FINITE ELEMENT MODEL UPDATING FOR

ROTARY MACHINERY

A Thesis

presented to

the Faculty of California Polytechnic State University,

San Luis Obispo

In Partial Fulfillment

of the Requirements for the Degree

Master of Science in Mechanical Engineering

by

Nathalie Gotin

September 2012

i
 © 2012

Nathalie Gotin

ALL RIGHTS RESERVED

ii
 COMMITTEE MEMBERSHIP

TITLE: Finite Element Updating for Rotary Machinery

AUTHOR: Nathalie Gotin

DATE SUBMITTED: September 2012

COMMITTEE CHAIR: James Meagher, Professor

COMMITTEE MEMBER: Hemanth Porumamilla, Assistant Professor

COMMITTEE MEMBER: Wolf-Immo Jutzler, Professor Dr.Ing.

iii
Abstract

Finite Element Updating for Rotary Machinery

Nathalie Gotin

The main approach of this thesis was to develop a mathematical model that

represents a rotary machine. Experimental data was used to define a finite

element model (FEM). In order to obtain the experimental data, the rotary

machine had to be balanced. An impact hammer test made it possible to obtain

frequency response functions (FRF). The frequency response functions were

curvefitted in order to obtain the mode shapes and natural frequencies.

Mathematical models have been created with ABAQUS and Matlab. For the

Matlab Model the assumption has been made that the rotor machine consists of

a specific number of beam elements. The FEM matrices have been reduced with

the Guyan Reduction Method to coincide with the DOFs of the experiment.

Applying the method of the least square to an Error Function made it possible to

obtain new values for the stiffness and damping of the bearings (

). This made

it possible to update the mathematical model. By applying the Model Assumption

Criterion the theoretical model and those detected from the experimental

measurement could be validated. The correlation for Mode Shapes 1 could be

improved from 0.6647 to 0.8186 and for Mode Shape 2 from 0.0209 to 0.4208.

Therefore, the created method could be proven to work. Additionally the whole

theory has been validated with a very simplified model.

iv
Contents

List of Tables ....................................................................................................... vii

List of Figures ..................................................................................................... viii
List of Abbreviations ............................................................................................. x
Chapter 1: Problem Statement and General Terms and Definitions ................. - 1 -
Problem Statement ....................................................................................... - 1 -
Finite-Element Modeling ............................................................................... - 2 -
Finite-Element Model Updating ..................................................................... - 3 -
Vibration Analysis ......................................................................................... - 4 -
Frequency Response Function.................................................................. - 6 -
Impact Test.............................................................................................. - 12 -
Curve fitting ............................................................................................. - 14 -
Data Plots ................................................................................................ - 14 -
Rotating machinery structures .................................................................... - 16 -
Dynamic Stiffness and Rotor Behaviour ..................................................... - 22 -
Dynamic stiffness .................................................................................... - 23 -
Synchronous Rotor Behavior ................................................................... - 24 -
Chapter 2 – Experimental Measurements ...................................................... - 25 -
Balancing .................................................................................................... - 25 -
Balancing with determining the Heavy Spot location ............................... - 26 -
Multiplane Balancing With Influence Vectors........................................... - 26 -
Weight Splitting ....................................................................................... - 27 -
Experimental Setup Balancing Rotary Machine .......................................... - 28 -
Obtaining Data ........................................................................................ - 31 -
Test: Unbalance Setting .......................................................................... - 32 -
Test: Obtaining Mode Shapes/ Natural Frequencies ............................... - 34 -
Mode Shapes .......................................................................................... - 35 -
Natural Frequencies ................................................................................ - 36 -
Experimental Setup Hammer Tip Testing ................................................... - 38 -
Dactron PhotonTM .................................................................................... - 40 -
FFT Windows .......................................................................................... - 42 -
Obtaining Data ........................................................................................ - 44 -
Animation with MEScope ............................................................................ - 47 -
Results .................................................................................................... - 48 -
v
Chapter 3 – Finite Element Model with Abaqus .............................................. - 51 -
Modeling – Introduction ............................................................................... - 51 -
Structured process to model the physical rotor system ............................... - 51 -
Statement of Assumptions....................................................................... - 51 -
Acting forces on the system .................................................................... - 52 -
Derivation of equation of motion .............................................................. - 53 -
Solution of equation of motion ................................................................. - 53 -
Modeling rotary machinery in ABAQUS CAE .............................................. - 53 -
Units ........................................................................................................ - 54 -
Part Module ............................................................................................. - 54 -
Property Module ...................................................................................... - 58 -
Assembly Module .................................................................................... - 60 -
Step Module ............................................................................................ - 61 -
Interaction Module ................................................................................... - 62 -
Meshing Module ...................................................................................... - 63 -
Results .................................................................................................... - 65 -
Chapter 4 – Matlab Finite Element Model Creation ........................................ - 67 -
Bernoulli-Euler Beam Theorem ................................................................... - 68 -
Assemblage of Matrices .............................................................................. - 70 -
Current Model ............................................................................................. - 71 -
Programming current model .................................................................... - 73 -
Guyan Reduction Method ........................................................................ - 76 -
Programming Guyan Reduction Method ................................................. - 79 -
Error function .............................................................................................. - 80 -
Verification of Error Function ................................................................... - 86 -
Model Assurance Criterion .......................................................................... - 88 -
Chapter 5 – Conclusions ................................................................................ - 92 -
References ..................................................................................................... - 94 -
Appendix A Start and use of ADRE for Windows/ DAIU ................................. - 96 -
Appendix B Balancing of two plane rotary machine ...................................... - 101 -
Appendix C Orbit Plots obtained with ADRE ................................................ - 102 -
Appendix D RT Pro Focus 6.21 .................................................................... - 104 -
Appendix E MEScope ................................................................................... - 108 -
Appendix E Remaining MATLAB code ......................................................... - 113 -

vi
List of Tables

Table 1 Recommendations for different window types [4] .............................. - 42 -

Table 2 System of Units ................................................................................. - 54 -
Table 3 Degree of Freedom/ Orientation ........................................................ - 58 -
Table 4 Comparison Natural Frequencies ...................................................... - 79 -
Table 5 Mass Normalized Shape Functions ................................................... - 84 -
Table 6 MAC Mode Shape 1 before updating ................................................ - 89 -
Table 7 MAC Mode Shape 2 before Updating ................................................ - 90 -
Table 8 MAC Mode Shape 1 after Updating ................................................... - 90 -
Table 9 MAC Mode Shape 2 after Updating ................................................... - 91 -

vii
List of Figures

Figure 1 Time Data ........................................................................................... - 6 -

Figure 2 Components of the Frequency Response Function ........................... - 7 -
Figure 3 Anatomy of an FFT Analyzer .............................................................. - 8 -
Figure 4 Response of Simple Line.................................................................. - 10 -
Figure 5 Mode 1 from fifth Row of a FRF Matrix ............................................. - 11 -
Figure 6 Cascade ........................................................................................... - 11 -
Figure 7 Excitation and response location ...................................................... - 14 -
Figure 8 Modal Parameter Estimation ............................................................ - 14 -
Figure 9 Different Measurement Techniques ................................................. - 14 -
Figure 10 Bode Plot ........................................................................................ - 15 -
Figure 11 Free Body Diagram of a Rotor ........................................................ - 17 -
Figure 12 Rotor position vector....................................................................... - 17 -
Figure 13 Vibration and Force response vectors ............................................ - 22 -
Figure 14 Conversion of dynamic input forces to output vibration .................. - 22 -
Figure 15 Weight Splitting .............................................................................. - 28 -
Figure 16 Rotor Kit Components .................................................................... - 29 -
Figure 17 Rotor Kit Speed Controller .............................................................. - 30 -
Figure 18 Rotor Kit ......................................................................................... - 31 -
Figure 19 Partial view of ADRE for Windows.................................................. - 32 -
Figure 20 Polar Plot ........................................................................................ - 33 -
Figure 21 RK 4 Rotor Kit Proximitor Assembly ............................................... - 35 -
Figure 22 First Mode Shape ........................................................................... - 36 -
Figure 23 Second Mode Shape ...................................................................... - 36 -
Figure 24 Bode Plot ........................................................................................ - 37 -
Figure 25 Impact hammer instrument ............................................................. - 38 -
Figure 26 Excitation and response location .................................................... - 40 -
Figure 27 Dactron Focus II Real-Time Analyzer and Dactron Photon ............ - 40 -
Figure 28 Input Dactron Signal Analyzer ........................................................ - 44 -
Figure 29 Exciting the rotary machine at location l and measuring at loc k .... - 45 -
Figure 30 Excitation at location 2.................................................................... - 46 -
Figure 31 Window MEScope .......................................................................... - 47 -
Figure 32 Simplified Model in Structur Window of MEScope .......................... - 48 -
Figure 33 Curve Fitting ................................................................................... - 49 -
Figure 34 Experimental Mode Shapes 1 and 2 .............................................. - 50 -
Figure 35 Definition of Coordinate System ..................................................... - 52 -
Figure 36 Abaqus Creat Part and Edit Feature(right figure) ........................... - 55 -
Figure 38 Sketcher of rotary machinery .......................................................... - 55 -
Figure 39 Edit Feature .................................................................................... - 56 -
Figure 40 Create DATUM CSYS .................................................................... - 56 -
Figure 41 Modeling Bearing and Coupling ..................................................... - 57 -
Figure 42 Create Partition .............................................................................. - 57 -
Figure 43 Partition Edge and Cell and Create Partition 2 (right figure) ........... - 58 -
Figure 45 Edit Section .................................................................................... - 60 -
Figure 46 Instance Part .................................................................................. - 61 -

viii
Figure 47 Edit Boundary Condition ................................................................. - 62 -
Figure 48 Create Constraint and Edit Constraint ............................................ - 62 -
Figure 49 Constrain on outer surface of beam ............................................... - 63 -
Figure 50 Partition of Rotor Disk..................................................................... - 63 -
Figure 51 Regular cube Region modeled with a Single Hexahedron ............. - 64 -
Figure 52 Results Step/Frame ........................................................................ - 65 -
Figure 53 Results Mode 3 which is the first Mode Shape ............................... - 66 -
Figure 54 Results Mode 5 which is the second Mode Shape ......................... - 66 -
Figure 55 Uniform element undergoing transverse deflection ........................ - 68 -
Figure 56 two beam elements with attached springs ...................................... - 70 -
Figure 57 Simplified model ............................................................................. - 72 -
Figure 58 Mode Shape 1 for natural frequency 67 rpm .................................. - 76 -
Figure 59 Mode Shape 2 for natural frequency 560 rpm ................................ - 76 -
Figure 60 Splitting Matrices ............................................................................ - 78 -
Figure 61 Trigger Event Configuration ............................................................ - 96 -
Figure 62 Sample Mode Configuration ........................................................... - 97 -
Figure 63 Channel Configuration .................................................................... - 98 -
Figure 64 Plot Control..................................................................................... - 98 -
Figure 65 Polar Plot Configuration.................................................................. - 99 -
Figure 66 Bode Plot Configuration................................................................ - 100 -
Figure 67 Real Time Plot Configuration ........................................................ - 100 -
Figure 68 Orbit Plot First Mode Shape, Probe 7 ........................................... - 102 -
Figure 69 Orbit Plot Second Mode Shape, Probe 7...................................... - 103 -
Figure 70 Setting Parameters for Forced/Exponential Window .................... - 106 -
Figure 71 Main Layout of ME Scope ............................................................ - 108 -
Figure 72 Units ............................................................................................. - 109 -
Figure 73 Assigning DOFs ........................................................................... - 110 -

ix
List of Abbreviations

ADC Analog to digital conversion

CAE Computer Aided Engineering

DOF Degree of Freedom

DSA Dynamic Signal Acquisition

DAQ Data Acquisition

DAIU Data Acquisition Interface Unit

FRF Frequency Response Function

FEM Finite-Element Modeling

FFT Fast Fourier Transformation

ODS Operation Deflection Shape

x
Chapter 1: Problem Statement and General Terms and

Definitions

Problem Statement
The primary approach of the work is to perform a finite-element modeling for a

rotary machine.

To perform the finite-element modeling the physical structure of the actual rotary

machine has to be translated into a mathematical model by means of a

mathematical and numerical process. The mathematical model is used to

estimate such dynamic characteristics as mode shapes and frequencies.

To better reflect the actual structure, the mathematical model is updated.

Updating involved the process through which the mathematical model is tuned in

a way to better reflect the actual structure. Therefore the assumption has to be

made that the experimental data is correct and that the mathematical model is

adapted.

The finite-element model matrices have to be equated to those of the measured

modal data; therefore the physical structure has to be translated into a

mathematical model with the Guyan Reduction Method.

Since the rotary machinery behavior is primary controlled by the dynamic

stiffness, the unknown parameters for the stiffness and damping matrices have to

be estimated. To estimate those an Error Function is applied to the data

(involving theoretical and reference data) and solved by using method of the

-1-
least square. Finally the theoretical and reference data are compared using the

Model Assurance Criterion. [1]

Finite-Element Modeling
Finite-Element Modeling is the mathematical and numerical process that

translates a physical structure into a mathematical model. The dynamic

characteristics can be estimated from this mathematical model. Those include

the natural frequencies and mode shapes.

