Mechanical Vibrations Ecole Centrale Nantes

AA242B: MECHANICAL VIBRATIONS

Direct Time-Integration Methods
These slides are based on the recommended textbook: M. G

eradin and D. Rixen, Mechanical
Vibrations: Theory and Applications to Structural Dynamics, Second Edition, Wiley, John &
Sons, Incorporated, ISBN-13:9780471975465
Outline
1 Stability and Accuracy of Time-Integration Operators
2 Newmarks Family of Methods
3 Explicit Time Integration Using the Central Difference Algorithm

Stability and Accuracy of Time-Integration Operators
Multistep Time-Integration Methods
Lagranges equations of dynamic equilibrium (p(t) = 0)

M
q + Cq + Kq = 0
q(0) = q0
q(0)
= q 0
First-order form

0 M
q
M 0
q
0
+
=
M C
0
K
q
0
q
|
{z
} | {z } |
{z
} | {z } | {z }
u
AB
= u = Au
AA
where
A = A1
B AA
Direct time-integration

General multistep time-integration method for first-order systems of

the form u = Au
m
m
X
X
un+1 =
j un+1j h
j u n+1j
j=1
j=0
where h = tn+1 tn is the computational time-step, un = u(t n ), and

qn+1
un+1 =
q n+1
is the state-vector calculated at tn+1 from the m preceding state
vectors and their derivatives as well as the derivative of the
state-vector at tn+1
0 6= 0 leads to an implicit scheme that is, a scheme where the
evaluation of un+1 requires the solution of a system of equations
0 = 0 corresponds to an explicit scheme that is, a scheme where
the evaluation of un+1 does not require the solution of any system of
equations and instead can be deduced directly from the results at the
previous time-steps

General multistep integration method for first-order systems

(continue)
m
m
X
X
j un+1j h
j u n+1j
un+1 =
j=1
j=0
trapezoidal rule (implicit)

un+1 = un +
h
h
h
(u n + u n+1 ) ( A I)un+1 = un u n
2
2
2
backward Euler formula (implicit)

un+1 = un + hu n+1 (hA I)un+1 = un
forward Euler formula (explicit)
un+1 = un + hu n un+1 = (I + hA)un

Numerical Example: the One-Degree-of-Freedom Oscillator
Consider an undamped one-degree-of-freedom oscillator

q + 02 q = 0
with 0 = rad/s and the initial displacement
q(0) = 1, q(0)
=0
exact solution
q(t) = cos 0 t
associated first-order system
u = Au
where

A=
02
0
0
1
u = [q,
q]T , and initial condition

u(0) =
0
1

Numerical Example: the One-Degree-of-Freedom Oscillator
Numerical solution
T
T = 3s, h = 32
3
Exact solution
Trapezoidal rule
Euler backward
Euler forward
4
0
0.5
1.5
t
2.5

Stability Behavior of Numerical Solutions
Analysis of the characteristic equation of a time-integration method

consider the first-order system u = Au
for this problem, the general multistep method can be written as
un+1 =
m
X
j un+1j h
j=1
m
X
j u n+1j
j=0
m
X
[j I hj A] un+1j = 0,
0 = 1
j=0
let r be the eigenvalues of A and X be the matrix of associated

eigenvectors
m
P
the characteristic equation associated with [j I hj A] un+1j = 0 is
j=0
obtained by searching for a solution of the form

un+1m
Xa
(decomposition on an eigen basis)
u(n+1m)+1
un+1m = Xa
un = = k un+1k = = m Xa
(solution form)
..
.
un+1
where C is called the solution amplification factor



(continue)
Hence
m
X
[j I hj A] mj Xa = 0
j=0
1
Since X
leads to
AX = diag(r ), premultiplying the above result by X1

" m
#
X
mj
[j I hj diag(r )]
a=0
j=0
m
X
[j hj r ] mj = 0, r = 1, 2
j=0
hence, the numerical response un+1 = m Xa remains bounded if each

solution of the above characteristic equation of degree m satisfies
|k | < 1, k = 1, , m


