Professional Documents
Culture Documents
Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct
a r t i c l e
i n f o
Article history:
Received 19 April 2011
Revised 11 February 2014
Accepted 17 March 2014
Keywords:
Magneto-rheological damper
Numerical model
Magneto-rheological uid
Energy dissipation devices
Semi-active dampers
Seismic response
High-rise buildings
a b s t r a c t
This investigation deals with the design, manufacturing, and testing of a large-capacity MR damper
prototype. The MR damper uses external coils that magnetize the MR-uid as it moves out of the main
cylinder through an external cylindrical gap. In its design, multi-physics numerical simulations are used
to better understand its forcevelocity constitutive behavior, and its eventual use in conjunction with
tuned mass dampers for vibration reduction of high-rise buildings. Multi-physics nite element models
are used to investigate the coupled magnetic and uid-dynamic behavior of these dampers and thus
facilitate the proof-of-concept testing of several new designs. In these models, the magnetic eld and
the dynamic behavior of the uid are represented through the well-known Maxwell and NavierStokes
equations. Both elds are coupled through the viscosity of the magneto-rheological uid used, which in
turn depends on the magnetic eld strength. Some parameters of the numerical model are adjusted using
cyclic and hybrid testing results on a 15 ton MR damper with internal coils. Numerical and experimental
results for the 15 ton MR damper showed very good agreement, which supports the use of the proposed
cascade magnetic-uid model. The construction of the 97 ton MR damper involved several technical
challenges, such as the use of a bimetallic cylinder for the external coils to conne the magnetic eld
within a predened magnetic circuit. As it should be expected, test results of the manufactured MR
damper show that the damping force increases with the applied current intensity. However, a larger
discrepancy between the predicted and measured force in the large damper is observed, which is studied
and discussed further herein.
2014 Elsevier Ltd. All rights reserved.
1. Introduction
In recent years large scale magneto-rheological dampers have
reached more attention in the area of structural response control
[18]. An example of such an MR damper is a large scale bypass type
MR damper with 400 kN force capacity and 47.5 cm stroke [19],
or a large scale annular orice type MR damper with a force capacity of 300 kN and a stroke of 28 cm [20]. MR dampers presented in
the literature have been designed using relatively simple analytical
expressions [1,16,17]. Detailed design and testing aspects have also
been thoroughly described in [1], as well as a semi-active seismic
clipped optimal control algorithm based on a linear square regulator [15]. Different phenomenological models can be found in the
literature, e.g. [5], where a modied BoucWen model has been
successfully proposed, and other non-parametric and neural network models have also been studied [6]. A good example of the
latter are the ANFIS models used to represent the forcevelocity
and displacement device behavior of MR dampers [13]. However,
all these are black-box models which present parameters that need
to be estimated experimentally. Therefore, these models are not
capable of capturing relevant design and practical topics such as
the connement of the magnetic ux within the damper coil,
among other things.
A mathematical model was presented elsewhere [1] to represent the device forcedisplacement and forcevelocity behavior
of an MR damper that includes variability on the damper geometry,
viscosity of the uid without magnetization, and level of shear
stress of the magnetized MR uid. In Ref. [8] mathematical expressions are provided, which consider the friction force effects
between the MR uid and the walls of the device, which also modify the behavior of the damper. Another important reference to this
work is a mathematical model of a double-tube MR damper using
the well-known annular ow solution and uid compressibility [9].
gc_ ; H
so H
g1 ; s > so H
c_
8 so H
< c_ g1;
: ag1;
c_ P c_ o
c_ < c_ o
where a is a non-dimensional impedance parameter in the preyield region of the uid, and c_ o corresponds to the critical shear
strain rate given by c_ o so H=a 1g1 . This alternative is compared schematically with the Bingham model in Fig. 1. The higher
the value of a, the better the consistency between the two models.