Originally FEM has been applied in complex elasticity and structural analysis

tasks in the field of aerospace and civil engineering.

When developing a mathematical model discrepancies occur to the actual

physical structure. This may be due to model structure errors and model

parameter errors. Structure errors occur among others due to the difficulty to

model damping and stiffness. Model parameter errors occur due to the

complexity of identifying the correct material properties. In the scope of this work

the assumption is made, that the measurements are correct, which excludes the

reason of the discrepancy between the mathematical model and the actual

physical structure due to measurement errors.

To model a structure, it is divided into a number of smaller bodies (finite

elements). These finite elements are connected to each other at points (nodes).

Therefore the whole structure can be analyzed without solving the problem for

the entire body, but by formulating equations for every single finite element,

which are then combined to receive the solution for the entire structure.

-2-
If the goal of developing a Finite-Element Model that represents the physical

structure entirely is achieved, analysis can be done in order to improve the actual

physical structure. [1] [10]

Finite-Element Model Updating

When performing an analysis of the mathematical obtained finite-element model

it can be seen that the results do not necessarily coincide with the results

obtained from the measured (reference) data.

There are two different methods to execute the updating of a mathematical

model;

 the direct method and the

 iterative method.

The direct method reproduces the measured data precisely. It uses the modal

properties.

The iterative method updates the finite-element model by changing its physical

parameters. Reviewing previous research papers elucidates that just lately

research has been done in the field of finite-element updating using an iterative

procedure.

One way to apply the iterative method is to update the structural model matrices

(e.g. the mass, stiffness and damping matrices). Thereby the distance between

the analytical and the measured data is minimized.

* + ( , - , - , -) * + (1)

-3-
Where , - is the mass matrix, , - the damping matrix, , - the stiffness matrix,

* + the ith mode shape, is the ith natural frequency, * + the the error vector

(residual force) and √ . The relevant parameters are changed until the error

is minimized.

To be able to execute the Error Function the Degrees-of-Freedom of the Finite-

Element Model and from the measurement have to be adapted. One way is to

use the reduction and expansion method, with which the measured modal data

and the Finite-Element System Matrices will be equated.

There are different reduction methods:

 static reduction (Guyan, 1965)

 dynamic reduction (Paz,1984)

 improved reduced system (O’Callahan,1989)

 system-equivalent-reduction process (O’Callahan et al., 1989)

To obtain the measured data, which will serve as the reference data for the

mathematical model, a Vibration Analysis has to be executed. [1] [10]

Vibration Analysis
To execute the Vibration Analysis, the structure of the rotary machine has to be

excited. Therefore a modal hammer is used to excite the structure. The raw data

is then measure in the time domain and by applying the Fast- Fourier Transform

the data is transformed into the frequency domain.

The modal properties are extracted from the Frequency Response Function

using the procedure modal analysis. The Modal Properties are

 frequency,

-4-
  damping

 and the mode shapes.

It is important to identify where the frequencies occur and in which way they

might affect the response of the present structure. Modal Analysis is used in all

fields of mechanical engineering, including the automotive, airplane, computer

and various other industries. By understanding where the frequencies and mode

shapes occur makes it possible for the engineer to find the best solution in terms

of noise and vibration applications.

The vibration deflection shape of a rotating structure is referred to as the mode

shape. For every natural frequency occurring in a structure there is a

characteristic deformation shape. It describes the axial distribution of the

vibration amplitude and phase along the structure. They are defined by the

material properties, mass, stiffness and damping.

Experimental modal parameters are obtained from measured Operating

Deflection Shapes (ODS). Operating Deflection Shapes are the visualization of

motion of two or more points on a structure. The motion at a point in a direction is

called a Degree of Freedom (DOF). Since both, location and direction are

correlated with motion it is a vector quantity.

In the experimental modal analysis the actual physical structure is excited and

the motion of each chosen point (DOF) is measured.

At a resonance peak the Operation Deflection Shape is dominated by the natural

frequency and therefore can be seen as a good approximation to the mode

shape. [5] [20]

-5-
Frequency Response Function
To execute a modal analysis a structure is excited with a measureable force. By

applying the Fast Fourier Transform to the time data, the Frequency Response

Functions (FRF) can be obtained. Figure 1 shows the time data of the

displacement at a sensor, while executing a hammer test.

Figure 1 Time Data

Therefore it can contain both, the real and imaginary parts and components.

FRFs allow isolating the dynamic properties and modal parameters of a

mechanical structure. A FRF is defined as a single input DOF (point and

direction) and a single output DOF. The Frequency Response Function describes

therefore how much displacement, velocity or acceleration has at an output DOF,

per unit excitation force at an input DOF.

-6-
Figure 2 Components of the Frequency Response Function

The peaks appear at the natural frequencies and can be seen in Figure 2. When

overlaying time trace with the response trace it can be seen that the amplitude

peaks occur at the same time. Therefore either the time trace can be used by

determining the frequency at which the amplitude reaches its maximum or the

response trace by determining where the resonant frequencies occur.

In the following Figure 3 the basic steps to obtain the FRF from the measured

data is explained. First the analog signals have to be filtered to prevent aliasing.

Then the analog signal is converted by the analog to digital converter (ADC) into

a digital representation of the actual signal. To prevent leakage weighting

functions (windows) are used. As the data is sampled the Fast Fourier

Transformation can be computed and forms linear spectra from the input

excitation and the output response. These are averaged and used to obtain the

Frequency Response Function and the Coherence. Coherence is a function that

represents how much the output signal is related to the input signal. FRF

-7-
contains information about the damping, the system frequency and important

information considering the mode shapes at the measured location.

Figure 3 Anatomy of an FFT Analyzer

If a structure is equipped with a certain amount of measurement locations, there

are the same numbers of possible places where the forces can be applied. To

assign the input and the output location of the FRF, subscripts are used: hout,in

(matrix notaion: hrow,column).

FRF contains information about the damping, the system frequency and

important information considering the mode shapes at the measured location. It

is the ratio of the output response (which can be measured as the displacement,

the velocity or the acceleration) as a result of the applied force. This means it

describes how a structure affected by a certain force causes it to respond.

-8-
The linear model for a vibratory structure can be written as

* ( )+ , ( )-* ( )+ (2)

* ( )+ is the -vector of Fourier transformed displacement response, , ( )- is

the(n by n) matrix of FRFs and * ( )+ is the -vector of Fourier transformed

inputs.

In the , ( )-, which is the ( ) matrix of FRFs the row corresponds to the

response points and the columns correspond to excitation points. Any particular

row or column contains sufficient information to compute the complete set of

frequencies, damping and mode shapes.

[ ]
The FRF matrix can be also expressed in terms of modal parameters:

, - [ ]
, ( )- ∑ (3)
( ) ( )

, - is the ( ) matrix of residues for mode (k), is the complex

pole location for mode ( ) and is the frequency for mode ( ). stands for the

complex conjugate.

From the matrix of residues for modes the mode shapes can be obtained. They

are related by , - * +* + , where is the mode shape of the

corresponding mode k and is the scaling vector for mode k.

Therefore the transfer function along the -axis in the S-plane is the Frequency

Response Function. The transfer function is of complex value, since the Laplace

variable is complex valued. [5] [11]

-9-
Mode Shapes
Depending on the frequency with which the structure is excited, the deformation

has different deformation patterns. At each one of the natural (resonant)

frequencies a specific deformation pattern in the structure is excited. As can be

seen in the Figure 4 below the deformation pattern at the first natural frequency

is bending, the second natural frequency has a twisting deformation pattern.

Figure 4 Response of Simple Line

Deformation patterns are referred to as mode shapes of a structure. The weight

and stiffness of a structure disposes where the natural frequencies and the mode

shapes will exist.

To extract the mode shapes from measured data the peak amplitude of the

imaginary part of the FRF has to be regarded. Here (Figure 5) all the imaginary

parts of the single FRFs for the the 5th row are shown. This means that the

structure was excited at every point and it was measured at one reference

location. The peak amplitude of the imaginary part (j ) at the same frequency in

every FRF is joint to show the shape of the first mode. Therefore it can be seen,

- 10 -
that just the measurement of a row can be used to describe the mode shape of

the system.

Figure 5 Mode 1 from fifth Row of a FRF Matrix

It is important to consider that the reference point cannot be at the node of a

mode. If this is the case the mode is not visual in the frequency response

function and therefore the mode cannot be obtained. To visualize the mode

shapes even better, it is convenient to show all frequency response functions in a

waterfall or Cascade style plot (Figure 6). [11]

Figure 6 Cascade

A drive point is where the structure is measured and excited at the same point.

In the experiment six measurement points are used. If the structure is excited

with the impact load at point 1 and the measurement is done at the same

- 11 -
location, the measurement is referred to as h1,1. Typical of a drive point

measurement is that the resonance frequencies are all separated by anti-

resonances. Furthermore the phase loses 180∘when the resonance is passed

and gains 180∘when the anti-resonance is passed. Additionally all peaks of the

imaginary part point in the same direction.

Usually in frequency response functions reciprocity is found. This means that the

frequency response function is symmetric (hi,j=hj,i). Therefore not all terms of the

frequency response function have to be measured. Since in this thesis the rotary

machinery is considered, the circumstances are different. [5]

Impact Test
Principally there are two different test cases to obtain vibration data, the shake

test and the impact test. In the scope of this work, the impact test will be

executed.

Usually when performing an impact test, accelerometers are mounted on the

structure. Accelerometers can affect the results, because their weights are

considerable in comparison to the effective weight of the structure. Especially for

light-weight structures this can be a concern and has to be considered. One way

is to mount all accelerometers on the structure even if just a few are used,

another possibility could be to add dummy weights to the structure, which will

eliminate the roving mass effect. In this thesis non-contacting distance sensors

are used which will not influence the rotary machinery.

The selection of the hammer tip is essential for the experimental measurement.

The harder the tip, the wider the frequency ranges. The hammer tip has to be

- 12 -
selected carefully to ensure that all modes of interest are excited. If a too soft

hammer tip is selected the modes are not excited sufficiently and the result of the

measurement will not be good enough. This means the input power spectrum

does not excite all the modes since there is a roll-off. A roll-off describes the

steepness of a transmission function over the frequency. The coherence function

is as well evidence for that since it is seen to deteriorate. The frequency

response function has the same behavior over the second half of the frequency

range. [12]

By executing the impact test on the rotary machinery, the Column Row Method

can be applied.

The response behavior of the rotor can be characterized by the complex

frequency response function, when the rotor is excited at a specific position

through by means of a harmonic force function ̃ (with exciter frequency ) and

the displacement ̃ measured at point . is the phase between two signals.

( )
( ) (4)

R. Nordmann showed in his paper “Modal Parameter Identification and

Sensitivity analysis in Rotating machinery” that the complex frequency response

function also can be written with respect to the eigenvalues and the

corresponding eigenvectors , . [6]

̃
( ) ∑ (5)

- 13 -
Figure 7 Excitation and response location

Curve fitting
After the frequency response measurements have been made parameter

estimation techniques (curve fittings) are applied to determine the modal

parameters.

Figure 8 Modal Parameter Estimation

Frequency Response Functions can be measured with Single-Point Excitation

and Multiple-Points Excitation. The following is a short overview of the different

measurement techniques.

Figure 9 Different Measurement Techniques

Data Plots
The most common data plot is the time-based plot. The instantaneous value of

one parameter (usually displacement, velocity or acceleration) from a single

- 14 -
transducer on a short scale is usually displayed. Filtered or unfiltered vibration

data is shown. They Keyphasor event is indicated by a blank or dot sequence.

The data for an Orbit Plot comes from a XY transducer (have to be

perpendicular to each other). The transducer observes the rotation of a rotor and

can be displaced in two independent timebased plots. When those two plots are

combined an orbit plot is produced. The notch is detected by a Keyphasor

transducer.

The Bode Plot seen in Figure 10 presents the data from a startup or shutdown of

a machine and shows the frequency response in dependency of amplitude and

phase respectively. [5]

Figure 10 Bode Plot

- 15 -
Rotating machinery structures
A rotating machine can be regarded as a structure, which consists of rotating and

non-rotating components such as a motor, turbine or pump. The dynamic

behavior is identified by the properties of the component parts and especially the

bearings since most machines operate below any bending modes.

Rotor Systems are subjected to static (constant in direction and time) and

dynamic forces (change in magnitude or direction through time). Static forces

produce a static deflection, whereas dynamic forces produce vibration, which can

be torsional, radial and axial. Radial is the most common vibration problem,

whereas axial is less frequent. Usually torsional vibration is neglected since it is

difficult to measure. Through cross-coupling mechanisms torsional and axial

vibration can produce radial vibration.

Rotor systems can usually be modeled with a small number of modes. With an

increasing machine speed, the number of signigicant modes increase. Each

mode can be modeled like a single degree of freedom if modal mass and modal

stiffness are used.

√ (6)

The natural frequency is also named mode, or natural mode. ”Each natural

frequency is associated with a particular vibration pattern, called a mode shape,

and each mode shape is independent of all the other mode shapes of the

system.”

Modes can be distinguished between in-phase and out-of-phase modes. In-

phase means that the masses move in the same direction, whereas out-of-phase

- 16 -
means that the masses move in opposite directions. The values of the mass and

stiffness elements of the system decide on the frequencies of the modes and the

relative amplitudes of vibration.

