See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/265538185

Use of genetic algorithms for fitting the Bouc-

Wen model to experimental hysteretic curves

Article · January 2009

CITATIONS READS

8 233

3 authors:

Marius Giuclea Tudor Sireteanu

Bucharest Univesity of Economic Studies Institutul de Mecanica Solidelor
38 PUBLICATIONS 248 CITATIONS 142 PUBLICATIONS 589 CITATIONS

SEE PROFILE SEE PROFILE

AM Mitu
Institute of Solid Mechanics Bucharest Roma…
37 PUBLICATIONS 86 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Novel linearization method for piecewiselinear systems with apllication to semi-active control with
variable dry friction View project

All content following this page was uploaded by Tudor Sireteanu on 08 January 2015.

The user has requested enhancement of the downloaded file.

 USE OF GENETIC ALGORITHMS FOR FITTING THE BOUC-
WEN MODEL TO EXPERIMENTAL HYSTERETIC CURVES

M. GIUCLEA, T. SIRETEANU, A.M. MITU*

In this paper a Genetic Algorithms (GA) method is developed to identify the Bouc-
Wen model parameters from the experimental data of periodic loading tests. It is
considered an extension of classical model in order to increase its capacity to
approximate experimental loops. From Bouc-Wen equation are derived integral
conditions that characterize the coordinates of hysteresis curve. The objective function
of GA is defined as a sum of relative errors obtained for a set of indices computed on
the predicted and experimental loops.

1. INTRODUCTION

The Bouc-Wen model, widely used in structural and mechanical engineering,

gives an analytical description of a smooth hysteretic behavior. It was introduced
by Bouc [1] and extended by Wen [2], who demonstrated its versatility by
producing a variety of hysteretic characteristics. The hysteretic behavior of
materials, structural elements or vibration isolators is treated in a unified manner
by a single nonlinear differential equation with no need to distinguish different
phases of the applied loading pattern. In practice, the Bouc-Wen model is mostly
used within the following inverse problem approach: given a set of experimental
input–output data, how to adjust the Bouc-Wen model parameters so that the
output of the model matches the experimental data. Once an identification method
has been applied to tune the Bouc-Wen model parameters, the resulting model is
considered as a “good” approximation of the true hysteresis when the error
between the experimental data and the output of the model is small enough from
practical point of view. Usually, the experimental data are obtained by imposing
cyclic relative motions between the mounting ends on the testing rig of a sample
material, structural element or vibration isolator and by recording the evolution of
the developed force versus the imposed displacement. Once the hysteresis model
was identified for a specific input, it should be validated for different types of
inputs that can be applied on the testing rig, such as to simulate as close as possible
the expected real inputs. Then this model can be used to study the dynamic

Institute of Solid Mechanics – Romanian Academy, Const. Mille, 15, RO-70701

Rev. Roum. Sci. Techn. – Méc. Appl., Tome 54, N° 1, P. 3–10, Bucarest, 2009
4 M. Giuclea, T. Sireteanu, A.M. Mitu 2

behavior of different systems containing the tested structural elements or devices

under different excitations.
Various methods where developed to identify the model parameters from the
experimental data of periodic vibration tests. A frequency domain method was
employed to model the hysteretic behavior of wire-cable isolators [3], iterative
procedures were proposed for the parametric identification of a smoothed
hysteretic model with slip [4], of a modified Bouc-Wen model to portray the
dynamic behavior of magnetorhological dampers [5], of a generalized model for
highly asymmetric hysteresis obtained in laboratory experiments for flexible
connectors [6], etc. The Genetic Algorithms were widely used for curve fitting the
Bouc-Wen model to experimentally obtained hysteresis loops for composite
materials [7], nonlinear degrading structures [8] or magnetorheological fluid
dampers [9-11].
In the present work, our primary focus is to use analytical relationships
derived from the Bouc-Wen differential equation to determine the parameters of
the model such as the predicted hysteresis curves and the experimental loops to
have same absolute values of the maximum force, same coordinates of the loop-
axes crossing points, and same slopes of the tangent at these key points of
hysteretic loops. As the measured data have a certain degree of imprecision, one of
the most suitable methods for approximating the model parameter is the genetic
algorithms (GA) approach.

