Dyrobes Rotordynamics Software

0 8S® https://dyrobes.com

AN INTRODUCTION TO VIBRATION ANALYSIS USING MATLAB

by
E. J. Gunter, Ph.D., Fellow, ASME
RODYN Vibration Analysis, Inc.
Charlottesville, VA

Introduction and Background

The development of the computer from the late 1950's has resulted in more powerful
computers of smaller size and of such low cost that extensive computational power is available to
engineers at an affordable cost. The recent extensive advances in the power of microprocessors has
resulted in a generation of new desktop and portable computers that rival mainframe, and even super-
computers, of only a decade ago. ·
With these high-powered personal computers have also come the development of powerful
engineering software programs for use in computer-aided design and manufacturing (CAD/CAM) and
extensive finite element packages for the analysis of various statics, dynamics, and stress problems
of interest to the engineer. Although there is an extensive array of pre-programmed software
available to the engineer and scientist for various problems and tasks, this does not preclude his need
or desire to perform his own specialized programming tasks. ·

Computer Languages

Computer languages can be described in terms of levels. The lowest level is machine
language, which is closely tied to the computer hardware and is a binary language, since the
instructions are written in binary sequences of zeros and ones. An assembly language is used to
generate a binary code and also is machine specific. The instructions may resemble written
statements instead of binary code, but the language does not have a rich array of commands or
statements at its disposal. Thus, writing programs in assembly language can be very tedious and is
generally not done by the engineer for normal problem solution. Programs written in assembly
language are done in order to achieve the maximum speed of execution. Instrumentation that
contains microprocessors are programmed in assembly language in order to run at maximum speed.
These programs are called real time programs. For example, a controller for a magnetic bearing
needs to run in real time to operate without failure.
Engineers usually write programs in terms of a higher level language such as FORTRAN
(FORmula TRANslation). The initial version of FORTRAN was developed in the late 1950's for
solving engineering and scientific problems. FORTRAN is a third-generation, high-level language,
as compared to machine and assembly language programs. FORTRAN must first be compiled in
order to execute on a computer. FORTRAN has been constantly updated, with new versions such
as FORTRAN 77 and the current standard, FORTRAN 90. FORTRAN 90 has strong numerical
computational capabilities and many of the features and structures found in C. Although FORTRAN
has been the past choice of the scientific community, it lacks many features such as built-in graphical
capabilities, GUI (graphical user interface), and advanced matrix and file manipulation capabilities.
In order to enhance the capabilities of FORTRAN, various extensive library subroutines have been
developed, such as the IMSL library. The main problem with FORTRAN is that it is a compiled
language, rather than an interactive language, and hence the program must be compiled first, and
errors in syntax corrected, before the program may be run. This process can be extremely time
consuming and some bugs are difficult to trace and correct.
 In order to teach computer programming with greater ease, BASIC (Beginner's All-Purpose
Symbolic Instruction Code) was developed in the mid-1960's as an educational tool. Many PC
computers have been supplied with a BASIC computer language, such as GW BASIC and MS BASIC.
These BASIC programs are hardly adequate for scientific programming because of their lack of
structure and inability to operate with large matrices in a rapid fashion. However, a major advantage
of BASIC is that the language is interpretive in nature. That is, it is computed line by line, and hence
one could identify the location of an error in logic and syntax with greater ease. This made
debugging extremely easy. However, the normal BASIC languages lack speed and robustness for
consideration as a serious engineering platform. Simultaneous with the development of BASIC,
various Computer Science Departments were pushing the teaching of ALGOL and PASCAL as
teaching languages. These languages had little appeal to the practicing engineer as they were simply
lacking in any desirable mathematical or graphical features.

Matrix Theory in Vibration Analysis

Matrix theory had been well established by the 1950's as evidenced by texts such as
Hildebrand, 1960. However, from an engineering standpoint, matrix theory had little practical
application since it was impossible to compute even small order matrices without the aid of a
computer. The idea of inversion of large matrices and eigenvalue extraction was not in the normal
toolbox of practicing engineers without the aid of mainframe computers.
In the late 1970's the Hewlett Packard Corp. introduced a remarkable BASIC language which
was to be later referred to as Rocky Mountain Basic or simply HP BASIC. This language could
multiply and invert matrices. This lead to the possibility of using the numerical and transfer matrix
methods ofMyklestad, Prohl, and Lund in the direct matrix formulation of rotor dynamics problems.
The author presented a paper at the Vibration Institute in 1981 on Rotor Dynamics on The
Minicomputer. The HP 9845 computer with the HP Basic language used at that time cost over
$35,000. Although newer and more powerful and cheaper HP workstations were developed, the HP
Basic was abandoned in favor of C ( HP BASIC, however, is available as HT BASIC.). The
abandonment of HP BASIC with its extensive matrix and graphical commands was felt as a
considerable loss to the vibration analysis community.
The MATLAB ( or Matrix Laboratory) language was developed originally for UNIX in C and
was later ported to the PC computers. The current Windows NT computers with 128Meg of memory
are equivalent to mainframe computers of the previous decade ( only better since they are more user
friendly and in direct control of the engineer). This has given the engineer a new set of tools that
may be applied to control theory, finite elements, graphics, as well as vibration analysis. Currently
matrix theory is of great interest to the engineer and scientist as it now is easy to apply. The analogy
is that one does not need to be a car mechanic to drive a car ( a little bit of training, however, does
help to keep one from crashing! ). One does not need to be an expert in matrix theory to use it.
There are currently a number of excellent texts on vibrations and dynamics that make use of matrix
theory as listed in the references such as Bathe, Diamarogonous, Genta, and Kramer. Of particular
interest is the text by Kwon & Bang on The Finite Element Method using MATLAB.
In this paper some 111A TLAB examples are presented to illustrate the application of MATLAB
to vibration analysis. The examples represent about 1 % of the commands and functions available.
The commands presented are sufficient to enable one to solve a large class of problems from forced
response, transient analysis and eigenvalue analysis. One may also develop a multi-plane balancing
program using the such matrix functions as singular value decomposition.

