Finite element method based analysis and modeling in rotordynamics

A thesis submitted to the

Graduate School

of the University of Cincinnati

in partial fulfillment of the

requirements for the degree of

Master of Science

In the Department of Mechanical Engineering

of the College of Engineering and Applied Sciences

by

Bradley Weiler

B.S. Purdue University

April 2012

Committee Chair: J. Kim, Ph.D.

Abstract

Rotordynamics refers to the analysis and study of the vibratory motion of rotating systems. The

application of finite element analysis (FEA) to rotordynamics allows for modeling rotors that have

complex geometry, which however requires sound understanding of basic concept and theory of

rotordynamics and the constraints of existing FEA software.

This study is to implement FEA based rotordynamics analysis to test, validate and compare capabilities

of the rotordynamics part of commercial software, ANSYS and NASTRAN. The formation of FEA matrices

containing concepts specific to rotordynamics such as gyroscopic effects, Coriolis force, spin softening

effects, internal and external damping, and circulatory matrices are discussed focusing on their effects

on the stability. A comparison between using the lumped parameter modeling approach and solid

elements to model a rotor disk, and the benefits of the respective approaches are also discussed. The

capabilities and limitations of current versions of ANSYS Workbench and NASTRAN are discussed with

suggested improvements.

While transient, harmonic, and modal analysis are all covered, the main focus is on modal analysis to

specifically discuss some unique issues that rotordynamic modal analysis presents. Analyses show that

the directivity information has to be used in the analysis for correct tracking of critical speeds and

instability, which is a problem not encountered in non-rotating systems. The method to handle rotating

structural damping in current commercial FEA software is discussed, and its implication is discussed

using an analytical model of a simple system. For example, it is shown that modal solvers of current

commercial software cannot properly handle frequency dependent structural damping.

ii
iii
Acknowledgements

I would like to thank my advisor Dr. Jay Kim for his guidance throughout my thesis work. His support and

out my thesis work. His support and valuable insights have been instrumental during this process. His

arrangement of financial support during my study at the University of Cincinnati is greatly appreciated

and has enabled me to be in the position that I am today.

I am grateful to both Dr. Kim and Dr. Randall Allemang for the wide range of courses they offer in

theoretical and experimental vibration analysis. Their teachings have allowed me to gain a wealth of

knowledge and experience in the field of study I am most passionate about. I would also like to thank

the faculty members for whom I have served as a Teaching Assistant and the people at the University of

Cincinnati Simulation Center who have given me great experiences.

Mention must also be made of the incredible support and encouragement that my parents and family

have made without which I may never have made the decision to attend graduate school. The advice

and support of my fellow graduate students here is something I am extremely grateful for and which has

helped me greatly. Finally, I must thank my first mentor in industry, Jim Gutknecht, who introduced me

to the challenges of analyzing vibrations of rotating structures and helped develop my fascination with

the subject.

iv
Table of Contents

Abstract ......................................................................................................................................................... ii
Acknowledgements...................................................................................................................................... iv
List of Figures .............................................................................................................................................. vii
List of Symbols ............................................................................................................................................. ix
1. Introduction .......................................................................................................................................... 1
1.1. Motivation and Significance.......................................................................................................... 1
1.2. Fundamental Concepts of Rotordynamics .................................................................................... 2
1.2.1. Stationary Coordinate System .............................................................................................. 2
1.2.2. Rotating Coordinate System ................................................................................................. 3
1.2.3. Gyroscopic Effect .................................................................................................................. 5
1.2.4. Coriolis Effect ........................................................................................................................ 6
1.2.5. Circulatory Matrix ................................................................................................................. 6
1.2.6. Spin Softening and Stress Stiffening ..................................................................................... 7
1.2.7. Rotor Whirl and Critical Speeds ............................................................................................ 8
1.3. Finite Element Modeling ............................................................................................................... 9
1.3.1. ANSYS Workbench .............................................................................................................. 10
1.3.2. NASTRAN ............................................................................................................................. 10
2. Finite Element Modeling ..................................................................................................................... 13
2.1. Finite Element Matrices .............................................................................................................. 13
2.2. Lumped Parameter Modeling ..................................................................................................... 14
2.3. Solid Element Modeling .............................................................................................................. 15
2.4. Element Selection ....................................................................................................................... 16
2.5. ANSYS Workbench and NASTRAN Rotor Model ......................................................................... 18
3. Stationary Reference Frame ............................................................................................................... 20
3.1. Inertia and Stiffness Matrices ..................................................................................................... 20
3.2. Gyroscopic Matrix ....................................................................................................................... 23
3.3. Bearings and the Non-Rotating Damping Matrix........................................................................ 24
3.3.1. Modeling Bearings as Springs and Dampers ....................................................................... 24
3.3.2. Bearing Elements with User-Defined Stiffness and Damping ............................................. 25
3.3.3. Modeling Journal Bearings.................................................................................................. 27

v
 3.4. Rotating Damping Matrix ............................................................................................................ 28
3.5. Circulatory Matrix ....................................................................................................................... 29
3.6. ANSYS Implementation of the Stationary Reference Frame ...................................................... 32
3.7. NASTRAN Implementation of the Stationary Reference Frame ................................................. 32
4. Rotating Reference Frame .................................................................................................................. 36
4.1. Similarities to the Stationary Frame ........................................................................................... 36
4.2. Coriolis Matrix ............................................................................................................................. 37
4.3. Spin Softening and Stress Stiffening Matrices ............................................................................ 38
4.4. Circulatory Matrix due to Non-Rotating Damping ...................................................................... 39
4.5. ANSYS Workbench Implementation of the Rotating Reference Frame...................................... 40
4.6. NASTRAN Implementation of the Rotating Reference Frame .................................................... 41
5. Rotordynamic Analysis ........................................................................................................................ 42
5.1 Modal Analysis ............................................................................................................................ 42
5.1.1 Complex Eigenvalues .......................................................................................................... 48
5.1.2 Campbell Diagrams and Critical Speeds.............................................................................. 49
5.1.3 Mode Tracking .................................................................................................................... 51
5.1.4 Directivity and Complex Coordinates Descriptions ............................................................ 52
5.2 Harmonic Analysis ....................................................................................................................... 55
5.3 Transient Analysis ....................................................................................................................... 59
6. Damping and Stability ......................................................................................................................... 62
6.1 Rotating Damping Mechanisms .................................................................................................. 62
6.1.1 Viscous Damping in Rotors ................................................................................................. 63
6.1.2 Structural Damping in Rotors .............................................................................................. 64
6.2 Viscous Damping and Effect on Stability..................................................................................... 67
6.3 Structural Damping and Effect on Stability ................................................................................. 72
7 Conclusion ........................................................................................................................................... 75
7.1 Summary and Recommendations ............................................................................................... 75
7.2 Future Work ................................................................................................................................ 76
References .................................................................................................................................................. 77

vi
List of Figures

Figure 1-1 - Stationary Reference Frame ...................................................................................................... 3

Figure 1-2 - Rotating Reference Frame ......................................................................................................... 4
Figure 1-3 - Centrifugal Force ....................................................................................................................... 7
Figure 1-4 – Whirling..................................................................................................................................... 8
Figure 1-5 - NASTRAN .bdf File Format ....................................................................................................... 11
Figure 2-1 - Beam Element.......................................................................................................................... 14
Figure 2-2 - Point Mass Element ................................................................................................................. 15
Figure 2-3 - Solid Elements ......................................................................................................................... 16
Figure 2-4 - Rotor Model............................................................................................................................. 19
Figure 3-1 - Bearing as Springs and Dampers ............................................................................................. 25
Figure 3-2 - CBUSH and PBUSH Data Entry Cards ....................................................................................... 26
Figure 3-3 - Rigid Body Element RBE2 Card ................................................................................................ 27
Figure 3-4 - Squeeze Film Damper NLRSFD Data Entry Card ...................................................................... 28
Figure 3-5 - Circulatory Matrix Effect on Non-Dimensionalized Natural Frequencies ............................... 32
Figure 3-6 - ROTORG Data Entry Card ......................................................................................................... 33
Figure 3-7 - ROTOR Data Entry Card ........................................................................................................... 33
Figure 3-8 - RGYRO Data Entry Card ........................................................................................................... 34
Figure 3-9 - RSPINR and RSPINT Data Entry Cards ...................................................................................... 35
Figure 5-1 – ANSYS Workbench Mode Shapes ........................................................................................... 44
Figure 5-2 - NASTRAN Mode Shapes........................................................................................................... 45
Figure 5-3 - EIGC Data Entry........................................................................................................................ 47
Figure 5-4 - Root Locus Plot ........................................................................................................................ 48
Figure 5-5 – ANSYS Workbench Campbell Diagram.................................................................................... 49
Figure 5-6 - Campbell Diagram with Positive and Negative Frequencies ................................................... 50
Figure 5-7 - CAMPBLL and DDVAL Data Entry Card .................................................................................... 51
Figure 5-8 - Campbell Diagram without Directivity Considered ................................................................. 53
Figure 5-9 - NASTRAN Complex Eigenvalue Results.................................................................................... 54
Figure 5-10 - ANSYS Workbench Frequency Response Function................................................................ 56
Figure 5-11 - APDL Commands for Unbalance Force .................................................................................. 57
Figure 5-12 - FREQ Data Entry Card ............................................................................................................ 58
Figure 5-13 - UNBALNC Data Entry Card ..................................................................................................... 58
Figure 5-14 - TSTEPNL Data Entry Card ....................................................................................................... 60
Figure 6-1 - Sign of Circular Structural Frequency ...................................................................................... 65
Figure 6-2 - Stability Effect of Viscous Rotating Damping in ANSYS Workbench Stationary Reference
Frame .......................................................................................................................................................... 69
Figure 6-3 – Stability Effect of Viscous Rotating Damping in NASTRAN Stationary Reference Frame ....... 70
Figure 6-4 – Stability Effect of Viscous Rotating Damping in ANSYS Workbench Rotating Reference Frame
.................................................................................................................................................................... 72
Figure 6-5 - Stability Effect of Structural Rotating Damping in NASTRAN Stationary Reference Frame .... 73

vii
Figure 6-6 - Stability Effect of Structural Rotating Damping in ANSYS Stationary Reference Frame ......... 74
Figure 6-7 - Stability Effect of Structural Rotating Damping in ANSYS Rotating Reference Frame ............ 74

viii
List of Symbols

[ ] - matrix containing all degrees of freedom at every node for all elements

[ ]𝑒𝑙 - matrix containing all degrees of freedom at every node of an element

[ ]𝑔 – generalized matrix containing all matrices with the same dimensionality

[ ]𝑛 - matrix containing all degrees of freedom for a node

[ ]𝑛𝑡 - matrix containing translational degrees of freedom for a node

[𝐴], [𝐵] – linearization matrices for modal analysis

[𝐵𝑛 ] – circulatory matrix due to non-rotating damping

[𝐵𝑟 ] – circulatory matrix due to rotating damping

[𝐶] – damping matrix

[𝐶𝑛 ] – non-rotating damping matrix

[𝐶𝑟 ] – Coriolis matrix

[𝐶𝑟 ] – rotating damping matrix

c - damping

𝑐𝑞𝑟 – damping in translational degree of freedom q due to velocity in translational degree of freedom r

D - diameter
𝜕
𝜕𝑡
( ) – derivative with respect to time

{𝐹} - nodal force vector

[𝐹𝐶/𝑆 ]𝐹/𝑅 – force due to damping/stiffness in the fixed/rotating frame

[𝐺] - gyroscopic matrix

j = √−1

𝐽𝑛𝑞 – rotational inertia for rotational degree of freedom about direction q at node n

k - damping

[𝐾] – stiffness matrix

[𝐾𝑐 ] – spin softening matrix

[𝐾𝑛 ] – non-rotating stiffness matrix

ix
[𝐾𝑟 ] – rotating stiffness matrix

[𝐼] – identity matrix

[𝑀] – inertia/mass matrix

m - mass

𝑚𝑛 – mass at node n

[𝑁] - shape function matrix, applied to a single element

{𝑞}, {𝑞̇ }, {𝑞̈ } – nodal vectors (displacements, velocities, and accelerations respectively) in the stationary
reference frame

R – radius

{𝑟} – complex coordinates description about displacement in X and Y directions

[𝑆] – stress stiffening matrix including centrifugal force terms

t – time

[𝑇] – transformation matrix to change from stationary coordinates to rotating coordinates

[𝑇]𝑇 – transpose of the transformation matrix

{𝑢}, {𝑢̇ }, {𝑢̈ } - nodal vectors (displacements, velocities, and accelerations respectively) in the rotating
reference frame

XYZ – global, stationary coordinate system

xyz – rotating coordinate system

α – mass matrix coefficient for Rayleigh damping

β – stiffness matrix coefficient for Rayleigh damping

η – Structural or hysteretic damping ratio

𝜙𝑞 – rotation about axis q

{𝜙} – complex coordinates description of rotation about axes X and Y

λ – complex eigenvalue

σ – stability, real part of complex eigenvalue

{𝜓} – eigenvector

Ω – rotor spin speed

{𝛺} – rotor spin speed vector: {𝛺𝑥 , 𝛺𝑦 , 𝛺𝑧 }

x
[𝛺] – rotor spin speed matrix

𝛺̇ – rotational acceleration

{𝛺̇} – rotational acceleration vector: {𝛺𝑥̇ , 𝛺𝑦̇ , 𝛺𝑧̇ }

ω – circular or whirl frequency, imaginary part of complex eigenvalue

xi
1. Introduction

1.1. Motivation and Significance

Analyzing and predicting vibration behavior of structures is important in the design and

development of mechanical systems. A rotor refers to a mechanical system in which at least one

part rotates with a very high angular momentum. Vibration analysis of rotor systems requires

special knowledge because of several unique behaviors of the structure stemming from rotation

effect, which is not observed in non-rotating structures. These concepts specific to rotors are the

gyroscopic effect, Coriolis effect, spin softening, rotating damping, and mode directivity. The

concepts of the gyroscopic effect, Coriolis effect, and spin softening are critical in the rotordynamics

field and have been discussed at length in literature[1-3].

Another unique aspect of rotordynamics is the effect of rotating damping. In non-rotating structures

damping is always a stabilizing factor. Damping reduces the magnitude of vibration at resonance

and makes the system more stable. In rotating systems, damping present in the rotating part of the

system can act as a destabilizing influence in certain situations[4,5]. In non-rotating vibrating systems,

the motion of the system oscillates harmonically along a linear path. In rotating systems, vibration is

actually a whirling, or circular, motion. Whether the whirling motion is in the same direction of the

rotation (forward motion) or opposite direction of the rotation (backward motion) is important in

the analysis of the system.

