ES386 Dynamics of Vibrating Systems

Part II: Multiple DoF systems (Units 7-12)

Dr Michael Faulkner
 Practical arrangements
Contact details:
Discussion forum on ES386 Moodle page
Email: [Link]@[Link]

Office hours:
Wednesdays 11:00-12:00 & Fridays 14:00-15:00 in D211
(exception: Friday 8 March is at 16:00-17:00)
Material:
i) Lectures on Tuesdays & worked examples on Fridays
ii) Additional material delivered in lecture videos
iii) All available on Moodle or via LectureCapture

JRK7: 2
 Moodle resources
Lecture videos, in small chunks of 15-20 mins
One PDF of lecture slides for each unit (Units 7-12)
PDFs of what I write in lectures
Weekly activities – e.g. quizzes, example questions
Some units have non-examinable ‘Example Sheets’ w/solutions
Past exam papers
Other resources – MATLAB examples, simulations
Reading list: Rao textbook (as in Part I with Oksana)
Assignment: Computational Assignment requires
i) modal analysis (Units 7 & 10)
ii) forced vibration (Unit 8)
iii) convolution method (Unit 9)

JRK7: 3
 Overview of Part II
Week 20: Unit 7. Modal analysis for multiple DOF systems
Week 21: Unit 8. Forced vibration and applications
Week 22: Unit 9. Non-periodic vibration
Unit 10: Modal analysis for forced and
damped systems
Week 23: Unit 11: Continuous systems
Unit 12: Approximate methods – Rayleigh and
Dunkerley approaches

JRK7: 4
 Notation
Scalar variables: x, x1
Angles: θ, θ1
Matrices: A, B, AT , A -1
some textbooks [A], [B], [A]T , [A]-1
% 1 !
Vectors: x, q, x1,
#
& 2 ",
#
{1 2 3}T
# 3 #
' $
some textbooks
{x}, {q}, {x1 }

Time-derivatives: x!, θ!, x! , !!x

Complex numbers: x + jy, j = −1
some textbooks x + iy, i = −1
JRK7: 5
Unit 7 – Modal Analysis for multiple
DoF systems
Dr Michael Faulkner
 Introduction and motivation
When a mechanical structure vibrates near one of its
resonant or natural frequencies, v. small forces can
cause significant deformation
This can damage the structure, eg, Tacoma Bridge
So resonant vibrations must be avoided in most
engineering applications
Each natural frequency and corresponding vibrational
shape defines a normal mode (often simply mode)
Key goal of modal analysis is to identify these
vibrational mode frequencies and modal shapes (as
well as the modal responses)

JRK7: 7
 Tacoma narrows bridge
The bridge collapsed on 7 Nov. 1940, a few months
after its opening due to wind-induced vibration.

Failure to predict transition from undulating mode of vibration to twisting

JRK7: 8
 Aeroplane flutter

JRK7: 9
[Link]
 Aeroplane flutter
Wings of aeroplanes can be subject to flutter
phenomena during flight
Before a new aeroplane is released, testing
needs to be performed to detect possible onset
of flutter.
– Test under similar flight conditions
in wind tunnel
– But this is expensive, and…
– …a few input parameters affect hundreds of
parts of wing structure
– è important to estimate resonant
frequencies and damping ratios beforehand
– Doesn’t completely eliminate need for test
flights, but saves a lot of time/money!

JRK7: 10
 Ferrybridge Power Station

Three cooling towers collapsed on 1st

November 1965 due to wind-induced
vibration in 85 mph storm conditions

Design of individual towers tested in wind

tunnel, but not grouping effects

JRK7: 11
 Modal analysis of a car body

Car door attached to Computer simulation

electromagnetic shaker of modal analysis
(Source: Wikipedia) (Source: OptiStruct)

JRK7: 12
 Modes
Modes are inherent properties
– Determined by material properties (mass m,
damping b, and stiffness k), and boundary
conditions of the structure
– Each mode is defined by a natural frequency
(eigenvalue) and mode shape (eigenvector)
– Can include some modal damping
– Modes are independent of initial conditions (ICs),
but ICs do decide when/if each mode activates

