1157

,4ctive Vibration Control of Rotor Systems

L.Y.Cheung t, R.W.Dunn t, A.R.Daniels tand T.Berry

University 0 1 Bath, U.K.

]:Westinghouse, Chippenham, U.K.

,4bstract each with a fixed structure and each operating in a

specified region of the state spacc. The combinalions
A study is presented in which the vibrations of a rotor of the subsystems according to some rules result in a
;Ire controlled under both synchronous and transient system that has a behaviour different from each of the
conditions. In this paper sliding mode control theory individual subsystems.
has been adopted to minimise rotor vibration due IO The algorithm and the implementation of this control-
inass imbalance. E.g., blade loss. The syslcm gain
ler to control the vibration of a Acxible :rotor suppcrled
matrix is obtained by choosing the desired pole posi-
by magnetic b a i n g is discussed. Using a finite ele-
lion and solving the Riccatti equation. Control forces
nient method. a mathematical model of a flexible iutor
;ire applied to the system via a feedback loop. Multi-
is formulated and sliding mode control is then applied
processor hardware is used to apply the controller a - to this model. Experiments arc cilrried out t o evaluate
gorithm on line to an experimental rig consisling of a
this performance of the controllcl.
rotorand twojonmalbearirigs witha magnetic actuator
10 apply thecontrol force. Experimental work has bccn
carried out to show the effectiveness of the proposed
control method. 2 Rotor Dynamics
The mathematical model for the tlexible rotor is 2:iven
1 Introduction by
M q + Cq Kq = 0 + (1)
Over the past few decades significant improvements where M , C and It are themass, damping and the:<tiI'f-
liave been made in many of the components present in ness matrices, respectively. q is the n* 1 displaceinent
I-otatingmachinery. Advance in rotor materials techno- vector and c m be expressed as
llogy has allowed machine designers to push equipment
!inch as turbines, compressors and engines to speeds 4' = ( I : , , y,, O l , @ l , .. . , J ' , b , y,t, On.@n) (2)
-
previonsly unattainable. Extensive studies of rotor
(dynamics have been carried out by many researchers
This flexible rotor is now divided into seven elements
.1 I, 21. These researchers showed that a rotor has certain
as shown in tigure 1. The flexible rotor is supportcd by
';peed ranges in which a large ruid unacceptable amp-
litude of vibration could be developed. These speed magnetic bearing at station 3 and sensors are located
at stations 2. 3 and 7.
;ranges are known as critical speeds (or critical frequcii-
cies) which could cause a bearing failure or result i n l'hcjonrnal hearing shown in figure 1 has thc following
tcxccssive rotor deflection. Under these circumstances, Carameler
ithe problem of ensuring that a rotor-bearing system
lperforms with a stable and low-level amplitude of V I - Hearing diameter 100 imn
bration becomes increasingly important. Heanng length 70 0 i-nm
In recent years, the electromagnetic bearing uscd 111 Radial clearance 146 / A m

:Such application has grown rapidly. It generates no Idmcant viscosity 0 013 V'SV-~
inechanical loss and needs 110 lubricants such as oil or
air as it supports the rotor without physical contac1. The two bearings have very similar stiffness and d;iml)-
'The electromagnets are open loop unstable and all ing coefficients. The two end discs each has a mass of
'designs require external electronic control to regulate 92.89 kg.
the forces acting on the bearing [ 3 ] .
Finite element method [4] is used to analyse the rotor,
I n this paper, a new control algorithm has been de- the uncontrolled synchronous responses for this model
veloped by using Sliding Mode Control(S1,C). Sliding are shown in figure 2. The critical speed is about 280
mode may be regarded as a combination of subsystems, rad aid 3 15. rads.

CONTROL'94.21-24March 1994. Conference Publication No.389,O IEE 1994

 1158

Reachability of the system

The switching logic of the controllex is operated ac-

cording to the signs of the signal of d(z)z, where d(z)
is defined as
u(t) = 7 2 (5)
where c is called the switching vector and u ( z ) = 0
is called the switching hyperplane. This switching
plane can be determined by minimising a quadratic
Fig 1: Rotor Geometry performance index. The control signal is given by

Magnitude(pn)

where
4., - wj, if zju, > 0, i = l,.. .,m
pij, ifzju,<O, j = 1 , . . . , n (7)

To find the switching vector c, the first step is to specify

a particular canonical form for the system to simplify
the development of the design scheme [SI.By assump-
tion, the matrix B has full rank m, so that there exists
an orthogonal n*n transformation matrix T [6] such
Rotational frequency(rad/s) that
Fig 2: Unbalanced Response TB= [ :2]

where B2 is m*m and non-singular.