Application of FEA modeling enables analysis of more complex rotor systems, however care must be

taken to ensure correct implementation of the analysis. Early rotor finite element models were

limited to beam and point mass elements because it is very simple to define rotational degrees of

freedom and gyroscopic effects. With the development of Geradin and Kill[6] of gyroscopic matrices

1
 for solid elements, finite element models of rotors have expanded to more complex models[7-10].

Finite element models of rotating structures utilize similar modal, frequency response, and transient

solvers as non-rotating models and have been developed to determine overall rotor motion and

predict natural frequencies and Campbell diagrams. Commercially available finite element programs

such as ANSYS Workbench and NASTRAN have been shown to match rotor natural frequencies

determined via rotor testing[11].

Discussion of rotor stability, the effect of rotor internal damping, and mode directivity are found in

literature[1-2], but discussions are mostly based on analytical models. The implementation of these

concepts into the finite element method requires additional knowledge and understanding. To

understand how these concepts should be implemented in the finite element method, a broader

discussion of rotordynamics finite element modeling is needed. The construction of ANSYS

Workbench and NASTRAN rotor models and the implementation of code internal matrices are

discussed in the opening sections of this thesis. Using this as a foundation for further analysis; rotor

internal damping, stability, and mode directivity in finite element modeling is then addressed.

Limitations and inaccuracies in the ANSYS Workbench and NASTRAN implementation of these

concepts are shown along with suggested changes that would allow for improved simulation

capabilities.

1.2. Fundamental Concepts of Rotordynamics

1.2.1. Stationary Coordinate System

The stationary coordinate system is a global, fixed reference frame. It is defined by using the three

principle axes X, Y, and Z. In rotordynamics, it is common for the axis of rotation for the rotor to

align along the Z axis (Figure 1-1).

2
 Figure 1-1 - Stationary Reference Frame

The equation of motion for the stationary reference frame can be derived using either the Newton-

Euler method or the Lagrange method and takes the form of Eq. 1.1 [1, 2, 12].

[𝑀]{𝑞̈ } + (𝛺[𝐺] + [𝐶𝑟 + 𝐶𝑛 ]){𝑞̇ } + ([𝐾] + 𝛺[𝐵𝑟 ]){𝑞} = {𝐹} 1.1

The above equation relates the motion of the rotor {𝑞} to its inertia [𝑀], stiffness [𝐾], and external

forces {𝐹}. The damping of the rotor is comprised of rotating damping [𝐶𝑟 ] and non-rotating

damping [𝐶𝑛 ]. The circulatory term 𝛺[𝐵𝑟 ] occurs due to the transformation of the rotating damping

term into the stationary reference frame. The gyroscopic moment in the system due to the rotation

of the system is accounted for by the term 𝛺[𝐺].

1.2.2. Rotating Coordinate System

The rotating coordinate system is a moving coordinate system that rotates with rotor spin speed

(Figure 1-2).

3
 Figure 1-2 - Rotating Reference Frame

The equation of motion in the rotating reference frame can be derived by transforming the

stationary reference frame equation of motion to the moving coordinate frame[1, 2, 12]. This

transformation relates the acceleration in the rotating reference frame to the acceleration in the

stationary reference frame.

{𝑢̈ } = {𝑞̈ } − 2{𝛺} × {𝑢̇ } − {𝛺} × ({𝛺} × {𝑢}) − {𝛺̇} × {𝑢} 1.2

When combined with the mass matrix, the three additional terms needed to describe the

acceleration in the rotating reference frame create frame dependent forces. These forces are not

actually acting on the body, but are present due to the constraint of viewing the rotor in the non-

inertial frame. The three forces are the Coriolis force, centrifugal force, and Euler force and are

created by the terms 2{𝛺} × {𝑢̇ }, {𝛺} × ({𝛺} × {𝑢}), and {𝛺̇} × {𝑢} respectively. The Euler force is

zero for rotors with a constant spin speed. The rotating reference frame equation of motion

(assuming a constant spin speed) is

[𝑀]{𝑢̈ } + ([𝐶𝑜𝑟 ] + [𝐶𝑟 ] + [𝐶𝑛 ]){𝑢̇ } + ([𝐾] − [𝐾𝑐 ] + [𝑆] + 𝛺[𝐵𝑛 ]){𝑢} = {𝐹} 1.3

4
 In the above equation of motion, the Coriolis force is referenced by the matrix [𝐶𝑜𝑟 ]. The centrifugal

force is represented by the spin softening matrix [𝐾𝑐 ]. The stress stiffening matrix [𝑆] represents

changes in the stiffness of the structure due to geometric changes or changes in the stress field of

the structure. The circulatory matrix 𝛺[𝐵𝑛 ] is due to the fact that non-rotating damping acts in the

stationary reference frame. When transformed into rotating reference frame a skew-symmetric

circulatory matrix term proportional to displacement vector appears.

1.2.3. Gyroscopic Effect

The gyroscopic effect refers to the gyroscopic moment that occurs due to the angular velocity of a

body. A body with rotational inertia that undergoes an angular velocity experiences this gyroscopic

moment. For a disk with polar inertia 𝐽𝑝 , rotation with angular velocity 𝛺 about the Z-axis, the

kinetic energy from the gyroscopic couple can be approximated[12] as

𝐸 𝐺𝑘 = −𝐽𝑝 𝛺∅𝑌̇ ∅𝑋 1.4

This approximation assumes that the rotation about the X and Y axis is small and also allows for the

gyroscopic couples from those rotations to be neglected. Translating the kinetic energy expression

for the gyroscopic couple to its moment equivalent results in the gyroscopic moment term

0 0 0 0 0 0 𝑋̇
0 0 0 0 0 0 𝑌̇
0 0 0 0 0 0 𝑍̇
𝛺[𝐺]{𝑞̇ } = 𝛺 −𝐽𝑝 0 ∅𝑋̇ 1.5
0 0 0 0
0 0 0 𝐽𝑝 0 0 ∅𝑌̇
[0 0 0 0 0 0] {∅𝑍̇ }

The gyroscopic effect is a skew-symmetric matrix and is directly proportional to the rotor spin speed

which results in rotor natural frequencies changing with spin speed. The gyroscopic effect is only

present in the stationary reference frame. When it is translated to the rotating reference frame, the

effect is part of the reference frame motion and thus does not show up in the equation of motion.

5
 For elements without rotational degrees of freedom, the gyroscopic effect is derived in a similar

manner. The kinetic energy from rotation is determined for the element, and this kinetic energy

term is then transformed into translational degrees of freedom.

1.2.4. Coriolis Effect

The Coriolis effect captures some of the kinetic energy associated with a mass as it rotates about an

axis of rotation in the rotating reference frame. When viewed in the rotating reference frame, the

entire system is rotating with rotor spin speed 𝛺 and the kinetic energy associated with this rotation

is added to the system. To account for this additional energy, the Coriolis force, spin softening force,

and Euler force are added to the rotating reference frame equation of motion[12]. The spin softening

and Euler forces are dependent on the displacement vector of the rotor and change the apparent

stiffness of the rotor. The Coriolis force 2{𝛺} × {𝑢̇ } is dependent on the velocity vector of the

system, but is skew-symmetric and therefore affects the rotor natural frequencies more than the

rotor stability. The Coriolis effect is significantly easier to implement than the gyroscopic effect for

finite element nodes that do not have rotational degrees of freedom.

1.2.5. Circulatory Matrix

Damping in rotordynamics is defined as two separate contributions: rotating and non-rotating

damping. Rotating damping results in two phenomena[1, 4, 5]. The first is the dissipation of energy

which is accounted for by the rotating damping matrix [𝐶𝑟 ]. The second phenomenon is the

transformation of energy from the rotation of the system to vibration. This transformation of energy

to vibration is accounted for by the circulatory matrix [𝐵𝑟 ] in the stationary reference frame and by

the rotating damping matrix itself in the rotating reference frame. Non-rotating damping always

results in the dissipation of energy from the system. In the stationary reference frame it is

6
 accounted for via the non-rotating damping matrix [𝐶𝑛 ], but in the rotating reference frame a

circulatory matrix due to non-rotating damping [𝐵𝑛 ] is required to counteract the implied rotation

that the non-rotating damping matrix is multiplied by. Both circulatory matrices are the result of the

transformation of their respective damping matrices from one reference frame to the other. For

both reference frames, the circulatory matrix is proportional to rotor spin speed and skew-

symmetric in nature. In the stationary reference frame, the circulatory matrix is proportional to the

rotating damping matrix and can be a destabilizing contribution leading to excessive system

vibration. In the rotating reference frame, the circulatory matrix is proportional to the non-rotating

damping matrix and is a stabilizing contribution used to correctly account for the effect of non-

rotating damping.

1.2.6. Spin Softening and Stress Stiffening

Spin softening is the apparent softening of the stiffness of a structure due to the centrifugal force

when viewed in the rotating reference frame. Spin softening is a negative contribution to the

generalized stiffness matrix and as a result lower the natural frequency of the system. The

centrifugal force is a frame dependent force that appears only when a system is viewed in the

rotating reference frame (Figure 1-3).

Figure 1-3 - Centrifugal Force

7
 Stress stiffening refers to the changing stiffness of a structure due to changes in the geometric

properties or stress field of the structure. In rotating structures, as a mass moves outward the

stiffness relating that mass to other points in the structure is deflected resulting in a changing

stiffness field. The change in the stress field due to a radial displacement away from the axis of

rotation is proportional to Ω squared[13].

When both spin softening and stress stiffening effects are active in a model, the stress stiffening

effects will be greater than the spin softening effects[14]. Lumped parameter rotor models will not

show the effects of spin softening or stress stiffening.

1.2.7. Rotor Whirl and Critical Speeds

Rotor whirling refers to the motion of the center of the geometry of the shaft[1]. Rotor whirling can

occur in the same or opposite direction as rotor spin. Rotor whirling in the same direction as rotor

spin is forward whirling; rotor whirling in the opposite direction is backward whirling (Figure 1-4).

Figure 1-4 – Whirling

A critical speed is defined as a rotor speed at which the frequency of a forcing function corresponds

to a natural frequency of the system[1, 15]. Not all intersections of forcing frequencies and natural

8
 frequencies are excited. It is possible for a forcing function and mode to be completely uncoupled

resulting in no resonance or for a mode to be sufficiently damped resulting in very little excitation[1,
15]
. Unbalance mass in a rotor provides an excitation at a frequency equal to rotor spin speed Ω.

Lateral modes, which are typically lower in frequency than torsional modes for realistic rotors, tend

to be well excited by unbalance mass[16]. Because of the particular danger of unbalance mass in

exciting modes, the spin speed at which an unbalance excitation frequency is equal to a natural

frequency of a rotor is referred to as a flexural critical speed[1]. Unbalance mass tends to only excite

forward whirl modes, but it is possible to excite a backward whirl mode with an unbalance

excitation when there is a mixed mode (two shafts spinning in opposite directions) or a ball bearing

with large differences in vertical and horizontal mount stiffness is present in the structure[15].

Subcritical speeds are rotor spin speeds below the flexural critical speed of a mode. Supercritical

speeds are rotor spin speeds above the flexural critical speed of a mode.

1.3. Finite Element Modeling

Finite element modeling is the process of dividing a structure into many subdivided elements. These

elements share points called nodes. The mass and rotating inertia of an element is distributed to its

nodes and the stiffness and damping relationship between nodes of the same element provide

internal forces that act upon the nodes. From the inertia of at the nodes, the internal forces acting

upon the nodes, and external forces acting upon the nodes an equation of motion for each degree

of freedom can be written for each node.

𝑚𝑞̈ = 𝑓𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 − 𝑓𝑖𝑛𝑡𝑒𝑟𝑛𝑎𝑙 = 𝑓𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 − 𝑐𝑞̇ − 𝑘𝑞 1.6

𝑚𝑞̈ + 𝑐𝑞̇ + 𝑘𝑞 = 𝑓𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 1.7

The equations of motion for all of the degrees of freedom at every node can be solved

simultaneously when combined into a matrix.

9
 Two commercially available finite element software programs are ANSYS and NASTRAN. Both

programs have developed rotordynamic specific code to handle the unique challenges of modeling

rotating systems. ANSYS Workbench 18.0 and NASTRAN version 2016 are examined in the following

chapters in order to discuss the specifics of finite element rotordynamics and the issues related to it.

While the process of building a rotating system model in these programs will be discussed, the focus

will be on the implementations of rotordynamic analysis equations.

1.3.1. ANSYS Workbench

ANSYS Workbench is a program that can be used to build a finite element model (pre-processor),

process a finite element analysis, and post-process the results. The pre-processor and post-

processor functions are handled by the program ANSYS Workbench and ANSYS Mechanical which is

a subprogram of ANSYS Workbench. When a model is built in ANSYS Workbench it is writing to an

input file using ANSYS Parametric Design Language (APDL) code. Several rotordynamic related

functions are not available in ANSYS Workbench and must be added as separate APDL command

lines. The APDL command lines specific to rotordynamic analysis will be discussed using the format

“command name, variable 1, variable 2, …” with the variables defined following the introduction of

the command.

1.3.2. NASTRAN

NASTRAN is a program utilized to perform analysis on finite element models. It does not have a pre-

processor or a post-processor although several programs are available to perform those functions

and create a NASTRAN input file or manipulate a NASTRAN output file. A NASTRAN input file, or .bdf

file, is a text file that defines the entire analysis. The input file is passed to NASTRAN where it is read

10
and the analysis is solved based on the instructions it contains. The .bdf file is separated into three

required sections in which the different parts of the analysis are defined. These sections are the

executive control section, case control section, and bulk data section. The executive control section

is used to define the type of analysis. The case control section is used to define analysis subcases,

define sets, and define output requests. The bulk data section defines the geometry, elements,

materials, loads, and boundary conditions.

Figure 1-5 - NASTRAN .bdf File Format

The executive control section is used to define controls for the entire analysis. The executive control

section is where the type of analysis is chosen and allows for optional input including diagnostics

and time limits on the analysis. The type of analysis is specified by the line “SOL #” where # is an

integer number indicating the solver being utilized. The line “CEND” indicates the end of the

executive control section and the beginning of the case control section.

Inputs to the case control section related to rotordynamic analysis are short inputs that take the

form of an equation. The case control section is used to define analysis subcases, define outputs,

and define sets that detailed in the bulk data section. The rotordynamic related case control entries

11
will be discussed in the following chapters with the format “case control entry = CCN” where CCN is

an integer value case control number that matches a bulk data entry.