(continue)
the stability limit is a circle of unit radius
in the complex plane of r h, the stability limit is therefore given by
writing = e i , 0 2
m
X
= r h =
j e i(mj)
j=0
m
X
j e i(mj)
j=0
one-step schemes (m = 1)
r h =
0 e i + 1
e i + 1
=
0 e i + 1
0 e i + 1


(continue)
one-step schemes (m = 1) (continue)
r h =
0 e i + 1
e i + 1
=
i
0 e + 1
0 e i + 1
forward Euler: 1 = 1, 0 = 0, 1 = 1 r h = e i 1
the solution is unstable in the entire plane except inside the circle of
unit radius and center 1
backward Euler: 1 = 1, 0 = 1, 1 = 0 r h = 1 e i
the solution is stable in the entire plane except inside the circle of
unit radius and center 1
1
1
2i sin
trapezoidal rule: 1 = 1, 0 = , 1 = r h = 1+cos
2
2
the solution is stable in the entire left-hand plane


(continue)
application to the single degree-of-freedom oscillator

0 02
q + 02 q = 0,
A=
1
0
the eigenvalues are r = i0
the roots r h are located in the unstable region of the forward Euler
scheme amplification of the numerical solution
the roots r h are located in the stable region of the backward Euler
scheme decay of the numerical solution
the roots r h are located on the stable boundary of the trapezoidal
rule scheme the amplitude of the oscillations is preserved

Newmarks Family of Methods
The Newmark Method
Taylors expansion of a function f

0
f (tn + h) = f (tn ) + hf (tn ) +
hs (s)
1
h2 00
f (tn ) + +
f (tn ) +
2
s!
s!
tn +h
(s+1)
Application to the velocities and displacements

Z tn+1
( )d
f = q, s = 0 qn+1 = qn +
q
tn
Z tn+1
( )(tn+1 )d
f = q, s = 1 qn+1 = qn + h q n +
q
tn
( )(tn + h ) d
tn

The Newmark Method
n and q
n+1 around [tn , tn+1 ]
Taylor expansions of q
n
q
n+1
q
(tn )2
+
(1)
2
(tn+1 )2
( ) + q(3) ( )(tn+1 ) + q(4) ( )
+ (2)
q
2
( ) + q(3) ( )(tn ) + q(4) ( )

q
( )
Combine (1 ) (1) + (2) and extract q
( ) = (1 )
= q
qn +
qn+1 + q(3) ( )( h tn ) + O(h2 q(4) )
( )
Combine (1 2) (1) + 2 (2) and extract q
( ) = (1 2)
= q
qn + 2
qn+1 + q(3) ( )( 2h tn ) + O(h2 q(4) )

The Newmark Method
tn+1
( ) in
Substitute the 1st expression of q
( )d
q
tn
Z
=
tn+1
tn+1
Z
( )d
q
tn
(1 )
qn +
qn+1 + q
( )( h tn ) + O(h q

) d
(3)
3 (4)
(3)
2 (4)
tn
Z
=
n + h q
n+1 +
(1 )h q
tn+1
( )( h tn )d + O(h q
tn
Apply the mean value theorem

Z
=
tn+1
"
( )d
q
n + h q
n+1 + q
(1 )h q
(3)
(
)
tn
n + h q
n+1 + (
(1 )h q
( ) in
Substitute the 2nd expression of q
( h tn )2
2
#t
n+1
3 (4)
+ O(h q
tn
1
2 (3)
3 (4)
)h q (
) + O(h q )
2
tn+1
( )(tn+1 )d
q
tn
Z
=
tn+1
tn
( )(tn+1 )d
q
1
1
2
4 (4)
n + h2 q
n+1 + ( )h3 q(3) (
)h q
) + O(h q )
2
6