The MR uid used in this research is the Lord MRF-132 DG and
has the nominal properties shown in Table 1. The magnetic
properties of the uid are given by the manufacturer and they
are characterized by a magnetization (BH) curve, and a yield
stressmagnetic eld strength relationship [11]. A comparison between these non-linear BH curves for the MR uid and steel SAE
1045 [12] used to manufacture the damper is plotted in a logarithmic axis in Fig. 2(a). It is apparent that magnetization of steel is larger, faster, and occurs for smaller values of H, reaching a saturated
state for lower H than the uid. Furthermore, the relationship between the yield stress in the uid and H is shown in Fig. 2(b),
where the yield stress gets essentially saturated at a value of
2. Problem formulation
In this research an MR damper with a cylindrical external orice
and external coils is designed, manufactured, and tested. The
design of an MR damper is an iterative process and different topics
such as magnetic ux connement, dynamic range optimization
for high capacity, and damper force rise time reduction have to
be considered to achieve an optimized design. MPFEM are used
in this research as an alternative to simplied design models in order to optimize the design of the damper, thus saving costs and
195
196
Table 1
Properties of Lord MRF 132DG uid.
Properties
Fluid density
Dynamic viscosity @ 40 C,
Solid content by weight
Operating temperature
Field dependent yield stress
3 [gr/cm3]
0.092 0.015 [Pa s]
80.98%
40 to +130 [C]
Fig. 2(b)
Fig. 2. Magnetic properties: (a) Magnetization curves for SAE 1045 steel and Lord MRF-132DG uid; and (b) MRF-132DG yield stress [kPa] versus magnetic eld strength [kA/
m].
197
Table 2
Field equations.
Fig. 4(b) and (c), respectively. The mesh consists of 801,590 elements with a total of 1,590,910 nodes. The MSM model includes
an external surrounding air chamber that has been intentionally
hidden in the gure. Finite elements are solid elements, specically
developed to simulate 3D magnetic elds [21]. Boundary conditions are given by zero voltage applied at the external surfaces of
the air chamber, and by the electric current intensity that runs inside each coil in opposite directions, as indicated in Fig. 4(c). The
arrows in Fig. 4(c) represent the direction of the current ow on
each coil. Each coil has 250 turns of AWG 16 copper wire and the
electric current varies at the beginning of a simulation from zero
to a steady state current of 3 [A], in ten equal intervals of 0.3 [A]
and h = 0.01 [s].
Fig. 4. Magnetostatic model and boundary conditions: (a) Geometry, (b) FEM mesh, and (c) direction of electric currents in each coil.
198
Fig. 6. Magnetic ux density around the third coil. (a) Radial ux on the external cylinder. (b) Axial ux on steel casing and internal cylinder.
magnetic ux that crosses both cylinders. The zones located between coils (Fig. 5) concentrate the larger H values because the
section where the magnetic ux crosses the uid is smaller. In
some places between coils, magnetic eld strength is above
300 kA/m], reaching saturation according to Fig. 2(b), which is
the target for the design of the damper. Symmetry of the results
is essentially explained by the axial symmetry of the problem. As
explained before, the results in the uid gap are taken and used
as an input for the dynamic model of the uid, taking into account
the direct relationship between H and the yield stress of the uid
already shown in Fig. 2(b).
Each one of the 4 coils has a measured resistance of 4 [X].
Therefore the voltage across one coil with a current intensity of 3
[A] is 12 [V], and the power consumption of a single coil is 36
[W]. Moreover, the total power consumption of the 4 coils is 144
[W]. In the damper test, the 4 coils are connected in parallel, so
the equivalent resistance of the 4 coils is 1 [X] and the total current
is 4 3 [A] = 12 [A]. Based on the model, L/R = 0.5 [s] (R = 4.19 [X],
L = 2.11 [Henry]) and V/L = di/dt = 6.16 [A/s]. The values provided
for L are nominal since the inductance was not measured
experimentally.
Fig. 8. Fluid domain: (a) Full damper geometry, (b) reduced geometry due to axial symmetry (used model), and (c) FEM mesh of the CFX model.
Fig. 9. Imposed boundary conditions for the ow: (a) Inlet velocity and opening pressure proles, and (b) uid-to-uid interface.
needs to be included in the model; such is the case since the damper has 12 orices in each side to allow the uid to get into the gap
(Fig. 8(a)). The full and the reduced domains can be compared in
Fig. 8(a) and (b), respectively. They are composed by a single uid
domain, including both chambers at either side of the piston, the
gap between the internal and external cylinder, and the 2 12
holes in the internal cylinder that connect the external gap passage
to the chambers. The number of orices in the internal cylinder
was selected by viscous force simulations of the MR damper, varying the number of orices while optimizing the viscous force. The
model was meshed using the ICEM CFD application, and consists of
hexahedral solid elements; the total number of elements is
520,456 and the total number of nodes is 553,434. The gap is
meshed separated from the rest of the domain, since it needs several layers of regular elements in order to obtain an accurate velocity prole for the uid (Fig. 8(c)).