JRK7: 13
Rao, Chapter 6
 Activities for Unit 7
Unit Title Learn Apply
7.1 Review of eigenvalue approach Today’s lecture Vevox/Quiz
7.2 Generalized eigenvalue problems Video 7.2
(13 mins)
7.3 Matrix Transformation I Video 7.3
(8 mins)
7.4 Matrix Transformation Questions Attempt Qs,
then review
solutions
7.5 Matrix Transformation II Video 7.5
(30 mins)
7.6 Modal Analysis Questions Attempt Qs,
then review
solutions

JRK7: 14
7.1 Standard Eigenvalue Approach
 Recap: Matrix Methods

3 DoF is enough to show the general patterns for n-DoF systems

T= 12 mx!12 + 12 mx!22 + 12 mx!32

2 2
V = kx + k ( x1 − x2 ) + k ( x2 − x3 ) + 12 kx32
1
2
2
1
1
2
1
2

JRK7: 16
 Recap: Matrix Methods

3 DoF is enough to show the general patterns for n-DoF systems

T= 12 mx!12 + 12 mx!22 + 12 mx!32

2 2
V = kx + k ( x1 − x2 ) + k ( x2 − x3 ) + 12 kx32
1
2
2
1
1
2
1
2

From Lagrange’s equation, problems of this type have solutions

involving diagonal & tri-diagonal matrices
M !!
x + Kx = 0
JRK7: 17
 Recap: Matrix Methods

This system is statically coupled (i.e. via stiffness)

The solution is a standard eigenvalue problem
Standard approach is to substitute A = M-1K

JRK7: 18
 Standard Eigenvalue Problem
Trial solution: x(t) = x 0 cosω t
Substituting gives: ( M + K) x = 0
−ω 2

A = M-1K gives standard eigenvalue problem: ( A − ω 2

I) x = 0
2
Eigenvalues λi = ω i and eigenvectors vi, 1 ≤ i ≤ n
General solution is superposition of n normal modes (n eigenvectors or
mode shapes, each at its natural frequency (square root of eigenvalue))
n
x(t) = ∑ Ai v i cos(ωi t + φi )
i=1
Ai and ϕi are amplitudes and phases of each mode vi

JRK7: 19
 How do we find the eigenvalues of an n x n square matrix A?
Ax − λ x = 0 (1)
( A − λ I) x = 0
For above set of equations to have a non-zero solution
det(A − λ I) = 0
Solving this characteristic equation gives eigenvalues

λn + c1 ln-1 + - - + cn = 0

This polynomial always has n roots, so there are n eigenvalues

(but they might not be distinct!)
Then substitute eigenvalues back in (1) to find eigenvectors

JRK7: 20
 Eigenvalue & Eigenvector Theorems
Theorem 1: If A is a n ´ n triangular matrix – upper triangular, lower
triangular or diagonal, the eigenvalues of A are the diagonal entries
of A.
Theorem 2: l = 0 is an eigenvalue of A if A is a singular (uninvertible)
matrix.
Theorem 3: A and AT have the same eigenvalues.
Theorem 4: Eigenvalues of a real symmetric matrix are real.
Theorem 5: Eigenvectors of a real symmetric matrix are orthogonal,
but only for distinct eigenvalues.
Theorem 6: det ( A) is the product of the absolute values of the
eigenvalues of A.
det(A) = λ1 λ2 λ3

JRK7: 21
 Find the eigenvalues and determinants of these
matrices by inspection
é 2 0 0ù
ê0 - 3 0ú
ê ú é3 5 7 ù
êë0 0 6úû ê0 - 2 1ú
ê ú
êë0 0 0úû é 2 0 0ù
ê 3 5 0ú
ê ú
êë2 1 6úû

JRK7: 22
 Find the eigenvalues and determinants of these
matrices by inspection
é 2 0 0ù
ê0 - 3 0ú
ê ú é3 5 7 ù
êë0 0 6úû ê0 - 2 1ú
ê ú
λ = {2, −3, 6} êë0 0 0úû é 2 0 0ù
| det(A1 ) | = 2 × 3× 6 = 36 ê 3 5 0ú
ê ú
λ = {3, −2, 0} êë2 1 6úû
| det(A 2 ) | = 0 (singular!)
λ = {2, 5, 6}
| det(A 3 ) | = 2 × 5 × 6 = 60
JRK7: 23
 3 DOF Example

