A Comparison of Semi-Active Control Strategies for the MR Damper

S.J. Dyke' and B.F. Spencer Jr.

'Dept. of Civil Engineering, Washington Univ., St. Louis, MO 63130, U.S.A.
2Dept. of Civil Engineering and Geol. Sci., Univ. of Notre Dame, Notre Dame, IN 46556, U.S.A.
Abstract

trolled responses are compared to the uncontrolled and

passive responses.

Because the behavior of semi-active devices is often highly

nonlineal; one of the main challenges in the application of
this technology is the development of appropriate control
algorilhms. Herein, a number of recently proposed semiactive control ulgorithms are evaluated f o r use wit?i the
magrietorheological (MR) damper; an innovative semi-active control device that appears to be particularly promising f o r civil engineering applications. The perj5ormance of
the resulting control systems are compared through simulation, and the advantages of each algorithm are discussed. The results demonstrate that the pelformanee of
the control system is highly dependent on the choice of algorithm employed.

2: Mechanical Model of the MR Damper

To evaluate the potential of MR dampers in structural
control applications, the simple mechanical model depicted in Fig. 1 was developed and shown to accurately predict
the behavior of an MR damper over a broad range of inputs [17, 181. The model is based on the response of a prototype MR damper, obtained for evaluation from the Lord
Corporation. The equations governing the force f predicted by this model are

1: Introduction
The magnetorheological (MR) damper is a semi-active control device that is capable of generating the magnitude of forces necessary for full-scale applications, while
requiring only a battery for power [2, 3, 211. Additionally,
this device is offers highly reliable operation at a modest
cost and its performance is relatively insensitive to temperature fluctuations or impurities in the fluid. However, because of the inherent nonlinear nature of these devices,
one of the challenging aspects of utilizing this technology
to achieve high levels of performance is in the development of appropriate control algorithm that can take advantage of the unique characteristics of the device.
A variety of semi-active devices and control algorithms have been proposed for seismic control [7-14, 201.
Because the characteristics of the various semi-active devices are different (e.g., variable friction, fluid-orificing,
controllable fluids, etc.), a control algorithm that performs
well for one device may not be suitable for use with another device. The focus of this paper is to investigate a number of recently proposed control strategies for use with the
MR dampcr. In simulation, one cxample is considered in
which these methods are applied to control a three-story
structure using an MR damper. To assess the effectiveness
of each of the control algorithms, they will be evaluated in
their ability to reduce the peak responses for the N-S component of the El Centro earthquake excitation. The con-

0-8186-8218-3/97 $10.00 0 1997 IEEE

(3)
where z is an evolutionary variable that accounts for the
history dependence of the response. The model parameters
depend on the voltage v to the current driver as follows

a
where

a,+abu,
U

c1

Cla+ClbUy

CO = C O a + C O b U

(4)

is given as the output of the first-order filter

li = q ( u - v ) .
(5)
Eq. (5) is necessary to model the dynamics involved in
reaching rheological equilibrium and in driving the electromagnet in the MR damper [ 17, 181.

3: Control Algorithms
Consider a seismically excited structure controlled
with a single MR damper. Assuming that the forces provided by the MR damper are adequate to keep the response of the primary structure from exiting the linear
region, then the equations of motion can be written as

M,X

+ C,X + K,x

Af - M,rZg

(6)

where x is a vector of the relative displacements of the

floors of the structure, xg is a one-dimensional ground acceleration, f is the measured control force, defined by Eqs.
(1-.5), r is a column vector of ones, and A is a vector de-

580

and P is a real, symmetric, positive definite matrix. In the

case of a linear system, to ensure V is negative definite,
the matrix P is found using the Lyapunov equation

A ' P + P A = -Qp

(1 1)

for a positive definite matrix Q p . The derivative of the

Lyapunov function for a solution of Eq. (7) is
V = - ;z'Q,z

Figure 1. Simple Mechanical MR Damper Model.

+ z'PBf + z'PEXg.