Free Body Diagram of a rotor machine

For a better comprehension of the forces on a rotor, consider the free body

diagram shown below in Figure 11:

Figure 11 Free Body Diagram of a Rotor

All forces are assumed to act through the rotor mass center. The position of the

center of the rotor relative to the equilibrium position is described through the

rotor position vector , which is defined as a complex rectangular plane.

Rotor position Vector

Figure 12 Rotor position vector

- 17 -
The vibration vector is described by the amplitude and the absolute phase of the

transducer (filtered) signal. The length of the vector demonstrates the amplitude,

whereas the angular position shows the absolute phase of the signal. The plot in

Figure 12 shows the X and Y axes, which represent the real and imaginary axes

of the complex plane. The equilibrium position of the rotor is the origin of the

coordinate system. They will coincide if no external forces act on the rotor. The

position of the center of the rotor relative to its equilibrium position is described

with the vibration vector r.

The vibration vectors are noted in polar representation as

(7)

where r is the displacement vibration vector, A is the vibration amplitude and Φ is

the phase angle.

The position of the center of the rotor in direction is represented by and in

direction by . It is defined in polar representation as

with

√ .

It follows that the magnitude of the rotor position vector is | | √ .

The position vector in exponential notation is

(8)

which was shown by Leonard Euler (1707-1783) and with (base of

the natural logarithms). becomes a function of time , where is the

absolute phase angle (at t=0), when the Keyphasor event occurs.

This leads to a general expression for r:

- 18 -
 ( )
(9)

Due to differentiation the velocity and acceleration are obtained:

( )
̇ (10)

̇ ( )
̈ (11)

Forces and Stiffnesses

is the spring stiffness. combines all spring stiffness of the rotor system and

is obtained from the series/ parallel combination of namely the shaft, the bearing,

bearing support stiffness and the foundation stiffness. The term contains a minus

sign because the bearing spring force points in the opposite direction to the rotor

position vector.

is the rotating perturbation force, which describes the disturbance of a

system. An example therefore can be an unbalance. The rotating perturbation

force has the magnitude , where is the unbalance mass (the heavy

spot), the distance from the center of the journal to the rotor center and is

the perturbation frequency.

is the damping force and acts in the opposite direction to the instantaneous

velocity vector. It is proportional to the viscous damping constant, D and the

magnitude of the instantaneous velocity vector.

The tangential stiffness force is expressed by the term (tangential

stiffness). is the angular velocity of the rotor, is the damping constant of the

bearings and is the Fluid Circumferential Average Velocity Ratio, which is

defined as where is the fluid average angular velocity.

- 19 -
Keyphasor Event

When working with rotor machinery, unbalances are very common. Therefore it is

necessary to have a fixed, timing reference signal in order to make phase

measurements. A once-per-turn reference signal has to be obtained. In the

experiment an eddy current displacement transducer is facing a keyway, but

other shaft configurations are also realizable. When the keyway is passing the

transducer, the distance will increase and as a result the signal voltage will

suddenly become less. After passing the keyway the voltage will get back to the

level it had before. The abrupt change in voltage is called keyphasor event. This

event is used to determine the lapsed time between the keyphasor event and the

event of another signal. The keyphasor is an important tool when determining the

characteristics of the dynamic response of the rotor and the rotor speed. [5]

Equation of Motion
By applying Newton’s Second law, the equation of motion is obtained. For

completeness also the tangential stiffness term is included.

̈ (12)

The expressions for the forces are substituted in the equation of motion:

( )
̇ ̈ (13)

By reorganizing the standard form of a differential equation is obtained:

( )
̈ ̇ ( ) (14)

With no fluid circulation is equal to zero and the equation reduces to:

( )
̈ ̇ (15)

- 20 -
The term on the right-hand side is equal to zero if a free (unforced) behavior of

the system is described. [5]

Solution of equation of motion

To solve the equation of motion, the general expression for is assumed to be

harmonic. Due to differentiation the velocity ̇ and acceleration ̈ are obtained

and by substitution and elimination of the time terms the following expression is

obtained:

,
(16)
( )-

This equation describes for the moment that the Keyphasor event occurs the

amplitudes and absolute phases. Simplified the equation states following:

(17)

Therefore can a change in vibration be caused either by a change in force, the

dynamic stiffness or both.

In the following Figure 13 the force and vibration response vectors are shown at

the moment the Keyphasor moment occurs. The two vectors (rotating with the

angular velocity ) are here shown in a way, that the response vector lags the

force vector, but also the opposite can be true. is the phase of the perturbation

force and is the phase of the response vector when the Keyphasor event

occurs at t=0. [5]

- 21 -
Figure 13 Vibration and Force response vectors

Dynamic Stiffness and Rotor Behaviour

When examining rotor machines it is beneficial to regard it as a system that takes

dynamic force as an input and produces vibration as an output.

Figure 14 Conversion of dynamic input forces to output vibration

The force is acting through the dynamic stiffness of the rotor system and

therefore is the reason for the vibrations. Vibration is therefore a ratio of the

applied dynamic force to the dynamic stiffness of the rotor stiffness. Therefore a

change in the vibration is either through a change in the applied force or the

dynamic stiffness.

Theoretically an infinite number of resonances are possible. Rotatory machines

exhibit numerous resonances during their startup and shutdown, but when

examining a machine, usually just the lower few resonances are important. [5]

- 22 -
Dynamic stiffness
Describing dynamic stiffness physically; it is a combination of static effects of the

springs and tangential stiffness accompanying the dynamic effects of mass and

damping.

The Nonsynchronous Dynamic Stiffness (if ) controls the rotor response

due to an applied dynamic force at any frequency and is expressed as

( ) (18)

With is the nonsynchronous Direct Dynamic Stiffness and

( ) is the Quadrature Dynamic Stiffness.

When writing the equation of the nonsynchronous stiffness in the following way,

the internal and external rotor system forces can be explained easier.

(19)

K is the spring stiffness of the rotor system (which is a combination of the fluid-

film bearing stiffness, the bearing support stiffness, the foundation stiffness and

the shaft stiffness) and will oppose the applied force. The second term – is

the mass stiffness. It appears as an dynamic term because of the inertia of the

rotor. The term is negative because it will create a force that will be opposite the

spring stiffness. The third term is called the damping stiffness. It acts 90° to

the direction of the applied force and stabilizes the rotor. The fourth term –

is a function of the fluid circumferential average angular velocity.

Synchronous stiffness (if )

( ) (20)

- 23 -
When applying a static load, the perturbation frequency is zero and just the

first and fourth term is remained.

For both, the synchronous and the nonsynchronous Direct Stiffness, the Direct

Dynamic Stiffness is the same. The Quadrate Dynamic Stiffness plots have in

both cases a straight line, but with different slopes. [5] [13]

Synchronous Rotor Behavior

If the rotor response is controlled by a synchronous (1X) force (e.g. unbalance) it

is called Synchronous Dynamic Stiffness. The terms in the Synchronous

Dynamic Stiffness equation are associated with the rotor response in three speed

ranges. These are at, below, and above the balance resonance.

At speeds well below a resonance the rotor high spot is in phase with the heavy

spot. The spring stiffness dominates. At resonance speed, just the Quadrature

Dynamic Stiffness is remained. The rotor amplitude is at its peak and the heavy

spot is lagged by the high spot by 90°. Above the resonance speed the mass

stiffness controls the response. The high spot lags now 180° and the vibration

amplitude approaches a constant level. [5] [13]

- 24 -
Chapter 2 – Experimental Measurements

Chapter 2 describes the experimental measurements that have been performed

to obtain the reference data. This will include the setup, the balancing of the

machine, the use of MEScope, and execution of the hammer test. Data obtained

will make it possible to update the mathematical models. [17] [19]

Balancing
Before the experimental testing can be done, the rotatory machine has to be

balanced. Ideally the geometric center is located at the center of all masses, but

in reality that is not necessarily the case, due to non-uniform radial mass

distribution. This may be due to holes in the casting, manufacturing tolerance,

(also containing machining and stacking errors), and nonsymmetric construction

(due to keys or pins).

The mass center of the rotor system is therefore some radial distance and a

specific direction away from the geometric center. The unbalance of the rotor is

the product of the rotor mass M and the radial distance re. The formula below

shows that the magnitude of the unbalance can also be calculated over the

unbalance mass m (the heavy spot) and a larger distance ru. The larger distance

ru is usually the radius of the balance circle.

(21)

The unbalance produces a force which is proportional to its mass, its radius, and

the square of the rotor speed Ω. The following force shows that the force is

generated due to the centripetal acceleration of the unbalanced mass.

(22)

- 25 -
To balance the rotatory machinery, two different techniques were used. First the

machine was balanced with determining the Heavy Spot location and attaching

the suitable calculated weight (for a suitable start of balancing the machine the

state rule was used), and then the machine was balanced with the help of

influence vectors. Note that the presented techniques are the most suitable for

multiplane balancing and multiple modes. In the case of singleplane balancing,

different techniques would be more appropriate. [13]

Balancing with determining the Heavy Spot location

The characteristic with multiplane balancing is that two or more planes have to

be balanced simultaneously. In the present case, of two disks with balance holes

on a shaft are used.

To determine the weight and angular position of an additional mass for the two-

plane balancing, the vibrations of two-rotor system have to be examined. This

analysis leads to the determination of the heavy spot. The angular position and

the magnitude of the influence vector are used to find the position for the

calibration weights. This calibration weight reduces, in theory, the amplitude of

the vibrations in the system.

Multiplane Balancing With Influence Vectors

When balancing with influence vectors, the rotatory machine has to be run at a

single operating speed.

The magnitude and the phase of the vibration response vector for plane 1 and

plane 2 have to be examined. It has to be considered that an added weight in

one plane (either plane 1 or plane 2) is influencing both, the response in plane 1,

- 26 -
as well as the response in plane 2. This means that two experiments have to be

run to collect all the data that is needed to finally balance the rotatory machine

sufficiently.

Weight has to be added in one plane, and after determining the influence vectors

it has to be removed. The same weight will be added to the other (second plane)

and the same procedure has to be realized. [13]

Weight Splitting
Often the exact angular position of the weight that has to be added is not

available, it is between two holes or the weight itself is too heavy. In these cases

the process of weight splitting is applied.

This means that if a specific weight m is need at the angular position Ɵ, the

weight has to be divided into at angular position Ɵ and at angular position

Ɵ .

The force produced by the weight is dependent on the mounting radii. In the

present case all holes are at the same radius, so the following derivations are

based on this circumstance.

When treating the masses and angles as vectors, the requirement becomes

(23)

and in complex numbers becomes

Ɵ Ɵ Ɵ

When equating real and imaginary parts, the following system of equations is

obtained

Ɵ Ɵ Ɵ
[ ] . / . /
Ɵ Ɵ Ɵ

- 27 -
Solving this system of equations numerically leads to the solution for the weight

needed. [5]

(Ɵ Ɵ)
[ ]
(Ɵ Ɵ )

(Ɵ Ɵ )
[ ]
(Ɵ Ɵ )

Figure 15 Weight Splitting

Experimental Setup Balancing Rotary Machine

To simulate the actual rotating machine behavior, the RK4 Rotor Kit from

Bently Nevada is used. It has a unique geometry and enables the user to isolate

and control individual machine characteristics.

To capture the specific machinery data, ADRE for Windows, a multi-channel

machinery data acquisition system is used. It combines the features and the

performing abilities of the oscilloscopes, the spectrum analyzers, filters and

recording instruments. Therefore additional equipment is seldom needed. The

system can be used with its real-time display capability, which demonstrates the

- 28 -
captured data on the screen as it is recorded. The system consists of the Data

Acquisition Interface Unit, the ADRE for Windows software and a computer

system which is capable of running the software. The system needs either an AC

or battery power to run. Since it is completely portable it allows operation in

almost every environment. It provides support for all standard and non-standard

input types. These are amongst others the dynamic transducer signals, the static

signals and the speed input signals.

The following Figure 16 demonstrates the RK4 Rotor Kit from Bently Nevada

consists of a Rotor Kit, the Motor Speed Controller, the TK 21 Keyphasor Signal

Conditioner, the Oscilloscopes and the 208 DAIU.

Figure 16 Rotor Kit Components

The V-frame design of the Rotor Kit provides an advanced control of the housing

dynamic stiffness properties. The electric motor can closely hold the desired

speed with changes in loading conditions, since a direct current motor and

control circuitry are combined. The motor can either run clockwise, or counter-

clockwise and it also has a slow roll speed capability.

- 29 -
Figure 17 Rotor Kit Speed Controller
The Figure 17 above displays the following items: The Rotor Kit Speed Controller

provides power for the rotor kit and the remaining devices. Ramp Rate allows the

user to regulate how many rpm/min the radial velocity of the rotor machine

increases until it reaches the maximal radial velocity, which is inserted over Max

Speed Setpoint. Whether the radial velocity increases or decreases can be set

over Ramp (Up/ down).

The function of the TK21 Keyphasor Signal conditioner is to manipulate the

analog signal, thus it coincides with the requirements for further processing.

