2%(u1 + (0+ $ ') , + $ ') , = ) , 2 &*u2 - *0#0 1&*u 2 - # .2 Assume harmonic response u(t) = [u1(t) u2(t)]T = ! cos!t . Substitute in for the ui and multiply by M1: % 12 0(% 4 + 2 "2( "# 2$ cos #t + ' *' *$ cos #t = 0 2) & 0 1)& "2 + % "# 2 %3 # $ 2 0 ( % 3 "1(. #1 ( *+' *+ = 0 *0$ cos #t = 0 " ' -' 0 "# 2 ) & "2 2 )0 2 #$2) & #2 ,& / Either ! is zero (i.e., the structure doesnt move) or the matrix in parentheses must be singular and have a zero determinant. !
3 "#2 "2
"1 = 0 = (3 " # 2 )(2 " # 2 ) " ( "2)( "1) 2 "#2 = # 4 " 5# 2 + 4 = (# 2 " 1)(# 2 " 4)
k2!=!2
2
k1!=!4
k1!=!4
So, the structure can oscillate either with ! = 1 or ! = 2 (or, as well see, both at the same time). It is customary to order the frequencies: !1 = 1 and !2 = 2. If we substitute back !, we find !: $3 # 12 $ 2 #1' ,1/ #1 ' "1 = 1: & )* = & )* = 0 + 2*11 = *21 + *1 = - 0 2 # 12 ( %#2 1 ( .2 1 % #2
$3 # 22 $ #1 #1' , 1/ #1 ' "2 = 2 : & )* = & )* = 0 + *12 = #*22 + *2 = - 0 2 # 22 ( % #2 #2( . #11 % #2 !1 and !2 are natural frequencies; !1 and !2 are called eigenvectors or mode shapes (and can be scaled arbitrarily).
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
While mode shapes !i can be scaled arbitrarily, there are some conventional scaling methods:
! Mass normalized: normalize so that Mi " !iTM!i = 1
#1& T "1 = c $ ' ) "1 M"1 = c [1 %2( #1 6 & *2 0-#1& 2 ' 2], /$ 'c = 6c = 1 ) "1 = $ +0 1.%2( %2 6 ( # # 1& * # & 2 0 1 1 3& T 2 ' "2 = c $ ' ) "2 M"2 = c [1 01], /$ 'c = 3c = 1 ) "2 = $ %01( +0 1.%01( %01 3 (
k2!=!2
2
k2!=!2
2
Define eigenmatrix # = [!1 !2]. Substitute u = #q into the equation of motion and premultiply by #T (we will use the mass-normalized mode shapes here): + " TK"q = 0 " T M"q
#1 6 =% $1 3 2 6 & #2 0 6 (% (% "1 3 ' $0 1'$2 6 1 , #1 6 1 3 &)q (* - + % 2 . $1 3 "1 3 '+q 2 6 &# 6 "2& #1 6 (% (% "1 3 '$ "2 2 ' $2 6 1 3 &)q1 , (* "1 3 '+q2 .
k1!=!4
!
! Normalize so maximum (absolute) value is one; i.e., maxj |!ji| = 1 T max " j 1 = max{c , 2c } = 2c = 1 # "1 = {1 2 1}
j
k1!=!4
"
# " 2 = {1 $1}
!
!
! Can also use #1M1 times the equation of motion and u = #q: 1 + "#1M#1K"q = 0 " q + #$1M$1K#q = q + % "#1M#1M"q '0 & !
#1 2 0& #1 1&)q 1 , #1 2&# 6 "2& #1 1&)q1 , "6 0%(q 1 + "6 0 %(q1 + (0+ % (% (% (* - + % (% (% (* - = $ ') , + $ ') , = ) , 2 ' $2 "1'+q2 . $1 "1'$0 1' $2 "1'+q #0 3&*q 2 . $1 "1' $ "2 2 - #0 12&*q2 - *0 0( *q 4)
=0
SDOF techniques can be used to solve for the qi (possibly using modal initial condition q(0)!=!#-1u(0)); then, u = #q is used to obtain the motion in the original coordinates. !