(12)

Thus, the control law which will minimize V is

termined by the position of the MR damper in the structure. This equation can be written in state-space form as
Z

= Az+Bf+Efg

y = Cz+Df+v

v =

(8)

3.2: Decentralized Bang-Bang Control

McClamroch and Gavin [ 161 used a similar approach
to develop the decentralized bang-bang control law. In this
approach, the Lyapunov function was chosen to represent
the total vibratory energy in the structure (kinetic plus potential energy), as in

V = ;x'K,x

3.1: Control Based on Lyapunov Stability Theory

V = ix'K,X

1/ 2

iTx,)'M,(x

+ rx,).

(14)

+ ( X + rig)'(C,X - K,x + Af).

(15)

In this expression, the only way to directly effect V is

through the last term. To achieve the goal of making V a
large and negative (maximizing the rate at which energy is
dissipated), the following control law is chosen
v = VmaxH(-(X+ T i J A f ) .

(16)

Note that, because the only non-zero terms in the A vector

are those corresponding to the location of the MR damper,
this control law requires only measurements of the floor
velocities and applied forces. When the damper is located
in the upper floors, the relative velocity is needed. Interestingly, when a semi-active device is located between the
ground and first floor, the absolute velocity of the first
floor is required, which is not readily available. Therefore,
to implement this control algorithm, one would approximate the absolute velocity (obtain the pseudo velocity) by
integrating the absolute acceleration, as in [ 191.

(9)

where I IzI I p is the P-norm of the states defined by

llzllp = [Z'PZI

+ :(X

Using Eq. (6) , the rate of change of the Lyapunov function

is then

In some cases it is possible to employ Lyapunov's direct approach to stability analysis in the design of a feedback controller [l]. The approach requires the use of a
Lyapunov function, denoted V(z) , which must be a positive definite function of the states of the system, z . According to Lyapunov stability theory, if the rate of change
of the Lyapunov function, V(z) , is negative semi-definite,
the origin is stable i.s.L. (in the sense of Lyapunov). Thus,
in determining the control law, the goal is to choose a control input which will result in making V as negative as
possible. An infinite number of Lyapunov functions may
be chosen, which may result in a variety of control laws.
Leitmann [ 141 applied Lyapunov's direct approach for
the design of a semi-active controller. In this approach, a
Lyapunov function is chosen of the form
= ~IlZllP

(13)

where H ( .) is the Heavisicle step function. Notice that this

algorithm is classified as a bang-bang controller, and is dependent on the sign of the measured control force and the
states of the system. To implement this algorithm, a Kalman filter is used to estimate the states based on the available measurements (i.e., :MR damper displacement and
structural accelerations). Thus, in this algorithm, better
performance is expected when measurements of the responses of the full structure are used.

(7)

where z is the state vector, y is the vector of measured

outputs, and v is the measurement noise vector. For these
applications, the measurements typically available for control force determination include the absolute acceleration
of selected points on the structure, the displacement of the
control device, and a measurement of the control force.
A variety of approaches have been proposed in the literature for the control of semi-active devices. Subsequently, a selection of these approaches will be presented and
evaluated for a numerical example. In developing the control laws, note that i) the control voltage is restricted to the
range v E [0, V,,,] , and ii) for a fixed set of states, the
magnitude of the applied force If1 increases when v increases, and decreases when v decreases.

V(Z>

V,,,H(-z'PBf),

(10)

581

to check if the force is greater than p N ( t ) , where p is the

coefficient of friction, because the MR damper is not subject to static friction, and ii) a force feedback loop was
used to induce the MR damper to produce approximately
the frictional force corresponding to the desired normal
force. Thus, the goal is to generate a desired control force
with a magnitude

3.3: Clipped-Optimal Control

One algorithm that has been shown to be effective for
use with the MR damper is the clipped-optimal control approach, proposed by Dyke, et al. [7-91. The clipped-optimal control approach is to design a linear optimal
controller K,( s) that calculates a desired control force f ,
based on the measured structural responses y and the
measured force f applied to the structure, i.e.,

f,

= L-l{ -K,OL{

where the proportionality constant g, has units of stiffness. The resulting control law is

;}}

where L { . } is the Laplace transform. Because the force

generated in the MR damper is dependent on the responses
of the structural system, the desired optimal control force
f , cannot always be produced by the MR damper. Only
the control voltage v can be directly controlled. Thus, a
force feedback loop is incorporated to induce the MR
damper to generate approximately the desired optimal
control force f , . To this end, the command signal v is selected according to the control law

v = V,,xH({f,-flf).