2. MATHEMATICAL MODEL

In this paper is assumed that the experimental hysteretic characteristic is a

symmetric loop − Fm ≤ F ( x ) ≤ Fm , obtained for a periodic motion
− xm ( t ) ≤ x ( t ) ≤ xm ( t ) , imposed between the mounting ends of the tested element.
In the xOz system of coordinates, the loop-axes crossing points are denoted by
( 0, F0 ) , ( x0 , 0 ) , ( 0, − F0 ) and ( − x0 , 0 ) . Next, the following dimensionless
variables are considered

τ = t T , ξ ( τ ) = x ( τT ) xu , ξ′ ( τ ) = d ξ d τ, z ( ξ ) = F ( xu ξ ) Fu ,
(1)
ξm = max ξ ( τ ) , zm = max z ( ξ ) , ξ0 = x0 xu , z0 = F0 Fu ,

where T is the period of the imposed cyclic motion and xu , Fu are displacement
and force reference units such as ξ m ≤ 1, zm ≤ 1 . Then, a generic plot of the
symmetric hysteresis loop z ( ξ ) can be represented as shown in Fig. 1.
3 Use of genetic algorithms for fitting Bouc-Wen model to experimental hysteretic curves 5

z
α1
B
zm
α2
z0 α

α0 C ξ
-ξm -ξ0 F 0 ξ0 ξm

D
-z0
-zm
E

Fig. 1 – Generic experimental hysteretic loop.

With the previous notations, the Bouc-Wen model is described by the

following non-linear differential equation
dz
= dξ , (2)
A − z β + γsgn ( ξ ' z ) 
n

where A, β, γ, n are loop parameters controlling the shape and magnitude of the
hysteresis loop z ( ξ ) . In the classical models the parameter n is assumed a natural
number but, this constraint doesn’t allow good approximation for some
experimental loops [12]. Therefore, in this paper the exponent in Bouc-Wen model
is not necessarily a natural number, being considered as a positive real number p
to be determined within the fitting procedure of experimental data.
In order to characterize the hysteresis loop there are defined some integral
and derivative conditions. These relations will be used for definition of genetic
algorithm objective function. By taking the integral on the branches AB , BC and
CD , one can write the following relations:
zm ξm
dz
∫ A − z p (β + γ )
= ∫ dξ = ξ m , (3)
z0 0
6 M. Giuclea, T. Sireteanu, A.M. Mitu 4

zm ξm
dz
∫ A − z p (β − γ )
= ∫ dξ = ξ m – ξ 0 , (4)
0 ξ0

0 ξ0
dz
∫ p
A − z (β + γ )
= ∫ dξ = ξ 0 . (5)
− z0 0

Next, there are denoted by α and α 0 the tangent slopes at the points
( −ξ0 , 0 ) and (0, z0), respectively. Then, using (2), one can derive the equations:

dz p
(0) = α ⇒ α = A − z (β + γ ) (6)
dξ ,

dz
(ξ0 ) = α 0 ⇒ α 0 = A . (7)
dξ
As z0 > 0 , then from (6) and (7) is obtained the relation:

α − α 0 + z0 p ( β + γ ) = 0 . (8)
In the same way, if α1 and α 2 are the slopes of left hand and right hand tangents
at the point ( ξm , zm ) then:

dz
= α1 ⇒ α1 = A − zm p (β + γ ) , (9)
dξ ξ=ξm
ξ '> 0

dz
= α 2 ⇒ α 2 = A − zm p ( β − γ ) . (10)
dξ ξ=ξm
ξ '< 0

For many experimental curves, the coefficients α1 and α 2 are difficult to be

accurately estimated, so they are used only when the sample time interval of
recorded force and displacement signals is sufficiently small.