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
k2!=!2
2
!
i + "i2qi = 0 is Review SDOF: modal free responses of q qi (t ) = ai sin"i t + bi cos "i t = ci sin("i t + #i ) i (0) where the ai & bi, or ci & "i, depend on qi (0) !!!!!!! and q !
For example, with initial conditions: q1 (0) = 1 = a1 sin 0 + b1 cos 0 = c1 sin("1) 1 (0) = 0 = a1#1 cos 0 $ b1#1 sin 0 = ! q c1#1 cos("1)
k1!=!4
q2 (0) = 0 = a2 sin 0 + b2 cos 0 = c 2 sin(" 2 ) 2 (0) = 1 = a2# 2 cos 0 $ b2# 2 sin 0 = c 2# 2 cos(" 2 ) q
then: a1=0, b1=1, c1=1, "1=!/2 and a2=1/2, b2=0, c2=1/2, "2=0. q1 (t ) = cos t q2 (t ) = 0.5 sin 2t
!
!
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
1 c2!=!0.1 k2!=!2 2
c1!=!0.2 k1!=!4 !
Lets add damping. (What is damping???) #c1 + c 2 "c 2 & # 0.3 "0.1& + Cu + Ku = 0 C=% Mu (=% ( c 2 ' $ "0.1 0.1' $ "c 2 $1 1' T Already know " = & ). Let u = "q and premultiply by " %2 #1(
+ " T C + " TK"q = 0 " T M"q "q !
#1 2 0& #1 1&)q 1 , #1 2&# 0.3 "0.1 1&)q 1 , #1 2& # 6 "2 & #1 1&)q1 , % (% (% (* - + % (% (% (* - + % (% (% (* 0.1'$2 "1'+q 2' $2 "1'+q2 . $1 "1'$0 1' $2 "1'+q 2 . $1 "1' $ "0.1 2 . $1 "1' $ "2 #6 0&)q 1 , #0.3 0 &)q 1 , #6 0 &)q1 , )0, * * * * =% + + = ( % ( % ( 0.6'+q $0 3'+q 2. $ 0 2 . $0 12'+q2 . +0 . 1 + 0.05q 1 + 1q1 = 0 q
qi (t ) = ai e "# i $ it sin $idt + bi e "# i $ it cos $idt = ci e "# i $ it sin($idt + %i ) i (0) , and where the ai & bi, or ci & "i, depend on qi (0) !!!!!!! and q 1/ 2 "id = "i [1 # $2 i] .
More generally, the responses in the original coordinates are: ! u(t ) = &e "# i $ it %i ai sin $idt + bi cos$idt = &ci %i e "# i $ it sin($idt + 'i )
i
!
!1
= &e
i
"# i $ i t
ai sin $idt
+ bi cos $idt
"
2 + 0.20q 2 + 4q2 = 0 q !
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
+ Cu + Ku = Ef Equation of motion: Mu
First use the undamped, unforced equation to find natural frequencies and mode shapes: (M" 2 #! K )$ = 0 or (" 2I # M#1K )$ = 0 The mode shapes have the properties that %Mi , i = j %Ki , i = j "iT M"j = Mi #ij = & "iTK"j = Ki #ij = & ' 0, i $ j ' 0, i $ j The fact that !iTM!j = 0 and !iTK!j = 0 for i ! j means the mode shapes are orthogonal. Note also that Ki /Mi = !i2.
!
Forced response of a linear system is the sum of the free response from any non-zero initial conditions and the effects of the external forcing. Let us assume the initial conditions are zero and consider a harmonic excitation. + Cu + Ku = Ef = Ef0 sin "t Mu Then the response is also harmonic. This is most easily demonstrated using a SDOF example
Note: some systems have K) be less than rank (n 1), which implies there are multiple modes at one or more frequencies. Note: some systems (e.g., satellites) may have ! = 0 be a solution; this implies one or more rigid body modes in which the whole structure moves or rotates, possibly without internal deformation.
(M!2
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
General MDOF Systems Forced Harmonic Response SDOF example with ej#t
(0) = 0 u (0) = 0, u + cu + ku = sin "t EOM: mu Assume: u (t ) = U sin("t + # ) Then: "m#2U sin(#t + $ ) + c#U cos(#t + $ ) + kU sin(#t + $ ) = sin #t ! Expanding sin and cos! terms and simplifying: ! [ "m#2 cos$ " c# sin $ + k cos$ ]U sin #t + [ "m#2 sin $ + c# cos $ + k sin $ ]U cos #t = sin #t !