If k = 1, m = 1:
⎡ 1 0 0 ⎤ ⎡ 2 −1 0 ⎤
⎢ ⎥ −1 ⎢ ⎥
M =⎢ 0 1 0 ⎥ = M ⇒ A = ⎢ −1 2 −1 ⎥ 2 − λ −1 0
⎢ 0 0 1 ⎥ ⎢ 0 −1 2 ⎥
⎣ ⎦ ⎣ ⎦ A − λ I = −1 2 − λ −1 =0
0 -1 2−λ

This is a special case, which can be (2 − λ ) ( λ 2 − 4λ + 2) = 0

solved by hand (NB: A is symmetric): λ1 = 0.58, λ2 = 2, λ3 = 3.41

JRK7: 24
 Mode Shapes

é0.5000 - 0.7071 - 0.5000ù

V = êê0.7071 0.0000 0.7071 úú
êë0.5000 0.7071 - 0.5000 úû

w1 = 0.76, w2 = 1.41, w3 = 1.85 rad / s

JRK7: 25
 Limitation 1: Loss of Symmetry of K and M

m 2m 3m

Only slightly more complex system, still have symmetric M and K

⎡m 0 0 ⎤ ⎡2k − k 0 ⎤
⎢ ⎥ ⎢ ⎥
M = ⎢0 2m 0 ⎥ K = ⎢ −k 2k − k ⎥
⎢⎣0 0 3m⎥⎦ ⎢⎣0 − k 2k ⎥⎦
• But with m = k = 1, A = M-1K is
⎡ ⎤ ⎡ ⎤
⎢1 0 0 ⎥⎡2 −1 0 ⎤ ⎢ 2 −1 0 ⎥
⎢ 1 ⎥⎢ ⎥ ⎢ 1 1 ⎥
A= 0 0 ⎢−1 2 −1⎥ = ⎢ − 1 − ⎥ ≠ AT
⎢ 2 ⎥ 2 2
⎢0 0 1 ⎥⎢⎣0 −1 2 ⎥⎦ ⎢ ⎥
⎢ 0 − 1 2
⎣ 3⎦ ⎣ 3 3 ⎥⎦

JRK7: 26
 Tongue, section 4.7

Limitation 2: Coupled Equations of Motion

So far the equations of motion have been coupled.
– to solve x1, we have to solve x2 at the same time.

é2 - 1ù ì x1 ü é5 0ù ì x1 ü ì f1 ü ¬mass-
ê- 1 4ú í x ý + ê0 8ú í x ý = í f ý coupled
ë ûî 2 þ ë ûî 2 þ î 2 þ
é1 0 ù ì x1 ü é7 - 2ù ì x1 ü ì f1 ü ¬stiffness-
ê0 2ú í x ý + ê- 2 8 ú í x ý = í f ý coupled
ë ûî 2 þ ë ûî 2 þ î 2 þ
é1 0 ù ì x1 ü é7 0ù ì x1 ü ì f1 ü ¬uncoupled
ê0 2ú í x ý + ê0 8 ú í x ý = í f ý
ë ûî 2 þ ë ûî 2 þ î 2 þ
– Can we change coordinates to decouple equations?
– Reduce N-DOF problem to N separate 1-DOF problems

JRK7: 27
 Vevox time!

JRK7: 28
 Limitations of using A = M-1K
1. Matrix algebra is much simpler with symmetric
matrices, but even when M and K are symmetric
A = M-1K may not be.

2. Coupling between DoFs è to find x1, we have to solve

for x2 at the same time, etc.

