Anda di halaman 1dari 13

Home

Search

Collections

Journals

About

Contact us

My IOPscience

BoucWen model-based real-time force tracking scheme for MR dampers

This article has been downloaded from IOPscience. Please scroll down to see the full text article. 2013 Smart Mater. Struct. 22 045012 (http://iopscience.iop.org/0964-1726/22/4/045012) View the table of contents for this issue, or go to the journal homepage for more

Download details: IP Address: 146.201.208.22 The article was downloaded on 29/07/2013 at 21:51

Please note that terms and conditions apply.

IOP PUBLISHING Smart Mater. Struct. 22 (2013) 045012 (12pp)

SMART MATERIALS AND STRUCTURES

doi:10.1088/0964-1726/22/4/045012

BoucWen model-based real-time force tracking scheme for MR dampers


F Weber
Empa, Swiss Federal Laboratories for Materials Science and Technology, Structural Engineering Research Laboratory, Uberlandstrasse 129, CH-8600 D ubendorf, Switzerland E-mail: felix.weber@empa.ch

Received 14 December 2012, in nal form 13 February 2013 Published 11 March 2013 Online at stacks.iop.org/SMS/22/045012 Abstract A BoucWen model-based control scheme is presented which allows tracking the desired control force in real-time with magnetorheological (MR) dampers without feedback from a force sensor. The control scheme estimates the MR damper force by parallel computing of several BoucWen models with different constant currents as inputs and for the actual MR damper displacement and velocity, respectively. Based on the estimated forces and the desired control force the MR damper current is determined by a piecewise linear interpolation scheme. The model-based feed-forward control scheme is numerically and experimentally validated. If the desired control force is not constrained by the pre-yield region, residual force at 0 A and force at maximum current, the very small force tracking error 0.0015 in the simulation is caused by the control-oriented simplication of the linear interpolation scheme. The tests reveal that the real-time control scheme is numerically stable and the force tracking error of 0.078 represents an acceptable accuracy. (Some gures may appear in colour only in the online journal)

1. Introduction
In recent years, controlled damping of civil structures with magnetorheological (MR) dampers has been investigated by many researchers and implemented on real civil structures. A selection of such control systems can be found in [111]. The main reasons for the use of MR dampers are that their force range is suitable for civil structures; they are reliable, low-power consuming, seen to be fail-safe and can combine the benets of active and passive control systems [12, 13]. The structural damping that can be expected from semi-actively controlled MR dampers depends on many aspects of the control scheme, of which some of the relevant items are the control law, number and location of MR dampers, number and type of sensors, accuracy of observer and control force tracking accuracy. The last item is determined by the control force tracking method that controls the MR damper current in real-time for minimum difference between the actual MR damper force and the desired semi-active control force. Existing control force tracking schemes for the real-time control of the MR damper force may be divided into the following classes.
0964-1726/13/045012+12$33.00 1

(1) Model-based feed-forward control force tracking schemes. (2) Nonlinear control force tracking schemes based on force feedback. (3) Combined feed-forward feedback control force tracking schemes. The approaches of the rst class estimate the MR damper current in real-time based on an MR damper model and the desired control force and thereby avoid the use of a force sensor. The Dahl model is adopted in [1, 6, 9, 14], the LuGre friction model is adopted in [15, 16], neural network models are developed in [7, 17, 18], fuzzy logic is used in [19] and a mapping approach in [8]. The Heaviside step function approach with feedback from a force sensor belongs to the second class [20]. The control force tracking schemes of the third class combine a model-based feed-forward with a force feedback to enhance the tracking accuracy [2123]. This paper presents a new approach to the rst class of control force tracking schemes. The real-time force tracking scheme is based on the BoucWen model [20, 2426] that is used as observer and is extended by the Stribeck effect [27]. The new approach is veried by simulation and the control
c 2013 IOP Publishing Ltd Printed in the UK & the USA

Smart Mater. Struct. 22 (2013) 045012

F Weber

where Z (s) is the complex evolutionary variable in the frequency domain, s is the complex radial frequency, 0 is the radial frequency and the damping ratio of the second-order lter. Figure 1 shows the force velocity trajectories of the proposed BoucWen model including the Stribeck effect. The modelled MR damper force is normalized because the proposed real-time force tracking scheme can be applied to any MR damper with any maximum force capacity. It is observed that the parameter settings of the second-order lter generate the typical force overshoot due to the Stribeck effect [2529]. The evolutionary variable z is governed by the differential equation according to the BoucWen approach z = | x|z|z|n1 x |z|n + Ax (3)

Figure 1. Normalized force response of the proposed BoucWen model with Stribeck effect.