1.2
2. MATLAB Language and Syntax

The syntax of MATLAB is very easy to master if one has familiarity with any of the higher
languages, such as FORTRAN, BASIC, or C. For engineers who are used to writing in FORTRAN,
a MATLAB file can be constructed to resemble FORTRAN. However, its power resides in its
extensive matrix capabilities. The language is constructed on the C platform and, hence, users who
are familiar with C or C++ will recognize some of the more advanced constructs of the language.
For simple problems, program commands may be generated at the MATLAB prompt. Since the
program is interpretive, an answer is given immediately. However, the real power of MATLAB
resides in the development of m files. The m file is simply an ASCII, or script, file representing a
set of commands with an extension of .m, instead of .txt for the normal text file. All MATLAB
script, or function files have the extension .m.

Common Functions

Table 2.1 represents some common functions used in MATLAB. The variable x may
represent a real or an imaginary number. In addition to the standard functions familiar to those used
in FORTRAN and BASIC, there are certain interesting functions in MATLAB that can operate on
complex numbers. These are ABS, ANGLE, REAL, and IMAG. These functions return the
absolute value of the complex number, the angle, and real and imaginary components. The ability
to deal with complex numbers makes the language particularly suitable for vibration analysis and
control theory. When entering a complex number with a real component of five and an imaginary
component of ten, one could use the following:

A = 5.0 + 10.0 i
or
A ~ 5.0 + 10.0*j

Note that either i or j could be used to indicate the imaginary value. Lower case must be used.
Complex numbers may be multiplied or divided, similar to real numbers.

Table 2.2 represents more advanced functions and operations. The stated functions, as shown
in Table 2.2, will provide the significant tools required for vibration analysis of a large class of
problems. Since most of the listed functions are m files, the functions may be viewed in detail and
even modified to suit a particular application.

Special Variables and Commands

ans - default name used for results

pi- default value= 3.1416
eps - smallest number when
whos - list of variables and sizes of arrays
clear - clears variables from work space
print - dwinc - send figure to color printer
help 'topics' - gives help on listed topics
diary (filename) - saves all terminal input and output to filename

2.1
Simple Matrix Commands

The ability to generate and manipulate matrices is the essence of MATLAB and is what
makes the language so powerful and attractive to use for engineering analysis. A few simple matrix
commands will be given to illustrate the ability to deal with matrices. The symbol>> represents the
interactive MATLAB command window.

>> A = -3:3 (specify array from -3 to 3)

A = -3 -2 -1 0 1 2 3
>> A = zeros (3) (specify 3 xJ matrix ofzeros)
A= 0 0 0
0 0 0
0 0 0
>> A =[ 1 -2 1 -2 3 -2 1 -2 4 ] (specify a 3 xJ matrix)
A =1 -2 1
-2 3 -2
1 -2 4
>> det(a) (determinant of a)
ans= -3
>> b = inv(a) (compute inverse matrix a)
b = -2.6607 -2.0000 -0.3333
-2.0000 -1.0000 0
-0.3233 0 0.3333
>> c = a*b (verify c = identity matrix)
C = 1.0000 0 0
0 1.0000 0
0 0 1.000
>> RT = [ 1 4 1 ] (generate row vector)
>> R=RT' (transpose ofRT to generate a column vector)
R= 1
4
1
>> X = A \R (solution ofR = A * A')
X = -11
- 6
0
>> X = inv(A) * R (solution of R = A * X by inversion ofA matrix)
X = -11
- 6
0

2.2
 Table 2.1 - Common Functions

angle(z) Angle of complex vector z, radians

abs(x) Absolute value or magnitude of complex number
acos(x) Inverse cosine
acosh(x) Inverse hyperbolic cosine
angle(x) Four-quadrant angle of complex number
asin(x) Inverse sine
asinh(x) Inverse hyperbolic sine
atan(x) Inverse tangent
atan2(x,y) Four-quadrant inverse tangent
atanh(x) Inverse hyperbolic tangent
ceil(x) Round toward plus infinity
conj(x) Complex conjugate
cos(x) Cosine
cosh(x) Hyperbolic cosine
det(A) Determinant of square matrix A, scaler
exp(x) Exponential: er
fu:(x) Round toward zero
floor(x) Round toward minus infinity
gcd(x,y) Greatest common divisor of integers x and y
imag(x) Complex imaginary part
lcm(x,y) Least common multiple of integers x and y
log(x) Natural logarithm
loglO(x) Common logarithm
max(x) Maximum value of vector x
real(x) Complex real part
rem(x,y) Remainder after division
Rem(x,y) gives the remainder ofx/y
round(x) Round toward nearest integer
sign(x) Signum function: return sign of argument, e.g.,
sign(l.2)=1,sign(-23.4)= -1,sign(0)=0
sin(x) Sine
sinh(x) Hyperbolic sine
sqrt(X) Square root
tan(x) Tangent
tanh(x) Hyperbolic tangent