For computation purposes, equation 1 can be written
in the first order state vector form The transformed state variable y = Tx is now defined,
in tenns of which the state equation (4) becomes

(K
[";]v!I[- -[MI-'[C]
[I1 ) ( )+
( 0
[Ml-L(Fl ).= ( 5) (3 and the sliding condition is

The above equation can be represented as the standard If the transformed state y is partitioned as
state equation of

z ( t ) = Az(l) + B u ( t ) (4)
and thematrices TAT- l,TB andCT-l are partitioned
where accordingly, then equations (9-10) my be rewritten as

and

where
3 Control System Design
The sliding mode controller has two important charac-
teristics: By defining F as

The switching hyperplane

 1159

For state feedback control, closing the loop gives By using equation 23, this may be transferred to

lii(t) = ( A i l - A I ~ F ) Y I ( ~ ) (17) ~2(1) = -Q,'[Q~I + A ; ~ P ] Y I ( =~ )- F Y I ( ~(28)

)
which indicates that the design of a stable sliding mode The switching hyperplane can be completely determ-
requires the determination of the gain matrix F such ined by equation( 28).
that ( A lI - A12F) has n-m left hand half plane eigen-
To ensure that the reachability condition, i.e. the con-
values.
ditions under which the system trajectory will hit the
Thesimplest method to solveC fmmFis that employed switching plane from any initial condition in the state
method in [7], by letting C2 = Im. This is equivalent space. The controller gain should be selected to satis-
to specify R = Bz, i.e fied the condition stated in [SI:

c = [F I,]T (18) kl; > (CTB)-.'CTai

This could minimise the amount of calculation in cal-

k2i < ( C T B ) -'CTtti (29)
culating from F to C and hence reduces the possibility where ai is the ith column of the matrix A in quation
of numerical errors. (4) and kli and k2i are theithcolumn controllergains.

Consider the quadratic performance index

d(u)= /: zT(t)Qz(t) dt (19)

4 Illustration
To illustrate the effectiveness of the sliding mode con-
troller, a switching vector is designed using the method
1 , denotes thetimeat which theslidingmodestart. The
described mlier.
matrix Q is positive and symmetric. The aim of this
technique is to minimise J subject to the system equa- 24.2 18.1 41.1 45.4 21.8 22.9
C.T =
tion, assuming that the initial conditions are known. 27.1 19.3 29.1 41.9 21.7 1.47
By partitioning the product
0.054 0.163 0.0035 0.112 0.05653 0.131
T-IQT= [ QII QIZ] ( 20)
0.025 0.112 0.124 0.123 --0.0158 0.102
(30)
]
Q21 Q22
and the controller gain is chosen as
Let
1 ~ 1 1 2 = - 1 ~ 2 ~ 2 = 2 0 0 , 1 ~ 1-1<21*=::
1~= 1.
Q' = QII - QIZQ~'QZI (21)
A' = Aii - AIzQ,~QzI (22) I C 1 2 . ~ = - K 2 2 s = 2 ~ , I ~ l z I I = - I ~:=21,2 , 1
(3 1)
4 1 ) = yz(t) + Q ~ L Q z ~ ~ ~ ( t(23)
) and all the other gains are zero.
where v(t) is the optimal control. Therefore, I can be The PI) controller using a pole-placement technique is
rewritten in the form of designed the feedback matrix of the s.ystem is

J ( 4= lY(yY(t)Q*y~(t) + vT(t)Q224t)) dt =- 62.0 232.7 7.85

18.9 8.54 7.84
79.4. 35.7
143.4 214.1
40.11
120.14
(24)
subject to 0.34 0.13 0.36 0.02 0.09 -0.15

$I@)= A*y(t) + A124t) (25)

0.08 0.09 0.08 0.13 0.14 0.03 ] (32)

The resulting eigenvalues are shown in Table 1. The

which is of the form of the standard linear quadratic eigenvalues obtained by the sliding mode control are all
optimal regulator problem. The controllability of A' stable and shifted more to the left hand plane. Figure 3
and A12 is proof in [7]. Thus, a positive define unlque shows the root loci of the open loop system, in which
solution P is guaranteed for the algebraic matrix Ric- the system is predicted as unstable at high speed.
catti equation.

Q'; P= 0
PA' t A * T P - P A ~ z Q ~ ~ ' A ~ + (26)
5 Implementation and Results
The optimal control 'v' is minimised to give
A rig was designed and constructed to allow a study
w = -Q;;A;PY,(~) (27) of the active control strategy described here [91. The
 1160

rotor was 1.2 mm. An emagency bearing p v i d e s

Open Loop Eigenvalues I Sliding Mode Eqenvalues protection when the amplitude exceeds 1.0 mm. The
-66.3 f 51 1.2j I -66.5 f+ 51 1.3i transducerswere connectedto the filter circuit to a data
acquisition card fitted with a PC.
-14.0 f 320.43' -32.1 f 320.3j
-29.7 k 243.7j -29.9 f 243.7j
-29.9 f 270.6j MoDue
-30.2 f 270.9j Inmface
-6.35 f 271.6j