The bulk data inputs are referred to as bulk data cards and have multiple variables to be defined

which must be placed in defined columns of the .bdf file. The order of the bulk data entries is not

important to the analysis. Bulk data cards that are utilized in rotordynamic analysis will be shown as

figures with the bulk data card name in the first column and the names of the other variables in the

other numbered columns. The relevant variables of the bulk data cards will also be discussed.

The NASTRAN solver produces several output files. The main output file is the .f06 file and contains

printed output such as displacements, stresses, eigenvalues, and eigenvectors. The .f06 file is a text

file and can be read using any text editor. The output data can also be read by a post-processing

software to analyze the results of the analysis.

12
2. Finite Element Modeling

2.1. Finite Element Matrices

Finite element analysis is the process of dividing a structure into discrete sections called elements

that share points called nodes. The nodes of a single element are related to each other by element

stiffness and damping. This stiffness and damping relationship between nodes of an element is

determined by material properties, the type of element, and the shape of the element. The mass of

the element is distributed to the node points and the rotational inertia of the nodes is determined

by the element shape and type. Elements can be beams, pipes, point masses, shells, or solid

element and each element type makes different assumptions in their stiffness relationship definition

between nodes. The motions (accelerations, velocities, and displacements) of the nodes due to

internal and external forces are written as a set of equations. These nodal equations are combined

into matrix form for easier manipulation.

̈ + [𝐶]{𝑞}
[𝑀]{𝑞} ̇ + [𝐾]{𝑞} = {𝐹} 2.1

When utilizing finite element analysis for rotor systems, the process is the same except that

rotordynamic analysis requires additional internal element relationships which form the gyroscopic,

Coriolis, circulatory, and spin softening terms. These additional internal element relationships are all

proportional to rotor spin speed Ω or rotor spin speed squared Ω2 . Only the gyroscopic matrix is

calculated directly using the element shape. The other rotordynamic specific matrices are

manipulations of existing mass, damping, and stiffness matrices.

Two modeling approaches are common for finite element analysis of rotors. The lumped parameter

approach models a rotor using beam, pipe, and point mass elements and is a simplifying

approximation of rotor behavior. The solid element approach uses solid and shell elements and is

more generally accurate but more calculation time intensive. Both of these modeling approaches

13
 can be utilized in ANSYS Workbench and NASTRAN. The matrix equations of each approach contain

the same terms but the matrices are much different.

2.2. Lumped Parameter Modeling

Lumped parameter modeling simplifies the spatial description of a model by distributing the physical

behavior into elements that approximate the behavior of the whole system. In rotordynamics, a

rotor shaft is approximated as a series of beam elements and rotor disks are approximated as point

mass elements located at a node of the shaft beam elements.

Beam elements consist of two nodes that each have six total degrees of freedom: three linear and

three rotational (Figure 2-1). The mass, rotational inertia, and gyroscopic effect of the beam element

are distributed to the two nodes. The relationship between the relative velocities and displacements

of the two nodes are described by the element’s damping and stiffness respectively. Beam element

stiffness definitions typically take the form of an Euler-Bernoulli beam or Timoshenko beam

definition. Pipe elements are identical to beam elements in their formulation but do not have a solid

cross-sectional area.

Figure 2-1 - Beam Element

14
 Point mass elements only add inertia terms of mass and rotational inertia to the system. A point

mass element does not have stiffness since it consists of a single node with six degrees of freedom

(Figure 2-2). Point mass elements have non-zero gyroscopic matrices in the stationary reference

frame if a polar inertia term is defined for the axis of rotation. Non-zero damping matrices are

present for point mass element if the Rayleigh damping definition is used and the α value in the

Rayleigh damping is non-zero.

Figure 2-2 - Point Mass Element

2.3. Solid Element Modeling

Solid element modeling or three dimensional modeling of rotor systems separates the rotor model

into discrete three dimensional elements. Solid elements consist of multiple nodes and form either

tetrahedrons or quadratic solids (Figure 2-3). Unlike beam and point mass elements, solid element

nodes only have translational degrees of freedom and do not have any rotational degrees of

freedom. The stiffness and damping definitions of solid elements define the relationships between

the nodes of the element.

15
 Figure 2-3 - Solid Elements

Shell elements are sometimes used in coordination with solid elements in finite element rotor

models. Shell elements can be used with the deformation of the element in one direction can be

neglected. The use of shell elements requires special care for attaching them to the surfaces of solid

elements to ensure that the degrees of freedom at shared nodes are the same.

2.4. Element Selection

There are several considerations to take into account when deciding which element types are best

for the simulation of rotor behavior. Element types have an effect on calculation time, the accuracy

of the model, and determine whether an existing geometric model can be used or a new model

must be created.

Matrix equations with fewer total degrees of freedom take less time to calculate solutions. This can

be important if computational power is limited or an approximate solution is needed quickly. Beam

and point mass models have six degrees of freedom at each node but the number of nodes on a

beam and point mass model is limited. Solid elements have three degrees of freedom at each node

but can have at least four nodes for each element. Models consisting of solid elements will have

many more elements than an equivalent lumped parameter model consisting of beam and point

16
mass elements. This results in significantly higher computational requirements for solid element

models.

Solid element models are generally more accurate than an equivalent lumped parameter model[8].

Lumped parameter models can have difficulty finding complex modes since they have fewer total

degrees of freedom than solid element models. Modeling large radius disks as point masses ignores

the effect that the stiffness of the disk and the deformation of the disk have.

The final advantage of solid element modeling is that is allows for the use of importing and meshing

existing geometry files that may be used to otherwise analyze the rotor parts. Lumped parameter

models mesh one dimensional line bodies into beams and add point masses along the line at points

where the beam elements are connected to each other. Rotor parts are likely already modeled as

three dimensional parts and three dimensional models when a rotordynamic analysis takes place.

These models can be easily meshed into solid elements while a new model would need to be

created to utilize beam and point mass elements.

ANSYS Workbench allows the importing of existing geometry into a model. Meshing follows the

geometry. Line bodies are meshed into beams, two dimensional bodies are meshed into shells, and

three dimensional bodies are meshed into solid elements. Point mass elements are added to a

model in ANSYS Workbench at the ends of line bodies or corners of two dimensional and three

dimensional bodies by inserting them into the geometry in ANSYS Mechanical. Rigid body sections

can alternatively be used to model disks as point masses considered as one element. Rigid bodies

are created in ANSYS Workbench by changing the stiffness behavior of a body from flexible to rigid

and prevent any deformation of the body. Rigid bodies in rotordynamic analysis can be used to

17
represent a disk or other non-shaft part that deform very little during the current analysis. However,

the connection of the rigid body to meshed parts can result in excessive penetration of meshed

elements and lead to errors in the simulation if this connection is not properly controlled. Rigid body

elements also cannot be used for rotating parts in the stationary reference frame within ANSYS

Workbench as they do not have a gyroscopic matrix definition. Due to these issues the use of rigid

body disks or other parts should be modeled as point masses in ANSYS Workbench instead.

In NASTRAN, elements are created by defining the position of the grid points of the element in the

coordinate system. Beam and pipe elements are created using at least two grid points, shell

elements are defined using several coplanar grid points, and solid elements are created using four or

more non-coplanar grid points. Point masses are added to a single grid point. Unlike ANSYS

Workbench, the types of beam, pipe, shell, or solid elements are specified by the user when

meshing geometry in NASTRAN. NASTRAN currently limits the rotating reference frame to solid

elements. Rigid bodies cannot be part of the rotating system in NASTRAN.

2.5. ANSYS Workbench and NASTRAN Rotor Model

The rotor model used to compare the rotordynamics capabilities of ANSYS Workbench and

NASTRAN is a modified Jeffcott rotor model shown in Figure 2-4. The original Jeffcott rotor was the

first rotor model developed to study some fundamental behaviors of rotors such as critical speed,

natural frequency and whirling of the rotor.

18
 Figure 2-4 - Rotor Model

This modified Jeffcott rotor model consists of a single shaft with a diameter of 0.01 m and length of

0.2 m is divided into twenty equal beam elements. At the midpoint of the rotor shaft there is a rotor

disk with a diameter of 0.1 m and a thickness of 0.02 m modeled as a point mass. Two isotropic

bearings are located at the ends of the shaft that are modeling as a combination of a spring with

stiffness 1.0x108 N/m and a damper with damping coefficient 5000 N/m-s. The shaft and disk have a

density of 7850 kg/m3, an elasticity of 2.0x1011 Pa, and a Poisson’s ratio of 0.3. Rotating damping

was given three alternative definitions for comparison purposes: no rotating damping, Rayleigh

damping with α of 0 and β of 5.0x10-6, and structural damping with the loss factor η of 5.0x10-4.

19
3. Stationary Reference Frame

The stationary reference frame is used when simulating axisymmetric rotating bodies, performing

multi-shaft analysis, or the rotating shaft is attached to a non-rotating support structure. In the

stationary reference frame each element’s matrices are formed with degrees of freedom that align

with a global fixed XYZ frame. The equations of motion relating the nodal accelerations {𝑞̈ } in the

global XYZ coordinates to the internal forces due to element deformations {𝑞̇ } and {𝑞} and external

forces {𝐹} are put in matrix form.

[𝑀]{𝑞̈ } + (𝛺[𝐺] + [𝐶𝑟 + 𝐶𝑛 ]){𝑞̇ } + ([𝐾] + 𝛺[𝐵𝑟 ]){𝑞} = {𝐹} 3.1

Any structure being analyzed via the finite element method requires an inertia matrix [𝑀] and

stiffness matrix [𝐾]. The damping matrix is divided into rotating damping [𝐶𝑟 ] and non-rotating

damping [𝐶𝑛 ] which contain the damping contributions of the rotating structure and non-rotating

support structure respectively. The gyroscopic matrix [𝐺] and circulatory matrix [𝐵𝑟 ] only contain

contributions for elements that are rotating.

3.1. Inertia and Stiffness Matrices

The inertia matrix [𝑀] for any element is a diagonal matrix consisting of the masses and rotational

inertias at the nodes of the element. The inertia matrix of each element is formed by distributing the

mass of the element to each node and directly applying the rotational inertias to the nodes where

required. The mass of beam elements, shell elements, and solid elements are calculated by

multiplying the density of the element’s material by the volume of the element. The mass is

distributed to the nodes of each element based on the shape of the element. The mass at each node

is applied equally to all three translational degrees of freedom at that node by default. Rotating

inertias are calculated and added to nodes of beam elements based on the geometry of the

element.

20
 𝑚1
𝑚1
𝑚1
𝐽1𝑋
𝐽1𝑌
𝐽1𝑍
[𝑀]𝑒𝑙,𝑏𝑒𝑎𝑚 = 𝑚2 3.2
𝑚2
𝑚2
𝐽2𝑋
𝐽2𝑌
[ 𝐽2𝑍 ]

Solid elements do not have rotating inertias since they do not have rotating degrees of freedom.

𝑚1 0 ⋯
0 𝑚1
⋮ 𝑚1
[𝑀]𝑒𝑙,𝑠𝑜𝑙𝑖𝑑 = ⋱ 3.3
𝑚𝑛
𝑚𝑛
[ 𝑚𝑛 ]

Point mass elements are defined with a mass that is applied to all three translational degrees of

freedom and with rotating inertias for all three rotational degrees of freedom.

𝑚1 0 ⋯
0 𝑚1
⋮ 𝑚1
[𝑀]𝑒𝑙,𝑝𝑡.𝑚𝑎𝑠𝑠 = 3.4
𝐽1𝑋
𝐽1𝑌
[ 𝐽1𝑍 ]

The stiffness matrix [𝐾] is defined as a function of each element’s material elasticity and shape

function. This matrix defines the stiffness between nodes of an element and is highly dependent on

the element shape for beam, shell, and solid elements. Point mass elements do not have a rotating

stiffness matrix [𝐾𝑟 ] since they only have one node. The stiffness matrix can be subdivided in

rotating [𝐾𝑟 ] and non-rotating [𝐾𝑛 ] matrices. The internal forces from the rotating stiffness matrix

[𝐾𝑟 ] are due to the displacement of nodes in the rotating structure.

[𝐹𝐾 ]𝑅 = [𝐾𝑟 ]{𝑢} 3.5

21
Since this internal force occurs in the rotating reference frame it must be transformed into

stationary coordinates using the transform matrix [𝑇] that defines the relationship between

stationary and rotating coordinate systems.

{𝑢} = [𝑇]{𝑞} 3.6

[𝐹𝐾 ]𝑅 = [𝑇][𝐹𝐾 ]𝐹 3.7

By applying Eqs. 3.6 and 3.7 the internal force in the stationary reference frame is determined.

[𝐹𝐾 ]𝐹 = [𝑇]𝑇 [𝐹𝐾 ]𝑅 = [𝑇]𝑇 [𝐾𝑟 ][𝑇]{𝑞} = [𝐾𝑟 ]{𝑞} 3.8

The transformation from the rotating reference frame to the stationary reference frame results in

no changes to the rotating stiffness term. Similarly the non-rotating stiffness contribution

transforms from the stationary reference frame to the rotating reference frame. While the

transformations of the stiffness terms from one reference frame to another does not change the

composition of the terms, this will not hold true for internal forces due to damping.

When using ANSYS Workbench, material properties such as elasticity and density can be defined for

a material under engineering data in the Workbench project schematic and then assigned to the

geometric bodies in ANSYS Mechanical. When no material properties are defined or they are not

assigned to the bodies, ANSYS Workbench will use default material properties for structural steel to

create the inertia and rotating stiffness matrices or each element.

In NASTRAN materials must defined and every element must be assigned a material. From these

material definitions and the element shapes, the inertia and stiffness matrices are created.

22
3.2. Gyroscopic Matrix

The gyroscopic term Ω[𝐺] is formed on an element by element basis in ANSYS Workbench and

NASTRAN. In ANSYS Workbench the gyroscopic effect is defined only for the point mass element

MASS21, beam and pipe elements BEAM188, BEAM189, PIPE288, PIPE289, shell elements SHELL181

and SHELL281, solid elements SOLID185, SOLID186, SOLID187, SOLID272, SOLID273, and a

superelement containing grouped together elements MATRIX50. Applying a rotational velocity to

any other element types in the stationary frame will result in an error and prevent the simulation

from successfully occurring.

NASTRAN defines the gyroscopic effect for point mass elements CONM1 and CONM2, beam

element CBEAM, axisymmetric elements CQUADX and CTRIAX, shell elements CQUAD4, CQUAD8,

CTRIA3, and CTRIA6, and solid elements CTETRA, CPENTA, and CHEXA. NASTRAN also allows the use

of beam elements CBAR in rotordynamic analyses but CBAR elements do not include any gyroscopic

effects. All other element types cannot be part of a rotor in NASTRAN.