This is easier if we use a complex-valued excitation: + cu + ku = e j"t (0) = 0 mu u (0) = 0, u Assume: u (t ) = Ue j"t Then: [m( j")2 + c ( j") + k ]Ue j"t = e j"t ! So: U = 1 [(k " m#2 ! ) + cj#] ! ! To get back the original real-valued u(t), recognize that j #t ! e = cos#t + j sin#t. Then a real valued response to excitation sin#t is just Im[Uej#t] or:
$ ' 1 u (t ) = Im& e j#t ) 2 % [(k " m# ) + cj#] ( $ ' $ ' 1 1 = Re& ) sin #t + Im& ) cos #t 2 2 % [(k " m# ) + cj#] ( % [(k " m# ) + cj#] (
! ! ! !
which means [ "m#2 cos$ " c# sin $ + k cos$ ]U = 1 [ "m#2 sin $ + c# cos $ + k sin $ ]U = 0
"
!
!
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
The response to an impulse can be used to find the p(t) t <0 response to arbitrary excitations. $ 0, & + Cu + Ku = ep (t ) p (t ) = %1/", 0 # t # " Mu 1/$% & t >" ' 0,
$%
[(K " #2M) + j#C ]Ue j#t = EFe j#t U = [(K ! " #2M) + j#C ]"1EF
Integrate both sides with respect to t from 0 to $ assuming the system is at rest at time 0. ! (" ) # u (0)] + C[u(" ) # u(0)] + K[uavg" ] = e M[u
(" ) + Cu(" ) + K[uavg" ] = e Mu
using mean value thm.
Take limit as $ $0: stiffness term $0; displacement cannot change instantaneously so u($)$0. Thus,
(0+ ) = M"1e u
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
Define the impulse response matrix function H as + CH + KH = E" (t ) " (t ) = Dirac delta function = lim p (t ) MH # $0 (0+ ) = M"1E H(0) = 0, H
! !
Note that H is the same size as E. The element hij(t) is the i-th response to an impulse in the ! shape of the j-th column of E.
Compute response of base-excited 2DOF structure to 1940 El Centro earthquake: 0.2 + Cu + Ku = "M1 g 0 Mu v
"2 0%(u 1 + " 3 .1%(u 1 + " 600 .200 %(u1 + (2+ $ ') , + $ ') , + $ ') , = .) , v 1&*u 200&*u2 - *1- g #0 1&*u 2 - # .1 2 - # .200
Mass in Mg, damping kN"s/dm and stiffness kN/cm. Natural frequencies are 10 rads/sec and 20 rads/sec (or 1.59 and 3.18 Hz), damping 2.5% and 5.0%.
vg
(0) = 0, u(0) = 0 u !
!
sin t sin t
1599 16 1599 16
"3 +3
sin t sin t
399 399
u(t ) = u(0) + ! !
A convolution of h with the ground acceleration produces the response of the structure to the earthquake.
!0.1 0
10 20 time [secs]
30
USC Viterbi
School of Engineering
Damping Models
USC Viterbi
School of Engineering
c1!=!0.2 k1!=!4
In the previous damped 2DOF example, we used specific damping coefficients in the model: c1 = 0.2, c2 = 0.1. "2 0%(u 1 + " 0.3 .0.1%(u 1 + " 4 + 2 .2 %(u1 + (0+ "1 = 1, #1 = 2.5% $ ') , + $ ') , + $ ') , = ) , 0.1&*u 2&*u2 - *0- " 2 = 2, #2 = 5.0% #0 1&*u 2 - # .0.1 2 - # .2 We found that the damping #TC# diagonalized.