2.3
 Table 2.2 - Advanced Functions and Operations

Plot(x,y) - plots arrays x and yin a graph and scales both x and y coordinates. X and y
must be the same dimension

y = det(A) - calculates the determinate of a matrix and returns scales value

b = Inv(A) - computes inverse of a matrix

xt = X ' - computes transpose of x

X=AIR - computes solution ofn simultaneous equations with {R} = [A] {X} where [A]
may be real or imaginary

{V, DJ= eig (A,B) - solves the general eigenvalue problem of A * V = B * V * D

where A and B are square matrices
V = eigenvectors and D is a diagonal matrix of the eigenvalues
or
{V, DJ= eig (A) - solves the eigenvalue problem of
A*V=V*D

[U,S, VJ = svd(A) - singular value decomposition of nxm A matrix

S = diagonal matrix of singular values
May be used for least squared error solutions and more accurate
eigenvalue compositions over QR

YI= Interpl (x,y,xl, 'cubic~ - Generates an interpolated set of points yl for given xl
vector using cubic spline interpolation

a = polyfit (x,y,n) - Generates the coefficients for an nth order polynomial such that
f(x) = ao xn + al xn- 2 + ... ¾-1 X + ¾

Y= polyval (a,x) - computes values ofy for vector x corresponding to polynomial with
coefficients a

Y = quad8( 'F ',A,B) - Numerical evaluation of integral F from A to B using 8-panel Newton-
Cotes 8-panel method

Y = FFT(x,n) - Performs FFT analysis over vector x with n points

[T,YJ = ODE23('F~Time,Yo) -Integrates the equations Y' = F(t,y) with initial condition Yo

2.4
. 3 MATLAB Applications For Vibration Analysis

In this section some simple examples will be presented on the use of MATLAB for vibration
analysis. The examples presented here are given in order to illustrate some of the power and ease
of use of the MATLAB command structure. One of the problems facing anyone attempting to use
a new programming language is the overwhelming number of commands available and the syntax
of the language. However, to perform many of the basic computations involved in vibration
analysis, one may be very effective with the use of only 1% of the commands available to the
programmer.

3.1 Fourier Approximation of a Square Wave

In Example 3.1 , a square wave is approximated by a Fourier series of 3 and 5 terms. It is a

simple procedure to extend the analysis to 10 or 20 terms by means of a recursion loop. The
equation for the square wave is given by:

4 cos(3ro *f) cos(Sro * f) cos(7 *ro *f) + ...)

x(t) = n x(cos(ro*f) - 4 + 5 7

Three & Five Term Approxmations of a Square Wave

1.5.------,-----,--------r- -- - , - - ---,--- - -, - - ---,i-----.

1 - --+ - - - - - - - 1- - - - - --4--- - - - - 1-- --

1 I

I I I I I I
0.5 - - _ _ _ T _ __ --, --- -- -r-- ---r-- -- - 7 - - - --,-- - - --r-- ---
1

I
I I I
0 ___ ___ .L_ _ _ _ _ _ I_ _ _ _ _ _ I__ _ ___ L ___ __ _ J. _ _ _ _ _ _I __ _
I I

I
-0.5 - - - - - .,. - - - - - ..., - - - - - -J- - - ---~-- - ---t- - _ __ _, ___ ---~ --
I I 1 I
I

-1

-1.5L___ _ ___..1__ _ ___J____ _ _. , _ __ _ _ , __ ___.__ _ _...,___ _ ---:----=:--- ~

-2 -1 .5 -1 -0.5 0 0.5 1 1.5 2
t
Figure 3.1 Three and Five Term Approximation of a Square Wave

3.1
 Table 3.l Fourier Series Script File

% sqrt3t.m
% Plot Fourier approximation to a square wave
t=-2: 0.005 2; % generate t vector with 801 elements
omega=2*pi; % pi is 3.1416 built in function
xl= cos (omega*t); % calculate xl vector with 801 elements
x2=-cos(3*omega*t) /3;
x3= cos (5*omega*t) /5;
x =4*(xl +x2 +x3) /pi;
plot(t,x, '-' ),grid % plot and scale the t vector vs x
title('Three & Five Term Approximations of a Square Wave')
xlabel ('t')
hold on % hold figure for a second plot

x4=-cos(7*omega*t) /7;
x5=cos(9*omega*t) /9;
x7=4*(xl +x2+x3+x4+x5) /pi;
plot(t,x7,'linewidth', 1)

Table 3 .1 represents the script file for the Fourier series representation of the square wave.
This simple file illustrates the generation of a vector t from -2 to 2 with steps of 0.005, having 801
elements. Therefore, the computations for xl, x2, and x3 represent vectors with the same dimension
as t. These elements also have 801 elements. The vectors of xl, x2, and x3 are automatically
generated by the calculations involving the vector t. Thus, the programmer does not need to specify
the dimension statement.

The graph is generated simply by the plot command ofplot (t,x), grid to generate a plot with
a grid. The scaling and labeling is automatically done. By using the command hold on, the graph
is kept open and a second plot is generated. A title may be added with the use of the title command.
Labels along the x and y axes are generated by xlable andylabel statements, respectively. If one
wishes to change the scaling in the x and y directions, one would use the command axis ([xmin,
xmax, ymin, ymax]). For example, if one wished to plot the x axis between -1 and+ I and they axis
from -2 to +2, the corresponding statement would be axis ([-1, 1, -2, 21). The capabilities of
MATLAB are such that one can use it for generating graphs, instead of using a standard plotting
program. There are many advantages to using MATLAB for graphics and plottip.g of functions, as
it is possible to plot in various colors, change line thickness, interpolate or curve fit experimental
data, or numerically differentiate or integrate the area under the curves.