The Newmark Method
In summary
Z
tn+1
( )d = (1 )h q
n + h q
n+1 + rn
q
tn
tn+1

( )(tn+1 )d =
q
tn

1
n + h2 q
n+1 + rn0
h2 q
2
where

rn
rn0

1
=
h2 q(3) (
) + O(h3 q(4) )
2

1
h3 q(3) (
) + O(h4 q(4) )
=
6
and tn < < tn+1


The Newmark Method
Hence, the approximation of each of the two previous integral terms

by a quadrature scheme leads to
q n+1
qn+1
n + h q
n+1
q n + (1 )h q

1
2
n + h2
qn + h q n + h
q
qn+1
2
where and are parameters associated with the quadrature

scheme
(3)
(4)

The Newmark Method
Particular values of the parameters and

=
1
1
( ) in [tn , tn+1 ]
and = leads to linearly interpolating q
2
6

n+1 q
n
q
ln ( ) = q
n + ( tn )
q
h
1
1
( ) in [tn , tn+1 ]
and = leads to averaging q
2
4
n+1 + q
n
q
av ( ) =
q
2

The Newmark Method
Application to the direct time-integration of M

q + Cq + Kq = p(t)
write the equilibrium equation at t n+1 and substitute the expressions
(3) and (4) into it
= [M + hC + h2 K]
qn+1 = pn+1 C[q n + (1 )h
qn ]

1
n
h2 q
K qn + hq n +
2
if the time-step h is uniform, M + hC + h2 K can be factored once
n+1
solve the above system of equations for q
substitute the result into the expressions (3) and (4) to obtain q n+1
and qn+1

Consistency of a Time-Integration Method
A time-integration scheme is said to be consistent if

un+1 un
n)
= u(t
h0
h
lim
The Newmark time-integration method is consistent

(1 )
qn +
qn+1

un+1 un
n
q
1
=
lim
= lim
n + h q
n+1
q n
hq
q n +
h0
h0
h
2
Consistency is a necessary condition for convergence

Stability of a Time-Integration Method
A time-integration scheme is said to be stable if there exists an

integration time-step h0 > 0 so that for any h [0, h0 ], a finite
variation of the state vector at time tn induces only a non-increasing
variation of the state-vector un+j calculated at a subsequent time
tn+j

The application of the Newmark scheme to M

q + Cq + Kq = p(t)
can be put under the form
un+1 = A(h)un + gn+1 (h)
where A is the amplification matrix associated with the integration
operator
1
A(h) = H1
1 (h)H0 (h), gn+1 = H1 (h)bn+1 (h)

(1 )hp
n + hpn+1
M + hC
hK
bn+1 = 1
,
H
=
1
h2 C
M + h2 K
h2 pn + h2 pn+1
2
(1 )hK
M +(1 )hC
1
H0 = 1
h2 C hM M +
h2 K
2
2

Effect of an initial disturbance

u0 = u00 u0
= un+1 = A(h)un = A2 (h)un1 = = A(h)n+1 u0
consider the eigenpairs of A(h)
(r , xr )
then
un+1 = An+1 (h)
2N
X
s=1
as xs =
2N
X
as n+1
xs
s
s=1
where N is the dimension of the semi-discrete second-order

dynamical system
= un+1 will be amplified by the time-integration operator only if
the moduli of an eigenvalue of A(h) is greater than unity
= un+1 will not be amplified by the time-integration operator if all
moduli of all eigenvalues of A(h) are less than unity

Undamped case
decouple the equations of equilibrium by writing them (for the
purpose of analysis) in the modal basis
q = Qy =
N
X
yi qai = yi + i2 yi = pi (t)
i=1
apply the Newmark scheme to the i-th modal equation recalled

above to obtain the amplification matrix

i2 h2
i2 h2
1 1+
i2 h2 1 2 1+
2 h2
2 h2
i
i
A(h) =
2 2
h
1 i h
1
2
2
2
2
2 1+ h
1+ h
i
characteristic equation is 2 ( +
i2 h2
1+i2 h2
1
) 2
2
+ 1 ( 12 ) 2 = 0
where 2 =
characteristic equation has a pair of conjugate roots 1 and 2 if
2