The gyroscopic matrix for point mass and beam elements can be determined directly from the polar

inertia of the elements, but solid elements do not have a polar inertia defined in their inertia matrix

and their gyroscopic effect must be determined using a different approach. Both ANSYS Workbench

and NASTRAN use the kinetic energy approach developed by Geradin & Kill[6] to determine the

gyroscopic matrix for their elements.

The kinetic energy of each element due to gyroscopic effects is defined and from the kinetic energy

expression the gyroscopic matrix of the element is then determined by taking the derivation of the

kinetic energy with respect to nodal displacement vector.

23
 𝜕𝐸 𝐺𝑘
𝜕{𝑞}
= 𝛺[𝐺]{𝑞̇ } 3.9

3.3. Bearings and the Non-Rotating Damping Matrix

Bearings connect rotor shafts to a non-rotating structure that support the structure and remove

energy from the rotating system by applying a source of non-rotating damping helping the stability

of the rotor system. There are several methods that can be used to model bearings for rotordynamic

analysis. Bearings can be modeled as a set of springs and dampers, a combined spring and damper

bearing element, or as a fluid bearing with changing stiffness and damping characteristics based on

the position of the rotor. For each rotor node that they are connected to, bearings add stiffness and

damping to the two translational degrees of freedom in the bearing’s plane which is orthogonal to

the rotor’s axis. For all of modeling approaches, the damping of a bearing is part of the non-rotating

damping matrix [𝐶𝑛 ] and the stiffness of the bearing is added to the stiffness matrix [𝐾] via its

external stiffness matrix [𝐾𝑛 ] contribution. Point masses can be added to the bearing nodes reflect

the inertia effect of the bearing housing.

3.3.1. Modeling Bearings as Springs and Dampers

The simplest modeling technique for bearings in rotordynamic analysis is to approximate the overall

stiffness and damping of a bearing as a set of springs and dampers that are orthogonal to the rotor

axis (Figure 3-1). If the shaft is modeled using beam elements a spring and a damper are attached to

a node of the beam element and a fixed point. Two sets of springs and dampers are required to

define the stiffness and damping in both translational directions. If the shaft is modeled using solid

elements, multiple sets of springs and dampers can be used to constrain in-plane nodes at the

location of the bearing. For simple models this approach is preferable in ANSYS Workbench.

24
 Figure 3-1 - Bearing as Springs and Dampers

3.3.2. Bearing Elements with User-Defined Stiffness and Damping

ANSYS Workbench contains a bearing element COMBI214 with two degrees of freedom at each

node and no bending or torsion consideration. The user defined stiffness and damping definition of

this bearing element is the default option when creating a bearing in ANSYS Workbench. The

bearing plane, stiffness, and damping definitions are all inputs for the element. Use of the

COMBI214 element also allows for cross-coupling the stiffness or the damping between the two

directions. The bearing element in ANSYS Workbench is scoped to two nodes. The first node is on

the shaft itself and the second node is a stationary non-rotating point which defaults to a coincident

ground. While the bearing element is not rotating in theory, because the nodes of the element are

coincident and one node is connected to the rotating shaft the motion of the bearing element is

determined to be rotating by ANSYS Workbench. As a result the damping of the bearing element is

added to the rotating damping matrix instead of the non-rotating damping matrix. To avoid this

incorrect application of bearing damping in FEA a rigid body element can be used. The rigid body

element is used to connect the translational motion but not rotational motion of a node on the

rotating structure to a coincident node that is not part of the rotating structure. This effectively

25
decouples the rotation of the rotor from the rotation of the support structure. Within ANSYS

Workbench this should be possible though the use of a remote point connection, but use of the

remote point connection still adds bearing damping to the rotating damping matrix instead of the

non-rotating damping matrix. Due to this issue, bearings in ANSYS Workbench are best modeled as a

combination of springs and dampers instead of using the built-in bearing elements.

In NASTRAN a CBUSH element is used to connect two coincident nodes together or a rotor node to

ground. The stiffness and damping of the CBUSH element are defined by its associated material

property definition PBUSH (Figure 3-2). This bearing element definition does not allow for cross-

coupling stiffness or damping terms.

Figure 3-2 - CBUSH and PBUSH Data Entry Cards

In the NASTRAN bearing element card CBUSH, EID is the element identification number, PID is the

property identification number, GA and GB are the primary method of defining the grid points of the

bearing, CID is coordinate system of the bearing, and the remainder of the card is used as an

alternative method to scope the bearing location. The NASTRAN bearing property card PBUSH

defines the stiffnesses Ki along each of the six primary directions along the coordinate system of the

element, as well as damping Bi and structural damping GEi along these directions. A mass M can be

added and recovery stress and strain components can be defined via the SA, ST, EA, and ET

26
 variables. NASTRAN also requires that the bearing not be explicitly connected to a point on the

rotor. A coincident node must be created with linked displacements but separate rotations using a

rigid body connection card RBE2 (Figure 3-3).

Figure 3-3 - Rigid Body Element RBE2 Card

The element identification number EID must be a new, unused element value, the rotor grid point is

GN, the linked degree of freedom are defined at CM as 123 for all translational degrees of freedom,

and GM1 is the coincident grid point that is not defined as part of the rotor. Separation of the rotor

from the bearing element is required to avoid adding the damping from the bearing to the rotating

damping matrix instead of the non-rotating damping matrix.

3.3.3. Modeling Journal Bearings

Journal bearings are non-linear bearings that change stiffness and damping based on the current

position of the center of the shaft. ANSYS Workbench uses a Reynold’s equation approach to model

journal bearings when a bearing element is created using the following two ANSYS APDL command

lines: “et, Node ID, element type” and “keyopt, KEYOPT1, KEYOPT2, KEYOPT3”. In these command

lines, the NodeID is an integer number specifying the rotor node the bearing connects to, a journal

bearing is created with element type combi214, KEYOPT1 is set to 1 for limited output results of a

journal bearing or 2 for full output results, KEYOPT2 is set to 0 for XY plane, 1 for YZ plane, or 2 for

XZ plane, and KEYOPT3 is set to 0 for a symmetric bearing or 1 for an asymmetric bearing.

27
 For either journal bearing output requirements, the geometry is defined by the radial clearance C,

bearing length L, and rotor radius R. These variables are defined as real constants using the ANSYS

APDL command “r, NodeID, Clearance, Length, Radius, Veloc1, Veloc2, PertInc, ThetaInc, OmgPrec”.

When the full output requirements are required, the perturbation increment for stiffness and

damping calculation PertInc and theta increment for integration ThetaInc must be defined. Veloc1

and Veloc2 are not used in rotordynamic analysis. OmgPrec is only used in the special case of a

squeeze film damper having synchronous precession and defines the rotational velocity for that

case. In ANSYS Workbench this journal bearing definition cannot be used in Modal analysis.

NASTRAN allows for squeeze film damper behavior to be approximated for transient analysis. The

squeeze film damper creates a force that is added to the force vector {𝐹}. Input data required to

calculate this force are the journal diameter BDIA, clearance BCLR, length BLEN, oil viscosity VISCO,

location of ports THETA1 and THETA2, and the boundary pressures PRES1 and PRES2. These

variables are input using the NLRSFD card (Figure 3-4).

Figure 3-4 - Squeeze Film Damper NLRSFD Data Entry Card

3.4. Rotating Damping Matrix

Rotating damping represents all of the damping within the rotating structure as opposed to damping

due to bearing supports and the stator. Rotor damping can be defined as Rayleigh damping applied

to the rotating structure, structural damping applied to the rotating structure, or a viscous damper

28
 element connecting two nodes that are part of the rotating structure. As will be shown, rotating

damping results in the rotating damping matrix [𝐶𝑟 ] and the circulatory matrix due to rotating

damping [𝐵𝑟 ]. These two matrices have a large effect on rotor stability and the different damping

models affect stability differently. Due to its implication to the system stability, rotating damping is

crucial in the analysis of rotordynamics.

The internal force at a node from rotating damping occurs in the rotating reference frame. Like the

internal force at a node from rotating stiffness, this force must be transformed to the stationary

reference frame for use in the stationary reference frame equation of motion.

[𝐹𝐶 ]𝑅 = [𝐶𝑟 ]{𝑢̇ } 3.10

Since the internal force from damping acts upon the velocity, the relationship between the velocity

vectors in the two coordinate frames is derived using the first time derivative of Eq. 3.6.

𝜕
{𝑢̇ } = [𝑇]{𝑞̇ } + ([𝑇]){𝑞} 3.11
𝑑𝑡

This results in an internal force due to rotating damping with two terms.

𝜕
[𝐹𝐶 ]𝐹 = [𝑇]𝑇 [𝐹𝐶 ]𝑅 = [𝑇]𝑇 [𝐶𝑟 ][𝑇]{𝑞̇ } + [𝑇]𝑇 [𝐶𝑟 ] ([𝑇]){𝑞} 3.12
𝑑𝑡

The first term in Eq. 3.12 results in the rotating damping matrix [𝐶𝑟 ] multiplied by the nodal velocity

vector {𝑞̇ }. The second term in the equation is the contribution from the circulatory matrix due to

rotating damping in the stationary reference frame.

3.5. Circulatory Matrix

As shown in the preceding section, the circulatory matrix in the stationary reference frame is a

result of the transformation of rotating damping from the rotating coordinate system to the

stationary coordinate system. For a rotor aligned along the z-direction the transformation matrix [𝑇]

relating the rotating xyz frame to the global XYZ frame can be written as

29
 𝑐𝑜𝑠𝛺𝑡 𝑠𝑖𝑛𝛺𝑡 0
[𝑇] = [−𝑠𝑖𝑛𝛺𝑡 𝑐𝑜𝑠𝛺𝑡 0] 3.13
0 0 1

Inserting the transformation matrix into the circulatory matrix term in Eq. 3.12 gives

𝑐𝑜𝑠𝛺𝑡 −𝑠𝑖𝑛𝛺𝑡 0 𝑐𝑜𝑠𝛺𝑡 𝑠𝑖𝑛𝛺𝑡 0

𝜕
𝛺[𝐵𝑟 ]𝑛𝑡 = [ 𝑠𝑖𝑛𝛺𝑡 𝑐𝑜𝑠𝛺𝑡 0] [𝐶𝑟 ]𝑛𝑡 𝑑𝑡 ([−𝑠𝑖𝑛𝛺𝑡 𝑐𝑜𝑠𝛺𝑡 0]) 3.14
0 0 1 0 0 1

𝑐𝑜𝑠𝛺𝑡 −𝑠𝑖𝑛𝛺𝑡 0 −𝑠𝑖𝑛𝛺𝑡 𝑐𝑜𝑠𝛺𝑡 0

= 𝛺 [ 𝑠𝑖𝑛𝛺𝑡 𝑐𝑜𝑠𝛺𝑡 0] [𝐶𝑟 ]𝑛𝑡 [−𝑐𝑜𝑠𝛺𝑡 −𝑠𝑖𝑛𝛺𝑡 0]
0 0 1 0 0 1

Note that the transformation of the rotational degrees of freedom has the same relationship as the

transformation of the translational degrees of freedom and that equation 3.14 can be expanded to

include all six degrees of freedom at each node. Most implementations of the circulatory matrix will

assume that the rotating damping matrix is symmetric about the rotation axis without cross-

coupling terms.

𝑐𝑟 0 0
[𝐶𝑟 ]𝑛𝑡 = [ 0 𝑐𝑟 0] 3.15
0 0 0

With this assumption, the circulatory matrix due to rotating damping can be further reduced

producing the skew-symmetric circulatory matrix.

0 1 0 𝑐𝑟 0 0 0 𝑐𝑟 0
[𝐵𝑟 ]𝑛𝑡 = [𝐽][𝐶𝑟 ]𝑛𝑡 = [−1 0 0] [ 0 𝑐𝑟 0] = [−𝑐𝑟 0 0] 3.16
0 0 0 0 0 0 0 0 0

The rotating damping defined for rotational degrees of freedom 𝜙𝑥 and 𝜙𝑦 are likewise multiplied

by the skew-symmetric matrix [𝐽].

In ANSYS Workbench the circulatory matrix is calculated on an element basis instead of a nodal

basis. Instead of defining damping at each node after the stiffness and inertia contributions of an

element have been distributed to the nodes, ANSYS Workbench defines damping for the element

based on the damping model and the element’s stiffness or inertia contributions. The damping of an

30
element is distributed to its nodes in the same manner as the stiffness matrix. Because of this,

ANSYS Workbench defines the circulatory matrix on an elemental basis and then distributes the

circulatory matrix to the nodes. In its implementation, ANSYS Workbench assumes that the damping

is symmetric, but allows for the rotor to rotate about an axis that does not align with one of the X, Y,

or Z axes. The circulatory matrix contribution is thus

0 𝛺𝑧 −𝛺𝑦
𝛺[𝐵𝑟 ]𝑒𝑙 = [𝐶𝑟 ]𝑒𝑙 [−𝛺𝑧 0 𝛺𝑥 ] 3.17
𝛺𝑦 −𝛺𝑥 0

This element based circulatory matrix definition is then distributed to the element’s nodes in the

same manner as the stiffness matrix.

NASTRAN calculates the circulatory matrix on a node by node basis. NASTRAN aligns rotors along

their local z axis and calculates the circulatory matrix at each node with the rotating damping matrix

[𝐶𝑟 ]𝑛 and a transformation matrix [𝑇𝐼 ] where

𝛺[𝐵𝑟 ]𝑛 = 𝛺([𝑇𝐼 ][𝐶𝑟 ]𝑛 + [𝐶𝑟 ]𝑛 [𝑇𝐼 ]) 3.18

0 0.5 0 0 0 0 0 0.5 0 0 0 0
−0.5 0 0 0 0 0 −0.5 0 0 0 0 0
0 0 0 0 0 0 [𝐶 ] 0 0 0 0 0 0
=𝛺 + [𝐶𝑟 ]𝑛
0 0 0 0 0.5 0 𝑟 𝑛 0 0 0 0 0.5 0
0 0 0 −0.5 0 0 0 0 0 −0.5 0 0
([ 0 0 0 0 0 0] [ 0 0 0 0 0 0])

This definition removes all cross-coupled damping terms from the circulatory matrix and preserves

the definition given in Eq. 3.16. For solid element nodes the transformation matrix [𝑇𝐼 ] is the three

by three upper left submatrix of the full six by six matrix. The nodal circulatory matrices are then

combined to form the global circulatory matrix in NASTRAN. Since the circulatory matrix is

multiplied by the rotational velocity, its effect on natural frequencies and stability increases with

spin speed 𝛺 (Figure 3-5).