Solutions:
1. Solve the generalised eigenvalue problem
2. Use matrix transformations to write EoM in terms of
symmetric matrices in coordinates w/decoupled DoFs

JRK7: 29
7.2 Generalised Eigenvalue Problems
 Generalised Eigenvalue Problem

M!!
x + Kx = 0

JRK7: 31
 M!!
x + Kx = 0
Solution is of form:
x(t) = u e jωt (or x(t) = u cos ω t)

( −ω 2
M + K ) =0
ue jω t

det(−ω 2 M + K) = 0
Direct solution of generalised eigenvalue equation
leads to algebraic characteristic equation for unknown w

N solutions for eigenvalues wi2 and eigenvectors ui

w1, u1 satisfies (−ω12 M + K) u1 = 0

w2, u2 satisfies (−ω 22 M + K) u 2 = 0

JRK7: 32
 For a 2 DOF system:
x(t) = c1u1 sin(ω1t + ϕ1 ) + c2 u 2 sin(ω 2 t + ϕ 2 )
w1 w2 are natural frequencies.
Mode amplitudes c1,c2 and phases f1, f2 can be determined from initial
conditions.
u1 is the first mode shape
u2 is the second mode shape

This approach works well in general BUT

– Simpler to deal with one matrix than two
– Much easier to deal with symmetric matrices

JRK7: 33
 Vevox time!

JRK7: 34
 Properties of Symmetric Matrices
If A is a real symmetric matrix (i.e. if A = AT)
– Eigenvalues of A are all real numbers
– Eigenvectors of A are all real vectors
– Eigenvalues are positive if and only if A is positive definite
– Eigenvectors of A are orthogonal (v T Av > 0 ∀v ≠ 0)
v Ti .v j = 0, i ≠ j
(NB: only true if corresponding eigenvalues are distinct!)
– Eigenvectors form a basis which can be used to expand arbitrary
functions – basis of modal expansion

K and M are usually symmetric matrices, but we have seen that

A = M-1 K may not be symmetric – what can we do?
JRK7: 35
 Cholesky Decomposition
If M is symmetric, M = MT
and positive definite, i.e. for every non-zero choice of vector x

x T Mx > 0
then M can be factorised into 2 terms
M = L LT

where L is a lower triangular matrix

JRK7: 36
 Matrix Square Root
For diagonal mass matrix, this is:
! m 0 0 $
# 1 &
M1 2 = #0 m2 0 &
# &
#"0 0 m3 &%

" 1 %
$ 0 0 '
$ m1 ' 𝐌!/# 𝐌!/# = 𝐌
$ ' in general, but…
so inverse is −1 2 1
M = $0 0 '
$ m2 '
$ ' 𝑀!/# $%
= 𝑀$%
$0 1 '
0
$# m3 '& is not true in general
JRK7: 37
7.3 Matrix transformation I:
Mass-Normalised Stiffness Matrix
 Mass-Normalised Stiffness
Equation of motion: M!!
x + Kx = 0 (1)

−1 2
Change to new coordinates q: x=M q, q = M1 2 x
Substitute, and pre-multiply resulting equation by M−1 2
( M −1 2
MM −1 2
!!
) (
q + M −1 2
KM −1 2
)q = 0
Noting that M−1 2 MM−1 2 = I
We have I q(t) ! q(t) = 0
!! + K
where K ! = M−1 2 K M−1 2

JRK7: 39
 Vevox time!

JRK7: 40
7.4 Matrix transformation I:
Worked Example
 Worked Example
For the 2 DOF system

m1 = 9 kg m2 = 1 kg k1 = 24 N/m k2 = 3 N/m

é9 0ù ìx1 ü é27 - 3ù ì x1 ü ì0ü

ê0 1 ú íx ý + ê- 3 3 ú í x ý = í0ý
ë ûî 2 þ ë ûî 2 þ î þ
(a) Solve the generalised eigenvalue problem
to determine the natural frequencies and det(−ω 2 M + K) = 0
mode shapes. Are they orthogonal?
(b) Construct the mass-normalised stiffness matrix
and compute its eigenvalues and eigenvectors. K! = M−1 2 K M−1 2
Are the new eigenvectors orthogonal?
(c) Are the natural frequencies the same with both approaches?

JRK7: 42
7.5 Matrix transformation II:
Normal Coordinates
 Decoupling Equations of Motion

Orthogonality allows decoupling of equations of N-DOF motion,

reducing analysis to that of N separate single DOF problems.

A matrix P can be made from normalised eigenvectors v

of the mass-normalized stiffness matrix, arranged as columns.
We then have:

P = [v1 v2 v3... vn ]
PT P = I
PT K! P = Λ = diag(λ )
i

JRK7: 48
 Vevox time!