3. GA OPTIMIZATION

The main goal of this paper is to find the values of parameters A, β, γ , p

such that the obtained Bouc-Wen loop represents a good approximation of the
experimental one. As the measured data have a certain degree of imprecision, one
of the most suitable methods for approximating the coefficients A, β, γ , p is the
5 Use of genetic algorithms for fitting Bouc-Wen model to experimental hysteretic curves 7

genetic algorithms approach. Here is proposed to be used a real coded GA with

four genes that are the codifications of parameters A, β, γ, p . The objective
function is defined as a sum of five indexes, three of them having integral
formulation and the other two obtained from (7) and (8). Hence, the partial
performance indices are defined by
zm zm
1 dz 1 dz
I1 =
ξm ∫ A − z (β + γ )
p
− ξm , I 2 =
ξ m − ξ0 ∫ A − z p (β − γ) − ( ξm − ξ0 ) ,
z0 0
(11)
0
1 dz
I3 =
ξ0 ∫ p
A − z (β + γ )
− ξ0 ,
− z0

I 4 = α 0 − A , I 5 = α − α 0 + z0 p ( β + γ ) . (12)

Therefore, the global performance index is given by

I = w1I1 + w2 I 2 + w3 I 3 + w4 I 4 + w5 I 5 , (13)

where wi , i = 1, 5 are weights that can be set such as the search is oriented to
minimize the most significant indices.
The other GA parameters used in numerical simulations are:
• average crossover with probability 0.7;
1
• uniform mutation with probability , where N is the number of genes;
N
• Monte Carlo selection.

4. APPLICATION OF GA FOR FITTING THE BOUC WEN MODEL TO EXPERIMENTAL

HYSTERETIC LOOPS OF SEISMIC DEVICES

Base isolation and dissipative bracing of buildings are modern and efficient
seismic protection strategies already implemented in many countries. While base
isolation is a more appealing solution in the case of new buildings, the dissipative
braces are use especially in the seismic retrofitting of the existing ones.
The force-displacement characteristic of most seismic protection devices is of
hysteretic type. Usually, the experimental hysteretic loops are obtained by
imposing cyclic relative motions between the device mounting ends on the testing
rig and by recording the evolution of the developed force versus the imposed
displacement. By fitting a Bouc-Wen model type to experimental data, one obtains
a single non-linear first order equation which can describe the evolution of force
developed by one device for almost any loading pattern (periodic, aperiodic or
random). All these equations are then added to the system of equations which
8 M. Giuclea, T. Sireteanu, A.M. Mitu 6

models the motion of the protected building. Thus, is obtained an enlarged system,
which can portray the dynamic behavior of the protected structure with a better
accuracy than it can be achieved by employing other methods (equivalent
linearization, phase description of hysteretic loops by piece-wise continuous
functions, etc).
In this paper, the proposed method for fitting the Bouc-Wen model to
experimental hysteretic loops is illustrated for two seismic devices, manufactured
by the Italian Company FIPP INDUSTRIALE:
– Lead Rubber Bearing (LRB), used for seismic protection by base isolation;
– Buckling-Restrained Axial Damper (BRAD), used for seismic protection by
inter-storey dissipative bracing.
The normalized force-displacement curves, obtained from those reported by
the manufacturer in the product technical notes [13], [14], and hysteretic loops
predicted by the developed fitting method, are shown comparatively in Figs. 2
and 3.
It should be mentioned that the force- displacement curve of LRB device was
obtained by a shear test conducting on a column of two devices, under a static
vertical load 1,500 KN and a cyclic horizontal load applied at the middle. The
reference values chosen from experimental curves in this case were xu = 50 mm
and Fu = 150 kN. After about 5.000 generations, the following values of model
parameters were obtained: A = 0.504, β = –13.165, γ = 13.05, p = 0.74.

1.0

Predicted hysteretic loop

Normalized experimental loop
0.5
Normalized force z

0.0

-0.5

-1.0
-1.0 -0.5 0.0 0.5 1.0
Normalized displacement ξ

Fig. 2 – The experimental and predicted hysteretic loops for LRB devices.
7 Use of genetic algorithms for fitting Bouc-Wen model to experimental hysteretic curves 9

1.5
Predicted hysteretic loops
Normalized experimental loops
1.0
Normalized force z

0.5

0.0

-0.5

-1.0

-1.5
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8
Normalized displacement ξ

Fig. 3 – The experimental and predicted hysteretic loops for BRAD device.