The 208 DAIU (Data Acquisition Interface Unit) is a portable system for mutli-

channel machinery data acquisition, that works together with the ADRE for

Windows® Software (Automated Diagnostics for Rotating Equipment). For

consistency with the Chapter 4, “Matlab Finite Element Model Creation”, the

measurement points are numbered as 2, 4, 5 and 7.

- 30 -
Figure 18 Rotor Kit

Obtaining Data
1. After everything has been prepared to collect data in ADRE for Windows

(see Appendix A) the measurement can be started.

2. At the Rotor Kit Motor Speed Controller the right lower was shifted to

SLOW ROLL

3. Store Enable was clicked in the main menu and the real time values,

aswell as the chosen plots are displayed

4. At the Rotor Kit Motor Speed Controller left lower switch was toggled to

RAMP UP. The rotor kit increased its radial velocity until it reached the

maximal radial velocity (8000 rpm)

5. In the 208 DAIU control menu, to finish collecting data Stop Sample was

toggled and then Close was clicked. At the Rotor Kit Motor Speed

Controller the left lower switch was toggled to RAMP DOWN. When it

- 31 -
 reached a radial velocity smaller than 300 rpm, the right lower switch was

toggled to STOPPED.

6. In the menu bar different buttons appeared:

Figure 19 Partial view of ADRE for Windows

Namely from left to right: Orbit, OTB, Timebase, Trend, XY Graph, Vector

Tabular, Polar, Bode, ShaftCL, HalfSpectrum Cascade, HalfSpectrum Waterfall,

HalfSpectrum and Orbit. By clicking the icons the different plots were displayed.

Test: Unbalance Setting

To test the unbalance setting, the radial velocity of the machine was increased

using a rotor speed controller to 8000rpm. Data is collected with the

measurement probes 2 and 7 (see Figure 18). The collected data is used to

determine the magnitude and phase of the vibration response in plane 1 and 2.

When first running the experiment, the collected data shows clearly that the rotor

machinery is very unbalanced. For the speed range close to the first natural

frequency, no value is displayed, because the amplitude of the natural frequency

is too high. While running the test, the whole rotor machine was making loud

noises and the displacement of the rotor is visually apparent without looking on

the plots.

- 32 -
Figure 20 Polar Plot
As seen in Figure 20, the experiment had to be shut down after 3000 rpm

because the amplitude of the vibration was already too high for the first mode,

and it is sufficient to determine the heavy spot and adjust the calibration weight at

the appropriate position. The experiment was repeated several times and with an

Excel Sheet provided by Professor Dr. Meagher (which contains the theory

described in the Chapter “Weight Splitting”) the appropriate weight for each

position could be found. The Excel Sheet is attached in Appendix B.

The rotor machine could be balanced to the point where none of the amplitudes

at the natural frequency would be higher than 10 mill pp.

- 33 -
Test: Obtaining Mode Shapes/ Natural Frequencies
At this point two additional measurement points were attached, namely

measurement point 4 and 5. The experiment was measured with two Bently

Nevada measurement kits simultaneously. The experiments were displayed with

ADRE® for Windows® / DAIU (see Figure 18).

The additional measurement points 4 and 5 were realized by using parts of an

additional RK4 Rotor Kit from Bently Nevada.

Setup for additional probes 4 and 5:

To measure the shaft rotative speed and have a reference for measuring the

vibration phase lag angle, the Keyphasor signal is used. Therefore the same

Keyphasor signal is used on both RK 4 Rotor KIT Proximior Assemblies.

The input from the Keyphasor Probe goes into the RK 4 Rotor KIT Proximior

Assembly. The output is then lead to the TK 21 oscilloscope and then to the 208

DAUI. To use the same Keyphasor signal a splitter is used. The Keyphasor

output signal is split and lead to the 208 DAIU of the second RK 4 Rotor Kit

- 34 -
Proximior Assembly. In this way the experiment can be run on both 208 DAUI

workstations simultaneously. The collected data is compiled after finishing the

experiment.

Figure 21 RK 4 Rotor Kit Proximitor Assembly

The experiment is conducted as described in the Chapter Obtaining Data.

Mode Shapes
From the obtained data the dynamic rotor behavior can be defined. To estimate

the mode shapes, there are two ways to obtain them, either with the 1X polar

plots or more precisely with the 1X orbit plots.

Polar Plots: This technique is using a single transducer per measurement plane.

The startup (usually) or shutdown data is displayed at points along the length of

the rotor. The plots are positioned in a way which represents the axial location on

the machine. Additionally, all plots have the same full scale range.

The rotor mode shape can more accurately be estimated using 1X orbits. This

requires two transducers. The orbits are plotted at the same scale and the

Keyphasor dots are connected.

- 35 -
Figure 22 First Mode Shape

Figure 23 Second Mode Shape

Natural Frequencies
From the Bode Plot the natural frequencies of the system can be determined.

- 36 -
Figure 24 Bode Plot
The natural frequency for mode 1 is at 1820 rpm, the natural frequency for mode

2 is at 5400 rpm.

- 37 -
Experimental Setup Hammer Tip Testing
For the hammer tip test the Dactron Focus II Real-time Signal Analyzer is

used. The structure is excited with a specialized measurement tool, here an

impact hammer, which produces short duration vibration level.

Figure 25 Impact hammer instrument

The impact hammer has a load cell (sensor which produces a signal proportional

to the force of impact) to measure the excitation force and different hardness tips.

The different hardness tips allow the adjustment of the frequency content of the

impact force. The hammer tip has to be selected carefully to ensure that all

modes of interest are excited. If a too soft hammer tip is selected, the modes are

not excited sufficiently and the result of the measurement will not be accurate

enough. Therefore selecting a hard hammer tip leads to an excitation that will

have higher frequency content and is impacted for a shorter time.

The response behavior of the rotor can be characterized by the complex

frequency response function, when the rotor is excited at a specific position

- 38 -
through by means of a harmonic force function ̃ (with exciter frequency ( ) and

the displacement ̃ measured at point . is the phase between two signals.

( )
( ) (4)

R. Nordmann showed in his paper “Modal Parameter Identification and

Sensitivity analysis in Rotating machinery” that the complex frequency response

function also can be written with respect to the eigenvalues and the

corresponding eigenvectors , .

̃
( ) ∑ (5)

Just a set of frequency response functions allow already to obtain the modal

parameters of a rotating structure. When executing the hammer tip testing a

broad frequency range and therefore it is a very fast tool. Sine testing for

example just excites one frequency at a time and is therefore a slow. To obtain

- 39 -
the column of the FRF Matrix the excitation has to be done at location and the

measurement of displacement takes place at location 2, 4, 5 and 7. When

obtaining a row of the FRF Matrix, the excitation takes place at location 2, 4, 5

and 7 and the displacement is measured at location . [6]

Figure 26 Excitation and response location

Dactron PhotonTM
To compute the time and frequency measurement a dynamic signal analyzer is

used. It consists of the Dactron Focus II Real-Time Signal Analyzer and the

Dactron PhotonTM.

Figure 27 Dactron Focus II Real-Time Analyzer and Dactron Photon

- 40 -
The Dactron Focus II Real-time Signal Analyzer computes the frequency domain.

Therefore time domain traces are transformed by using the Fast Fourier

Transformation (FFT). Before the FFT can be performed on the signal, an analog

to digital conversion (ADC) has to be carried out. Therefore the Nyquist sampling

theory has to be fulfilled.

( ) (24)

is the sampling rate and fmax is the highest measured frequency. The error

that occurs when the sampling rate is too low is called aliasing. Therefore an

anti-aliasing filter is used, it prevents an error when sampling with too high

frequencies.

The FFT is necessary to properly analyze complex signals. When just the time

domain trace is available it is difficult to determine the frequencies of a signal. By

transforming this signal the frequency components can be seen simpler.

After the rotor machinery has been setup, one column of ( )is obtained by

measuring the response at the same place and exciting the rotor on four different

spots.

The frequency measurement computed by the Dynamic Signal Analyzer includes

the Fast Fourier Transform, Power Spectral Density, Frequency response

Functions, Coherences and various others. Signals are computed in the Dactron

Phonton with the RT Pro Focus 6.06 Software, where the time data is digitized.

The time data is sampled into the DSP block by block, which is also called a time

record or time window. [14]

- 41 -
FFT Windows
The FFT computation assumes that signals satisfy a periodic requirement, which

means that the signal is identical every time. The transient signal also satisfies

this requirement. A transient signal starts at zero at the beginning, rinses to a

maximum and decays to zero before the end of the time window. When the FFT

of a non-periodic signal is computed, leakage might occur. Therefore window

functions have to be applied. Windows are shaped that it is zero at the beginning

and end.

Windowing functions must be applied to correct measurement errors, which are

known as leakage (occurs when FFT is estimated from a non-periodic data

block). When windowing is not applied, an error might occur in the FFT

amplitude, frequency or overall shape of the spectrum.

Windows: Weighting functions are applied to the measured data to minimize

leakage. For that reason the distortion effects of leakage are reduced by forcing

the data to be more fitting to the periodicity requirements of the Fourier

Transform process. Since windows (weighting functions) cause distortion as well

so they should be avoided if possible. Usually the following windows are used to

minimize the leakage of experimental data:

 Uniform

 Hanning

 Flat Top

 Force/ Exponential

Table 1 Recommendations for different window types [4]

Type of Signal Exponential Force Hanning Flat Top Uniform
Transients whose duration is yes yes

- 42 -
longer than the length of the
window
General-purpose applications yes
Spectral analysis (frequency- Yes Yes (random
response measurements) (pseudorandom excitation)
excitation)
Accurate single-tone amplitude Yes
measurements
Sine wave or combination of sine Yes
waves
Sine wave and amplitude accuracy Yes
is important
Narrowband random signal Yes
(vibration data)
Broadband random (white noise) Yes
Closely spaced sine waves yes Yes
Excitation signals (hammer blow) yes
Response signals yes
Unknown content yes
The force and exponential windows are used when an impact test is performed.

The exponential window is used for transients whose duration is longer than the

length of the window. The force window is applied on excitation signals (like the

impact hammer).

The Hanning Window forces the sample in the beginning and end of the interval

to be zero. This offers advantages when the signal does not fulfill the

requirements of the FFT process and is the most commonly

The Flat Top window is the most common function used for sinusoidal signals

which do not fulfill the periodicity requirement of the FTT process.

Uniform signals are applied when the entire signal is captured in one record or

when the data is assured to satisfy the periodicity requirements of the FFT

process. For impact testing this window can be used just if the response signals

are monitored within one sample of collected data. Sine chirp, digital stepped

sine and pseudo-random are all signals that fulfill the periodicity requirements of

the FFT process. [14]

- 43 -
Generally all windows will cause degradation of the peak amplitude of the

function and therefore the actual measurement will appear to have more

damping than what actually exists.

Window for Impact Measurements

When using an impact hammer, the exponential windowing is the most suitable.

It is specifically developed for modal analysis and has two parts, which are the

pre-window at the beginning of the time frame, which removes any noise before

the impact (length of this hold-off period is chosen by the user over the pre-

trigger time) and the exponential window. The exponential window forces the

response data to zero by the end of the frame.

Obtaining Data
Data is collected with the same measurement probes that were used in the

previous experiment for the balancing of the rotary machine. The output signals

of the four probes (for vertical and horizontal testing respectively) of the Rotor Kit

Proximitor Assembly were split and lead to the Dactron Focus II Real-Time

Analyzer .

Figure 28 Input Dactron Signal Analyzer

- 44 -
To obtain the Data the program RT Pro Focus is used; the setup therefore is

described in Appendix D. After the data has been captured it was imported as an

*.uff file in MEScope for further analysis and examination. [16]

To execute the hammer test for obtaining a row of the ( ) matrix following

procedure has to be done:

1. Signal Setup>Auto Channel Signals for Channel 1 (hammer) and 2

(displacement probe) Time Capture, FFT and Power Spectra was chosen.

2. Signal Setup>Cross Channel Signals for Channel 2 Power spectra,

FRF/ Coh and Excitation 1 was pressed. was chosen, since a

row will be captured.

3. To start capturing data Start has been pressed and the rotor machine has

been excited at the appropriate position; Since probe 2 is on the input

Channel 2 has been obtained.

Figure 29 Exciting the rotary machine at location l and measuring at loc k

4. After the data has been collected, Stop was pressed. To measure the

next FRF Next Point is pressed, Figure 29 shows the excitation of

- 45 -
 location 2. The procedure is repeated until the whole row has been

captured.

Figure 30 Excitation at location 2

5. It is also possible to obtain the whole FRF Matrix by choosing under

Signal Setup>Auto Channel Signals Channel 2, 3, 4 and 5 and Signal

Setup>Cross Channel Signals for Channel 2, 3, 4 and 5 Power spectra,

FRF/ Coh and Excitation 1.

The results are shown in the RT Pro Focus Window and for further examination

are imported in MEScope.

- 46 -
Figure 31 Window MEScope

Animation with MEScope

MEScope is used to build and animate a simplified model. ME Scope offers three

different options for the Modal Analysis. They all build up on the performance of

the former option and are Modal Analysis, Multi-Reference Modal Analysis and

Operating Modal Analysis. For the scope of this thesis just the Modal Analysis

option has been used. The Modal Analysis option offers the possibility to obtain

the mode shapes and operation deflection shapes which gives the opportunity to

determine if a resonance is excited. The operation deflection shapes

demonstrate how the examined structure vibrates and shows for various points

and directions where the exaggerated vibration occurs.