Table 1: Eigenvalues
44 '

h v *
Conlrol scmr
Plot * anphi
smu
I I I I I I

m - .. . . .:. . . . .. : . . . . . . . .:, . . . . .-
,/./-------

400- " ' . . . (. .<. . . . . . . . . . . -

- - - - - - -. - . _ _ _ _ -c;;-.-
--- -
_.-.
_.e..
,. . -
200 - ,;>.;;e._.-. . . . . . Fig 5: Parallel Tasks
.,i -'- ~

, ---- ?'---
The Pc was linked to a parallel processing computer
that had 8 Inmos T800 transputers. The Inmos T800
transputer is a single chip microprocessor consisting
of a 32 bit CPU,a 64 bit floating point unit, 4KB of
fast on chip RAM, an extemal memory interface, and
4 serial links.

overall system configuration is shown in figure 4. The

rotor consists of a l00mm diameter mild steel shaft of
nominal length 2358 mm supportingtwo 9 0" thick,
406 mm diameter overhung steel disks. The rotor was
driven by a 25kW dc variable speed drive via a flexible
tooth belt and universal coupling. The oil film bear-
ings supporting the rotor were of the circumferential
grmve type. The bearing parameters are listed in the
previous section. The whole assembly was mounted
on a concrete bedplate.
Vibration control is provided by a magnetic bearing
situated at shown in figure 1. The nominal air gap
between the electromagnetic force actuator stator and

I I
I I
I I
I I
I I
I I
I I
I I
L- - -1

Fig 6:Control Scheme - Block Diagram

The vibration control softwareis written in C program-

Fig 4: Experimental Rig ming language and runs on the Helios parallel operat-
 1161

such test carried out at 300 rad/= using sliding mode

control method and PD control method. By contrast,
the sliding mode control has a better responses than
the fixed gain system.

Displacement(pm)
1601 II I I 6 II - 1

..

- Uncontrolled
-
100 - .PD control
-- ShdingMcdeControl
-
-
80 -

Ti”$)
Fig 9: Transient Mass Loss (Horizontal)

Displacement(pm)
I

100

60
I
I
-
-- -------_-- --
I I L L
2o0 02 04 06 08 1
Time(s)
Fig 10 Transient Mass L x s (Vertical)

6 Conclusion
This paper has described the modelling and control of
a magnetic bearing for a flexible rotor. Application
Rotational Speed(rad/s) of sliding mode control to rotor dynamic vibrations
has b e n investigated. The general vibration level can
Fig 8: Synchronous Response (Vertical) be significantly reduced by the sliding mode method.
Comparisons between the PD control and the sliding
Transient rotor mass loss tests were then perlormed. mode methods have been carried out. The results in-
This involves attachment of a mass to the non-driven dicate that the sliding mode control has better overall
end disc of the rotor using simple tie on technique. The response. The sliding mode control has also shown
mass was then removed at any selected time using a much better parameter uncertainty response than the
blade mechanism. Figures 9 and 10 show the results of fixed gain system.
 1162

7 Acknowledgement
This work has bem supported by SERC grant number
GR/F/63213. The authors would also l i e to thank
Prof. C.R. B m w s and Dr. I?S Keogh for their
comments and suggestions.

References
[ l ] SUBBIAH, R. AND -ER, N., ‘On the transient
analysis of rotor bearing systems’, ASME Journal
of Wbmtion, Acousrics, Stress and Reliability in
Design, vol. 110, 1988.
Z. AND PORAT,I., ‘An optimal con-
[2] VIDERMAN,
trol method for passage of a flexible rotor through
resonances’, ASME Journal of Dynamic Systems,
Measurement and Control, vol. 109,1987.

[3] F.J. KErrH, R.W. AND ALLAIRE, I?, ‘Digitalcon-

trol of magnetic bearings supportinga multimass
flexible rotor’, Tribology Transcations, vol. 33,
1990.

[4] RUHL, R. AND BOOKER,J., ‘A finite element

model for distributed parameter turborotor sys-
tems’, Trans. ASME, J.Eng had, vol. 94, 1972.
D., ‘Canonical forms for linear
[SI LUENBERGER,
multivariable systems’, IEEE Trans. Automatic
Control, 1%7.
[6] C.M.DORLINGAND A.S.I.ZINOBER,‘Two ap-
proaches to hyperplane design in multivariable
variable structure control systems’, Inr. J. Con-
trol, vol. 44,1986.
[7] UmIN, V. AND YANG, K., ‘Variable structure
of model following model’, Automation Remote
Control, 1978.

[8] UTKIN,V., ‘Varaiblestructuresystem with sliding

modes‘, IEEE Trans. Automatic Control, vol. 22,
1977.

[9] C.R.BuRRows, M. AND CLEMENTS,s., ‘Active

vibration control of flexible rotors: an experi-
mental and theoretical study’, Proc. Ro.val Socierv
London A, vol. 422,1989.

[lo] PERIHELION SOFTWARE LTD.,The Helios @er-

aring Sysrem. Prentice Hall, 1989. ISBN 0-13-
386004-3.
R y e I. Fleuble mbor arm

- . .. ... . . .... ,