31
 Figure 3-5 - Circulatory Matrix Effect on Non-Dimensionalized Natural Frequencies

3.6. ANSYS Implementation of the Stationary Reference Frame

ANSYS Workbench implementation of the stationary reference frame can only be accomplished by

utilizing APDL commands. The command used is “CORIOLIS, 1, --, --, 1, RotDamp” where RotDamp is

an option to activate the circulatory matrix due to rotating damping. To activate the circulatory

matrix RotDamp is set to 1, to deactivate it RotDamp is set to 0. By default the circulatory matrix is

not activated in ANSYS Workbench. The axis of rotation and the bodies included in the rotating

structure are selected during the process of defining a rotational velocity.

3.7. NASTRAN Implementation of the Stationary Reference Frame

Within NASTRAN the stationary reference is utilized whenever the ROTORG card is present in the

input file or the ROTOR card is present and the fixed reference frame is specified as part of the

ROTOR card. The ROTORG card (Figure 3-6) is used for lumped parameter modeling.

32
 Figure 3-6 - ROTORG Data Entry Card

In the ROTORG card, ROTORID is a positive integer number that serves as a rotor identification

number, and GRID1, GRID2, and GRIDn are grid point (nodes) that are present in the rotor. The first

format of the ROTORG card for identified the nodes requires individually listed all of the grid points.

The second format defines minimum (GRID1) and maximum (GRID2) grid point identification

numbers and defines the increment (INC) by which successive grid points are defined. All grid points

in a ROTORG rotor must be collinear.

The ROTOR card (Figure 3-7) is used for solid element modeling and is similar to the ROTORG card in

that it requires a rotor identification number ROTORID and grid points to be defined.

Figure 3-7 - ROTOR Data Entry Card

The FRAME is set to “FIX” for a stationary reference frame or “ROT” for a rotating reference frame.

LYTPE is the list type that is entered as either ELEM for elements or PROP for properties and is

followed by element or property numbers (ID1, ID2, etc.) that define the rotor. The rotor axis is

defined by two nodes identified as grid points GID1 and GID2.

33
Rotors defined by ROTOR and ROTORG cards require addition NASTRAN cards to have rotordynamic

effects calculated. RGYRO is both a case control card, “RGYRO = n”, and a bulk data entry card

(Figure 3-8) and is used to define the rotation speed of a reference rotor.

Figure 3-8 - RGYRO Data Entry Card

In the RGYRO bulk data entry card, RID is an integer number equal to that of the RGYRO case control

card. REFROTR is used to refer to the rotor identification number defined in a ROTOR or ROTORG

card. SYNCFLG specifies whether the analysis is asynchronous or synchronous and is set to “ASYNC”

or “SYNC” respectively. The minimum and maximum rotation speeds of a synchronous analysis are

set via SPDLOW and SPDHIGH, while the rotation speed of the rotor is defined by SPEED for an

asynchronous analysis. SPDUNIT determines the units of the rotation speed as either revolutions per

minute or hertz, and is input as “RPM” or “FREQ”. ROTRSEID is only used if the rotor is part of a

superelement and matches the identification number of that superelement.

Each rotor in an analysis must be given a RSPINR card for modal analysis and frequency response

analysis or a RSPINT card for transient analysis. The RSPINR and RSPINT cards (Figure 3-9) must be

defined for rotordynamic effects to be calculated even if the rotor is the same as the reference rotor

defined by the RGYRO card.

34
 Figure 3-9 - RSPINR and RSPINT Data Entry Cards

The RSPINR and RSPINT data cards define a positive spin direction from grid point 1 (GRIDA) to grid

point 2 (GRIDB) for a rotor identified by its rotor identification number ROTORID. SPTID refers to a

table of spin speeds if it is an integer value or it is a multiplier of the reference rotor spin speed

defined in the RGYRO card if it is a real value. If the rotor is the same as the reference rotor SPTID

should be set to 1.0. For transient analysis, SPTID must be a table. SPDOUT in the transient analysis

card is a variable used to output rotor speed versus time. SPDUNT defines the units for the spin

speed the same as the RGYRO card. ROTRSEID is only required if the rotor is part of a superelement

and matches that superelement identification number. GR is used to define the structural damping

ratio for the rotor, ALPHAR1 and ALPHAR2 define the α and β Rayleigh damping coefficients for the

rotor.

The circulatory matrix contribution can be turned off within a NASTRAN analysis using the bulk data

entry “MDLPRM RDBOTH 2”. If this line is not added to the NASTRAN input file, the circulatory

matrix will be considered.

35
4. Rotating Reference Frame

The rotating reference frame is used when simulating non-axisymmetric rotating bodies, performing

single shaft analysis, or there is no non-rotating support structure included in the model. In the

rotating reference frame each element’s matrices are formed with degrees of freedom that align

with a rotating xyz frame that moves with the rotor’s angular velocity Ω. The equations of motion at

each node are put into matrix form for the rotating reference frame resulting in Eq. 4.1.

[𝑀]{𝑢̈ } + ([𝐶𝑜𝑟 ] + [𝐶𝑟 ] + [𝐶𝑛 ]){𝑢̇ } + ([𝐾] − [𝐾𝑐 ] + [𝑆] + 𝛺[𝐵𝑛 ]){𝑢} = {𝐹} 4.1

The internal forces in this equation of motion are aligned with the rotating coordinate system

instead of the fixed reference frame. The Coriolis effect [𝐶𝑜𝑟 ], spin softening matrix [𝐾𝑐 ], and

circulatory matrix due to non-rotating damping 𝛺[𝐵𝑛 ] are unique to the rotating reference frame

while the gyroscopic matrix and circulatory matrix due to rotating damping seen in the stationary

reference frame vanish. The stress stiffening matrix [𝑆] can also be applied in the stationary

reference frame, but is discussed for the rotating reference frame where it is more often

implemented by finite element models.

NASTRAN limits the rotating reference frame to solid elements only while ANSYS Workbench allows

any element type to be used. ANSYS Workbench does not take into account the circulatory matrix

due to non-rotating damping. Neglecting this term contributes to differences seen when comparing

results between a stationary reference frame analysis and rotating reference frame analysis.

4.1. Similarities to the Stationary Frame

The inertia matrix [𝑀] and stiffness matrix [𝐾] are formed similarly to their stationary reference

frame counterparts. The internal force due to rotating damping occurs in the rotating reference

36
 frame and doesn’t need to be transformed like in the stationary reference frame. This removes the

circulatory matrix due to rotating damping and allows the internal force from rotating damping to

be defined by only the rotating damping matrix [𝐶𝑟 ]. The internal force from non-rotating damping

occurs in the stationary reference frame. It is transformed to the rotating reference frame using the

same transformation relationship used for transforming the rotating damping force to the stationary

frame. This transformation results in a non-rotating damping matrix [𝐶𝑛 ] formed the same as in the

stationary reference frame and a circulatory matrix term with an internal force equivalent to

𝛺[𝐵𝑛 ]{𝑢}.

4.2. Coriolis Matrix

The Coriolis matrix of each element is a skew-symmetric matrix representing the kinetic energy of

an element as it rotates about the axis of rotation. The Coriolis matrix is calculated on an elemental

basis and this calculation takes into account the mass of the element. For both ANSYS Workbench

and NASTRAN the Coriolis matrix of an element is calculated by the equation

0 −𝛺𝑧 𝛺𝑦 𝑚 0 0
[𝐶𝑜𝑟 ]𝑒𝑙 = 2[𝛺][𝑀𝑚 ]𝑒𝑙 = 2 [ 𝛺𝑧 0 −𝛺𝑥 ] [ 0 𝑚 0] 4.1
−𝛺𝑦 𝛺𝑥 0 0 0 𝑚

where 𝛺𝑥 , 𝛺𝑦 , and 𝛺𝑧 represent the x, y, and z contributions of the rotational speed. For simulations

where the rotational velocity 𝛺 is aligned solely along the z-direction this simplifies to the skew-

symmetric form

0 −𝑚 0
[𝐶𝑜𝑟 ]𝑒𝑙 = 2𝛺 [𝑚 0 0] 4.3
0 0 0

37
 The element’s shape function is then used to distribute the x, y, and z components of the elemental

Coriolis matrix to each of the element’s nodes similarly to how the inertia and stiffness matrices are

created. ANSYS Workbench calculates the Coriolis matrix for any element type while NASTRAN limits

the calculation of the Coriolis matrix only to elements that are allowed in rotordynamic simulations.

4.3. Spin Softening and Stress Stiffening Matrices

The spin softening matrix, sometimes referred to as the centrifugal softening matrix, accounts for

the effect of centrifugal load in the rotating reference frame. The spin softening matrix is derived

from the term {𝛺} × ({𝛺} × {𝑢}). In ANSYS Workbench the spin softening matrix is calculated on an

elemental basis using the elemental mass matrix [𝑀𝑚 ]𝑒𝑙 and the rotational velocity matrix [𝛺].

0 −𝛺𝑧 𝛺𝑦 0 −𝛺𝑧 𝛺𝑦 𝑚 0 0
−[𝐾𝑐 ]𝑒𝑙 = [𝛺][𝛺][𝑀𝑚 ]𝑒𝑙 = [ 𝛺𝑧 0 −𝛺𝑥 ] [ 𝛺𝑧 0 −𝛺𝑥 ] [ 0 𝑚 0] 4.4
−𝛺𝑦 𝛺𝑥 0 −𝛺𝑦 𝛺𝑥 0 0 0 𝑚

For simulations where the rotational velocity 𝛺 is aligned solely along the z-direction this simplifies

to

𝑚 0 0
−[𝐾𝑐 ]𝑒𝑙 = − 𝛺2 [ 0 𝑚 0] 4.5
0 0 0

The element’s shape function is then used to distribute these terms to the nodal version of the

element’s spin softening matrix.

NASTRAN calculates the spin softening matrix on a nodal basis for the translational degrees of

freedom using the same procedure. In ANSYS Workbench and NASTRAN the spin softening effect is

included for all analyses that take place in the rotating reference frame.

38
 The stress stiffening matrix is used to consider the effects that the stress state or geometric changes

of an element has on its stiffness. When an element sees large amounts of deformation the increase

of the element’s stiffness due to the stress state must be considered for accurate results. This

increase in stiffness is accounted for in the stress stiffening matrix [𝑆]. In the rotating reference

frame, this is most likely to occur due to centrifugal load and which results in a differential increase

in stiffness proportional to Ω2. In ANSYS Workbench an analysis calculating the stress field of the

elements must be simulated separately and added as a pre-stress for stress stiffening effects to be

included. NASTRAN automatically accounts for the differential increase in stiffness due to centrifugal

load but will not account for any other changes to the stiffness matrix due to the stress field of the

elements.

The effects of spin softening and stress stiffening due to centrifugal load are only seen in solid

element models. Stress stiffening increases natural frequencies while spin softening decreases

natural frequencies. For models that incorporate both stress stiffening and spin softening the stress

stiffening effect is larger than that of spin-softening.

4.4. Circulatory Matrix due to Non-Rotating Damping

The circulatory matrix in the rotating frame is due to non-rotating damping. This circulatory matrix

has a stabilizing effect on rotor motion. It is required to offset the non-rotating damping matrix

which is implied to rotate with a spin speed of Ω in the rotating reference frame since it creates a

force with the rotating coordinates velocity vector {𝑢̇ }. Using the same transformation relationship

between the rotating reference frame and stationary reference frame defined by Eq. 3.6 and the

39
 same procedure as used for determining the circulatory matrix due to rotating damping in the

stationary reference frame results in a similar circulatory matrix [𝐵𝑛 ].

𝛺[𝐵𝑛 ]𝑛𝑡 = 𝛺[𝐽][𝐶𝑛 ]𝑛𝑡 4.6

In NASTRAN, the circulatory matrix in the rotating reference frame is calculated using the same

transformation matrix [𝑇𝐼 ] used for the circulatory matrix in the stationary reference frame.

𝛺[𝐵𝑛 ]𝑛 = 𝛺([𝑇𝐼 ][𝐶𝑛 ]𝑛 + [𝐶𝑛 ]𝑛 [𝑇𝐼 ]) 4.7

ANSYS Workbench does not account for the circulatory matrix due to non-rotating damping. The

circulatory matrix due to non-rotating damping contributes stability to a system. Without this

stability in ANSYS Workbench models, the stability results of simulations in the rotating reference

frame are inaccurate.

4.5. ANSYS Workbench Implementation of the Rotating Reference Frame

In ANSYS Workbench, the rotating reference frame is the default reference frame when

rotordynamic effects are activated. Spin softening effects are included whenever rotordynamic

effects are active in ANSYS Workbench. Stress stiffening effects can only be included as a pre-

stressed field when the results from a previously simulated stress analysis are attached to the

current simulation. The circulatory matrix due to non-rotating damping is never considered by

ANSYS Workbench in its rotating reference frame equation of motion.

40
4.6. NASTRAN Implementation of the Rotating Reference Frame

In NASTRAN, the rotating reference frame can only be used for solid element models. It is

implemented when the FRAME variable of the ROTOR card is set to “FIX”. RGYRO and RSPINR or

RSPINT bulk data cards must also be included in the simulation for rotating effects to be included.

Spin softening effects and stress stiffening due to centrifugal load are automatically included in a

NASTRAN rotating reference frame analysis. The circulatory matrix due to non-rotating damping is

considered unless specifically turned off.

41
5. Rotordynamic Analysis

With finite element models modal analysis, harmonic analysis, or transient analysis can be used.

Modal analysis is used to determine rotor modes, critical speeds, and mode stability. Harmonic

analysis determines rotor response to a harmonic forcing function that is either synchronous with

rotor speed or asynchronous. Transient analysis models rotor behavior with time and can be

particularly useful in modeling rotor response during start up or following a sudden change to the

rotor system such as a blade loss. Between these three analysis types most rotordynamic related

issues can be studied. In ANSYS Workbench the analysis type is chosen when a new model is

created. Each analysis type has several solvers that can then be utilized. NASTRAN determines the

analysis type based on the solver which is explicitly stated in the executive control section of a

NASTRAN input file.

5.1 Modal Analysis

Modal analysis is the process of determining the natural frequencies of a system and the

corresponding mode shapes of the system. The natural frequencies and mode shapes are

determined by finding the solutions to the equation of motion when the external forces are zero.

The nodal displacement vector is written as consisting of a vector of coefficients {𝜓} multiplied by a

frequency dependent term 𝑒 𝑗𝜆𝑡 .