The Modal Parameters frequency, damping and residues for each mode are

estimated by curve fitting an analytical FRF parametric model to a set of

experimental FRF data. The estimated modal parameters are collected into a

Shape Table and can be animated on a 3D model.

- 47 -
In the Appendix E the procedure to build and animate the model is given.

Furthermore the exact steps to perform the curvefitting in ME Scope are

described.

Results
Following the results for vertical excitation and vertical measurement are shown

and discussed. The reference DOF is the accelerometer point (here the

displacement point).

Figure 32 Simplified Model in Structur Window of MEScope

With the results from the measurement the simplified model can be animated.

The modes can be visually seen. To obtain more information, curvefitting has to

be done. The modal parameter estimation phase (curvefitting) is used to obtain

the modal characteristics from the measured data. Therefore a mathematical

algorithm is used. The FRF is broken down into various single DOF systems.

- 48 -
Different techniques to do this are possible, but generally all of them break down

measured data into principal components (which are the dynamic characteristics

frequency, damping and mode shapes).

Figure 33 Curve Fitting

- 49 -
Figure 34 Experimental Mode Shapes 1 and 2

The result for Mode Shape 1 looks like expected, unfortunately Mode Shape 2

shows a sudden decrease, which can be due to a measurement mistake.

The results of the mode shapes which were obtained by curvefitting the data

from the experimental measurements are used for further processing to update

the mathematical model developed with a matlab code and an ABAQUS model.

The assumption is made that the experimental data is correct and the models are

adopted to those.

- 50 -
Chapter 3 – Finite Element Model with Abaqus

Modeling – Introduction
Often a mathematical representation of a rotor system is used to model how an

applied force is transformed into vibration. It is important to simplify the model

and decide which assumptions have to be made. Obviously the mathematical

representation shows variations to reality. Then it is necessary to evaluate

whether the assumptions are reasonable.

In the context of this thesis a finite element model has been modeled with

ABAQUS and Matlab. The following is a description of the Finite Element Model

in ABAQUS with relevant terms defined. [7]

Structured process to model the physical rotor system

Statement of Assumptions
1. Parts of rotor system: The rotor system includes the attached disks and

the bearings that support the shaft. The bearings are represented by

attached dampers and springs. Structures that support the bearings, the

foundation system, coupled machines and the cabling are neglected.

2. The model has 1-Complex-Degree-of-Freedom (1-CDOF): Since the

model has just one moving element, the system is described just with one

position measurement variable (independent, lateral and complex), no

angular deflection and one differential.

3. The model is isentropic; therefore all properties are radial symmetric and

a compact mathematical model description is allowed.

4. Neglecting gyroscopic effects

- 51 -
 5. No fluid interaction; since no parts of the model rotate with a fluid-filled

cylinder

6. Damping will be due only within the bearings.

7. Model is linear; the duplication of the input by a constant factor causes a

duplication of the output by the same factor. Real rotor machines behave

in a nonlinear manner, but the added complexity is not necessary for small

motions. [5]

Definition of Coordinate System:

Figure 35 Definition of Coordinate System

The X-plane goes through the geometry center of the rotor model.

Acting forces on the system

Rotor Systems are subjected to static (constant in direction and time) and

dynamic forces (change in magnitude or direction through time). Static forces

produce a static deflection, whereas dynamic forces produce vibration, which can

be torsional, radial and axial. Radial is the most common vibration problem,

whereas axial is less frequent. Usually torsional vibration is neglected since it is

- 52 -
difficult to measure. Through cross-coupling mechanisms torsional and axial

vibration can produce radial vibration. [7]

Derivation of equation of motion

As shown in Chapter 1 – Equation of motion, the equation of motion with no fluid

reduces to:

( )
̈ ̇ (25)

The term on the right-hand side is equal to zero if a free (unformed)behavior of

the system is described.

Solution of equation of motion

The solution of the equation of motion (shown in Chapter 1 – Equation of motion)

with the use of the general expression for and various substitutions and

eliminations is:

,
, (26)
( )-

which describes for the moment that the Keyphasor event occurs the amplitudes

and absolute phases.

Modeling rotary machinery in ABAQUS CAE

Abaqus/CAE consists of different modules which accomplish different tasks and

therefore define an aspect of the modeling process. When modeling, the user

moves from module to module. After the model is build, the input file is submitted

for the analysis and the results can be visualized. All the required information to

perform the analysis is contained in the Abaqus/ CAE model container.

These modules include:

- 53 -
  Part (the geometry for the problem is created, for 3-D first the 2-D profile

has to be created)

 Property

 Assembly

 Step

 Interaction

 Load

 Mesh

 Job

 Visualization

 Sketch

Units
It is very important to decide which system of units will be used, since Abaqus

has a no built-in system of units. In this project the Unit system SI(mm) is used.

Table 2 System of Units

Quantity SI SI(mm) US Unit (ft) US Unit (inch)
Length m mm ft In
Force N N lbf lbf
Mass kg tonne (10^3 kg) slug lbf s^2/in
Time s s s s
Stress Pa (N/m^2) Mpa (N/mm^2) lbf/ft^2 psi(lbf/in^2)
Energy J mJ (10^-3 J) ft lbf in lbf
Density kg/m^3 tonne/mm^3 slug/ft lbf s^2/in^4

Part Module
The rotor machinery first was sketched in the Part Module

1. ABAQUS/CAE and then File> New Model Database> With Standard/

Explicit Model was opened.

- 54 -
 2. From Module> Part was chosen.

3. Create Part> Type: Deformable> Shape: Solid Shape> Type: Extrusion>

Approximate size: 560

A deformable body can change its shape (deformation: motion in which a change of
shape takes place). The contrast is a rigid body, which can only undergo rigid body
motions.
4. With the Edit Feature, the model can be revolve with an 360 degree angle

Figure 36 Abaqus Creat Part and Edit Feature(right figure)

Figure 37 Sketcher of rotary machinery

5. With Part> Create Round or Fillet the edges of the disks are rounded.

- 55 -
Figure 38 Edit Feature
6. To define the local coordinate system with Part> Create Datum CSYS: 3

Points> Coordinate System Type: Rectangular> Continue…> Select a

Point to be the origin – or enter X, Y, Z: (0.0, 0.0, 0.0)> Press Create

Datum is chosen.

Figure 39 Create DATUM CSYS

Modeling Springs
To model the bearing following assumptions are made: The bearing on the left

will be modeled with springs and dampers in Y- and Z-direction and the coupling

from the shaft to the electric motor will be modeled with springs and dampers in

X-,Y- and Z-direction.

- 56 -
Figure 40 Modeling Bearing and Coupling
The springs will be attached in the center of the outer surface of the rotor.

Therefore a reference point on each side has to be created.

Creating the reference point to attach the springs and dampers:

1. Part> Create Datum Plane: Offset from principal plane> XY Plane> Offset:

0. The created plane will be used to partition the model.

2. Part> Partition Cell: Define Cutting Plane> Point & Normal, Pick Point and

the axis normal to the point. Another way of doing it: Create Partition>

Cell> Use datum plane

Figure 41 Create Partition

3. To create a point on the partition: In the dropdown menu: Tools>

Reference Point> Select point to act as a reference point – or enter X, Y,

Z: First (0.0,0.0,0.0) for the left surface and (579.0, 0.0, 0.0) for the right

surface. Then Create Partition> Edge> Specify Parameter by location.

- 57 -
 Click on partition and then the created point. Detail from the rotor with

partition edge and cell in red.

Figure 42 Partition Edge and Cell and Create Partition 2 (right figure)
4. In the model tree> Engineering Features> Springs/Dashpots

Table 3 Degree of Freedom/ Orientation

Degree of Orientation Left Side Ride Side (Coupling)
Freedom
Spring Dashpot Spring Stiffness Dashpot
Stiffness Coefficient Coefficient
1 Yes, value: 80 10 N/mm
N/mm^2
2 Yes, value: 10 N/mm Yes, value: 80 10 N/mm
80 N/mm^2 N/mm^2
3 Yes, value: 10 N/mm Yes, value: 80 10 N/mm
80 N/mm^2 N/mm^2
4
5
6

Property Module
1. Select Create Material from the Property Toolbox

- 58 -
 While building the whole model it is important to maintain one type of unit.

Here the SI (mm) system of unit is used. As material name enter Steel and

following values have been typed in:

General> Density from drop-down menu and type in Density 7.5440533E-009

(Calculated over weight and geometry of rotary machinery)

Obtaining density of rotary machinery:

Disk Beam Total

Weight
Volume
Density
Calculations:

( ) ( )

a. Youngs modulus 209000 and Poisson’s ratio 0.3 (Mechanical> Elasticity>

Elastic from drop-down menu). Also reference temperature (273.15) and

expansion coefficient 13.0*10^-6.

2. Select Create Section from the Property Toolbox. Type in for name

RotorSection. The Rotary Machine is of solid and homogeneous type.

Homogeneous means that the property of the part is common throughout.

- 59 -
Figure 43 Edit Section
3. On property toolbox> Assign Section and select the whole part from the

viewport. Then select created Section “RotorSection”

Assembly Module
Assembly usually means that all created parts are assembled in this module

together, but even if just one single part has been created, it has to be included

to the assembly.

1. Assembly Toolbox> Instance Part

2. Select the created part. Instance type: Independent.

Instance Part will make a copy of the part. By choosing dependent, all of the

mesh information comes of the same part. Independent means that different

mesh information comes from different copies of the part.

- 60 -
Figure 44 Instance Part

Step Module
In this module the variables that are included in the output files can be selected.

Furthermore loads over a step can be applied.

Creating analytical step:

1. In module tree> Steps> Create Step> name: NaturalFreq. From the list:

>Linear Perturbation > Frequency> Continue> Lanzos as Eigensolver>

Number of eigenvalues> Values:10> OK

The initial step is defined, in which the boundary condition which constrains the

rotation of both ends is applied. The rotation in and is fixed.

Applying boundary condition:

2. Model Tree> BCs container> Create Boundary Condition> name: Fixed>

Step: Initial> Type: Displacement/ Rotation> Chose UR2 and UR3 for

and

- 61 -
Figure 45 Edit Boundary Condition

Interaction Module
The outer surfaces of the beam are defined as rigid. This means they will

maintain the same shape, but they can rotate and translate. No local high

stresses and no large local deformations will occur.

Figure 46 Create Constraint and Edit Constraint

1. In the model tree> double click on Constraints> Rigid body> Continue…

>Region type: Tie (nodes)

- 62 -
Figure 47 Constrain on outer surface of beam

Meshing Module
To mesh the module, more partitions have to be assigned. The meshing of the model
cannot be carried out in one step, when the three-dimensional solid element type
hexahedral wants to be applied.
Therefore more partitions have to be assigned to the part. The partitions have to be
executed at the rotor disks and the meshing of the part has therefore to be done
separately.

Figure 48 Partition of Rotor Disk

1. Assigning mesh controls, which examines the technique, which

Abaqus/CAE is using to mesh the model and the shape of the elements

that will be generated. In the main menu bar> Mesh> Controls> acceptHex

- 63 -
 2. Assigning the Abaqus element type: Mesh> Element Type> Standard as

default Eleement Library, Linear as Geometric Order, 3D Stress as Family

of elements

3. Creating the mesh

Figure 49 Regular cube Region modeled with a Single Hexahedron

The approximate global size of 10 will generate 3087 elements on the part with

4896 nodes. The hexahedron is of the type C3D8R, which means that it is three

dimensional, with 8 nodes and with reduced integration. The results are attached

for in the Appendix.

Job Module

1. On the Job Container, double click on Job to create a job. Create Job box

appears> name: Modes

2. Click Job Manager> Click Submit. After the Job is submitted, the job’s

statues is indicated next to the job. The status is as following

Submitted, which means that the analysis input file is being

generated

Running, which means that Abaqus is analyzing the model

Completed, which means that the analysis is completed and the

output has been written to the output database. Status aborted is

- 64 -
 written if there is a problem in the Abaqus input file. Additionally the

a problem report is displayed.

Results

Figure 50 Results Step/Frame

- 65 -
Figure 51 Results Mode 3 which is the first Mode Shape

Figure 52 Results Mode 5 which is the second Mode Shape

From the Finite-Element Model the natural frequency for the first mode can be

found at 2018 rpm and the natural frequency for the second mode is at 4153

rpm. Compared to the experimental obtained data that is for the first mode

90.23% and for the second mode it is 74,40%.

- 66 -
Chapter 4 – Matlab Finite Element Model Creation

A mathematical model of the rotatory machinery is created by using the MATLAB

program. MATLAB is a high-performance programming language based

environment that integrates computation, visualization, and programming.

Problems and solutions can be expressed in common mathematical notation.

The described method could as well be used in a FEA code such as ABAQUS.

This simplified model illustrates the details on a smaller number of elements.