The boundary conditions applied to the CFX model are schematically shown in Fig. 9, and correspond to: (i) no-slipping of the uid
at the cylinders and piston walls; (ii) an inlet at the initial end and
an opening at the nal end of the domain with a predened uid
velocity prole and relative pressure, respectively (Fig. 9 (a)); (iii)
rotational periodicity at the uiduid interface for the lateral surfaces of the domain (Fig. 9(b)); and (iv) the yield-stress eld obtained from the magneto-static model that controls the uid
viscosity.
An imposed ow at the ends of the model has been nally selected given its simplicity and the negligible differences relative
to a more sophisticated model with a moving mesh. Consequently,
at the inlet the specied velocity prole in the X-direction is
vow,x = v0 sin(xt), where v0 corresponds to the peak velocity, x is
the circular frequency [rad/s], and t is time. Laminar ow is considered in the model, and a transient analysis is performed with a
peak velocity of vmax = 0.17 [m/s], a frequency x = 10 [rad/s], and
a time-step h = 5 104 [s]. The selection of the time-step is the result of a large number of trials and errors. This process showed that
the time step required to ensure convergence is about h = 1 106
[s]. However, computational costs associated with this time step
are excessive, and hence, an integration time step h = 5 104 [s]
was selected as a reasonable compromise to obtain a better
trade-off between accuracy and computational costs especially
for parametric analyses. Results are ltered in order to reduce
eventual noise due to coarseness of the time interval. To control
the accuracy of the simulation, RMS values are calculated for the
three momentum conservation equations and continuity equation.
RMS values of the numerical equilibrium errors, or residuals, at
each time-step of the integration of the governing equations are
199
an overall measure of how well the numerical solution has converged. For adequate accuracy, the converged solution requires a
maximum residual level of 1.0e4. These RMS residuals of the simulation are usually an order of magnitude lower than this tolerance. In case that the RMS values do not achieve this limit,
simulation continues to the next time-step after 10 iterations.
Resulting damper force is computed by integrating the total
pressure applied to the two faces of the damper piston. Thus, the
device forcevelocity (fv) and forcedisplacement (fd) behavior
can be obtained as shown in Fig. 10(a) and (b), respectively. Please
note that the compressibility of the uid was not included in the
model. By using this model and assumptions, the maximum damper force obtained for the given parameter values was about 97.1
[ton]. The numerical noise in the gures is produced by the time
step selected, which despite being small (h = 5 104 [s]), is coarser
than the one required to assure perfect convergence of the CFX
simulation. The model data was ltered by a smoothing ve-point
moving average (MA) scheme. Even with this MA scheme, the
numerical noise could not be eliminated completely, although differences in the maximum force value are negligible relative to the
output given by the ideal time-step h = 1 106 [s].
5. Model validation
The model developed was rst validated using as a benchmark
the experimental results of an existent MR damper [2,3]. The complete geometry of this proof-of-case damper can be seen in
Fig. 11(a) and consists of: (1) a piston; (2) a cylinder; (3) two coils
wound around the piston; (4) two steel head plates; and (5) the MR
uid. The design considers a 1.5 [mm] gap between the piston and
the cylinder, and a measured capacity of 15 [ton]. The procedure is
the same as explained in previous sections, but for this model a
reduction of the uid domain was not necessary, since the computational cost of this problem was manageable. The mesh of the
magneto-static model consisted of 61,503 elements and 94,289
nodes, while the mesh of the CFX model has 272,988 elements
and 77,485 nodes (Fig. 11(b) and (c)). A picture of the damper is
shown in Fig. 11(d); further details of this solution may be found
elsewhere [2,3].
The rst set of experiments consisted of harmonic motions applied to the piston at different constant current intensities i = 0, 0.5,
1, 1.5, 2 and 2.5 [A]. Testing included different displacement amplitudes and cycling frequencies, and the tests used to validate the
FEM model corresponds to a maximum displacement of the
damper of 5 [cm], a period of the cycles of 2.7 [s], and a current
Fig. 10. Capacity of the MR damper estimated numerically using the CFX model: (a) device forcevelocity, and (b) device forcedisplacement.