{𝑞 } = {𝜓}𝑒 𝑗𝜆𝑡 5.1

By taking the time derivative of Eq. 5.1 and substituting it into the equation of motion for the

stationary reference frame (Eq. 3.1), the equation of motion becomes

(−𝜆2 [𝑀] + 𝑗𝜆(𝛺[𝐺] + [𝐶𝑟 + 𝐶𝑛 ]) + ([𝐾] + 𝛺[𝐵𝑟 ])){𝜓}𝑒 𝑗𝜆𝑡 = {0} 5.2

42
In order to solve for the frequencies 𝜆 and vectors {𝜓} that are the non-trivial solutions to Eq. 5.2,

terms must be grouped into generalized mass, damping, and stiffness matrices containing all of the

terms with the same dimensions. For the stationary reference frame these generalized matrices are

defined as

[𝑀]𝑔 = [𝑀] 5.3

[𝐶]𝑔 = 𝛺[𝐺] + [𝐶𝑟 + 𝐶𝑛 ] 5.4

[𝐾]𝑔 = [𝐾] + 𝛺[𝐵𝑟 ] 5.5

The rotating reference frame equation of motion is similarly grouped into generalized mass,

damping, and stiffness matrices. The modal equation is now a classic eigenvalue equation.

(−𝜆2 [𝑀]𝑔 + 𝑗𝜆[𝐶]𝑔 + [𝐾]𝑔 ){𝜓} = {0} 5.6

The eigenvalue 𝜆 solutions can now be extracted utilizing an eigensolver method. Several of these

methods exist including the complex Lanczos method, Upper Hessenberg method, and QR reduction

method.

Once the eigenvalues have been determined, the corresponding eigenvectors {𝜓} can be solved for

when the eigenvalues are put back into the eigenvalue equation. The eigenvectors are the non-zero

solutions to Eq. 5.6 and represent the mode shapes of the system. In some eigenvalue solvers, the

eigenvectors are solved for in the complex modal subspace and then transformed back.

Eigenvectors can be multiplied by a constant and still be valid solutions to the eigenvalue equation.

As a result eigenvectors are often normalized using a specific method so that they can be more

readily compared. The eigenvector can be normalized to the mass matrix, so that the largest

43
component of the eigenvector is a unit value, or so that the sum of the squares of each component

of the eigenvector is equal to a unit value.

ANSYS Workbench outputs mode shapes as animations (Figure 5-1) but does not include the

numerical values of the eigenvectors in the solver output. The mode shapes in ANSYS Workbench

are normalized to the mass matrix by default.

Figure 5-1 – ANSYS Workbench Mode Shapes

NASTRAN outputs mode shape data as the complex eigenvector consisting of translational and

rotational motions in the six degrees of freedom at each node point (Figure 5-2). NASTRAN

normalizes the eigenvectors so that one of the components of the eigenvector has a unit value.

44
 Figure 5-2 - NASTRAN Mode Shapes

To perform a modal analysis of a rotordynamic system in ANSYS Workbench, the option for a

damped system must be selected as “Yes” under the analysis settings and solver controls. If this

option is set to “No” instead, the gyroscopic matrix, damping matrices, circulatory matrices, and the

Coriolis matrix are not included in the analysis and the analysis is that of an undamped, non-rotating

system. The number of frequencies to be extracted and the number of rotational speeds at which

these frequencies are to be extracted at are also set in the analysis settings.

ANSYS Workbench utilizes two algorithms for finding eigenvalues of damped systems. The first

algorithm is referred to by ANSYS Documentation as the damped method and utilizes the full

generalized damping matrix. The quadratic equation 5.6 is transformed to a linear version.

[𝐶] [𝑀]𝑔
[𝐴] = [ 𝑔 ] 5.7
−[𝐼] [0]

45
 [𝐾]𝑔 [0]
[𝐵] = [ ] 5.8
[0] [𝐼]

(𝜆̅[𝐴] + [𝐵]){𝜙} = 0, 𝑤ℎ𝑒𝑟𝑒 𝜆 = 𝑗√𝜆̅ 5.9

This linearized version of the eigenvalue equation is then solved using the Block Lanczos eigenvalue

and eigenvector extraction method.

The second method used by ANSYS Workbench is the QR damped method. The QR damped method

finds the eigenvalue solutions to the undamped non-rotating system, transforms the equation of

motion to the modal coordinates of the undamped modes, and then extracts the modes of the

transformed damped equation of motion. The QR algorithm is utilized to reduce the degrees of

freedom using in solving the damped equation of motion by eliminating degrees of freedom that are

not well activated by the undamped modes. The QR damped method is best for models with a large

number of degrees of freedom and models that are lightly damped with symmetric stiffness

matrices.

The eigenvalue extraction method is selected in the solver controls portion of the analysis settings in

ANSYS Workbench. The “Full Damped” option will utilize the damped method, the “Reduced

Damped” option will utilize the QR damped method, and the “Programmed Controlled” option

allows ANSYS Workbench to internally determine which of the two methods to use based on the

size of the matrices and the number of frequencies being extracted.

In NASTRAN two methods are used for rotordynamic modal analysis. These two methods are the

complex Lanczos and Upper Hessenberg algorithms. The complex Lanczos method is similar to

damped method utilized by ANSYS Workbench while the Upper Hessenberg method reduces the full

46
generalized matrices to Upper Hessenberg matrices before further reducing them to triangular

matrices using QR transformation. Both methods are options under solver SOL 107 and are selected

along with the number of frequencies to extract using the complex eigenvalue extraction method

selection case control entry “CMETHOD = CCN” and the complex eigenvalue extraction bulk data

entry EIGC (Figure 5-3).

Figure 5-3 - EIGC Data Entry

In the complex eigenvalue extraction bulk data entry, the set identification number (SID) must

match the integer CCN of the complex extraction method selection case control entry. The method

of extraction (METHOD) is set to “CLAN” for the complex Lanczos method and “HESS” for the Upper

Hessenberg method. The eigenvectors are normalized (NORM) either by scaling the eigenvector

such that the maximum component has a unit value for the real part and a zero value for the

imaginary part or by scaling the eigenvalue so that a component at a point identified by point G and

degree of freedom C has a unit value for the real part and a zero value for the imaginary part. The

first normalization option is selected by setting NORM to “MAX” and is the default setting if no input

is entered for NORM. The second option is selected by setting NORM to “POINT”. Inputs for point G

and degree of freedom C are only required if the normalization method is the point method. The

convergence criterion is defined as E. ND0 is the number of eigenfrequencies to be extracted. The

second line of the complex eigenvalue extraction bulk data entry is used to specify the shift points

for the eigenvalue extraction search of the complex Lanczos method by defining the real and

imaginary parts of the shift point j (ALPHAAJ and OMEGAAJ), the maximum block size (MBLKSZ),

initial blocksize (IBLKSZ), and frequency of solve (KSTEPS). This continuation line may be repeated for

47
 multiple shift points and NJi is used to define the number of frequencies to be extracted for the shift

and ND0 is not required. If the continuation line is not used, shifts will be automatically calculated

by the solver.

5.1.1 Complex Eigenvalues

The complex eigenvalues extracted by an eigensolver contain information of both the modal

frequencies of the system and the stability of each mode. The imaginary value of the complex

eigenvalue is the modal frequency ω and real value of the complex eigenvalue is the stability value

σ. Each modal frequency represents a damped natural frequency of the system at a given rotor spin

speed. The stability value determines whether that particular mode is stable or unstable at that

rotor spin speed. A negative stability value indicates that the mode is stable at that spin speed and a

positive stability value indicates that the mode is unstable. A root locus plot shows the change in

complex frequency of a mode with increasing rotor spin speeds (Figure 5-4).

Figure 5-4 - Root Locus Plot

48
5.1.2 Campbell Diagrams and Critical Speeds

A Campbell diagram plots the natural frequencies of the rotor as a function of rotor spin speed. All

the points of the same mode are connected to each other to form a line showing how the natural

frequencies change with rotor spin speed. Critical speeds are the rotor spin speeds at which the

resonance of a natural frequency occurs due to an excitation force. Campbell diagrams typically plot

the excitation frequency equal to the rotor spin speed (first order excitation) to determine the

flexural critical speeds of the rotor. The flexural critical speeds are also important in analyzing mode

stability. When the rotor spin speed is above the flexural critical speed of a forward whirl mode,

rotating damping is destabilizing for that mode. If this mode is then excited by an asynchronous

forcing function, the rotor itself can become unstable. On a Campbell diagram it is possible to

identify flexural critical speeds as a crossing between a natural frequency line and the first order

excitation order line (Figure 5-5).

Figure 5-5 – ANSYS Workbench Campbell Diagram

49
Most Campbell diagrams do not consider the sign of the natural frequency and plot only the

magnitude of the frequency. Backward whirl modes are typically not excited by rotor spin speed and

therefore the crossing between a backward whirl frequency line and the first order excitation line is

usually not a flexural critical speed[15]. Therefore it is necessary to plot the natural frequencies with

the sign of the natural frequency to better understand the modal frequencies of the system (Figure

5-6).

Figure 5-6 - Campbell Diagram with Positive and Negative Frequencies

The data for a Campbell diagram is created in ANSYS Workbench when the Campbell diagram option

in the analysis settings is activated and the number of rotor speeds and the frequencies to be

extracted are defined. ANSYS Workbench creates a Campbell diagram using the absolute value of

natural frequencies and identifies flexural critical speeds for both forward whirl and backward whirl

modes.

50
 A Campbell diagram is not created directly by NASTRAN. Instead the data required for producing a

Campbell diagram is determined and this data can then be post-processed to create a Campbell

diagram. To create the data for a Campbell diagram in NASTRAN three additional entries are

required. The case control entry “CAMPBELL = CCN” must be added as well as the bulk data entries

CAMPBLL and DDVAL (Figure 5-7).

Figure 5-7 - CAMPBLL and DDVAL Data Entry Card

For the CAMPBLL bulk data card “CID” is the case identification number and must match the integer

value in the case control entry CAMPBELL, VPARM is the variable parameter and must be set to

“SPEED”, DDVALID is the identification number for the DDVAL table entry (a table of discrete digital

values), and TYPE should be set to either “FREQ” or “RPM”. The DDVAL bulk entry card is a list of

discrete design variable values which in this case is the list of rotor spin speeds used to find the

eigenvalues used to create the Campbell diagram. The “ID” must match the “DDVALID” of the

CAMPBLL card and the DVAL entries are a list of the rotor spin speeds.

5.1.3 Mode Tracking

One issue with Campbell diagrams is accurately applying mode tracking to ensure that the natural

frequencies of the same mode at different rotor spin speeds are grouped together. This is necessary

51
 to accurately produce lines connecting the multiple points of the same mode. If the mode shapes

are highly complex or mode tracking is not used, the natural frequencies of different modes may be

grouped together resulting in inaccurate critical speed estimations.

ANSYS Workbench does not use mode tracking. When creating a Campbell diagram, ANSYS

Workbench sorts modes based on the mode shapes to group together different points of the same

mode. If the same modes are not calculated at each rotor spin speed, ANSYS Workbench will not be

able to accurately group modes and will have an inaccurate Campbell diagram.

NASTAN utilizes a looping system to determine the natural frequencies of the same mode at each

rotor spin speed. Only the natural frequencies found at the first rotor spin speed are returned by the

eigensolver for the later rotor spin speed values. The similarity of a mode shape from one spin

speed to the next is used to accurately track the modes. Issues in accurate mode tracking can still

arise if the mode shape changes significantly from one rotor spin speed to the next which can

happen if two consecutive rotor spin speeds are far apart or the rotating reference frame is used

and a forward whirl mode changes to a backward whirl mode.

5.1.4 Directivity and Complex Coordinates Descriptions

Eigenvalue solutions to the modal equation are paired and take the form 𝑒 𝑗𝜔1 𝑡 where 𝜔1 > 0 and

𝑒 𝑗𝜔2 𝑡 where 𝜔2 < 0. These solutions imply that 𝜔1 is a forward whirl mode since it is a positive

frequency and that 𝜔2 is a backward whirl mode since it is a negative frequency. An alternative

mathematical solution to the modal equation is the pair of 𝑒 −𝑗𝜔3 𝑡 where 𝜔3 > 0 and 𝑒 −𝑗𝜔4 𝑡 where

𝜔4 < 0. These solutions lack the implied directional information of 𝜔1 and 𝜔2 due to the added

52
negative sign in the solution. Further examination of these two sets of paired solutions reveals that

the solutions are mathematically equivalent[17].

|𝜔1 | = |𝜔4 | 5.10

|𝜔2 | = |𝜔3 | 5.11

When solving a rotordynamic analysis, eigenvalue solvers will find both sets of results. When plotted

this leads to misleading and overcrowded Campbell diagram (Figure 5-8).

Figure 5-8 - Campbell Diagram without Directivity Considered

ANSYS Workbench and NASTRAN both solve for both sets of solutions. Despite solving for both sets

of solutions, ANSYS Workbench outputs only positive frequency results and determines the direction

of each mode from their eigenvector. NASTRAN outputs both sets of solutions (Figure 5-9) and the

data must then be post-processed to identify the rotor modes with the correct directivity.

53
Recognizing that eigenvalues occur in pairs and that the absolute value of a backward whirl mode in

the stationary reference frame will be lower than its companion forward whirl mode allows this

determination to be easily made. In Figure 5-9, it can be determined that eigenvalues 1, 4, 5, and 8

are the modes with the correct directivity.

Figure 5-9 - NASTRAN Complex Eigenvalue Results

Directivity issues can be avoided by utilizing complex variables to model the rotor[17-20]. For a

complex variable definition the translational deformations are coupled together as one complex

variable and the rotational deformations about the translational directions are coupled together by

another complex variable.

{𝑟} = {𝑥} + 𝑗{𝑦} 5.12

{𝜙} = {𝜙𝑦 } − 𝑗{𝜙𝑥 } 5.13

The use of complex coordinates also reduces the number of degrees of freedom at each node from

six to four which by halving the number of rotor modes that are related to lateral deformation and

non-torsional rotation. The reduction in overall degrees of freedom leads to reduced solver time. By

utilizing complex coordinates the direction of the positive whirl frequency is explicitly defined and

the duplicated solutions are not calculated when extracted eigenvalues which reduces post-

processing time. ANSYS Workbench and NASTRAN do not utilize complex coordinates when

modeling rotors but instead apply all six degrees of freedom to each node (or three degrees of

freedom for solid elements). Complex coordinates can only be used for rotors with symmetric mass,

54
 damping, and stiffness matrices. Since most rotors found in turbomachinery are symmetric, this

requirement does not reduce the value of implementing complex coordinates for modal analysis

very much.

5.2 Harmonic Analysis

Harmonic analysis is used to determine the response of the system due to a harmonic excitation

force. All forces in a harmonic analysis are a function of frequency. Synchronous response analysis

means that the force excitation frequency ω is the same as the rotor spin speed Ω. Synchronous

response analysis is often used to determine the response of a rotor due to mass unbalance.