For the BRAD experimental loops the reference units are xu = 20 mm and
Fu = 200 kN and Fig. 3 depicts the predicted and experimental loops for three
different amplitudes of the imposed displacement, for the following set of model
parameters A = 4.1, β = –7.72, γ = 12.07, p = 1.054. The fitting method was
applied for experimental data corresponding to the maximum relative displacement
allowed for this device.

5. CONCLUSIONS

The proposed method for fitting Bouc-Wen model to experimental hysteretic

loops is based on analytical relationships. These relations are expressed in terms of
integral and differential conditions which describe the essential properties of
hysteretic loop. The objective function defined as a function of the analytical
performance indices is suitable to be efficiently used within a genetic algorithms
procedure. The obtained results show the extension of the classical Bouc-Wen
model by taking the parameter n as a positive real number could be very useful in
increasing the approximation accuracy of experimental loops.

Acknowledgment. The authors wish to express their gratitude to the Romanian Academy for
supporting this work through the Grant no. 326/2007-2008 “Theoretical and Experimental Analysis of
the Dynamic Behavior of Degrading Structures”.

Received on April 22, 2009.

 10 M. Giuclea, T. Sireteanu, A.M. Mitu 8

REFERENCES

1. BOUC, R., Forced vibration of mechanical systems with hysteresis, Proceedings of the Fourth
Conference on Non-linear oscillation, Prague, Czechoslovakia, 1967.
2. WEN, Y.K., Method for random vibration of hysteretic systems, Journal of the Engineering
Mechanics Division, 102, 2, pp. 249–263, 1976.
3. NI, Y.Q., KO, J.M., WONG, C.W., Identification of non-linear hysteretic isolators from periodic
vibration tests, Journal of Sound and Vibration, 217, 4, pp. 737–756, 1998.
4. LI, S.J., YU, H., SUZUKI, Y., Identification of non-linear hysteretic systems with slip, Computers
and Structures, 82, pp. 157–165, 2004.
5. SPENCER, B.F., DYKE, S.J., SAIN, M.K., CARLSON, J.D., Phenomenological Model for
Magnetorheological Dampers, J. Eng. Mech., ASCE, 123, 3, 1997.
6. HORNIG, K. H., Parameter characterization of the Bouc-Wen mechanical hysteresis model for
sandwich composite materials by using Real Coded Genetic Algorithms, Auburn University,
Mechanical Engineering Department, Auburn, AL 36849.
7. SONG J., ARMEN Der KIUREGHIAN, Generalized Bouc-Wen Model for Highly asymmetric
Hysteresis, Journal of Engineering Mechanics, 132, 6, pp. 610–618, 2006.
8. AJAVAKOM, N., NG, C.H., MA, F., Performance of nonlinear degrading structures:
Identification, validation, and prediction, Computers and Structures, 86, pp. 652–662, 2008.
9. GIUCLEA, M., SIRETEANU, T., STANCIOIU, D., STAMMERS, C.W., Modelling of
magnetorheological damper dynamic behavior by genetic algorithms based inverse method,
Proc. Ro. Acad., Series A, 5, 1, pp. 55–63, 2004.
10. GIUCLEA, M., SIRETEANU, T., STANCIOIU, D., STAMMERS, C.W., Model parameter
identification for vehicle vibration control with magnetorheological dampers using
computational intelligence methods, Proc. Instn. Mech. Engers., 218, Part I: J. Systems and
Control Engineering, 17, pp. 569–581, 2004.
11. KWOK, N.M., HA, Q.P., NGUYEN, M.T., LI, J., SAMALI, B., Bouc–Wen model parameter
identification for a MR fluid damper using computationally efficient GA, ISA Transactions,
46, pp. 167–179, 2007.
12. SIRETEANU, T., GIUCLEA, M., MITU, A.M., An analytical approach for approximation of
experimental hysteretic loops by Bouc-Wen model, Proceedings of Romanian Academy,
Series A, 10, 1, 2009.
13. FIPP INDUSTRIALE, Isolation Devices- LRB series, Technical Sheet.
14. FIPP INDUSTRIALE, Buckling-Restrained Axial Dampers-BRAD-series, Technical Sheet.

View publication stats