JRK7: 49
 Recap: 1st coordinate transformation

Matrix M-1/2 can be used in place of the ‘simple’ inverse matrix M-1
to preserve the symmetry of the stiffness matrix. This leads to the
mass-normalised stiffness:
! = M−1 2 K M−1 2
K
This is an example of matrix transformation
(NB: this is actually a special case of the Cholesky decomposition of
M into lower- and upper-triangular matrices
L and LT)
Matrix P and its transpose PT formed from normalised eigenvectors
of mass-normalised stiffness matrix can be used to decouple an n-
DOF system into n-single DOF systems
PT P = I

JRK7: 50
 Recap: 1st coordinate transformation
Equation of motion M!!
x + Kx = 0 (1)

with boundary conditions x(0) and x! (0)

First co-ordinate change:
x = M−1 2 q
Substitute into (1), & pre-multiply resulting equation by M−1 2

!! + M−1 2 KM−1 2 q = 0
M−1 2 MM−1 2 q
Then we have
Iq(t) !
!! + Kq(t) =0
q(t) !
!! + Kq(t) =0 (2)

JRK7: 51
 2nd transformation: Normal coordinates
Normal coordinates r: q(t) = P r(t)
r(t) = P −1q(t) = P T q(t)
Substitute into (2) P !! ! r(t) = 0
r(t) + KP
T !
Multiply by PT ( )
P T
P !!
r(t) + P ( )
K P r(t) = 0
(NB: PT = P-1)

We have: !!
r(t) + Λ r(t) = 0

JRK7: 52
 Final eqs. of motion can be written out

!!!
r1 (t) % (λ1 0 0 ... 0 +!r1 (t) %
# # * -# #
#!!
r2 (t)# *0 λ2 0 ... 0-#r2 (t)#
" &+ " &=0
#. # *. . . ... . -#. #
#$!! * -#
rn (t)' )0 0 0 ... λn ,$rn (t)#'
#

r1 (t ) + w12 r1 (t ) = 0
λi = ωi2
r2 (t ) + w22 r2 (t ) = 0
.....
rn (t ) + wn2 rn (t ) = 0
These equations are the modal equations. Note: uncoupled form

JRK7: 53
 Modal matrix
The modal matrix can be formed as
−1 2
S=M P
and its transpose and inverse are given by
T −1 2 T T −1 2
S = (M P) = P M
S−1 = (M−1 2 P)−1 = P −1M1 2 = P T M1 2

S combines 1st & 2nd transformations:

x(t) = M−1/2 q(t) = M−1/2 P r(t) = S r(t)

JRK7: 54
 The coordinate system defined by r(t) is
the modal coordinate system (a.k.a. normal coordinate system)

The initial conditions must also be transformed

r(0) = P T q(0) = P T M1 2 x(0) = S−1x(0)

r! (0) = P T q(0)
! = P T M1 2 x! (0) = S−1x! (0)
where S−1 = P T M1/2
The equations are decoupled, allowing known solutions
for a single DOF system to be employed.

ωi2 ri02 + r!i02 " −1 ω i ri0

%
ri (t) = Ai sin (ω i t + φi ) = sin $ω i t + tan '
ωi # r!i0 &

JRK7: 55
 ω12 r102 + r!102 " −1 ω1r10
%
r1 (t) = sin $ω1t + tan '
ω1 # r!10 &
ω 22 r202 + r!202 " −1 ω 2 r20
%
r2 (t) = sin $ω 2 t + tan '
ω2 # r!20 &
....
ω n2 rn02 + r!n02 " −1 ω n rn0
%
rn (t) = sin $ω n t + tan '
ωn # r!n0 &

JRK7: 56
 After deriving the solutions to the modal equations,
inverse transformation can be used to obtain x(t)

x(t) = M−1 2 q(t) = M−1 2 P r(t) = S r(t)

where S = M−1 2 P is the modal matrix

JRK7: 57
 Comparison of Coordinate Systems
Original Mass normalised Normal
coordinates coordinates q(t) coordinates
x(t) r(t)