! However, a different choice (c1 = 0.2, c2 = 0.2) may not: $1 2'$ 0.4 #0.2 '$1 1' $ 0.4 #0.2' T " C" = & )& )& )=& ) 0.2(%2 #1( % #0.2 1( %1 #1(% #0.2
!
c1!=!0.2 m1!=!2 k1!=!4
"
We often do not know the damping coefficients (while mass and stiffness can be computed or easily measured, damping is much harder to quantify). Damping ratios &i in individual modes are easier to estimate (e.g., excite the structure in a particular mode, and estimate & from the decay e&!t; repeat for other modes). So, one way to construct the full damping matrix C is by assuming a modal decomposition %! 0( ' * " T C" = ' 2#i $i Mi * '0 !* & )
&! ( C = "# T ( 2$i %i Mi (0 ' &! ( #1 2$i %i +" = M" ( + (0 !* '
0) +
#1 +" + !*
0) +
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
2# i $ i
!* )
( + Ku = 0 *"+1u
2# i $ i
!* )
( + Ku = 0 *"+1u
Decouple method 1: let u = #q and premultiply by #T: %! ( + " T M" ' 2# i $ i *"+1"q + " TK"q = 0 " T M"q ' &! !* ) %! ( %! ( %! ( %! ( + ' M i * ' 2# i $ i *q + ' K i *q = 0 ' M i *q ' ' ' & !* ) & !* )' & !* ) & !* ) %! ( %! ( %! ( %! ( + ' M i * ' 2# i $ i *q + ' K i *q = 0 ' M i *q ' * ' * ' * ' & !) & !) & !) & !* ) %! ( %! ( 2 q + ' 2# i $ i *q + ' $ i *q = 0 ' ' & !* ) & !* )
i + 2#i $i q i + $i2qi = 0, i = 1,...,n q
Decouple method 2: let u = #q and premultiply by #1M1: &! ) + "#1M#1M" ( 2$ i % i +"#1"q + "#1M#1K"q = 0 "#1M#1M"q ! ( !+ ' * &! ) + "#1" ( 2$ i % i +"#1"q + "#1M#1K"q = 0 "#1"q ( ' !+ *
$! + & q & %
2" i # i
!) (
# i2
!) (
' )q = 0
Let " = #$1M$1K# % M#" = K# % # T M#" = # TK# &! ) &! ) % ( Mi +" = ( K i + ( ( ' !+ * ' !+ * &! ) &! % " = ( K i M i + = ( , i2 ( !+ ' * ( '
!+ *
) +
!
!
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
Rayleigh damping (1877) is given by C = 'M + (K. Since both M and K diagonalize with #, it is easy to see that #TC# = '#TM# + (#TK# is diagonal. The result: 2&i!i = ' + (!i2 or &i = ('/!i + (!i)/2. Since there are only two parameters, ' and (, we can choose the damping of two modes; all other modal damping ratios are given by the above equation. If we know (or choose) &r and &s then ' = 2!r!s(&r!s &s!r) / (!s2 !r2) (= 2(&s!s &r!r) / (!s2 !r2) &i = ('/!i + (!i)/2
It can be shown that the system is classically damped if either of the following are true: C = M F(M1K) + K G(K1M) C = F(KM1)!M + G(MK1) K for some matrix functions F(") and G(").
Note: should add references here.
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
Time history p(t) is related to its Laplace transform P(s) and Fourier Transform P(j!) as
#
P (s ) =
$ p (t )e "st dt
$
+ cu + ku = f Consider single degree of freedom system mu Laplace transform both sides: ms 2U (s ) + csU (s ) + kU (s ) = F (s )
(ms 2 + cs + k )U (s ) = F (s ) ! U (s ) 1 These are called = Or, rearranging: transfer functions. F (s ) ms 2 + cs + k ! U ( j" ) 1 Could do similar using Fourier transform: = F ( j" ) m( j" )2 + cj" + k ! 1 If f(t) = F0 sin!t, then = k # m" 2 + cj" 1 u (t ) = F0 sin(#t + $ ) k " m# 2 + cj# % ( 1 c #1 " = angle' ! * = tan m$ 2 # k 2 & k # m$ + cj$ )
P ( j" ) =
% p (t )e # j"t dt
%
"#
#$
! !
1 1 p (t ) = % P (s )e st ds & P ( j# )e j#t d# 2" #$ 2" $% ! Laplace transform P(s) may also be denoted L{p(t)} and Fourier Transform ! P(j!) by F{p(t)}.
p (t ) =
(t )} = Note: L{p
"#
integration by parts
(t )} = j"F {p (t )} ! Similarly: F {p
! !