An appropriate choice of g, will keep the force f ,within

the operating envelope of the MR damper a majority of the
time, allowing the MR damper force to closely approximate the desired force. Additionally, notice that this control law requires only measurements of applied force and
the relative displacements of the control device.

4: Numerical Example
The performance of the control algorithms presented
previously are now evaluated in one example through numerical simulation. A model of a three-story building configured with a single MR damper is considered. The MR
damper is rigidly connected between the ground and the
first floor, as shown in Fig. 2. The governing equations can
be written in the form of Eq. (6) by defining

(18)

Although a variety of approaches may be used to design

the optimal controller, H , L Q G methods are advocated
because of their successful application in previous studies
[5-111. The approach to optimal control design is discussed further in [8,9].

3.4: Modulated Homogeneous Friction

Ms =

Another semi-active control algorithm was proposed

for use with a variable friction damper [13]. In this approach, at every occurrence of a local extrema in the deformation of the device (i.e., when the relative velocity
between the ends of the semi-active device is zero), the
normal force applied at the frictional interface is updated
to a new value. At each local minima or maxima in the deformation, the normal force, N ( t ) , is chosen to be proportional to the absolute value of the deformation of the semiactive device. The control law is written [ 131
N ( t ) = glP[A(t)ll

98.3 0
0 98.3
0
0

175 -50
0

IN I:],13
1 0 -50

5112.0 -6.84

K, = 10 -6.84 13.7 -6.84

0 N.sec

k g , Cs = -50 100 -50

-6.84 6.84

?;

A =

This system is a simple model of the scaled, three-story,

test structure, described in [6, 91 which has been used in
previous active control studies at the Structural Dynamics
and Control / Earthquake Engineering Laboratory at the
University of Notre Dame. Because the MR damper is attached between the first floor and the ground, its displacement is equal to the first floor relative displacement, i.e.,
x = x1 in Eqs. (1-3).
In this example, the structural measurements available
for calculating the control action include the absolute accelerations of the structure, and the MR damper displacement (i.e., y = [fal xa2 jin3 x,]). Thus, the governing
equations can be written in the form of Eqs. (7-8) by defining

(19)

where g is a positive gain, and the operator P[ a] (referred

to as the prior-local-peak operator) is defined as

P[A(t)] = A ( t - s ) , where
s = {min x>O:A(t-x)=O},
(20)
defining A(t - s) as the most recent local extrema in the
deformation.
Because this algorithm was developed for use with a
variable friction device, the following modifications were
necessary to apply it to the MR damper: i) there is no need

582

largest damping forces may not be the most effective approach to protective system design.
A variety of Q p matrices were tested for the
Lyapunov-based controlle:r, and the best results were
achieved when the Q p matrix was chosen as a 6 x 6 matrix, with nonzero values only in the (l,l), (2,2), and (3,3)
positions. While using a lower control force, this controller reduces the peak interstory displaccment and peak absolute accclcration by an additional 6.9% and 19%,
respectively, over the best passive case.
In this example, the decentralized bang-bang controller appears to be quite effective at reducing the maximum
acceleration (33% below the better passive case). However, this controller was not effective at reducing the relative
displacements or interstory displacements. The maximum
interstory displacement is slightly larger than that of the
best passive case and the peak force is quite large. Because
this control algorithm is attempts to minimize the total energy, it appears to be imitating a base isolation system, resulting in a large relative displacement at the first floor.
The results in Table I show that the performance of
the system employing a clipped-optimal controller surpasses that of both passive systems considered. The
clipped-optimal controlleir reduces the maximum third
floor relative displacement and interstory displacement by
an additional 31% and 28%, respectively, as compared to
the best passive response. In addition, this controller reduces the maximum floor acceleration more than the passive-on case, although not as well as in the passive-off
case.
The controlled responses using the modulated homogeneous friction control algorithm indicate that this control algorithm is also quite effective at reducing both the
relative displacements and accelerations of the structure.
The maximum third floor displacement and interstory displacement are reduces by an additional 20% and 23%, respectively, over the best passive case. Furthermore, this
controller achieves a 10% reduction in the maximum absolute acceleration as compared to the best passive case,
while using a significantly smaller peak force than all other control algorithms.