A value at any point on the curve may be obtained by the command

[x,y] = ginput(n) ;
where:
n =no.of desired points
[x,y] = x, y vectors of coordinates of points

3.2
3.2 Forced Response of Single-Mass Systems

In this example, the forced response of a single degree-of-freedom system is presented. The
equations of motion for the system is given by

M X+ CX+KX = F0 cos co t

Figure 3.2 represents the dimensionless amplitude and dimensionless forces transmitted for a range
of~= 0.005 to 0.5.

Forced Response of Single Mass System For Various Damping Values

10 z=.05
1
E 8

I I I
Cl
w 6
Cl
I i ! I
:::>
r- 4 ··························I···················· ..•.:.. ·······················l····························l····························l···························.I.···························'··························
...J
a..
~
<( 2

0
! 1
! l ! I I
0 0.5 1 1.5 2 2.5 3 3.5 4
Frequency Ratio ,Dim
Forces Transmitted for Single Mass System For Various Damping Values
10
E
Cl
8
"O
.....
Q)
.....
.E 6
Cl)
C
(lJ
I,... 4
.....
(/)
Q)
0
I,...
2
0
LL
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Frequency Ratio ,Dim

Figure 3.2 Forced Response and Transmitted Forces for

a Single Degree-of-Freedom System

3.3
 Table 3.2 Single Degree-of-Freedom Unbalance Response Script File

% Unbalance Response of single Degree of Freedom System

% constant Forcing Function
% E.J. Gunter Dept of Mech & Aerospace Engineering
% A= amplitude vector for the various damping cases
% TRD = dynamic transmissibility = Ft/Fo
zeta=[0.05 0.1 0.2s 0.333 0.5];
%freq=o:o.os:2;
freq=linspace(0,4,200);

for k=1:s;
z=zeta(k);
for 1=1:200 % frequency sweep
f=freq(l) ;
f2=f•f;
n1=(1. 0 -f2) A2;
n2:(2*Z*f)-2;
d =sqrt(n1 +n2);
A(k,l)= 1.0/d ; % amplitude
n=sqrt( 1 + n2);
TRD(k,l)=n/d;
end;
end;
subpl.ot(2,1,1)
plot(freq,A)
grid
xlabel(' Frequency Ratio ,Dim');
ylabel('AMPLITUDE, Dim');
title('Forced Response of Single Mass system For Various Dam.ping
text(.B,.75, •z=.S');
text(1.1s,s,•z=.l');
text(l.OS,9.6,"z=.OS')

pause
subplot(2,1,2)
pl.ot(freq,TRD);
hold on
set(gca,'DefaultTextFont','Times','DefaultFontsize',14)
grid
xlabel(' Frequency Ratio ,Dim');
ylabel('Forces Transmitted, Dim');
title('Forces Transmitted for Single Mass system For various Dam.ping

In Table 3 .2, the calculation for the response was computed by using two nested do loops for k = 1:5
for the five values of damping, and l = 1:200 for the 200 values of the frequency. With 11M TLAB,
one or more of the do or for loops may be condensed, but for those familiar with FORTRAN, one
can write the program to be similar to FORTRAN. This procedure is recommended until one
becomes more familiar with the compact 11M TLAB matrix statements.

In Figure 3.2, two plots are generated using the subplot command. Subplot(2,1,1)
represents the first plot of two. The first plot is in the first column and first row. Subplot(2,1,2)
represents the second plot in the second row. By using the subplot command, one may have as
many plots on a page as desired. It should be noted that with the set command, one may change the
font type and size.

3.4
 Table 3.2 Single Degree-of-Freedom Unbalance Response Script File

% Unbalance Response of single Degree of Freedom system

% constant Forcing Function
% E.J. Gunter Dept of Mech & Aerospace Engineering
% A= amplitude vector for the various damping cases
% TRD = dynamic transmissibility = Ft/Po
zeta=[o.os 0.1 0.25 0.333 0.5];
%freq=O:O.OS:2;
freq=linspace(0,4,200)i
for k=1:5;
z=zeta C:lt) ;
for 1=1:200 % frequency sweep
f=freg(l) ;
f2=f•f;
n1=(1.o -f2)-2;
D2:(2*Z*f) ~2;
d =sqrt(n1 +n2);
A(k,l)= 1.0/d ; % amplitude
n=sqrt( 1 + n2);
TRD(k,l)=n/d;
end;
end;
subplot(2,1,1)
plot(freg,A)
grid
xlabel(' Frequency Ratio ,Dim');
ylabel('AHPLITUDE, Dim');
title('Forced Response of Single Mass system For various Damping
text(.B,.75, •z=.S');
text(1.15,s,•z=.1');
text(1.05,9.6,'z=.0S')
pause
subplot(2,1,2)
plot(freq,TRD);
hold on
set(gca,'DefaultTextFont','Times','DefaultFontBize',14)
grid
xlabel{' Frequency Ratio ,Dim');
ylabel('Forces Transmitted, Dim');
title('Forces Transmitted for single Mass system For various Damping

In Table 3 .2, the calculation for the response was computed by using two nested do loops fork= 1:5
for the five values of damping, and I= 1:200 for the 200 values of the frequency. With MATLAB,
one or more of the do or for loops may be condensed, but for those familiar with FORTRAN, one
can write the program to be similar to FORTRAN. This procedure is recommended until one
becomes more familiar with the compact MATLAB matrix statements.