To obtain the mathematical model, first the matrix for a single beam element has

to be assembled. To derive the matrix, the Bernoulli-Euler Theorem is used. The

natural frequencies and mode shapes can be determined. The values for the

stiffness and damping parameters for the bearings are unknown and for that

reason assumptions are used. Therefore, the result is a very simplified model,

however, the solutions are likely to be different from those obtained in the

experimental measurement. Since the assumption for the damping paramerters

are made, the results from the experimental measurements are correct, and the

mathematical model will be adapted. The obtained global matrices, with their

degrees of freedom, have to coincide with the degrees of freedom from the

experminetal measurements . The Guyan Static Reduction Method is executed.

Finally the reduced matrices lead to an error function from which you can obtain

the unknown bearing parameters and allow the adaptation of the mathematical

model to the experimental model. [1] [18] [21]

- 67 -
Bernoulli-Euler Beam Theorem
In this method a simplified uniform element is considered. The beam element has

the length L, the mass density ( ), the modulous of elasticity I, the cross –

sectional area (A), and the moment of inertia (I).

The beam element experiences a transverse motion. The displacement can be

simply interpolated over :

( ) ∑ ( ) ( ) (27)

The shape functions have to satisfy following boundary conditions:

( ) , ( ) ( ) ( )

( ) , ( ) ( ) ( )

( ) , ( ) ( ) ( )

( ) , ( ) ( ) ( )

Figure 53 Uniform element undergoing transverse deflection

The shown beam in Figure 53 shows an element of a regarded beam statically

loaded by end shears and bending moments. The result produces numerous

deflection shapes. Therefore the equilibrium equation for an uniform end loaded

beam is:

- 68 -
( ) , the general solution for a uniform beam therefore is

( ) . / . / (28)

By substituting the boundary condition in this cubic polynomial solution of the

equilibrium equation, following shape functions are obtained:

( ) . / . /

( ) . / . /

( ) . / . /

( ) . / . /

By inserting the shape functions in the expressions for and for Bernoulli-

Euler Beams,

∫ (29)

∫ (30)

the following stiffness and mass matrices are obtained

[ ]

[ ]

As can been seen, both of the matrices are symmetrical. [8]

- 69 -
Assemblage of Matrices
The assemblage for two beam elements is shown below. This is necessary to

obtain the global stiffness and mass matrices.

Figure 54 two beam elements with attached springs

The elements have each two nodes and share one of them with each other.

Since neighboring elements share nodes, the contributions need to be

assembled in global matrix.

One node has two degree of freedoms; one translational and the other rotatory.

This means every element has four degrees of freedom and in total the system

has six degrees of freedom. To obtain the global matrix, the local matrices have

to be assembled. Each element has a local coordinate. In this case the local

coordinate axes coincide with the global x and y axes of the overall beam.

The matrices obtain from the Bernoulli-Euler Beam Theorem are used.

[ ]

[ ]

- 70 -
The stiffness for the spring k1 and k2 has been added to the stiffness matrix at

the appropriate degree of freedom. [9]

The direct stiffness method is used to assemble the stiffness matrix. The element

matrices are therefore added to a global matrix in the row and column position

determined by the nodes (DOF). The governing equation for the beam is then

given by:

[ ]

Current Model
Through the experiments and measurements the natural frequency and node

shapes of rotatory machinery are obtained. These values are used to invent a

mathematical model. Therefore, the rotatory machine is split in seven elements

and eight nodes. Each node has two degrees of freedom. The stiffness matrix for

element one, will include the stiffness of the bearing k1, element 7 will include the

stiffness of the bearing k2 at the appropriate degree of freedom. The mass matrix

for element two, three, five and six will include the lumped disk mass. Also in the

figure, are measurement points from the experiment shown; two, four, five and

seven. The deflection at point 1 and 8 is not measured but is obtained from a

- 71 -
curve fit of the experimental mode shapes.

Figure 55 Simplified model

[ ]

[ ]

[ ]

The seven (4x4) local stiffness matrices assemble to the global (16x16) stiffness

matrix.

The mass matrix is calculated in the same way.

[ ]

- 72 -
 [ ]

The seven (4x4) local mass matrices assemble to the global (16x16) mass

matrix.

Programming current model

Stiffness matrices
The Bernoulli-Euler Beam Theorem provides the stiffness matrices for the

individual elements. These stiffness matrices can be adopted for element two

through six . For element one and seven, the stiffness of the bearing has to be

added at the appropriate degree of freedom.

Therefore, first the general stiffness matrix has been programmed. For the

stiffness matrices for element one and seven the bearing stiffness have to added.

Syms A %cross section

syms roh % density
syms m % lumped disk mass
syms k1 % stiffness spring 1
syms k2 % stiffness spring 2
syms c1 % damping spring 1
syms c2 % damping spring 2
syms I % moment of inertia
syms Ivar % moment of inertia with rotor element
syms E % modulus of elasticity
syms l % length element
syms d % diameter beam element

l= 0.08 % in m
E=210*10^9 % steel in N/m^2
I= 4.9087*10^-10% pi*d^4/64in m^4, calculated with d=10mm%
k1= 90*10^6% N/m
k2= 90*10^6% N/m

% stiffness matrix obtained from Bernoulli-Euler Beam Theorem

Kn = [ 12 6*l -12 6*l
6*l 4*l^2 -6*l 2*l^2
-12 -6*l 12 -6*l
6*l 2*l^2 -6*l 4*l^2 ];

a = E*I/l^3;

% stiffness matrices for each beam element of the mathematical model

- 73 -
% k1 and k2 are included at appropriate degree of freedom

K1 = a * (Kn + [ k1 0 0 0
22 0 0 0
22 0 0 0
22 0 0 0 ])
K2 = a * Kn
K3 = a * Kn
K4 = a * Kn
K5 = a * Kn
K6 = a * Kn
K7 = a * (Kn + [ 0 0 0 0
22 0 0 0
22 0 k2 0
22 0 0 0 ])

% mass matrices for each beam element of the mathematical model

% the lumped disk mass m is included at appropriate degree of freedom
Mn = [ 156 22*l 54 -13*l;
22*l 4*l^2 13*l -3*l^2;
54 13*l 156 -22*l;
-13*l -3*l^2 -22*l 4*l^2];
roh= 7850 %kg/ m^3;
;
d = 0.01% mm;
A=pi*d/4 %mm^2;
m=0.5 % in kg;

b=roh*A*l/420

M1 = b * Mn;
M2 = b * ( Mn + blkdiag(zeros(2),m,zeros(1)));
M3 = b * Mn;
M4 = b * Mn;
M5 = b * ( Mn + blkdiag(zeros(2),m,zeros(1)));
M6 = b * Mn;
M7 = b * Mn;

c1=10;
c2=10;
C = [blkdiag(c1,zeros(5))+blkdiag(zeros(5),c2)];

Assemblage of stiffness and mass matrices

To assemble the stiffness matrices a function is programmed which can be called

out:

K = - 74 -ssemblage( {K1, K2, K3, K4, K5, K6, K7} )

The assemblage-function is built according to:

function [ K ] = - 74 -ssemblage( arrK )

- 74 -
 shiftSize = 2; %shifts matrices for 2 columns and 2 rows

numberInputMatrix = length(arrK); %number of element stiffness

matrices
sizeInputMatrix = length(arrK{numberInputMatrix}); %size of the
%element stiffness
matrices

sizeResultMatrix = shiftSize*(numberInputMatrix-1) +
sizeInputMatrix;
%size of global stiffness
matrix

K = zeros(sizeResultMatrix, sizeResultMatrix); %fills the global

matrix
%with zeros
sizeZerosLeft = 0;
sizeZerosRight = shiftSize*(numberInputMatrix-1);
for i=(1:length(arrK)) %number of element stiffness matrices
K = K + blkdiag( zeros(sizeZerosLeft), arrK{i},
zeros(sizeZerosRight));
%fills the global matrix with element stiffness matrices
values
sizeZerosLeft = sizeZerosLeft + shiftSize; %how many zeros
are
%left to the left when filling in every individual el.
Stiffness matrix
sizeZerosRight = sizeZerosRight – shiftSize; %how many zeros
are
%left to the right when filling in every individual el.
Stiffness matrix
end

end

C = assemblage( {C1, [0], [0], [0], [0], [0], [0], [0], [0], [0],
[0], [0], [0], [0], [0], C2}, 1 )

The Matlab Code , - ( ) yields to the natural frequencies and mode

shapes.

Mode Shapes and Natural Frequencies

[shape,frequency]=eig(K,M) %frequency are generalized eigenvalues in
omega^2

Wn_full=sqrt(frequency) %rad/s
Vn_full=sqrt(frequency)/(2*pi)*60 %rpm
Then various natural frequencies and mode shapes can be compared. This code
produces a diagonal matrix D of generalized eigenvalues and a full matrix V. The
columns of the full matrix V are the corresponding eigenvectors .

- 75 -
The mode shapes:

Figure 56 Mode Shape 1 for natural frequency 67 rpm

Figure 57 Mode Shape 2 for natural frequency 560 rpm

Guyan Reduction Method

From the mathematical model 16x16 matrices are obtained. Unfortunately this

mathematical model is not a sufficient representation for the real rotatory

machinery because the stiffness values k1 and k2 and the damping values c1

and c2 are unknown. One method to obtain those and update the mathematical

- 76 -
model is to first match the coordinates of the reference (measured) model and

the mathematical model and then use a minimized objective function that

measure the distance between the analytical and measured matrices. To realize

this, the assumption will be made by using the results of the reference model for

the natural frequencies and mode shapes are correct and from there the k1, k2,

c1 and c2 values in the stiffness and damping matrices can be evaluated.

Since there is always a mismatch between the coordinates of the finite-element

model and those of the measured data, it is an advantage to reduce the systems’

matrices so they meet the measured degrees of freedom. Another possibility is to

expand the measured coordinates, so that they correspond to the finite-element

model. [3]

In this case the degrees of freedom of the mathematical model will be reduced to

meet the degrees of freedom of the reference model. Therefore the Guyan

Static Reduction (GSR) method is used. It reduces the , -, , -, , - matrices

so that they meet the six Degree of Freedoms of the measurement.

In the Guyan Static Reduction method the state and force vector, x and f, and the

mass and stiffness matrices (K and M) are split into the measured (master) and

unmeasured (slave) subvectors and matrices. The matrices related to the master

degrees of freedom are retained and those related to the slave degrees of

freedom are eliminated. Since just force is applied to the master degrees, the

equation becomes:

̈
[ ]{ } [ ]2 3 2 3 (29)
̈

In the following matrix it can be seen, how the matrices are split:

- 77 -
Figure 58 Splitting Matrices

Neglecting inertia terms for the second set of equations leads to:

(30)

This equation is used for the elimination of the slave coordinates and leads to

following equation:

, -
2 3 [ ]* + , -* + (31)
, - , -

The transformation from the full state vector to the reduced state vector can be

obtained from

, -
[ ] , - (32)
, - , -

The transformation vector leads then to the reduced mass, stiffness and damping

matrices:

, - , - , -, -

, - , - , -, -

, - , - , -, -

- 78 -
Programming Guyan Reduction Method
For reduced stiffness matrix
Guyan Reduction Method
% to reduce the matrix the transformation matrix has to be found first
% therefore from K Kss (just with the slave rows and columns, here 2 4
5 6
% 8 10 11 12 14 16 has to be build)

Kss=K([2 4 5 6 8 10 11 12 14 16],[2 4 5 6 8 10 11 12 14 16] )

%Ksm has to be foundm where the master row and columns are 1 3 7 9 13
15

Ksm=K([2 4 5 6 8 10 11 12 14 16],[1 3 7 9 13 15])

%to obtain T

Ksssm= -inv(Kss)*Ksm % which is the lower part of the matrix

%and identity matrix (length(Ksssm(1,) = columns number)

Ieye = eye( length(Ksssm(1,) )

% this two have to be put together vertically and it is Ts

Ts = vertcat(Ieye, Ksssm)

%then the reduced stiffness matrix can be found out

Kr=Ts’*K*Ts

%then the reduced mass matrix can be found out

Mr=Ts’*M*Ts

Leads to following reduced stiffness matrix:

Comparison of the resonant frequency in comparison to the reference

(experimental) model:
Table 4 Comparison Natural Frequencies
Mode Reference Mathemat %
model in ical model difference
rpm in rpm
1 1800 1115 61.94444
2 1980 1548 78.18182
3 5988 3725 62.20775
4 17280 6532 37.80093
5 39660 10039 25.31266
6 43020 19855 46.15295

- 79 -
As can be seen from the correlation between the experimental and computed

model data is for the first three modes acceptable, for the fourth, fifth and sixth

mode it differ significantly. Therefore a error function was used with the aim to

reduce the error in the mathematical model and obtain more compatible results

to the reference model. [1]

Error function
To modify the structural model matrices, the distances between the mathematical

and measured matrices have to be minimized. After Friswell and Mottershead

(1995) this is:

* + ( , - , - , -) * + (33)

Where , - is the mass matrix, , - the damping matrix, , - the stiffness matrix,

* + the ith mode shape, is the ith natural frequency, * + the the error vector

(residual force) and √ .

In this method the relevant parameters have to change until the error is

minimized. [1]

Structural Dynamics: The modal properties (the natural frequencies and the

mode shapes) which are related to the physical properties of a structure can be

described in the matter of their distributed mass, damping and stiffness matrices

in the time domain as following:

, - { ̈} , - { ̇} , -* + * + (34)

Where { ̈ } is the acceleration vector, { ̇ } the velocity vector, * + the

displacement vector and * + is the applied force vector.