A = [ -M, 0,K,-M, 1,C,] , B = [ MI' A] , E = - /

r

The MR damper parameters used in this study are cO,

= 8 Nseckm, Cob = 6 Nsec/cmN, k , = 50 Nkm, c l a =
290 Nseckm, C l b = 5 NseclcmN, k , = 12 Nkm, xo = 0
cm, a, = 100, ab = 450 v', y = 363 cm-2, p = 363 cm', A = 301, n = 2 , q = 190 sec-'. These parameters were
identified based on the prototype MR damper tested at the
University of Notre Dame [9, lo].
In simulation, the model of the structure is subjected
to the NS component of the 1940 El Centro earthquake.
The simulations were performed in MATLAB [15]. Because the system under consideration is a scaled model,
the earthquake was reproduced at five times the recorded
rate. The maximum structural responses of each system
are presented in Table 1. Here, xi is the displacement of
the i th floor relative to the ground, d j is the interstory
drift (i.e., xi- xi- ), .fa, is the absolute acceleration of the
i th floor, and f is the applied control force.
As a basis for comparison, two cases are considered in
which the MR damper is employed in a passive mode. In
the first case, designated pussive-off, the command voltage
to the MR damper is held at 0 V. The second passive case
the voltage to the MR damper is held at the maximum
voltage level (2.25 V) and is denoted as passive-on. From
the results, both passive systems are able to achieve a reasonable level of performance. However, notice that the
passive-on system results in larger maximum accelerations
and interstory displacements than the passive-off system.
Apparently, choosing a passive device that produces the

5: Conclusion

In this paper a number of recently proposed, semi-active control strategies were evaluated for use with the MR
damper through a numerical simulation. Here, a model of
a three-story structure was controlled using a single MR
damper. The results demonstrated that the performance of
the resulting controlled syistem and the requirements of the
control device are highly dependent on the control algorithm employed. Each semi-active controller performed
noticeably better than the passive controllers in some way.

' x.

.I

Figure 2. Diagram of MR Damper Implementation.

583

tion to Dynamic Hazard Mitigation, Proc. 2nd Int. Wkshp.

on Struc. Control, Hong Kong, pp. 99-109, Dec. (1996).
[4] Dyke, S.J., Spencer Jr., B.F., Quast, P., and Sain, M.K. Role
of Control-Structure Interaction in Protective System Design. J. of Engrg. Mech., Vol. 121 NO. 2, pp. 322-38
(1995).
[5] Dyke, S.J., Spencer Jr., B.F., Quast, P., Sain, M.K., Kaspari
Jr., D.C. and Soong, T.T. Acceleration Feedback Control of
MDOF Structures, J. of Engrg. Mech., ASCE, Vol. 122, No.
9, pp. 907-918 (1996).
[6] Dyke, S.J., Spencer Jr., B.F., Quast, P., Kaspari Jr., D.C., and
Sain, M.K., Implementation of an AMD Using Acceleration
Feedback Control, Microcomputers in Civil Engrg., Vol.
11, pp. 305-323 (1996).
[7] Dyke, S.J., Spencer Jr., B.F., Sain, M.K. and Carlson, J.D.
Seismic Response Reduction Using Magnetorheological
Dampers. Proc. of the IFAC World Congress, San Francisco, CA, June 30-July 5 (1996).
[8] Dyke, S.J., Spencer Jr., B.F., Sain, M.K. and Carlson, J.D.
Modeling and Control of Magnetorheological Dampers for
Seismic Response Reduction, Smart Materials and Structures, Vol. 5, pp. 565-575 (1996).
[9] Dyke, S.J. AccelerationFeedback Control Strategies for Active and Semi-Active Systems: Modeling, Algorithm Development and Experimental Verification., Ph.D. Dissertation,
University of Notre Dame, Notre Dame, IN (1996).
[10]Dyke, S.J., Spencer Jr., B.F., Sain, M.K., and Carlson, J.D.,
Experimental Verification of Semi-Active Structural Control Strategies Using Acceleration Feedback, Proc. of the
3rd Intl. Con. on Motion and Ebz Control, Vol. 3, pp. 291296, Chiba, JAPAN, September (1996).
[l l]Dyke, S.J. and Spencer Jr. B.F., Seismic Response Control
Using Multiple MR Dampers, Proc. of the 2nd Intl. Workshop on Struc. Control, Hong Kong, pp. 163-173 (1996).
[12]Gavin, H.P., Hanson, R.D. and McClamroch, N.H. Control
of Structures Using Electrorheological Dampers, Proc. 1Ith
World Con$ on Earthquake Engrg., Mexico, (1996).
[ 13]Inaudi, J.A., Modulated Homogeneous Friction (MHF),
submitted (1997).
[14]Leitmann, G., Semiactive Control for Vibration Attenuation, J. of Intelligent Material Systems and Structures, Vol.
5 September, pp. 841-846 (1994).
[ISIMATLAB. The Math Works, Inc. Natick, Mass. (1994).
[16]McClamroch, N.H. and Gavin, H.P. Closed Loop Structural
Control Using Electrorheological Dampers, Proc. of the
Amel: Ctrl. Con$, Seattle, Washington, pp. 4173-77 (1995).
[17]Spencer Jr., B.F., Dyke, S.J., Sain, M.K. and Carlson, J.D.,
Idealized Model of a Magnetorheological Damper, Proc.
of the 12th Con$ on Analysis and Computation, ASCE, Chicago, Illinois, pp. 361-370 (1996).
[18]Spencer Jr., B.F., Dyke, S.J., Sain, M.K., and Carlson, J.D.
Phenomenological Model for Magnetorheological Dampers, J. Engrg. Mech., Vol. 123, No. 3, pp. 230-238 (1996).
[19]Spencer Jr., B.F., Dyke, S.J., and Deoskar, H.S. Benchmark
Problems in Structural Control-Part I: Active Mass Driver
System,Proc. of the ASCE Struc. Cong. XV, Oregon (1997).
[ZOISpencer Jr., B.F. Recent Trends in Vibration Control in the
U.S.A., Proc. of the 3rd lnt. Con$ on Motion and Vibl: Control, Chiba, Japan (1996).
[21]Spencer Jr. B.F., Carlson, J.D., Sain, M.K., and Yang, G. .
On the Current Status of Magnetorheological Dampers:
Seismic Protection of Full-scale Structures, Proc. of the
Amel: Control Con$, pp. 458-62 (1997).