In Figure 3.2, two plots are generated using the subplot command. Subplot(2,1,1)
represents the first plot of two. The first plot is in the first column and first row. Subplot(2,1,2)
reptesents the second plot in the second row. By using the subplot command, one may have as
many plots on a page as desired. It should be noted that with the set command, one may change the
font type and size.

3.4
3.3 Tuned Vibration Absorber

Figure 3 .3 represents a schematic figure of a tuned vibration absorber shown in Thomson,

Theory of Vibration With Application, page 167. Although the system has only two degrees-of-
freedom, the technique used in the solution is applicable to multiple degrees-of-freedom systems
with damping.

0
Figure 3.3 Tuned Vibrations Absorber
(Thomson)

In the problem shown by Thomson for the vibration absorber, no damping is included. This is easily
accomplished by first writing the kinetic, potential, and dissipative energies of the system.

·2 ·2
T=½MX
1 1
+½MX
2 2

(3.3.1)

•2 • • 2
D=½CX
I I
+½C2 (X-X)
2 I

The system equations of motion may be generated by LaGrange' s Equation of Motion, shown below
in its most generalized form, including the Rayleigh dissipation function and generalized forcing
function.

d ( ar) ar an av_F
- - - + - -+ - - -
dt ax ax ax ax x
(3.3.2)

Application ofLaGrange's Equation to (3.3.1) leads to the generalization of two coupled equations
in X 1 and X 2, as follows:

M1 .xI + ( ci + c2 ) xI - c2 x2 (3.3.3)
+ (K + K) X - K X
1 2 1 2 2
= U
1
al eirot

3.5
The general matrix form of the coupled linear equations of motion is

[M] {.X} + [C] {.X} + [K] {X} = 00 2 { U} eiwt (3.3.4)

Let X (t) =Xe icot for harmonic motion. The complex equations for forced response are

[ [K] - ro2 [ M] + i ro [ C] ] {X} = ro 2 { U} (3.3.5)

or

[A] {X} = {R} (3.3.6)

For the two degree-of-freedom problem shown with included damping, the complex a coefficients
are given by

Table 3.3 represents the script file for the dynamic vibration absorber. The system is assumed to
have a natural frequency at 1,850 RPM. The tuner is designed to be in resonance at 1,800 RPM,
which is the operating speed of the rotor. The weight of the absorber is 50 lb and the weight of the
rotor is 200 lb. The response is completed using two for loops for clarity to make the file similar
to FORTRAN. The computation of the absolute value of the complex motion is achieved by using
the·function abs.

3.6
 Table 3.3 Vibration Absorber Script File

% V2dABS.M Vibration Absorber For Rotor Operating at 1800 rpm

% Prob 5.37 pg 151 Thomson
% Rotor Weight W1 =200 l b , W2=50 lb absorber weight
% Absorber tuned for 1800 RPM
% kl value tuned for critical speed at 1850 RPM
% c2= absorber damper
elf; clear
W1:200; W2=50; % Main weight & absorber weight
m2=W2/386; m1=W1/386;
c2= [ 2 5 10 20 50 ]; % Values of damping-Lb-Sec/In
Ntuned=1800 ; % absorber frequency tuned to running speed
om2=(Ntuned/9.55)A2 ; %
k2= m2*om2 % Absorber spring rate tuned to 1800 RPM
omcr=1850/9.55
kl=ml*omcr"'2 % kl= major spring rate system tuned to 1850 RPM
um::2.0/386 % lb-in/g unbalance
freq=linspace(1000,3000,200); % Starting speed=1000 ,fina1=3000,200 cases
for n=1:5; % Loop thru 5 damping values
for l=l:200; % Loop thru 200 speed cases
omega=freq(l)/9.55; % Speed in Rad/sec
omega2=(freq(l)/9.55 )"'2 ; %
fu=omega2*um ; % Rotating unbalance force, Lb
a(1,1)= k1 + k2 -m1*omega2 + i*omega*c2(n) ; % Computer complex coef
a(2,1)=-k2 -i*omega*c2(n)
a(l,2)= a(2,l) ;
a(2,2)= k2-m2*omega2+ i*omega*c2(n) ; % influence coefficient matr:
b=inv(a); % Invert complex matrix
zl(n,l)=b(l,1)*fu ;
zlmils(n,l)=abs(z1(n,1))*1000 ;
z2(n,l)=b(2,l)*fu ;
z2r(n,l)=zl(n,l)-z2(n,l) ; % Relative absorber motion
z2rmils(n,l)=abs(z2r(n,1))*1000; % Amplitude in mils
z2mils(n,l)=abs(z2(n,1))*1000;
ftr=zl(n,l).*kl ;
ftrans(n,l)=abs( £tr) ; % force traaansmmm.itted
% trd(n,l)=k(n)/kd; % ftran(n,1)./fu ; % dynamic traansmissibility
end;
end;
subplot(2,1,1)
plot(freq,zlmils, 1 linewidth 1 ,l), grid
xlabel('Rotor Speed, RPM' )
ylabel( 'Amplitude, Mils')
title( 1 RESPONSE With VARIOUS TUNED ABSORBER DAMPING VALUES')
legend('C2=2','C2=5','C2=10','C2=20 1 , 1 C2=50 1 )
axis([l000 3000 0 800])

subplot(2,2,l)
plot(freq,ftrans, 1 linewidth 1 ,l), grid
xlabel('Rotor Speed, RPM' )
ylabel( 'Forces Transmitted , Lbs•)
title(• FORCES TRANSMITTED With Various ABSORBER DAMPER RATES')
axis( [ 1000 3000 0 3000 ] )