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
For example:
+ 0.2u + 8u = f 2u U (s ) 1/ 2 = 2 F (s ) s + 0.1s + 4
2
+ 0.2u + 8u = f 2u U (s ) 1/ 2 = 2 F (s ) s + 0.1s + 4
Magnitude
Real Part
10
! 0
!1
c!=!0.2 k!=!8
10 ! 10
!2 !1
c!=!0.2 k!=!8
Roots ! of the denominator polynomial are called the poles of the system: ! "0.1 0.12 " 4 spoles = = "0.05 j 3.9975 # "0.05 1.9994 j 2
Im
10
10 10 Frequency [rads/sec]
!2 !1 0 1 10 10 10 Frequency [rads/sec]
Phase [degrees]
Imaginary Part
0 !1 !2 !3 !1 0 1 10 10 10 Frequency [rads/sec]
!
Re
10 10 Frequency [rads/sec]
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
Some TFs have a numerator that is also a polynomial in s. For example, the transfer function from ground acceleration to the absolute acceleration of a SDOF system:
+ 0.2u + 8u = "2v g 2u
2
For a MDOF system, must be careful to handle matrices correctly in determining the transfer function: + Cu + Ku = Ef Mu
c!=!0.2 k!=!8 !
+ v g Lu
g
[Ms
!
Im
+ Cs + K U(s ) = EF(s )
U(s ) = Ms + Cs + K
"1
EF(s )
= ( "0.1s " 4)
L{u} g (t ) Lv
0.1s + 4 s 2 + 0.1s + 4
!
Re
Now, the transfer function [Ms2 + Cs + K]1E is a matrix, each element of which is a scalar transfer function.
Roots of the numerator polynomial are called the zeros of the system: sroot = "40
USC Viterbi
School of Engineering
State-Space Formulation
USC Viterbi
School of Engineering
Rather than using n second-order differential equations + Cu + Ku = Ef Mu It is convenient (for simulation, control design, etc.) to instead use 2n first-order differential equations. Define state vector %
! #u & " =$ ' ( %u
A Laplace transform of the state-space matrices can be used to find the transfer function: = A" + Bf " & ' s" (s ) = A" (s ) + BF(s )
y = C y" + D y f " (s ) = (sI # A)#1BF(s )
d dt
u=u
!
state (or system) matrix
output matrix
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
The exact choice of state-space matrices depends on how one defines (. For example, let ( = T%. = A" + Bf = AT" + Bf = ( T#1AT)" + ( T#1B) f " " T"
y = C y" + D y f
&'
y = C y T" + D y f
&'
To see that the state-space response corresponds to our = A" with " = #e $t previous, consider " "#e "t = A#e "t & ' ( "I # A)$e "t = 0 For a non-trivial solution the determinant of ()I A) must ! structure, ! be 0. For our " I #I P Q ! but = SP " RP"1QP "I # A = #1 R S M K "I + M#1C
= ( "I + M#1C )( "I) # (M#1K )( "I)#1( #I)( "I) = "2I + "M#1C + M#1K ! !
!
y = (C y T)" + (D y ) f !
This gives a different set of state-space matrices but still describes the same I/O relationship from f to y.
! ! ! We can also see this from the transfer function: Y(s ) = C y T(sI " T AT) T B + D y F(s )
y (sTT y (sI " "1
[ = [C = [C
"1
"1 "1
= " #1( $2I + $M#1C + M#1K)" = "2I + "# $1M$1C# + # $1M$1K# ! If #1M1C# decouples to a diagonal matrix of 2&i!i, ! 2 "I # A = & ( "2 + 2$i %i " + %i2 ) = 0 "i , "* i = #$i % i j%i 1 # $i ! i !
A)"1B + D y F(s )
! !
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
Partition the eigenvector ) into displacement and velocity $ "I '+*d . +0 . "#d = # v #I & #1 #1 ), v / = , / & ' "1 M K#d + $# v + M"1C# v = 0 %M K "I + M C (-* 0 -0 0
( " I + "M C + M K)$ = 0 ! Let )d = #* and premultiply by #1: #1 2 #1 #1 ! " ( $ I + $M C + M K )"% = 0 ( "2I + "# $1M$1C# + # $1M$1K#) % = 0
2 #1 #1 d
! ! ! ! !