Table 1: El Centro Earthauake Peak Responses.

Control
Strategy

(cm)

di

(cm)

Uncontrolled
Passive-Off
Passive-On
Decentralized
Bang-Bang
Control
Lyapunov
Control
ClippedOptimal
Control
Modulated
Homogeneous
Friction

0.211
0.357
0.455
0.080
0.196
0.306
0.168
0.281
0.330
0.117
0.243
0.309
0.114
0.185
0.219
0.114
0.203
0.245

I 1
0.211
0.153
0.103
0.080
0.158
0.110
0.168
0.114
0.069
0.117
0.147
0.084
0.114
0.090
0.101
0.114
0.121
0.093

I I

21
(cdsec )

1400
420

4;

f
(N)

258

1050
477
390
580
703
738
702
417

25

947

589

The largest reduction in the maximum acceleration was

achieved with the decentralized bang-bang approach. The
largest reduction in the third floor displacement and peak
interstory displacement was achieved with the clipped-optimal acceleration feedback algorithm. Furthermore, in
this example the modulated homogeneous friction approach performed quite well, achieving a significant reduction in both the interstory displacement and the
maximum absolute acceleration.
Note that none of the control methods discussed here
requires a model for the M R damper, although a model is
important for system analysis. Algorithms that explicitly
incorporate actuator dynamics and control-structure interaction into the design process may offer additional performance gains [4]. Efforts are currently underway to
investigate this possibility.

Acknowledgment
This research is partially supported by National Science Foundation Grant Nos. CMS 93-01584 & CMS 9528083.

References
[l] Brogan, W.L. Modem Control Theory, Prentice Hall, Englewood Cliffs, New Jersey (1991).
[2] Carlson, J.D. and Spencer Jr. B.F. Magneto-Rheological
Fluid Dampers for Semi-Active Seismic Control, Proc. of
the 3rd Int. Con$ on Motion and Vibl: Control, Chiba, Japan,
Vol. 3, pp. 35-40 (1996).
[3] Carlson, J.D. and Spencer Jr., B.F. Magneto-Rheological
Fluid Dampers: Scalability and Design Issues for Applica-

584