200
Fig. 11. MR damper used in the validation of the numerical model: (a) damper geometry; (b) magneto-static mesh; (c) CFX FEM model mesh; and (d) picture of the 15 [ton]
MR damper.
intensity of 2 [A], which correspond to typical building design values. The force measured with these parameters was 13.8 [ton],
201
Fig. 13. Results from parametric analyses: (a) device controllable force Fs [ton] versus spacing between coils b [mm]; (b) device controllable force versus electric current
intensity for b = 50 [mm]; (c) device forcevelocity curves for different values of peak velocity, v0; and (d) dynamic range versus gap size for v0 = 25 [cm/s].
6. Parametric analyses
One of the main advantages of developing a numerical model of
the MR damper is to try out many analyses with different damper
parameters before manufacturing. Parametric analyses were performed on the model of the 97 [ton] MR damper by varying the following independent design parameters: (i) axial spacing between
coils b; (ii) current intensity i; (iii) maximum piston velocity v0;
and (iv) gap size h. By varying axial spacing of the coils, an optimal
distribution of the magnetic eld strength (H) can be obtained, i.e.,
an optimum volume of saturated uid. Current intensity helps us
understand the importance of the power source used, while velocity conditions are related to different ground motion conditions
and available damper hardware. Moreover, since the gap size has
a strong inuence on the dynamic range of the damper, this effect
needs to be evaluated in the damper response. For this case, Yangs
formula [1] for the MR damper was used since a change in the gap
size of the model would imply a change in the FEM mesh, which
would imply a signicant increase in computational costs.
Although we found that this formula may differ by about 30% relative to CFX results in external coil congurations, it shows the
same trends in the results as obtained from changes in the gap size
of the model, and hence, it can be used to predict the optimum gap
size to maximize the dynamic range of the MR damper.
The variation of the axial spacing between coils, b, was generated by keeping all other independent geometric parameters constant, except the length of the outer steel casing that needs to be
elongated in order to cover the coils and generate the magnetic
ux circuit. The values of b that have been considered range from
40 to 200 [mm] (Fig. 13(a)). The second analysis keeps b = 50 [mm],
and changes the current intensity in the coils from 0 to 3 [A]
(Fig. 13(b)). The damper force has been obtained for every current
intensity value using the CFX model. Third, the considered variation in the maximum velocity ranges from 17 to 46 [cm/s] in intervals of 3 [cm/s], corresponding to the expected range of velocities
in the building where the damper is to be installed. Finally, a sensitivity analysis of the gap size h was done with values ranging in
the interval between 0.9 and 5.5 [mm] and a maximum velocity of
25 [cm/s]. The gap size has an inuence mainly in the viscous force,
and controls the dynamic range D. The viscous component of the
damper force, Fg was also obtained using the CFX model by representing the uid properties without the magnetic eld (Table 1).
The controllable force of the damper is computed by subtracting
Fg from the total damper force. The dynamic range D (Fig. 13(d))
is dened as the ratio of the controllable force Fs and the passive
viscous and frictional damper force, Fg + Ff [1]. The friction force
Ff is constant and can be measured by testing the device; during
the design phase, this friction force needs to be blindly estimated.
It is apparent from part (a) of Fig. 13 that the value of the
controllable force increases as the spacing b between coils also increases. This is so because the uid becomes magnetically saturated in every model analyzed and, hence, the magnetic ux
density is kept constant for different values of b. Therefore, the
202
Fig. 14. Schematic geometry of the additional models used to compare with the proposed model: (a) three coils; (b) ve coils; (c) diamagnetic rings on the external cylinder
only; and (d) diamagnetic rings in the external and internal cylinder.
Table 3
Summary of the results of comparative analyses.
Model
Original model
(a) 3-Coils
(b) 5-Coils
sav
23.43 [kPa]
97.08 [ton]
17.75 [kPa]
91.98 [ton]
5.25%
34.41 [kPa]
120.37 [ton]
23.99%
26.56 [kPa]
106.52 [ton]
9.72%
27.74 [kPa]
110.92 [ton]
14.26%
Fmax (CFX)
Variation relative to base model
Table 4
Chosen parameters for the design of MRDA-UC.