Lets return to the 2DOF system and see it in state-space. "2 0%(u 1 + " 0.3 .0.1%(u 1 + " 6 .2 %(u1 + (2+ $ ') , + $ ') , + $ ') , = .) , v 0.1&*u 2&*u2 - *1- g #0 1&*u 2 - # .0.1 2 - # .2 )0 0 " 0% 1 0 , d "u % + 0 0 0 1 ."u % / / # & = + (3 1 (0.15 0.05 .# & + # (0 &f ' ' / 1/ dt $u $u + $ (1 ' * 2 (2 0.10 (0.10 . -
If #1M1C# decouples to a diagonal matrix of 2&i!i, then 2 [ "1 ... " n ] = I "i , "* i = #$i % i j%i 1 # $i The state-space eigenmatrix then becomes &# # ) "=( %+ (where * denotes complex conjugate). '#$ #$ * ! The first n columns of + correspond to $! ' eigenvectors ,i, the second n columns to )i*. " = & #i ) & % !) ( !
!
vg
!
An eigenvalue analysis with MATLAB gives: " = [ #0.025 j 1599 1600 # 0.050 j 399 100 ] $ or, with alternate scaling, 0.3162 0 j 0.016 0.3158 j '
A = [zeros(2) eye(2); -[2 0;0 1]\[[6 -2;-2 2] [.3 -.1;-.1 .1]]]; [V,D] = eig(A); [lambda,ii]=sort(diag(D)); V=V(:,ii); % sort by increasing frequency T
!
!
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
If the damping term does not decouple, then the statespace approach must be used and ( "2I + "M#1C + M#1K)$d = 0 must be solved directly to get the complex eigenvalues and complex eigenvectors. Once the the eigenvalues )i,)i* are found, then !i = |)i| and &i = Re{)i}/(2!i). If some eigenvalues are purely real, then no oscillator mode corresponds to that eigenvalue.
Lets look at an example of a 3DOF structure with classical damping (-c = 0). The natural frequencies & damping are: !1 = 1 , &1 = 3.14%; !2 = 2.80, &2 = 8.80%; !3 = 4.05, &3 = 12.72%. A non-classically damped version with additional damping in the first floor (-c = 15c) has natural frequencies & damping: !1 = 1.15, &1 = 23.6%; !2 = 2.71, &2 = 85.9%; !3 = 3.64, &3 = 15.7%.
!
c!=!0.01a a!=!199.3233
(so first natural frequency is 1!Hz)
#"0.233 ! 0.19 j % %"0.557 ! 0.08 j %"0.737 % %"0.966 1.97 j % 0.409 4.04 j % $ 1.260 5.17 j
"0.545 ! 3.99 j "0.977 ! 0.93 j "0.591 "26.85 63.2 j 6.213 22.1j 8.655 5.2 j
0.003 0.17 j & ( 0.537 ! 0.07 j ( ( "0.328 ( 3.749 ! 0.66 j ( "3.575 ! 11.9 j ( ( 1.174 7.40 j '
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
Proportionally Damped:
= A" The unforced free response of the state-space system " could be found through modal decomposition, but there is an easier way. To compute the response at time t, break the time up into r smaller steps. Given the definition of a derivative: ! t 1 1 1 "( r t ) # "( 0 r t ) + " r t # " (0) + A" (0) r t = [I + A r ]" (0)
The same can be used to approximate ( after each t/r. t 1 t 2 "( 2 r t ) # [I + A r ]" ( r $t ) = [I + A r ] " (0) ! t r %1 t r "( r r t ) # [I + A r ]" ( r $t ) = [I + A r ] " (0) To eliminate the approximation, let r$#. 2 3 1 1 " (t ) = [I + ( At ) + 2 ! ( At ) + 3! ( At ) + !]" (0)
" (t ) = e At " (0)
NonProportionally Damped:
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
eAt is a matrix; its elements are not the exponential of the elements of At. The computation of eAt can be computed through the power series formulation, though that is not computationally efficient. An efficient computation uses the fact that the eigenvectors of a power of a matrix are the same as those of the matrix itself. If A+ = +., where + is the eigenmatrix and . is a diagonal matrix of eigenvalues, then it can be shown that An+ = +.n. 2 3 1 1 Thus, e At = I + ( At ) + 2 ! ( At ) + 3! ( At ) + !