3.7
 The solution to the complex vector displacements {X} is given by

(3.3.7)

In MATLAB, this is performed either by direct inversion, such as the statement

X=Inv(A) * R

or

X=A/R

In the first method of solution, an inversion of the complex matrix A is performed similar to the
theoretical solution as shown in Eq. (3.3.7). Such a solution method can be time consuming for a
system with hundreds of degrees-of-freedom. A quicker solution is obtained by the second method.
In this method (shown with the backslash), an LU decomposition is performed, rather than a direct
inversion of the A matrix. This procedure is very efficient and may be used on systems with
thousands of degrees-of-freedom. This is the procedure of choice used in the majority oflarge finite
element system solvers.
Figure 3.4 represents the motion of the machine or rotor mass over a speed range with the
vibration absorber tuned to 1,800 RPM.. The response was run with five cases of damping, ranging
from 2 to 50 lb-sec/in damping in the absorber. Without any damping in the absorber, the motion
at 1,800 RPM would be zero. However, the mass of the absorber M 2 creates resonance frequencies
below and above the tuned absorber frequency of 1,800 RPM. The new frequencies are
approximately 1,430 RPM and 2,330 RPM. For the case ofno absorber damping, the amplitude of
motion at these two resonance frequencies would be unbounded. These responses become finite
with absorber damping applied.
As the absorber damping is increased, the amplitude of motion at the tuned frequency of
1,800 RPM increases. Therefore, the introduction of absorber damping is a compromise in order to
limit the motion at 1,430 and 2,330 RPM. There are two speeds in which the amplitude of motion
is independent of damping. These two speeds are at 1,500 RPM and 2,300 RPM. These two points
were referred to as the P and Q points for an absorber as first described by Den Hartog. An optimum
damping for the system could be chosen such that the slope is zero through the first (or P) point.
This would require a damping of about 12 lb-sec/in in the absorber. As the absorber damping
exceeds 50 lb-sec/in, the damper becomes locked up. The new natural frequency of the locked-up
system drops to 1,650 RPM due to the added mass of the absorber. Damping values in excess of 50
lb-sec/in would result in large amplitudes of motion and bearing forces transmitted.
Figure 3.5 represents the forces transmitted to the foundation. At running speed, the
unbalance force is 184 lb. Note that with only 2 lb-sec/in absorber damping, the transmitted force
at the 1st resonance is over 2500 lb. However if the absorber could be controlled magnetically, then
the absorber could be locked until the P point is reached. It is then unlocked until the Q point is
reached. At speeds above the Q point the dynamic vibration absorber is again locked up before it
moves into the second resonance speed. The transient motion of the rotor and absorber at P & Q
points due to unlocking and locking may be studied by the method outlined in the next section.

3.8
 RESPONSE Wrth VARIOUS TUNED ABSORBER DAMPING VALUES
140r---.--.----.--.---.---.-----,---.--~-~

- ! - ---+----+-:
!
+- c2k
+-c2::i5
~C2~10
~C2~0
I -t-~so

i:__ r~=~---
'
.!!l

..;....__ -I'-- - j - --1 ·- -r---

i

Pooo 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000
Rotor Speed, RPM

Figure 3.4 Amplitude of Rotor vs. Speed With Tuned Vibration

Absorber for Various Values of Absorber Damping

FORCES TRANSMITTED Wrth Various ABSORBER DAMPER RATES

2500 .- - - + - - ·i ~~~----.-----\
i .
Furb=t~ Lb t'eoo_
R
,t_
__ u_--+-_
.
I
~2000
l
...J

-0
4l!
.E
~
,
1500 --·--·-·-···-· ;.....- - ------+-+--il------'---
Ii--
l! . i
I- . 1

~ . ___ ____ ! - -···-· -

&1000
!;

500 ----·---··· ...

1'boo 1200 1400 1soo 1aoo 2000 2200 2400 2600 2aoo 3000
Rotor Speed, RPM

Figure 3.5 Forces Transmitted vs. Speed For Various Values of Absorber Damping

3.9
3.4 Transient Response of Vibration Absorber

The initial transient motion of the absorber may be computed by writing the equations in state space
formulation with four coupled first order equations in Y as follows

Y=f(Y,t) (3.4.1)

The derivatives of Y are given by

(3.4.2)

and the Y vector is given by

(3.4.3)

.MATLAB has a number of numerical integration procedures. In this example, the numerical
integration function ODE23 will be used. This is a low order numerical integration routine based
on a 2nd order Runge-Kutta integration method. For more information on this and other numerical
methods, at the .MATLAB >> prompt type

>>help ODE23

In order to apply the function ODE23 or any ofthe other numerical integration procedures, a separate
function file must be generated to have the derivatives of Y. The function file, called absorber.m
is shown in Table 3.4. The first line of the function file must start with the word "function" and the
file must be saved corresponding the function call. In this case, the file is saved as absorber.m. The
function call differentiates the function file from the main body, which is a script file. The majority
of the function calls of .MATLAB are ASCII files that may be modified.

Table 3.4 represents the functional file, which is stored as absorber.m. Note the important
statements
[rows, cols] = size (y)
yd= zeros (rows, cols)

These statements set dimension size and initialize the derivative matrix yd. The values of y(l)
through y(4) are not single-valued functions, but are variable vectors with the dimension oft. As
the absorber function is called, the variables are increased in dimension. The number of rows in the
matrices yd and y are determined by the number of time steps. One has only to set the initial time
(0) and the final time (I sec) and the increment and number of time steps are automatically
computed. This is in considerable contrast to a FORTRAN program in which the size of the matrices
must be initially set. This variable or dynamic dimensioning of the matrices is one of the powerful
features of MATLAB. The author has encountered on many occasions the statement of DIMENSION
OUT OF BOUNDS in a FORTRAN transient analysis program.