Asynchronous response means that the force excitation frequency ω is not related to the rotor spin

speed Ω. Asynchronous response analysis is typically used for when the rotor is excited by an

external force. Both types of harmonic analysis determine the frequency response function between

the displacement of a node and a forcing function at a node.

The stationary reference frame and rotating reference frame apply harmonic analysis in a similar

manner. The displacement at each degree of freedom is considered as a magnitude and phase

component.

𝑞(𝑡) = 𝑞(𝜔)𝑒 𝑗𝜔𝑡 5.14

Applying this definition to the acceleration and velocity terms results in Eq. 5.15 for the stationary

reference frame and Eq. 5.16 for the rotating reference frame.

(−𝜔2 [𝑀] + 𝑗𝜔([𝐺] + [𝐶𝑛 ] + [𝐶𝑟 ]) + ([𝐾] + 𝛺[𝐵𝑟 ])){𝑞(𝜔)} = {𝐹(𝜔)} 5.15

(−𝜔2 [𝑀] + 𝑗𝜔([𝐶𝑜𝑟 ] + [𝐶𝑛 ] + [𝐶𝑟 ]) + ([𝐾] − [𝐾𝑐 ] + [𝑆] + 𝛺[𝐵𝑛 ])){𝑢(𝜔)} = {𝐹(𝜔)} 5.16

55
The magnitude and phase of the displacement vector due to the force vector is then calculated to

determine the harmonic response. The frequency response function is the ratio between the force

at one node and the displacement at one node. The frequency response function has a magnitude

and a phase. In ANSYS Workbench, the response solution can be used to create the frequency

response function (Figure 5-10).

Figure 5-10 - ANSYS Workbench Frequency Response Function

For a realistic rotor, the center of mass of the rotor is not aligned with the axis of rotation and the

distance between the two of them is defined as the eccentricity ε. As the rotor spins, the center of

mass rotates about the rotor’s axis of rotation which creates an unbalance force that is synchronous

with the rotor spin speed. This mass unbalance leads to rotor vibration. The amount of vibration due

to mass unbalance can be simulated using synchronous response analysis.

Harmonic response analysis is one of the options available at model creation in ANSYS Workbench.

ANSYS Workbench allows for two types of harmonic response analyses, full and mode-

superposition. The full analysis utilizes the entire finite element model to determine the response.

56
The mode-superposition determines the modal responses of a specified number of modes and adds

the modal responses together. The mode-superposition method requires that the results of an

existing ANSYS Workbench modal analysis precede the harmonic analysis.

A harmonic response analysis in ANSYS Workbench will apply any forces in the model as harmonic

forces. The response is determined for a range of frequencies defined in the analysis settings by a

minimum frequency, maximum frequency, and the number of increments. By default ANSYS

Workbench assumes that the harmonic response is asynchronous. For synchronous response the

APDL command SYNCHRO must be added. The SYNCHRO command takes the form of “SYNCHRO,

Ratio, Cname” where Ratio is the ratio between the harmonic frequency ω and the rotor spin speed

Ω and Cname is the component name of rotor. An unbalance force can be added to a node using

APDL commands where f0 is the magnitude of the unbalance force (Figure 5-11).

Figure 5-11 - APDL Commands for Unbalance Force

For the full matrix harmonic response analysis ANSYS Workbench utilizes three solvers, one direct

matrix solver and two iterative, convergence based solvers. The direct solver is the SPARSE solver

and uses the most amount of memory but can take the least amount of time. The iterative solvers

are the Jacobi Conjugate Gradient (JCG) solver and the Incomplete Cholesky Conjugate Gradient

(ICCG) solver which is more robust than the JCG solver but more time consuming.

57
In NASTRAN the solver SOL 108 is used for harmonic response analysis in rotordynamics. SOL 108 is

a direct matrix solver and will use the entire finite element matrix to calculate the response vector.

For asynchronous response analysis, the case control “FREQUENCY = SID” and bulk data entry FREQ

(Figure 5-12) must be used where SID is the set identification number and must match for the case

control and bulk data entries. Any static forces in the NASTRAN model will be applied at those

frequencies.

Figure 5-12 - FREQ Data Entry Card

For synchronous response analysis, the synchronous option must be selected in the RGYRO bulk

data entry. All static forces will be converted to harmonic forces with a frequency equal to the rotor

spin speeds. The minimum and maximum frequency values are identified in the RGYRO data entry

and the spin speeds are created in a DDVAL table that is referenced by the RSPINR data entry.

For unbalance mass loads, NASTRAN has developed a specific bulk data card. The case control

“DLOAD = SID” and bulk data entry UNBALNC are utilized for unbalance mass (Figure 5-13).

Figure 5-13 - UNBALNC Data Entry Card

58
 In the UNBALNC card MASS is the mass of the unbalance, GRID is the grid point at which the

unbalance is attached to, and X1/X2/X3 are the components of a vector from the grid point that are

used to define the coordinate system for the displacement of the unbalance mass. ROFFSET is the

radial offset distance of the mass unbalance from the grid point, THETA specifies the angle of the

mass unbalance in the plane defined by the vector X1/X2/X3 and has a default value of 0.0, ZOFFSET

is the offset of the mass unbalance in the z-direction of the mass unbalance coordinate system, and

T/Fon and T/Foff are the start and stop times/frequencies for applying and terminating the unbalance

forces respectively. The remaining variables in the UNBALNC bulk data entry card are not used for

harmonic response analysis. From the unbalance mass and the offset, NASTRAN will create a force

that is synchronous with the rotor spin speeds defined by the RGYRO and RSPINR data entries.

5.3 Transient Analysis

Transient analysis of rotors is typically used to model rotor start-up, changes in rotor spin speed,

rotor shutdown, or changes to the rotor structure such as a blade failure or varying damping from a

squeeze film damper. Nonlinear properties of rotors are best analyzed using a transient response

model. Transient analysis can be used to determine rotor motion, energy, stresses, and strains.

Transient analysis is a time based analysis and the underlying equations of motion are the same as

for modal and harmonic response analyses except that the Euler force due to angular acceleration is

present. Transient analysis can have issues with convergence.

59
In ANSYS Workbench transient response analysis requires that loads, displacements, and rotor spin

speeds are defined as functions of time. Step increments are defined to create time points at which

the rotor’s motion is calculated. Successive time points are continually solved until a stop time is

reached. In ANSYS Workbench structural damping cannot be utilized for transient analysis since

structural damping is defined on a per cycle basis and the excitation frequency is derived from the

frequencies defined or solver for in modal and harmonic response analysis. This is because ANSYS

Workbench creates an equivalent viscous damping based on the excitation frequency and for

transient response analysis there is no excitation frequency like there is in modal and harmonic

response analysis.

NASTRAN transient analysis functions in much the same way as ANSYS. The transient solver in

NASTRAN for rotordynamics is SOL 400. Time steps are defined using case control entry “TSTEP = ID”

and the bulk data entry for non-linear time points TSTEPNL (Figure 5-14) where ID must match that

of the case control entry.

Figure 5-14 - TSTEPNL Data Entry Card

The other required variables for the TSTEPNL data entry card are the number of time steps (NDT),

time increment (DT). The remaining variables determine the output intervals, method of

incrementation, convergence criteria, tolerances, and interval criteria. The table referenced by the

RSPINT data entry card is a list of times at which the spin speeds defined by the RGYRO data entry

card occur. Structural damping is allowed for transient analysis in NASTRAN since the excitation

60
frequency for structural damping is explicitly stated in the model by parameters WR3 and WR4. The

parameters WR3 and WR4 are defined as an average excitation frequency rather than only for a

single frequency of interest.

61
6. Damping and Stability

In realistic structures undergoing an oscillatory cycle of motion energy is dissipated during each

cycle. Damping is the macro-scale definition used to describe the mechanism of energy dissipation.

The dissipation of the energy is desirable in many systems since the removal of excess energy

reduces the amplitude of vibration especially at the resonance frequency. In rotating systems

however, the presence of damping in the rotating structure can result in a net supply of energy to

the system, rather than removing the energy.

Mathematically this additional energy appears as the skew-symmetric circulatory matrix in the

stationary frame. In the rotating reference frame, the energy from rotating damping is included in

the non-inertial reference frame itself which rotates with the rotational velocity Ω. ANSYS

Workbench does not define the circulatory matrix due to non-rotating damping 𝛺[𝐵𝑛 ] when the

rotating reference frame is used, which results in inaccurate depictions of rotor stability.

In a rotating system damping is separated into rotating and non-rotating damping. Rotating damping

in rotors is primarily due to dry friction between rotating parts that make up the rotor[4]. Other types

of friction including internal friction resisting the deformation of a material also contribute to

rotating damping.

6.1 Rotating Damping Mechanisms

There are several definitions of damping available within both ANSYS Workbench and NASTRAN.

These definitions can be grouped into two damping models: viscous damping and structural

damping. Viscous damping is proportional to the velocity of and structural damping is proportional

62
 to the deformation of the system. Viscous damping can be defined as Rayleigh damping or as an

independent viscous damper element (elements without stiffness or mass that have a specified

damping value). Structural damping includes elastomeric materials or other internal resistance to

material deformation. Both viscous damping and structural damping definitions can be used for

stability analysis of rotors.

6.1.1 Viscous Damping in Rotors

The Rayleigh damping model defines the damping matrix as a linear combination of the mass and

stiffness matrices as follows.

[Cr ] = 𝛼[𝑀] + 𝛽[K r ] 6.1

The Rayleigh damping model is also referred to as proportional damping and results in a symmetric

rotating damping matrix when the mass and stiffness matrices of the rotor are both symmetric.

Viscous damper elements are rarely used to model rotor rotating damping and are usually more

appropriate for non-rotating damping modeling. The can be utilized for non-proportional viscous

damping of a rotor and are defined as the damping relationship between two nodes.

Within ANSYS Workbench there are several methods to apply viscous damping. Viscous damper

elements can be added to the rotor, or Rayleigh damping can be applied to the entire structure or

via material properties. Applying Rayleigh damping to the entire structure in ANSYS Workbench will

create a viscous damping matrix for the entire mass matrix [𝑀] and stiffness matrix [𝐾] not just the

rotor. This global definition of Rayleigh damping is added to the simulation via the damping options

in the model. Applying Rayleigh damping via material properties allows the rotor and the stator to

be given different damping coefficients. Rayleigh damping for materials is defined in ANSYS

63
 Workbench as constant α and β coefficients or as variables defined in a table. Rayleigh damping with

constant α and β values is defined in the engineering data of a model. Rayleigh damping that

changes based on temperature, time, frequency, or other user-defined variables can only be defined

using APDL commands and is accomplished by created a data table and applying the data table to

the material property.

In NASTRAN, viscous damping is defined for the rotor via two methods. Viscous damper elements

that are part of the rotor are created with a specified damping value. Rayleigh damping is defined

for the entire rotor using the RSPINR card for modal and harmonic analysis or the RSPINT card for

transient analysis.

6.1.2 Structural Damping in Rotors

Some materials have internal friction that resists deformation. This internal friction can be modeled

as structural damping which is also referred to as hysteretic damping. Structural damping is

modeled as the imaginary term of a complex stiffness definition and is defined by the loss factor 𝜂

where the loss factor is the ratio between the amount of energy dissipated and the amount of the

potential energy from stiffness on a per cycle basis. The cycle through which the structural damping

energy of a rotor is dissipated is called the circular structural frequency 𝜔𝑠𝑡𝑟𝑢 . For the rotating

reference frame, the circular structural frequency 𝜔𝑠𝑡𝑟𝑢 is equal to that of the natural frequency 𝜔.

For the stationary reference frame, the circular structural frequency 𝜔𝑠𝑡𝑟𝑢 is equal to 𝜔 − 𝛺. As the

imaginary term of a complex stiffness model, the structural rotating damping term is orthogonal to

the rotor’s stiffness matrix and is defined as ±𝑗𝜂[𝐾𝑟 ]. The sign of this term depends on the direction

of the cycle through which the energy is dissipated and is determined by the circular structural

frequency (Figure 6-1).

64
 Figure 6-1 - Sign of Circular Structural Frequency

The structural damping term is written as 𝑠𝑔𝑛(𝜔 − 𝛺)𝑗𝜂[𝐾𝑟 ] for the stationary frame or

𝑠𝑔𝑛(𝜔)𝑗𝜂[𝐾𝑟 ] for the rotating frame[21]. The sign of the structural damping term is important as it

indicates the direction in which the material goes through the hysteresis cycle.

Since structural damping is by definition a per cycle phenomenon, its contribution to the

rotordynamics equation of motion must be defined as an equivalent viscous damping term for

transient analysis, modal analysis, and harmonic analysis of backward whirl inputs[22]. The equivalent

viscous damping contribution of structural damping is defined for the stationary reference frame as
𝜂
[𝐶𝑟 ]𝑒𝑞 = [𝐾𝑟 ] 6.2
|𝜔−𝛺|

65
For the rotating reference frame in which the whirl frequency ω includes the rotor spin speed Ω, the

structural damping equivalent viscous damping is defined as

𝜂
[𝐶𝑟 ]𝑒𝑞 = [𝐾𝑟 ] 6.3
|𝜔|

In both the stationary reference frame and rotating reference frame, these equivalent viscous

damping contributions preserve the sign of the structural damping term. These equivalent viscous

damping contributions are also used to calculate the contribution to the circulatory matrix due to

rotating damping.

[𝐵𝑟 ] = −𝑗𝛺[𝐶𝑟 ]𝑒𝑞 6.4

ANSYS Workbench defines structural damping as either a global property of the analysis or as a

material property but does not take the sign of the structural damping term into consideration. The

global definition defines the loss factor η using the APDL command line “DMPSTR, η” where η is the

real constant value of the loss factor. The material definition can be defined in the engineering data

of the model. ANSYS Workbench does not allow for structural damping to be included in a transient

analysis. For damped modal analysis and harmonic response analysis ANSYS Workbench uses an

equivalent viscous damping for the rotating reference frame and the stationary reference frame

that it defines as
𝜂
[𝐶𝑟 ]𝑒𝑞 = [𝐾𝑟 ] 6.5
𝜔

This definition of structural damping results in the equivalent viscous damping terms having the

incorrect sign for backward whirl frequencies and does not account for the rotor spin speed Ω term

in the stationary reference frame.