- 80 -
To form an eigenvalue equation of the ith mode, the expression has to be

transposed into the modal domain.

(̅̅̅̅̅ , - ̅, - , -)* ̅ + * + (35)

Where * + is the null vector, ̅ is the ith complex eigenvalue which correlates to

the natural frequency . * ̅+ correlates to the normalized mode shape * + . It

can be seen that the change of the mass and stiffness matrices will change the

natural frequencies and mode shapes of the structure.

As a result the detection of the right mass and stiffness matrices leads to the

identification of the model properties. As described in the chapter xxx the

frequency response function is the ratio of the output response (which can be

measured as the displacement, the velocity or the acceleration) as a result of the

applied force. The FRF can be expressed over the modal properties by using the

modal summation equation.

( ) ∑ (36)

Whereas k is the excitation position and l is the position where the response is

measured. N is the number of modes and is the damping ratio. The measured

modal properties can be reproduced if the right mass and stiffness matrices are

identified. Following the minimized objective function that measures the distance

between the analytical and measured matrices is given, with which the

parameters k1, k2, c1 and c2 can be upgraded. For , -, , - and , - estimated

parameters are inserted. , - and , - contain unknown parameters. To find out

the unknown parameters that are contained in , - and , - the error function is

used.

- 81 -
 ‖∑ . , - (, - , -) (, - , -)/ * + ‖

Where ‖ ‖ is the Euclidean form. The mode shapes and natural frequencies are

measured. The mass, stiffness and damping matrices are received from the finite

element model. If the mode shapes and natural frequencies is expressed over

the matrices, then E is equal to zero.

The minimization of the Error function with respect to k1, k2, c1 and c2 will lead

to four linear equations whose solutions gives updated values for the bearing

stiffness and damping. [3] [8]

For the M, C and K matrices the matrices which were obtain with the Guyan

Static Reduction method. For the change of the stiffness and damping values

, - and , - following matrices are used:

[ ]

- 82 -
 [ ]

Let .– , - (, - , -)/ * + * +

Then

‖( , - (, - , -)) * +‖ * + * +

∑[( ( )) ( )

( ) ( ) ( )

( ) ]

∑[( ) ( ) ( )

( ) ( )

( ( )) ]

Let .– , - (, - , -)/ * + * +

Then

‖( , - (, - , -)) * +‖ * + * +

- 83 -
 ∑ , ( )

∑ , ( )

Before the values of the mode shapes can be inserted in the Error Function, they

have to be mass normalized, therefore:

{ }

With √
Table 5 Mass Normalized Shape Functions
shape shape shape shape
function function 1 function function 2
1 normalized 2 normalized
0.494 0.494 0.172 0.172
1.19 1.19 0.294 0.294
2.84 2.84 0.11 0.11
2.59 2.59 0.136 0.136
0.808 0.808 0.247 0.247
0.379 0.379 0.0513 0.0513

Programming normalized shape function in Matlab:

syms v1 % mode shape 1

v1=[0.172

- 84 -
0.294
0.11
0.136
0.247
0.0513
]
beta = sqrt(1/(v1’*Mr*v1))
v1_gen=beta*v1

Programming Error Function

syms deltak1
syms deltak2
deltaK = [blkdiag(deltak1,zeros(5))+blkdiag(zeros(5),deltak2)]

syms deltac1
syms deltac2
deltaC = [blkdiag(deltac1,zeros(5))+blkdiag(zeros(5),deltac2)]

syms Kr_delta
syms v1
syms v2
syms K_E
syms deltaE
syms diffdeltaE
syms diffdeltaE_deltak1
syms diffdeltaE_deltak2
Kr_delta=Kr+deltaK

%the modeshape 1 obtained from the experiment

w1=29.8*(2*pi)
w2=99.6*(2*pi)

phi1=[ 0.494
1.19
2.84
2.59
0.808
0.379]

phi2=[
0.172
0.294
0.11
0.136
0.247
0.0513]

K_E=Kr+deltaK

K_E_K_1=(-w1^2 * Mr + K_E)*phi1
K_E_K_2=(-w2^2 * Mr + K_E)*phi2

deltaE_K_1=K_E_K_1.’*K_E_K_1

- 85 -
deltaE_K_2=K_E_K_2.’*K_E_K_2

deltaE_K=deltaE_K_1+deltaE_K_2
diffdeltaE_deltak1=diff(deltaE_K,deltak1)
solve(diffdeltaE_deltak1, deltak1)

diffdeltaE_deltak2=diff(deltaE_K,deltak2)
solve(diffdeltaE_deltak2, deltak2)

syms deltaE_C
syms deltac1
syms deltac2
K_E_C_1=(-w1^2 * Mr + j*w1*deltaC)*phi1
K_E_C_2=(-w2^2 * Mr + j*w2*deltaC)*phi2
deltaE_C_1= K_E_C_1.’* K_E_C_1
deltaE_C_2= K_E_C_2.’* K_E_C_2
deltaE_C=deltaE_C_1+deltaE_C_2
diffdeltaE_deltac1=diff(deltaE_C,deltac1)
solve(diffdeltaE_deltac1, deltac1)

diffdeltaE_deltac2=diff(deltaE_C,deltac2)
solve(diffdeltaE_deltac2, deltac2)

leads to

This was then inserted in the stiffness and damping matrices for a

single beam and the matrices were assembled and reduced again. [1]

Verification of Error Function

On a very simplified Problem the Error Function will be verified.

The following mass and stiffness matrices

0 1

0 1

lead to those shapes and eigen frequencies:

0 1

0 1

- 86 -
When adding [ ] to the stiffness matric and applying

( , - (, - , -)) * + * +

Following equations are obtained:

[ 0 1 [ ]] 2 3 2 3

With ‖– , - (, - , -) * +‖ * + * + this equations become

‖ ‖ 2 3 2 3 for inserting the first mode shape and first natural

frequency.

‖ ‖ { } { } for inserting the second mode shape and second

natural frequency.

Considering ‖∑ ‖ + .

leads to a =0. This proves that the Error Function is equal to zero,

when the mode shapes and natural frequency of the system are inserted.

For the further verification

When the mode shapes are changed with just one value, here instead of “0” the

value “1” is inserted.

0 1

Solves for , when adding to the original stiffness matrix

[ ] and applying again the method of the least square with the Error

Function it can be proven that now the Error is 0. This proves that this method

- 87 -
uploads the finite-element model in a way that it matches the mode shapes and

natural frequencies of experimental obtained data.

Model Assurance Criterion

To validate the theoretical model, the computed properties and those detected

from the experimental measurement have to be compared. First the predicted/

obtained results from the Abaqus finite element model and the Matlab finite

element model are compared to the results obtained from the experimental

measurement. Then the results from the Matlab finite element are compared to

the experimental measurement. This will show the improvement of the Matlab

finite element model due to the use of the error function, which allowed to obtain

improved values for the stiffness and damping coefficients k1, k2, c1 and c2.

To compare the mode shapes the Model Assurance Criterion (MAC) is used. The

MAC method compares the theoretical mode shapes and those obtained from

the Abaqus finite element model and the Matlab finite element model. The model

assurance criterion is well-known and has been used already by various experts

within the last 20 years. After the equation of Allemang and Brown the MAC can

be mathematical expressed as:

|2 3{ }|
* + * +* + {
(37)
}

The indices have following meaning; c is the reference, d is the degrees-of-

freedom; r is the mode; T is the transpose; * is the complex conjugate; and * + is

a vector.

The points from the straight-line correlation are used and the least-squares

deviation is measured. Therefore a value close to 1 demonstrates that the two

- 88 -
compared mode shapes are well-correlated, whereas a value closes to 0

indicates that the mode shapes are less well-correlated. [1]

Programming MAC in Matlab

syms phi_cr
syms phi_dr
syms MAC

phi_cr=[0.494 %mode shape from experiment

1.19
2.84
2.59
0.808
0.379]

phi_dr=[0
0.1588
0.9813
0.8871
1
0.9117] %mathematical obtained mode shape

MAC=(phi_cr’*phi_dr)^2/(phi_cr’*phi_cr*phi_dr’*phi_dr)

Results
Mode Shape 1 and 2 obtained from the mathematical model before updating

the parameters :

Table 6 MAC Mode Shape 1 before updating

0.494 0.0

1.19 0.1588

2.84 0.9813

2.59 0.8871

0.808 1.0000

0.379 0.9117

Leads to an MAC of 0.6647 for the comparison of Mode Shape 1. \

- 89 -
Table 7 MAC Mode Shape 2 before Updating

0.494 0.172

1.19 0.294

2.84 0.11

2.59 0.136

0.808 0.247

0.379 0.0513

Leads to an MAC of 0.0209 for the comparison of Mode Shape 2.

Mode Shape 1 and 2 obtained from the mathematical model after updating the

parameters :

Table 8 MAC Mode Shape 1 after Updating

0.494 0

1.19 0.1243

2.84 1

2.59 0.7927

0.808 0.5786

0.379 0.5932

Leads to an MAC of 0.8186 for the comparison of Mode Shape 1.

- 90 -
Table 9 MAC Mode Shape 2 after Updating

0.494 1

1.19 0.8551

2.84 -0.269

2.59 0.0376

0.808 -0.0144

0.379 0.1097

Leads to an MAC of 0.4208 for the comparison of Mode Shape 2.

- 91 -
Chapter 5 – Conclusions

The main approach of this work to develop and update a finite-element model

has been achieved. It has been shown that a very simplified beam model can be

used to represent a complex rotary machinery structure. The main reason was to

be able to show every step in the overall process of generating and updating a

finite-element Model

Unknown bearing parameters have been added to the global matrices of the

mathematical model. Those have been obtained by applying the method of

assemblage.

To equate the degree of freedoms of the mathematical model to those obtained

from the experimental measurement, the Guyan Reduction Method has been

applied.

By applying the method of the least square to the Error Function, which

measures the distance of the experimental obtained mode shapes and natural

frequencies and the mathematical obtained matrices, the unknown bearing

parameters could be obtained and inserted in the originally mathematical

matrices.

To validate the theoretical model and those detected from the experimental

measurements, they have been compared with the Model Assurance Criterion

before and after the updating. For Mode Shape 1 an improvement from 0.6647 to

0.8186 and for Mode Shape 2 from 0.0209 to 0.4208 could have been obtained.

Since the values obtained from the comparison of the updated theoretical model

- 92 -
and the experimental measurements are closer to 1, it has been proven that the

mode shapes correlate more than before with each other.

This means that in the scope of this work a specific way of how a finite-element

model can be uploaded has been successfully shown. Generally by using the

Guyan Static Reduction Method every Finite-Element-Model can be equated to

the measurement data. Using the method of the least square with the Error

Function is a very reliable iterative method to obtain parameters. Not only the

stiffness and damping parameters can be updated, but every parameter that is

involved in the system matrices.

For further projects it is very interesting to upload the matrices of the Finite-

Element Model obtained from Abaqus. The Abaqus Finite-Element Model

consists of many more elements and since the geometry of the whole rotary

machinery is considered; it will be a better representation of the actual rotary

machinery than a simplified beam model.

- 93 -
References

[1] Tshilidizi Marwala, Finite-Element-Model Updating using Comptational

Intelligence Techniques. Springer Verlag London Limited 2010
[2] W Wang, JE Mottershead, C. Mares (2009) Mode-shape Recognition
and Finite Element Model Updating Using the Zernike Moment
Descriptor. Mech Syst and Signal Process 23:2088–2112
[3] RJ Guyan (1965) Reduction of Stiffness and Mass Matrices. Am Inst of
Aeronaut and
Astronaut J 3:380
[4] Windowing: Optimizing FFTs Using Window, http://www.ni.com/white-
paper/4844/en, 08.07.2012
[5] Donald E. Bently, Fundamentals of Rotating Machinery Diagnostics.
Bently Pressurized Bearing Company 2002
[6] R. Nordmann, Identification of Modal Parameters of an Elastic Rotor
With Oil Film Bearings. University of Kaiserslautern,
Federal Republic of Germany 1984
[7] ABAQUS/ Explicit User’s Material Volume I. Hibbitt, karlsson &
Sorensen, Inc. 2002
[8] Roy R. Craig Jr. Andrew J. Kurdila, Fundamentals of Structural
Dynamics Second Edition. John Wiley & Sons, Inc. 2006
[9] Daryl L. Logan, A First Course in the Finite Element Method. University
of Wisconsin–Platteville, Nelson 2002
[10] R. Nordmann, “Modal Parameter Identification and Sensitivity Analysis
in Rotating Machinery,” International Conference “Rotordynamic
Problems in Power Plants,” Proceedings of the IFTOMM Conference,
Rome, Italy, Sept. 29-Oct. 1, 1982, pp. 95-102.
[11] Michel Verhaegen, Filtering and System Identification. Cambridge
University Press 2007
[12] Dan Biddle, Adverse Impact and Test Validation, Biddle Consulting
Group, Inc. 2002
[13] Brian Murphy, John M. Vance, Fouad Y. Zeiden, Machinery Vibration
and Rotordynamics, John Wiley & Sons 2010
[14] LDS Dactron, RT Pro Signal Analysis and Waveform Source, RT Pro
Brochure US 2005 09
[15] Dr. Jim Meagher, Introduction to Spectral Analysis. Lecture Notes for
ME 517 provided Fall 2012
[16] 22-403 Laboratory Experiment Introduction to Spectral Analysis,
University of Massachusetts J.B. Francis College of Engineering
Mechanical Eng. Department 2011

- 94 -
[17] Peter Avitabile, Teaching Experimetnal Structural Dynamics
Applications. University of Massachussetts Lowell.
[18] P. Avitabile, J.Hodgkins, Numerical Evaluation of Dispolacement and
Acceleration for a Mass, Spring, Dashpot System. ASSEE Conference,
Salt Lake, UT 2004
[19] A. Jauhola, E. Kinzel, D. Reding, N. Hunter, Modal Parameter for a flat
Plate supported by an Oil Film, Departement of Mechanical
Engineering, 2011
[20] D.J. Ewins, Modal Testing: Theory and Practice. Research Studies
Press LTD 1984
[21] P. Avitabile, B. Birdsong, Development of Efficient Reduced Models for
Flexible Body Dynamic Simulations, Structural Dynamics and Acoustics
Systems Laboratory University of Massachusetts Loweell

- 95 -
Appendix A Start and use of ADRE for Windows/ DAIU

1. Click on File on the ADRE for Windows main menu> New> No (Continue

without saving – in case no important experiments have been done

beforehand)

2. File> Open Database> to create a new database and determine a

subdirectory, where everything should be saved. By creating a new

database all configuration settings and sample data will be erased. The

collection is then carried out by the DAIU (Data Acquisition Interface Unit).