3.10
 Table 3.4 Function File for Absorber

function yd=absorber{t,y)
% 2 mass system vibration absorber,Ml main mass, Kl main stiffness
% M2 & K2 tuned to running speed at 1,800 RPM (omega=l94 rad/sec)
% Her= system natural frequency at 1,850 RPM
% y(2)=Xl motion, y(4)= X2 absorber absolute motion
% y(l)=Xl velocity, y(3)= X2 absorber absolute velocity
%global Ml M2 Kl K2 Cl C2 omega Fu
Ml:200/386; M2=50/386;
Cl:0.05; %Damping in rotor
C2=2 ; %Relative damping in absorber
Kl= 19444 % System spring lb/in-resonance at 1850 RPM
omega=lS00/9.55; % Operating speed, rad/sec
K2=M2*omega~2; % Absorber spring tuned to 1800 rpm
Fu:200; % Applied Forcing function, Lb
[rows,cols]=size{y); % * Set Matrix size, trows, 4 columns
yd=zeros(rows,cols); % * Set size matrix

% Specify the 1st order functional form

yd(l)=-{(Cl+C2)*y(l)-C2*y{3) +(Kl+K2)*y(2)-K2*y{4))/Ml+Fu*sin(omega*t)/Ml;
yd(2)= y{l); % Xl velocity
yd{3)=-( C2*( y{3)-y{l) ) + K2*( y{4) -y{2) ) )/M2; % X2 acceleration
yd(4)=y(3); % X2 velocity
% end absorber function

Table 3.5 represents the main program or script file that calls the numerical integration
procedure ODE23. Note that ODE23 calls the function absorber which contains the system first
order derivatives. The response is plotted for one sec of motion, as shown in Figure 3.6. The first
plot, using the sllbplo( command, is for the motion of the major mass. The array y (:;2) represents
the second column, which is the motion ofX1 for all the time values. In Figure 3.6, the top figure
represents the major mass.

It is seen that the maximum amplitude of motion after two cycles is approximately± 20 mils.
After one sec of motion, the amplitude reduces to ± 4 mils. In this example, an absorber damper
value of C2 = 2 lb-sec/in was assumed. Although the steady state response of the rotor mass is less
than 4 mils, it is seen that the sudden application of the dynamic load may cause excessive
amplitudes of motion. Therefore the final damping selected for the absorber may be a compromise
based on range of operation and transient considerations.

The second figure represents the absolute motion of the absorber. Note that after two cycles
of motion, the absorber peak amplitude is over ± 70 mils of motion. Thus, the absorber transient
motion is almost four times as great as the main mass. After one sec of motion, the system is
approaching steady-state conditions in which the absorber is moving with± 40 mils of motion. In
Fig 3.6 (b) there appears to be a beating motion caused by the suddenly applied unbalance. This
nature of the beating motion may be evaluated by performing an FFT analysis on the absorber.

3.11
 Table 3.5 Script File Tranabs.m for Transient Analysis

% tranabs.m transient motion of two mass absorber with suddenly

% applied unbalance
% 11/19/96 E.J. Gunter
% y(2)= xl main mass motion, y(4)= x2 absorber motion
% global Ml M2 Kl K2 Cl C2 omega Fu
elf
clear
Ml=200/386; % main mass
M2=50/386 ;
Cl=0.05 ;
C2=2.0 ;
N= 1800; omega=N/9.55 ; % speed rad/sec
IC2=M2*omega"'2 ; % tuned absorber to running speed
Kl:19444 ; % resonance at 1850 RPM original
Fu=200

t0=0.0 ; tf= 1.0 ; % initial & final time

% set initial conditions as column vector
yo = [ 0 • 0 ; 0 • 0 ; 0 • 0 ; O • 0] ; % v1 , xl , v2 , x2
[t,y]=ode23( 1 absorber 1 ,t0,tf,y0) ; % Call Numerical integrater

subplot(2,l,l)
plot(t,y(:,2)*1000, 1 linewidth 1 ,l.S),grid
xlabel('Time, Sec•)
ylabel( 1 Amplitude, Mils')
title('Motion With Suddenly Applied Unbalance With Tuned Absorber,C2=2')

subplot(2,l,2)
plot(t,y(:,4)*1000,•linewidth•,l.S),grid
xlabel('Time, Sec•)
ylabel('Amplitude, Mils')
title ( '~sorb~r Motion With !3ulf<ienlv Jlppl:i,P<i Unbalance , C2=2 •)

The classical method of solution of a transient response is by means of a convolution integral of the
form:

For the case of a damped single degree of freedom system, the convolution function of the integral
equation, h(t) is given by:

The convolution integral is difficult to evaluate for even the simplest of forcing functions. The
theory also is applicable only to linear systems. For more complex systems, a damped eigenvalue
analysis of the roots is first required. Therefore the classical method of transient analysis by
convolution integral is obsolete. It should also be pointed out that a computer program is always
required to plot out the results. MATLAB has a variety of equation solvers to handle nonlinear and
stiff equations.
3.12
 Motion With Suddenly Applied Unbalance With Tuned Absorber,C2=2
20..---..----,----,----..----.---.-----.-----.---....--~

10 -ilw---•-· .·..
.
-..... .......~_ ..;..................... --.j ...... . ............... i.--.................... c---------------.. ··-· · ...................... ;. .......................................... .. .
: i ' ! ! ! ! !
0 .. .... ---- .. ... .... ... .. ...... .. ... .. --1 .. ......