1
4
+
4 2 2 , i = 1, , N
2
i h

Undamped case (continue)

the eigenvalues 1 and 2 can be written as
1,2
e i
where
s

1
2
1
2
q
1 14 ( + 12 )2 2
arctan
1 12 ( + 21 ) 2
then, the Newmark scheme is stable if

1
but recall that this is assuming

2
1
4
+
4 2 2 ,
2
i h
1
2
i = 1, , N
= limitation on the maximum time-step



the algorithm is conditionally stable if
1
2
it is unconditionally stable if furthermore

1

2
1
+
2
1
1
the choice = and = leads to an unconditionally stable
2
4
time-integration operator of maximum accuracy

Stability of the Newmark scheme

Damped case (C 6= 0)
consider the case of modal damping
then, the uncoupled equations of motion are
yi + 2i i yi + i2 yi = pi (t)
where i is the modal damping coefficient
1
1
consider the case = , =
2
4
an analysis similar to that performed in the undamped case reveals
that in this case, the Newmark scheme remains stable as long as
<1
in general, damping has a stabilizing effect for moderate values of

Amplitude and Periodicity Errors
Free-vibration of an undamped linear oscillator

y + y = 0
and
y (0) = y0 , y (0) = 0
A=
0
1
02
0
the above problem has an exact solution y (t) = y0 cos t which can
be written in complex discrete form as yn+1 = e ih yn the exact
amplification factor is ex = 1 and the exact phase is ex = h
the numerical solution satisfies

yn+1
un+1 =
= A(h)un
yn+1
let 1,2 (, ) be the eigenvalues of A(h, , )
2
when + 21 4 24h2 , 1 and 2 are complex-conjugate
i
1,2 (, ) = (, )e i(,)
where
s
=

1

2 ,
1
1 2 2
1 4 ( + 2 )
= arctan
,
1 12 ( + 12 ) 2
2 h2
1 + 2 h2

Free-vibration of an undamped linear oscillator (continue)

amplitude error
ex = 1 =
1
2

1
2 h2 + O(h4 )
2
relative periodicity error

1
T
= 1 =
T
1
ex
1
ex
h
1
1=
1
12
2 h2 + O(h3 )

Algorithm
Stability
limit
h
Purely explicit
Central difference
0
1
2
0
0
0
2
Fox & Goodwin
1
2
1
12
Linear acceleration
1
2
Average constant
acceleration
1
2
Table:
Amplitude
error
1
2 h2
4
Periodicity
error
T
T
2 2
24h
2.45
O(h3 )
1
6
3.46
2 h2
24
1
4
2 h2
12
Time-integration schemes of the Newmark family
The purely explicit scheme ( = 0, = 0) is useless

The Fox & Godwin scheme has asymptotically the smallest phase
error but is only conditionally stable
1
1
The average constant acceleration scheme ( = , = ) is the
2
4
unconditionally stable scheme with asymptotically the highest
accuracy

Total Energy Conservation
Conservation of total energy

dynamic system with scleronomic constraints
ns
X
d
(T + V) = mD +
Qs q s
dt
s=1
1
1
T = q T Mq and V = qT Kq
2
2
the dissipation function D is a quadratic function of the velocities
(m = 2)
1
D = q T Cq
2
external force component of the power balance
ns
X
Qs q s = q T p
s=1
integration over a time-step [tn , tn+1 ]

Z tn+1
t
[T + V]tn+1
=
(q T Cq + q T p)dt
n
tn

Total Energy Conservation
Conservation of total energy (continue)

note that because M and K are symmetric (MT = M and KT = K)
1
T
(q n+1 q n ) M(q n+1 + q n )
2
1
T
+
(qn+1 qn ) K(qn+1 + qn )
2
when time-integration is performed using the Newmark algorithm with