3. Trigger Event determines at with which event the DAIU begins to collect

data. Tab Configure…> Data Acquisition> Trigger Event… >Speed> Press

Configure> Keyphasor 1> Insert Over 200 and under 60000 RPM

Figure 59 Trigger Event Configuration

4. Sample Mode Configuration is used to specifiy which mode to use for

the data acquisition and to configure the interval for each mode. It is also

possible to use both sampling modes at once. With a trigger event (here

- 96 -
 the speed), the sampling begins and is taken at each interval, which is

specified in the dialog box. Sample Mode Configuration> Click on Delta

Rpm > Delta Rpm: 25 (Click on increasing and decreasing) >Delta Time: 2

sec. > Select Soure: Keyphasor 1

Figure 60 Sample Mode Configuration

5. The location each channel is identified under Channel Configuration.

Additionally channel pairs are created that must use the same Keyphasor,

since they are required for specific plots (Orbit, Orbit/ Timebase).

Transducer configuration identifies the type of transducer used by each

channel. Channel Configuration > Active Ch. > Click 1, 2, 3 & 4. Pick BRG

1 Vertical and horizontal with the same Keyphasor 1 and BRG 2 Vertical

and Horizontal with the same Keyphasor 1. The Angel for the horizontal

Channels must be set to 90R. Tot.> CCW (counter-clockwise).

- 97 -
Figure 61 Channel Configuration
6. Tab Plots gives the possibly to select the plots that are generated after the

data has been collected. Usually Polar, Bode and Real Time are the most

important, but it is always also interesting and sometimes necessary

(depending what the user is looking at) to generate other plots. Usually all

plots are selected.

Figure 62 Plot Control

- 98 -
When clicking configure next to Polar, the window with the Polar Plot

Configuration is opened. For the polar Plot the following configurations have

been chosen:

Figure 63 Polar Plot Configuration

When clicking configure next to Bode, the window with the Bode Plot

Configuration is opened. For the polar Plot the following configurations have

been chosen

- 99 -
Figure 64 Bode Plot Configuration
When clicking configure next to Real Time, the window with the Real Time Plot

Configuration is opened. For the polar Plot the following configurations have

been chosen

Figure 65 Real Time Plot Configuration

Tab Store Enable is used to start collecting the data in the DAIU. DAIU begins to

collect the data at the set trigger event.

- 100 -
Appendix B Balancing of two plane rotary machine

- 101 -
Appendix C Orbit Plots obtained with ADRE

Figure 66 Orbit Plot First Mode Shape, Probe 7

Detailed Orbit Plot for Mode Shape 2, probe 7

- 102 -
Figure 67 Orbit Plot Second Mode Shape, Probe 7

- 103 -
Appendix D RT Pro Focus 6.21

1. After the hardware for the experiment is set up, turn on the DACTRON

and the computer and open RT-Pro Focus 6.21.

2. Than the impact hammer output has been connected to one of the input

channels and the probes for the X-Y relative displacement measurements

have been connected to four different channels.

3. From the RT-Pro Focus 6.21 menu a new project has been started and

Modal Data Acquisition has been selected.

4. Than the Signals icon has been selected. In Auto Channel Signals for

the five used channels time capture, FFT and power spectra has to be

checked. In Cross Channel Signals FRF/Coh has been selected for the

input channel for the input channels. The excitation channel default is the

channel of the hammer. Click to determine that a row

5. Channel icon has been selected to set acquisition parameters for each

channel. All sensors for all tests are described here. For Max Volts the

highest to expect voltage amplitude has been selected. For this

experiment 10 Volts were chosen. Quantity specifies the type of input

signal; therefore displacement has been selected for the probes and

acceleration for the impact hammer. In mv/EU the sensitivity level has

been selected. The impact hammer has a toggle which allows choosing

between 1, 10 and 100. For this experiment it has been set to 100 mv/EU.

The exact values can be seen from the manufacturer’s calibration

specifications. The probes have a calibration specification of 100 mv/EU.

- 104 -
 For the Coupling AC, DC or ICP can be set. The coupling ICP stands for

integrated circuit piezoelectric sensor. It enables the AC coupling and

applies power to the ICP type sensors. ICP AC 0.7 Hz has been chosen

for this experiment. For the Axis the Z+ has been selected. I.D. permits to

name the single channels.

6. From the main menu Setup > Preferences > Miscellaneous has been

selected. There “use channel ID label for signal name” has been enabled.

Thus the names selected in the channel settings are used for all signal

names. They will appear in all legends for all different windows.

7. Than from the main menu Setup > Measurement Parameters >

Sampling has been selected. For the frequency span 12800 Hz has been

evaluated. The frequency span defines the upper frequency range for the

measurement. The windowing type selects the spectral window which is

applied to each frame of the data. Here Forced/Exponential has been

selected which is the most suitable window for modal data acquisition with

an impact hammer. The forced/exponential window applies damping to the

time history which guarantees that the response signals completely decay

within the data frame period.

- 105 -
Figure 68 Setting Parameters for Forced/Exponential Window
8. From Setup > Measurement Parameters > Averaging as the averaging

type Linear has been selected. As the average frame number 5 has

been determined. The reason for averaging is to increase the statistical

accuracy or restrain external noise. When selecting linear averaging each

set of data elements (data frames) contributes evenly to the average.

9. In the frequency range dropdown menu 12600 Hz has been selected and

1024 points.

10. From Setup > Measurement Parameters > Triggering the following

settings have been selected; Analog Input which guarantees the use of a

signal from any enabled input channel. The pre/post trigger has been set

to -33. This defines the pre- or the post-trigger time in terms of the number

of points. This is very important especially for impact analysis since it

assures that also the beginning of an impact impulse is captured. Level

defines the trigger event. The value is a percentage of the full EU range.

1% has been selected here. For Run Mode Auto Arm every frame has

been evaluated as suitable since this run mode waits for a trigger event

- 106 -
 after a start command, obtains a data frame and then pauses until the

next frame command is to re-arm the trigger to collect another data frame.

11. The project has been saved than under Project > Save As… and a

suitable file name has been evaluated.

12. From Window > 2D Displays > Two-Pane Horizontal which creates a

customized two-pane window and contains two separate and independent

user-defined plots. The upper window was used to display the hammer (t)

for the impact time trace. Therefore right click and contents has been

selected. The same procedure in the lower window with the difference that

one of the input signals has been selected.

13. From Window > 2D Displays > FRF Bode Window. The top pane

defines the phase in degrees and the bottom pane the magnitude in

decibel. With the right click contents have been selected and the FRF

signal H1_2,1(f) chosen.

14. From Window > New Window an additional single window has been

created. From right click > contents the coherence between the impact

hammer and one of the probes, Coh2,1(f). Than Window > Tile has been

selected. The coherence gives the possibility to observe the accuracy

after every test. A value closer to one means an accurate FRF at that

frequency. [15]

- 107 -
Appendix E MEScope

Following the procedure to building and animate a simplified model of the rotor

machinery in ME Scope is described.

Therefore first the model including all its testing points is build. Through the

Modal Data Acquisition with DACTRON four testing points have been recorded.

The determined FRF measurements are than imported into a Data Block file.

Than the FRF measurements have been assigned to the DOFs (Points and

Directions) on the model where the measurements were made. Afterwards the

model was animated and the approximate mode shapes for each resonance can

be displayed.

Figure 69 Main Layout of ME Scope

1. File > Project > New

- 108 -
 2. File > New > Structure

3. In Structure Window: File > Options… > Units and chose the suitable

units, in this project the metric system is used.

Figure 70 Units

4. Draw > Drawing Assistant and chose the icon line in the substructure. In

the tab Dimensions the length of the line (560 mm) and the number of

points (six) can be entered. Under Position global has been chosen.

5. File > Import > Data Block…, which is a *.uff file. This action imports the

FRF measurements (which are calculated from the data taken during the

roving impact test), from the Modal Data Acquisition and makes it able to

display the Operation deflection shapes (ODS).

6. A window with the translate files is opened. Manually the DOFs have to be

changed. The Trace DOF is equal to the Roving DOF divided through the

Reference DOF; the impact points (where the structure has been excited)

are the roving DOFs. The reference DOF is the accelerometer point (here

the displacement point). Since there is just one reference DOF this is

- 109 -
 called a Single Reference Test. For more accurate results several rows

have been recorded.

Figure 71 Assigning DOFs

7. To animate the measured DOFs the measured animation equations are

created. Therefore the measurement numbers and measurement

directions have to be matched. The measurement axes have to be

oriented in the direction of the measurement.

8. In the structure window: Draw > Number Points and then on each point

of the model has to be clicked. Click close when finished.

9. Edit > Objects > Points

- 110 -
 10. Display > Point labels. As can be seen each point on the line has its own

local measurement axes.

11. Draw > Animation equations > equation editor

12. Edit > select object > select all

13. Draw > assin measurement numbers > select match structure and

source DOFs

14. Window > Arrange > for animation

15. Click on animation of structure

Procedure for Curve Fitting

In a frequency band of measurements the numbers of modes have to be

determined. Within this frequency band the modal frequency and damping has

to be estimated. With this estimates the modal residues can be determined. All

modal parameters have to be saved then in a Shape Table file.

1. Format > Overlay the FRFs can be overlaid and the resonance peaks at

the similar frequency in all FRFs can be determined.

2. By using the band cursor only data within the cursor band is curve fitted.

This is of advantage because just the valid resonance peaks should be

curve fitted and not the entire data of each trace.

3. With the mode indicator the number of modes in a cursor band can be

approved. Method>Modal Peaks Function and then Count Peaks

- 111 -
1. Frequency & Damping> Method: Orthogonal Polynomial> Global> Press
Freq. & Damp

2. Residues & Save Shapes> Method: Orthogonal Polynomial> Single

Reference> Residues

3. With Save Shapes the values of the Mode Shapes are obtained

- 112 -
Appendix E Remaining MATLAB code

Programming and assemblage of mass matrices

Mn = [156 22*l 54 -13*l; 22*l 4*l^2 13*l -3*l^2; 54 13*l 156 -22*l; -
13*l -3*l^2 -22*l 4*l^2]

b=roh*A*l/420;

M1= b* Mn
M2 = b * ( Mn + blkdiag(zeros(2),m,zeros(1)))
M3 = b * ( Mn + blkdiag(m,zeros(3)))
M4= b* Mn
M5 = b * ( Mn + blkdiag(zeros(2),m,zeros(1)))
M6 = b * ( Mn + blkdiag(m,zeros(3)))
M7= b* Mn

M = assemblageK( {M1, M2, M3, M4, M5, M6, M7} )

function [ M ] = assemblageM( arrM )

shiftSize = 2; %shifts matrices for 2 columns and 2 rows

numberInputMatrix = length(arrM); %number of element mass matrices

sizeInputMatrix = length(arrM{numberInputMatrix}); %size of the
%element mass matrices

sizeResultMatrix = shiftSize*(numberInputMatrix-1) +
sizeInputMatrix;
%size of global mass matrix

M = zeros(sizeResultMatrix, sizeResultMatrix); %fills the mass

matrix
%with zeros
sizeZerosLeft = 0;
sizeZerosRight = shiftSize*(numberInputMatrix-1);
for i=(1:length(arrM)) %number of element mass matrices
M = M + blkdiag( zeros(sizeZerosLeft), arrK{i},
zeros(sizeZerosRight) );
%fills the global matrix with element mass matrices values
sizeZerosLeft = sizeZerosLeft + shiftSize; %how many zeros
are
%left to the left when filling in every individual el. mass
matrix
sizeZerosRight = sizeZerosRight - shiftSize; %how many zeros
are
%left to the right when filling in every individual el. mass
matrix
end

end

- 113 -