Design parameter
Dimension
Design parameter
Dimension
Design parameter
Dimension
Stroke
Piston radius
Coils width
No. of holes in internal cylinder
Steel casing thickness
200 [mm]
130 [mm]
60 [mm]
2 12
25 [mm]
Gap size, h
Rod radius
Spacing between coils, b
Internal cylinder holes diameter
Ext. cylinder thickness (steel)
1.5 [mm]
35 [mm]
50 [mm]
20 [mm]
41 [mm]
Cylinder length
No. of coils
Int. cylinder thickness
Head plates thickness
Ext. cylinder thickness (diamagnetic)
640 [mm]
4
30 [mm]
40 [mm]
24 [mm]
(a)
(b)
(c)
(d)
Amplitude [mm]
Frequency [Hz]
Current intensities [A]
10
0.48
03
60
0.212
03
100
0.21
03
140
0.19
02.5
203
Fig. 16. Measured device fv and fd behavior for different amplitudes, frequencies, and current intensities i = 0, 0.6, 1.3, 2.0, 2.5 and 3.0 [A].
204
Fig. 17. Results of Test 21: (a) measured pressure of both chambers and temperature measured in one chamber (lower curve); and (b) total and estimated damper force by
measuring the pressure difference between chambers.
the gure with the force measured by the load cell F cell . This is a
clear proof that the applied external force on the damper can be
computed as the resultant of the pressure differences.
8. Conclusions
This paper deals with the design, manufacturing, and experimental testing of a nominal 97 [ton] large-capacity MR damper.
As a design tool a multi-physics FEM model is used for numerical
simulation of the response of the MR damper, which solves the
magneto-static and uid dynamic problems as cascade systems.
It has been shown in this article that the numerical boundary conditions imposed to the problem lead to a well posed numerical
problem. The numerical solution of the magneto-static and uid
problems is coupled through the uid viscosity, which depends
on the magnetic eld strength.
It is concluded from the comparison between numerical and
experimental results for a 15 [ton] damper that the capacity of
an MR damper with coils wound around the piston can be correctly
predicted by the use of a FEM numerical model like the one presented herein. The model was validated by using dynamic tests
on a 15 [ton] MR damper that was later implemented in a 21-story
building in Santiago, Chile. It is also true that the numerical simulation proposed that uses the Bingham model is not capable of representing the real hysteretic behavior observed in the MR damper
due to assumption of incompressibility of the uid. Even with that
assumption, the maximum damper capacity and post-yield capacity differ in less than 0.5%.
The numerical model was basically developed to optimize the
design of a prototype of a real-scale 97 [ton] MR damper with
external coils, which was manufactured and tested. The scaling
of the device from 15 [ton] to 97 [ton] presented several design
and practical construction challenges that have being mostly resolved but still require ne-tuning. A detailed explanation of these
aspects goes beyond the scope of this article and will be presented
in a sequel publication. In general terms, the critical construction
aspects can be traced to the proper connement of the magnetic
eld, and the increase in pressure within the uid. The former
was achieved by the use of a bimetallic cylinder. For the latter,
since pressures inside the damper are signicant and rise as the
temperature in the uid increases, a careful design of the joint
between the two materials for the cylinder was considered. A complete set of pressure tests were performed to prove that this
bi-metallic joint works properly.
Numerical results have shown that the inclusion of diamagnetic
components can actually improve the efciency of an MR damper.
Also, as the number of coils increases in the damper, the capacity
increases if the cylinder length is kept constant. Furthermore, it
is derived from parametric analyses that the spacing between coils
is a critical design parameter due to its inuence in the active
length and the magnetic ux saturation within the uid domain.
These analyses also proved that the uid saturates for the design
of the 97 [ton] MR damper with four coils after a current intensity
of 5 [A]; the dynamic range for the proposed conguration is smaller than the optimally predicted one due to the fact that the target
passive force for the damper was larger, and it required a smaller
dynamic range.
Initial test results on the large scale damper show that the concept of the external coil and other damper design aspects work
well in practice, but that there are still several manufacturing
and model details that need to be improved in order to get a better
agreement between the FEM model simulations and the proof-ofconcept experimental results in this case. Several physical and
205