✗ ✓ ✓
Symmetric
matrices

✗ ✓ ✓
Orthogonal
eigenvectors

✗ ✗ ✓
Decoupled
equations of motion

JRK7: 58
 Steps in solving multiple DOF by modal analysis
1. Calculate M−1 2

2. Calculate mass-normalised stiffness matrix: ! = M−1 2 K M−1 2

K

3. Solve symmetric eigenvalue problem ! = λv

Kv
4. Normalise vi to form P: P = [ v̂1 v̂ 2 v̂ 3... v̂ n ]
PT P = I
5. Calculate modal matrix S and its inverse S-1: S = M−1 2 P
S−1 = P T M1 2

JRK7: 59
 Steps in solving multiple DOF by modal analysis

6. Calculate the modal initial conditions

r(0) = S−1 x(0)
r! (0) = S−1 x! (0)
7. Compute components of r(t)
ωi2 ri02 + r!i02 " −1 ω i ri0
%
ri (t) = sin $ω i t + tan '
ωi # r!i0 &
8. Multiply r(t) by S to get final solution for x(t)

x(t) = S r(t)

JRK7: 60
 % Step 1. Calculate square root and inv square root of M
Msqrt = M^(0.5)
Minvsqrt = M^(-0.5)

% Step 2. Calculate mass normalised stiffness matrix

Kt = Minvsqrt * K * Minvsqrt

% Step 3. solve the symmetric eigenvalue problem

[P, D] = eig(Kt);
w = sqrt(diag(D)) % convert eigenvalues to frequencies

% Step 4. Normalise Vi to form P (already done by MATLAB)

P' * P

% Step 5. Calculate modal matrix S and its inverse

S = Minvsqrt * P;
Sinv = P' * Msqrt;

JRK7: 61
 % Step 6. Calculate modal initial conditions
x0 = [0.2 0. 0.]';
xdot0 = [0., 0., 0.]';
r0 = Sinv * x0;
rdot0 = Sinv * xdot0;

% Step 7. Compute components of r(t)

A = zeros(N,1);
phi = zeros(N,1);
for i = 1:N
A(i) = sqrt(w(i).^2*r0(i).^2 + rdot0(i).^2)/w(i);
phi(i) = atan2(w(i)*r0(i), rdot0(i));
end

% Step 8. Compute x(t)

t = 1.0 % or any arbitrary time
r = A .* sin(w*t + phi); % normal coordinates
x = S * r; % original coordinate system

JRK7: 62
7.6 Modal Analysis Worked Example
JRK7: 64
 Pure
Mode 1

! 0.5 %
# #
r0 = " 0 &
# 0 #
$ '
! 0 %
# #
r!0 = " 0 &
# 0 #
$ '

JRK7: 72
 Pure
Mode 2

! 0 %
# #
r0 = " 0.5 &
# 0 #
$ '
! 0 %
# #
r!0 == " 0 &
# 0 #
$ '

JRK7: 73
 Pure
Mode 3

! 0 %
# #
r0 = " 0 &
# 0.3 #
$ '
! 0 %
# #
!r0 = " 0 &
# 0 #
$ '

JRK7: 74
 Mixed
mode

! 0.2 %
# #
x0 = " 0 &
# 0 #
$ '
! 0 %
# #
x! 0 = " 0 &
# 0 #
$ '

JRK7: 75
 Example question for Friday session

JRK7: 76
 Summary
Modal analysis allows us to change a series of
coupled MDOF equations to a set of uncoupled
1DOF equations
– Change co-ordinates (from x to q)
– Solve symmetric eigenvalue problem
– Change co-ordinates again (from q to r)
– Solve equations using transformed initial conditions
– Get results back in original co-ordinates (from r to x)

JRK7: 77
 Activities for Unit 7
Unit Title Learn Apply
7.1 Overview of modal analysis & Today’s lecture Vevox +
review of eigenvalue approach Moodle Quiz
7.2 Generalized eigenvalue problems Video 7.2
(13 mins)
7.3 Matrix Transformation I Video 7.3
(8 mins)
7.4 Matrix Transformation Questions Attempt Qs,
then review
solutions
7.5 Matrix Transformation II Video 7.5
(30 mins)
7.6 Modal Analysis Questions Attempt Qs,
then review
solutions

JRK7: 78