#1 2 2 #1 3 3 1 1 = "#1" + "#1$"t + 2 ! " $" t + 3! " $ "t + ! 2 3 1 1 = "#1[I + ( $t ) + 2 ! ( $t ) + 3! ( $t ) + !]"
Integrating from t = 0 to t = 0+, will give the initial conditions that make a free response equal to the impulse response.
!
+ # H " (0+ ) # H " (0# ) = AHavg " (0 # 0 ) + B
H " (0+ ) = B ! !
) + +" "+ *
For t > 0, the impulse response is just an unforced free response (which was solved several slides ago), so
H " (t ) = e At H " (0+ ) H " (t ) = e At B
USC Viterbi
School of Engineering
Add Example!
USC Viterbi
School of Engineering
The forced response is, then, the combination of the initial condition free response and the effects of the force
" (t ) = e At " (0) +
t
Weve already seen the free response of the unforced state = A" . So, ((t + -t) = eA-t ((t). space system " = A" + Bf in [t, t + -t] The effect of the forced response of " is the superposition of the free response with the impulse response of f during the time step. !
" (t + #t ) = e A#t " (t ) + ! & e A(t + #t $% )Bf(% )d%
t + #t t
In the interval [t, t + -t), if f($) is constant (a zero-order hold) "A '#t * $ "Ad Bd % #0 " (t + #t ) = e A#t " (t ) + ) & e A( #t $% )d% ,Bf(t ) $ ' =e 0 I & # ( + 0 Ad
Bd
B% ' (t 0I&
Defining ((k) = ((k-t), then the discrete-time state space form is ! " (k + 1) = Ad" (k ) + Bd f(k ) Note: one can also assume a first-order y(k ) = C y" (k ) + D y f(k ) (linear) hold on f to determine B .
d
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
! !
Continuous systems (e.g., beams, plates, shells, or more complex structures) also have modes of vibration (generally infinitely many). Lets look at an Euler-Bernoulli beam; transverse displacement v(x,t) must satisfy the partial differential equation of motion where (")/ = 0/0x: = f ( x ,t ) (EIv "")"" + #Av v(x,t) Substitute v(x,t) = V(x)T(t) into the unforced, undamped system. = 0 x (EIV "")""T + #AVT EI, *A L (EIV "")"" T 2 = $ =" #AV T LHS is a function of x, the RHS of t, so both must be constant. 4 2 1/2 Assuming ! EI and *A are constant, let ) = [*A! /EI] so: V """" # $4V = 0 V ( x ) = c1 sinh "x + c 2 cosh "x + c 3 sin "x + c 4 cos "x + " 2T = 0 T T (t ) = c sin("t + # )
! !
c1 sinh "L + c 3 sin "L = 0 sinh "L sin "L $ 2 =0 "2c1 sinh "L # "2c 3 sin "L = 0 " sinh "L # "2 sin "L
The non-trivial solution is )iL = i! where n = 1, 2, ; the resulting eigenfunctions are Vi(x) = C sin i!x/L and the natural frequencies are !i = (i!/L)2 [*A/EI]1/2.
!1
!3!=!9!1
!2!=!4!1
! !
USC Viterbi
School of Engineering
USC Viterbi
School of Engineering
which has solutions )L = 1.8751, 4.6941, 7.8548, 10.996, ; larger solutions are approximately !(1"+ 2n)/2.
"1 =
3.516 EI #A L2
"2 =
22.03 EI #A L2
cosh "i L + cos "i L sinh "i L + sin "i L
"3 =
61.70 EI #A L2
Sometimes solving for the exact mode shapes from the partial differential equations of motion is difficult. An alternate approach is to assume a set of mode shapes Vi(x) that satisfy the boundary conditions. Then, Lagranges equations can be used to determine the equations of motion. For example, for an Euler-Bernoulli beam: v ( x ,t ) = "Vi ( x )v i (t )
V = T=
! !1
1 2
kij =
mij =
# "AVi (x )V j (x )dx
0
!2!=!6.27!1
!
!
! !