1
1
= , = , the above variation becomes see (3) and (4)
2
4
t
[T + V]tn+1
= [Tn+1 Tn ] + [Vn+1 Vn ]
n
=
[T + V]tn+1
n
1
h
T
T
(qn+1 qn ) (pn + pn+1 ) (q n+1 + q n ) C(q n+1 + q n )
2
4
when applied to a conservative system (C = 0 and p = 0), preserves the total energy
Rt
t
= tnn+1 (q T Cq + q T p)dt and therefore
for non-conservative systems, [T + V]tn+1
n
both terms in the right-hand side of the above formula result from numerical
quadrature relationships that are consistent with the time-integration operator

Z t

Z t
n+1 T
n+1 T
pn + pn+1
1
T
q dt
q pdt
= (qn+1 qn ) (pn + pn+1 )
2
2
tn
tn
Z

Z t
tn+1
n+1 T
q n + q n+1
1
q n + q n+1
T
T
q Cqdt
q dt C
= (qn+1 qn ) C
2
2
2
tn
tn
=
h
T
(q n+1 + q n ) C(q n+1 + q n )
4

Explicit Time Integration Using the Central Difference Algorithm
Algorithm in Terms of Velocities
Central Difference (CD) scheme = Newmarks scheme with = 12 ,

=0
q n+1
qn+1
n + q
n+1
q
)
2
h2
n
qn + hn+1 q n + n+1 q
2
q n + hn+1 (
where hn+1 = tn+1 tn

Equivalent three-step form
h2
n1
start with qn = qn1 + hn q n1 + n q
2
divide by hn
subtract the result from qn+1 divided by hn+1
account for the relationship (5)
n =
= q
hn (qn+1 qn ) hn+1 (qn qn1 )

hn+ 1 hn hn+1
2
where hn+ 1
hn + hn+1
=
2
(5)

Algorithm in Terms of Velocities
Case of a constant time-step h
n =
q
qn+1 2qn + qn1

h2
Efficient implementation
compute the velocity at half time-step
n+ 1 ) = q n +
q n+ 1 = q(t
2
h
1
n = (qn+1 qn )
q
2
h
compute
n =
q
1
(q 1 q n 1 )
h n+ 2
2
Stability condition
cr h 2
where cr is the highest frequency contained in the model: this condition is also known as
the Courant condition, and
2
hcr =
cr
is referred to here as the maximum Courant stability time-step

Application Example: the Clamped-Free Bar Excited by an End Load
Clamped bar subjected to a step load at its free end

Model made of N = 20 finite elements with equal length l =
19
17
18
L
N
20
19
20
lumped mass matrix
Eigenfrequencies of the continuous system

r

r

EA
2r 1 EA
2r 1
contr = (2r 1)
=
=
2 mL2
N
2 ml 2
N
2

Finite element stiffness and mass matrices
ml
M=
2
2
1
EA
K=
2
..
2
1
1
2
1
1
2
..
0
..
..
.
1
1
2
1
1
1
(6)
Analytical frequencies of the discrete system

r
r = 2
EA
sin
ml 2

2r 1
2N

2r 1
2N
2 sin
cr < cr (N ) = 2
r = 1, 2, N
Critical time-step for the CD algorithm

cr hcr = 2 hcr = 1

h = 1, h = 0.707
Node 1
Node 10
25
h=1
h=0.707
2.5
h=1
h=0.707
20
Displacement
Displacement
2
1.5
1
0.5
15
10
0
0
0.5
1
0
50
100
5
0
150
Time
Node 20
40
100
150
100
150
Time
Node 10
1.5
h=1
h=0.707
35
50
h=1
h=0.707
0.5
25
Velocity
Displacement
30
20
15
0.5
10
1
5
0
0
50
100
Time
150
1.5
0
50
Time

h = 1.0012
Node 10
60
Node 20
100
h = 1.0012
80
h = 1.0012
40
Displacement
Displacement
60
20
20
40
20
0
20
40
40
60
60
0
50
100
Time
150
80
0
50
100
Time
150