I l
1

-20
i I ! \ ;

I _ _.,___
-30 .___ _.,___ I _ _I.,____ _.,____ _!,....__ _,....__ _i,....__ _.,___
I _ _.,___ i_ ___,
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time, Sec

(a) Initial Transient Motion of Rotor For 1 Second

(b) Initial Transient Motion of Absorber For 1 Second

Figure 3.6 Transient Motion of Rotor and Absorber With Suddenly-Applied Unbalance
3.13
3.5 fft Analysis of Transient Motion

In order to better understand the transient motion and the beating effect observed in the
absorber motion, an/ft (fast Fourier Transform) analysis using MATLAB was performed . Table 3.6
represents the additional commands required to perform the fft. Since this function is included in
only the student version, the Signal Processing Toolbox is required. Figure 3.7 shows the fft
analysis of the absorber. The frequencies seen in the signal are 30 Hz, running speed, and the two
resonance frequencies at approximately 24 Hz and 39 Hz. Without the use of the fft it would be
difficult to determine the presence of the second resonance frequency at 39 Hz.

Table 3.6 FFT Analysis Statements

% perform fft on main mass using 1024 points, use whos to insure that
% ntotal >1024 to use fast fourier transform, else must use slow dff

n:1024;
ttotal=t(n); fs=l024/ttotal; % sample frequecy
ts=l/fs ; % time per sample
k=0:n-1 % counter for plotting
dhz=fs/n ; % frequency increment of £ft
flcomplex=fft(y(:,4),n) ; % compute complex fft for frequency content
fl=abs(flcomplex) ; % convert to absolute magnitude
hertz=k*dhz ; % frequency steps
subplot(2,1,2) ; % plot fft vs hZ
stem( hertz(l:n/2) , fl(l:n/2) ) % plot using vertical lines
xlabel('Frequency ,Hz')
ylabel('Amplitude')
title('FFT Analysis of Absorber Mass•)
axis(r 0 60 o 20])

FFT of TRANSIENT MOTION of MAIN MASS

4.--------,r------~-------,---------.------.-------,
24 Hz -1st Resonance
3.5

3
30 Hz -Running Speed
2.5

1.5

Figure 3. 7 FFT Analysis of Transient Motion of Dynamic Vibration Absorber

3.14
ACKNOWLEDGMENT

The author wishes to acknowledge Robert L. (Rob) Howes, Senior Engineer- Acoustics and
Vibrations of Cessna Aircraft, Wichita, Kansas for demonstrating to me the various applications
of MATLAB in the aerospace industry and directing me towards the study and application ofthis
language to vibration analysis and data processing.

REFERENCES

1. Bathe, K., and Wilson, E.L., Numerical Methods in Finite Element Analysis, Prentice-Hall,
Englewood Cliffs,NJ, 1976
2. Biran, A. and Breiner, M .. , MATLAB For Engineers, Addison-Wesley Publishers, Ltd,
Great Britain, 1995 ..
3. Den Hartog, J.P., Mechanical Vibrations, McGraw-Hill, Inc.,New York,NY, pp 112-132
4. Diamarogonas,A., Vibration For Engineers, Prentice Hall, Upper Saddle River, NJ, 1995
5. Ehrich, F. F., editor, Handbook of Rotordynamics, McGraw-Hill, Inc., New York, NY,
1992.
6. Genta, G., Vibration of Structures and Machines, Springer-Verlag ,New York,NY, 1995
7. Gunter, E.J., "Rotor Dynamics on The Minicomputer", Proceedings Vibration Institute,
April 1981
8. Gunter,E.J., and Sharp,W.L.,"Minicomputer Aids Rotor Research ",Industrial Research and
Development, May 1981,pp. 122-127
9. Hanselman, D. and Littlefield, B., Mastering MATLAB 5, Prentice-Hall, Inc., Englewood
Cliffs, NJ, 1998.
10. Hildebrand, F.B., Methods ofApplied Mathematics, Prentice Hall Englewood Cliffs, NJ,
1960
11. Kramer,E., Dynamics of Rotors and Foundations Springer-Verlag ,New Y ork,NY, 1993
12. Kwon, Y. W. and Bang, H., The Finite Element Method Using MATLAB, CRC Press, Inc.,
Boca Raton, FL, 1997.
13. Leonard, N. E. and Levine, W. S., Using MATLAB to Analyze and Design Control
Systems, Addison-Wesley Publishers, Redwood City, CA, 1995.
14. Myklestad, N. 0., Fundamentals of Vibration Analysis, McGraw-Hill, New York, NY,
1956
15. Moler, C. and Costa, P. J., MATLAB Symbolic MathToolbox, The MathWorks, Inc.,
Natick, MA, 1997 . .
16. Sigmon, Kermit, MATLAB Primer, Fifth Edition, CRC Press, Boca Raton, FL, 1998.
17. The MathWorks, Inc., MATLAB User's Guide, The MathWorks, Inc., Natick, MA, 1993.
18. Thomson, W. T., Theory of Vibration With Applications, Fourth Edition, Prentice-Hall,
Inc., Englewood Cliffs, NJ, 1993.
19. Wilson, H.B., and Turcotte, Louis H., Advanced Mathematics and Mechanics Applications
Using MATLAB, CRC Press, Inc., Boca Raton, FL, 1994.