NASTRAN allows structural damping to be defined as a material property or as a property of the

rotor. To define structural damping as a material property, the loss factor η is input in the field “GE”

for any material property definition. Alternatively, the loss factor η can be defined for the entire

66
 rotor using the field “GR” in the RSPINR or RSPINT cards. In addition to the loss factor, NASTRAN

requires that the |𝜔 − 𝛺| and |𝜔| terms be specified for each analysis. These terms are defined as

parameters WR3 and WR4 in the NASTRAN bulk data section for the rotor definition and material

definition respectively. Since these parameters are defined for a specific frequency of interest,

modal analyses concerned with structural damping occurring at multiple frequencies require

multiple simulations with the parameters WR3 or WR4 varied.

For harmonic response analysis with forward whirl excitation frequencies, NASTRAN uses the

equivalent viscous damping to define the circulatory matrix due to rotating damping [𝐵𝑟 ] but does

not add the equivalent viscous damping term to the equation of motion. Instead the structural

damping is added as an imaginary stiffness term 𝑗𝜂[𝐾𝑟 ] where the sign of the term is always

positive. For transient analysis, modal analysis, and harmonic analysis with backward excitations,

NASTRAN uses the equivalent viscous damping definitions defined in Eq. 6.2 and 6.3.

6.2 Viscous Damping and Effect on Stability

Rotating damping is only destabilizing for supercritical forward whirl modes. Supercritical forward

whirl modes are any forward whirl mode where the rotor spin speed Ω is higher than the natural

frequency of the mode. In the stationary reference frame this criteria is met when a positive natural

frequency ω has a magnitude less than the rotor spin speed Ω. In the rotating reference frame, the

natural frequency ω includes the rotor spin speed Ω.

𝜔 = 𝜔𝑠𝑡𝑎𝑡𝑖𝑜𝑛𝑎𝑟𝑦 − 𝛺 6.6

Since the natural frequency accounts for the rotor spin speed in the rotating frame, the flexural

critical speed of a forward whirl mode in the rotating frame occurs when ω is zero. Supercritical

forward whirl modes in the rotating reference frame are modes with a negative natural frequency ω

67
value at the current rotor spin speed Ω that had a positive natural frequency ω value at a rotor spin

speed of zero.

The destabilizing effect of rotating damping at supercritical forward whirl modes can be seen by

examining the rotor’s equation of motion. Consider a symmetric rotor with mass, stiffness, rotating

and non-rotating damping. In the stationary reference frame the equation of motion for this rotor is

defined as

𝑚𝑢̈ + (𝑐𝑛 + 𝑐𝑟 )𝑢̇ + (𝑘 − 𝑗𝛺𝑐𝑟 )𝑢 = 𝑓 6.7

Setting the force vector to zero and solving for the complex eigenfrequencies results in

−𝑚𝜆2 + 𝑗𝜆(𝑐𝑛 + 𝑐𝑟 ) + (𝑘 − 𝑗𝛺𝑐𝑟 ) = 0 6.8

−(𝑐𝑛 +𝑐𝑟 ) 𝑘−𝑗𝛺𝑐𝑟 (𝑐𝑛 +𝑐𝑟 ) 2

𝜆= 2𝑚
± 𝑗√ 𝑚
−( 2𝑚
) 6.9

Since damping is much less than stiffness, a pair of complex eigenfrequencies can be approximated

as

−(𝑐𝑛 +𝑐𝑟 ) 𝛺𝑐
𝜆= + 𝑟 ± 𝑗𝜔 6.10
2𝑚 2𝑚𝜔

Looking only at the real part of the complex eigenfrequency pair, the stability values of the pair are

−(𝑐𝑛 +𝑐𝑟 ) 𝛺𝑐𝑟

𝑆𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 2𝑚
+ 2𝑚𝜔 6.11

The sign of the second term is determined by the direction of the whirl. From Eq. 6.11, it becomes

clear that for backward whirl modes the stability value will also be negative indicating a stable

mode. It also shows that when viscous damping is present in the rotor, it is providing a destabilizing

contribution to the forward whirl modes due to the circulatory matrix. The destabilizing contribution

68
 to the forward whirl mode from the circulatory matrix is greater than the stabilizing influence of the

rotating damping and non-rotating damping when

𝑐
𝛺 > 𝜔(1 + 𝑐𝑛 ) 6.12
𝑟

This instability condition and the effect of viscous damping in stationary reference frame is

confirmed by the modal analysis stability results of the example rotor defined in ANSYS Workbench

(Figure 6-2) and NASTRAN (Figure 6-3).

Figure 6-2 – Stability Effect of Viscous Rotating Damping in ANSYS Workbench Stationary Reference Frame

69
 Figure 6-3 – Stability Effect of Viscous Rotating Damping in NASTRAN Stationary Reference Frame

When the spin speed Ω is greater than the natural frequency ω, the destabilizing contribution from

the circulatory matrix is greater than the stabilizing contribution from the rotating damping matrix.

However, the destabilizing contribution from the circulatory matrix must be greater than both the

stabilizing contribution of the rotating damping matrix and the non-rotating damping matrix in order

for the mode to be unstable.

In the stationary reference frame, backward whirl modes are always stable, subcritical forward
𝑐𝑛
modes are stable, supercritical forward whirl modes are unstable when 𝛺 > 𝜔(1 + ). There is no
𝑐𝑟

difference in how the stability is calculated in the stationary reference frame for ANSYS Workbench

and NASTRAN.

70
The factors for stability in the rotating reference frame are harder to identify directly from the

rotating reference frame equation of motion. Translating Eq. 6.12 to the rotating reference frame

results in a instability condition for forward whirl modes when

𝛺𝑐𝑟 > (𝜔 + 𝛺)(𝑐𝑟 + 𝑐𝑛 ) 6.13

Rearranging Eq. 6.13 further it can be seen that the instability criteria for forward whirl modes only

includes damping matrix terms and the circulatory matrix due to non-rotating damping.

𝜔(𝑐𝑟 + 𝑐𝑛 ) + 𝛺𝑐𝑛 = 𝜔(𝑐𝑟 + 𝑐𝑛 ) + 𝐵𝑛 < 0 6.14

From Eq. 6.13, it is shown that forward whirl modes are unstable when the sign of the natural

frequency is negative. As a result, a stability plot of a rotor due to viscous damping in a rotating

reference frame should look similar to a stationary reference frame version.

In ANSYS Workbench, the circulatory matrix due to non-rotating damping is not included in the

equation of motion of the rotating reference frame. As a result forward whirl modes appear to

always have positive stability values above the flexural critical speed of the mode whether there is

or isn’t any rotating damping in the model (Figure 6-4)

71
 Figure 6-4 – Stability Effect of Viscous Rotating Damping in ANSYS Workbench Rotating Reference Frame

6.3 Structural Damping and Effect on Stability

The stability contribution from structural rotating damping is similar to that of viscous damping

except that the sign of the rotating damping term is dependent on the value of ω or ω-Ω. Structural

damping always has a stabilizing contribution to backward whirling modes and to subcritical forward

whirling modes. At the onset of instability, the sign of structural rotating damping flips resulting in a

sharp change in the contribution of structural rotating damping to mode stability as can be seen in

the results from the NASTRAN model (Figure 6-5). Similar to viscous damping however, the mode

𝑐𝑛
remains stable unstable until 𝛺 > 𝜔𝑛 (1 + ).
𝑐𝑟

72
 Figure 6-5 - Stability Effect of Structural Rotating Damping in NASTRAN Stationary Reference Frame

In the rotating reference frame, the effect of structural rotating damping is the same and the change

from stabilizing to destabilizing occurs for each mode at the same rotor spin speeds as in the

stationary reference frame.

As previously shown ANSYS Workbench does not correctly apply structural damping in the

stationary reference frame. ANSYS Workbench converts structural damping to an equivalent viscous

damping contribution by dividing by the circular frequency ω in both the stationary reference frame

and the rotating reference frame. In the stationary reference frame the term should instead be

divided by |𝜔 − 𝛺|. This results in incorrect stability results when structural damping is included in a

stationary reference frame model in ANSYS Workbench (Figure 6-6).

73
 Figure 6-6 - Stability Effect of Structural Rotating Damping in ANSYS Stationary Reference Frame

In the rotating reference frame ANSYS Workbench correctly accounts for structural rotating

damping resulting in the characteristic switch from a stabilizing contribution to a destabilizing

contribution at the critical speed (Figure 6-7).

Figure 6-7 - Stability Effect of Structural Rotating Damping in ANSYS Rotating Reference Frame

74
7 Conclusion

7.1 Summary and Recommendations

The current state of rotordynamics FEA programs allows for accurate analysis and prediction of

rotor behavior but requires a sound understanding of the rotordynamics theory in order to correctly

create models and interpret results. Two commercially available software, ANSYS Workbench and

NASTRAN, are both able to capture critical speeds and natural frequencies, but will also identify

additional invalid critical speeds for backward whirling modes because the way their analysis

method is set up does not allow to track correct directional information for all modes. While some

experienced rotordynamics specialists may distinguish backward modes from the gyroscopic

stiffening / softening trends, this shortcoming has good potential to lead to wrong engineering

conclusions. Further improvement of mode directivity could be made by filtering out mode shapes

of lateral vibration modes with rotation opposite that of the rotational spin and thereby removing

duplicate data. However, the complete solution will be adopting complex variable notation in the

FEA procedure.

It has been shown that ANSYS Workbench does not currently implement contributions from rotating

structural damping correctly and does not have a clear method to apply external damping (such as

stator damping) and internal damping (such as structural damping of the rotor), which can lead to

wrong conclusions in mode stability analysis. Structural damping for finite element rotors is best

implemented as an equivalent viscous damping contribution using the process shown in Chapter 6.

Element calculations add an extra layer of calculations for the Coriolis, spin softening, and

circulatory matrices when transforming them back to nodal coordinates. While element calculations

are necessary and appropriate for the inertia, stiffness, and gyroscopic matrices due to the manner

75
 in which they must be calculated, the additional matrices can be calculated on a nodal basis saving

computational time. NASTRAN creates these matrices in their nodal form while ANSYS Workbench

creates all matrices on elemental basis.

7.2 Future Work

The current commercial FEA software can handle rotordynamics analysis well for the forced

response analysis, while stability analysis and handling of rotating damping should be improved.

Clearly the most effective way for improvement in current commercial finite element analysis code

is the adoption of the complex notation in describing the motion of the rotor system so that

directivity of the rotor motion is correctly accounted for. More systematic implementation of

modeling of internal damping is also desired. Future areas for capability improvement include multi-

physics modeling and applications, for example heat transfer and fluid interactions with

rotordynamics that will enable more accurate simulation of realistic rotor systems such as aircraft

engines.

76
References

[1] Genta, G., 2005, Dynamics of Rotating Systems, Springer-Verlag, pp. 658.

[2] Muszynska, A., 2005, Rotordynamics, CRC Press.

[3] Vance, J., 1988, Rotordynamics of Turbomachinery, John Wiley & Sons.

[4] Kandil, M., 2004, “On Rotor Internal Damping Instability, Doctorate of Philosophy Dissertation,”

University of London.

[5] Nelson, 2007, “Rotor dynamics without equations,” International Journal of COMADEM, 10(3), pp. 2

[6] Geradin, M., and Kill, M., 1984, “A New Approach to Finite Element Modelling of Flexible Rotors.”

Engineering Computations, Vol. 1, pp. 52-64

[7] Kumar, D., 2016, "Rotordynamic Analysis using 3D Elements in Fixed and Rotating Reference Frame"

[8] Genta, G., Silvagni, M., and Qingwen, C., September 2013, "Dynamic Analysis of Rotors: Comparison

between the Simplified One-Dimensional Results and those obtained through 3-D Modeling,"

Congresso Nazionale AIMETA, XXI.

[9] Ma, W., Wang, J., 2012, “3D Solid Finite Element Modeling and Rotordynamics of Large

Machines: Application to an Industrial Turbo Engine,” Advanced Materials Research, 591-593,

pp. 1879-1885

[10] Thiery, F., 2010. “Evaluation of 3D rotordynamics capabilities within NX Nastran, Master of Science

Thesis”, Lulea University of Technology.

[11] Srikrishnanivas, D., 2012, “Rotor Dynamic Analysis of RM12 Jet Engine Rotor using ANSYS, Master of

Science Thesis,” Blekinge Institute of Technology.

[12] Vollan, A., and Komzsik, L., 2012, Computational Techniques of Rotor Dynamics with the Finite

Element Method, CRC Press

77
[13] Friswell, M., Penny, J., Garvey, S., 2010, Dynamics of Rotating Machines, Cambridge University

Press.

[14] ANSYS Inc, "ANSYS Help Documentation," 2016.

[15] Swanson, E., Powell, C., and Weissman, S., 2005, “A Practical Review of Rotating Machinery Critical

Speeds and Modes,” Sound Vibration, 39(5), pp. 10-17

[16] Vance, J., Zeidan F., Murphy B., 2010, Machinery vibration and rotordynamics, John Wiley & Sons.

[17] Kessler, C. L., 1999, "Complex Modal Analysis of Rotating Machinery, Doctorate of Philosophy

Dissertation," University of Cincinnati.

[18] Dimentberg, F.M., 1961, Flexural Vibrations of Rotating Shafts, Butterworths.

[19] Lee, C., 1993, Vibration Analysis of Rotors, Kluwer Academic Press

[20] Kessler, C., and Kim, J., 1998, "Complex Modal Analysis and Interpretation for Rotating Machinery,"

IMAC, XVI, ed.

[21] Genta, G., 1995, Vibration of Structures and Machines, Springer-Verlag, pp. 474.

[22] MSC Software Corporation, "MSC NASTRAN Documentation," 2016.

[23] Myklestad N., 1944, "A New Method of Calculating Natural Modes of Uncoupled Bending Vibration

of Airplane Wings and Other Types of Beams," Journal of the Aeronautical Sciences, 11(2), pp. 153-

162.

[24] Prohl M., 1945, "A General Method for Calculating Critical Speeds of Flexible Rotors," Journal of

Applied Mechanics, 12, pp. 142.

[25] Murphy, B., and Vance, J., 1983, "An Improved Method for Calculating Critical Speeds and

Rotordynamic Stability of Turbomachinery," Journal of Engineering for Power, 105(3), pp. 591-595.

[26] Vance, J., Murphy, B., and Tripp, H., 1987, "Critical Speeds of Turbomachinery: Computer

Predictions Vs. Experimental Measurements—Part I: The Rotor Mass—Elastic Model," Journal of

Vibration, Acoustics, Stress, and Reliability in Design, 109(1), pp. 1-7.

78
[27] Barrett, L., Gunter, E., and Allaire, P., 1978, "Optimum Bearing and Support Damping for Unbalance

Response and Stability of Rotating Machinery," Journal of Engineering for Power, 100(1), pp. 89-94.

79