Restitution of the Exact Solution by the Central Difference Method
For the clamped-free bar example, the CD method computes the

exact solution when h = hcr
Comparison of the exact solution of the continuous free-vibration bar
problem and the analytical expression of the numerical solution
denote by qj,n the value of the j-th d.o.f. at time tn

if qj,n is not located at the boundary, it satisfies see (6)
EA
ml
(qj,n+1 2qj,n + qj,n1 ) +
(qj1,n + 2qj,n qj+1,n ) = 0
h2
l
the general solution of the above problem is
qj,n = sin(j + ) [a cos n + b sin n]
|
{z
} |
{z
}
spatial component
temporal component
comparing the above expression to the exact harmonic solution of

the continuous form of this free-vibration problem (which can be
derived analytically)
= n = t = nh
= num
h
(7)

Restitution of the Exact Solution by the Central Difference Method
Comparison of the exact solution of the free-vibration bar problem

and the analytical expression of the numerical solution (continue)
introduce the exact expression for qj,n in the CD scheme
2[(1 cos ) 2 (1 cos )]qj,n = 0
2
1
1
ml
1
where 2 =
= 2 1 cos = 2 (1 cos )
EA h2
h
make use of the boundary conditions in space q0,n = 0,

and plug (7)
in the last equation in (6) = = 0 and r = 2rN1 2 , r N
=1 cos r =
1
(1 cos r )
2
special case 2 = 1 (h = hcr = 1) r = r and

r

r
2r 1 EA
2r 1
numr =
= r =
=
h
N
2 ml 2
N
2
= the numerical frequency coincides with the r -th eigenfrequency
of the continuous system

Mechanical Vibrations Ecole Centrale Nantes

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Mechanical Vibrations Ecole Centrale Nantes

Enviado por

Direitos autorais:

Formatos disponíveis

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

These slides are based on the recommended textbook: M. G

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

1 Stability and Accuracy of Time-Integration Operators

2 Newmarks Family of Methods

3 Explicit Time Integration Using the Central Difference Algorithm

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Lagranges equations of dynamic equilibrium (p(t) = 0)

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

General multistep time-integration method for first-order systems of

where h = tn+1 tn is the computational time-step, un = u(t n ), and

AA242B: MECHANICAL VIBRATIONS

General multistep integration method for first-order systems

trapezoidal rule (implicit)

backward Euler formula (implicit)

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Consider an undamped one-degree-of-freedom oscillator

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Analysis of the characteristic equation of a time-integration method

let r be the eigenvalues of A and X be the matrix of associated

obtained by searching for a solution of the form

(decomposition on an eigen basis)

where C is called the solution amplification factor

AA242B: MECHANICAL VIBRATIONS

Analysis of the characteristic equation of a time-integration method

AX = diag(r ), premultiplying the above result by X1

hence, the numerical response un+1 = m Xa remains bounded if each

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Analysis of the characteristic equation of a time-integration method

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Analysis of the characteristic equation of a time-integration method

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Analysis of the characteristic equation of a time-integration method

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Taylors expansion of a function f

f (tn + h) = f (tn ) + hf (tn ) +

Application to the velocities and displacements

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

( ) + q(3) ( )(tn ) + q(4) ( )

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Apply the mean value theorem

AA242B: MECHANICAL VIBRATIONS

and tn < < tn+1

AA242B: MECHANICAL VIBRATIONS

Hence, the approximation of each of the two previous integral terms

where and are parameters associated with the quadrature

AA242B: MECHANICAL VIBRATIONS

AA242B: MECHANICAL VIBRATIONS

Particular values of the parameters and

AA242B: MECHANICAL VIBRATIONS