where , , n and A are parameters to shape the hysteretic force part of the MR damper force. (i) is modelled as a quadratic function of the MR damper current as follows = a + b (i 1.2imax )2 (4)

force tracking errors are discussed. It is shown that the errors can be split into errors due to physical force constraints of MR dampers and very small errors due to the proposed control scheme. The BoucWen model-based real-time force tracking scheme is then experimentally validated. It is explained how the BoucWen model can be run numerically stable in real-time in the Matlab R /dSPACE R environment. The measured control force tracking accuracies are discussed and conclusions are drawn.

where a and b are shaping parameters and imax is the maximum MR damper current. Approach (4) with negative b is able to model the nonlinear increase of the hysteretic force part for increasing current, as is typically observed in case of many MR dampers [29]. The nonlinear increase in the hysteretic force part is visible in gure 1 from the fact that the MR damper force increase from, e.g., 0.2imax to 0.4imax , is greater than from, e.g., 0.8imax to imax . c(i) is modelled as a linear function of current c(i) = ca + cb i (5)

2. BoucWen model-based real-time force tracking scheme


2.1. BoucWen model including Stribeck effect The BoucWen model proposed here is based on the original approach and is extended by a second-order lter to model the Stribeck effect [27]. If the MR damper does not include an accumulator, as is the case for the rotational MR damper described in [8, 9], the current-dependent MR damper force f (i) becomes f (i) = (i)zStribeck + c(i)x (1)

where zStribeck denotes the evolutionary variable including the Stribeck effect and x is the damper velocity. (i) and the viscous coefcient c(i) are modelled as functions of the MR damper current i and not as functions of the command voltage u to the current driver as done in, e.g. [23], because the current driver of type KEPCO R that is used for the validation of the proposed real-time force tracking scheme is capable of almost fully compensating for the coil impedance of the MR damper; hence, the actual current is almost equal to the desired current at all instants (see section 4, gures 19 and 22). As a consequence, current driver dynamics can be neglected. The Stribeck effect is modelled by adopting the second-order lter Z Stribeck (s) = Z (s)
2 0 2 s2 + 2 0 s + 0

to take into account the effect that the MR damper viscous force part increases with increasing current. This feature is visible in gure 1 from the increasing slope of the force velocity trajectories in the post-yield region [29]. The parameters of the proposed BoucWen model equation (1)equation (5) are selected such that the simulated force resulting from a sinusoidal displacement of 1 Hz and 50 mm shows the behaviour that is described above and is typical for many cylindrical MR dampers. The parameters are = 200 1 m1 , = 200 1 m1 , n = 2, A = 5000 1 m1 , a = 4250 N, b = 27.7778 N A2 , ca = 1400 N s m1 , cb = 1000 N s m1 A1 , 0 = 2 25 rad s1 and = 0.585. The given units assume that z is a non-dimensional variable. Please notice that the simulation of the proposed MR damper model does not target to model a specic MR damper but is required in section 2.2 and later in section 3 to explain the proposed BoucWen model-based real-time force tracking scheme. 2.2. Real-time force tracking scheme Figure 2 shows the entire real-time force tracking scheme. The desired control force fdes due to the selected control laws of viscous damping and clipped viscous damping with negative stiffness is computed based on the actual MR damper displacement x and velocity x . The desired control force, the displacement and velocity are input state variables to the
2

(2)

Smart Mater. Struct. 22 (2013) 045012

F Weber

3. Simulation results
The BoucWen model-based real-time force tracking scheme is adopted to track in real time the desired control forces due to the control laws: (a) linear viscous damping [9, 30], and (b) clipped viscous damping with negative stiffness [9, 31, 32] with an MR damper that is modelled by the BoucWen approach with Stribeck effect (15). The control law (a) is chosen as a test case because this benchmark damping strategy is appropriate to explain the working behaviour of the proposed real-time force tracking scheme. The second test scenario is the control law (b), which can double the structural damping compared to linear viscous damping, since it is a promising approach for semi-active structural control and the limitations of the force tracking accuracy due to the pre-yield stiffness and residual force at 0 A of MR dampers can be demonstrated well. It is emphasized that the proposed real-time force tracking scheme can be adopted to track any desired control force in real time as given by, e.g., clipped optimal control [2, 3], sliding mode control [33, 34], fuzzy logic [35, 36] and neural network-based control schemes [17, 37, 38] that are often used for structural control. 3.1. Simulated real-time force tracking of viscous damping The simulation of the system as depicted in gure 2 is made in Matlab R /Simulink R using the solver ode3 (BogackiShampine) with a xed time step size of 0.5 ms. The input to the Simulink R le is a sinusoidal displacement with 5 mm amplitude and constant frequency of 1 Hz. The desired control force as function of time t is [9, 30] fdes (t) = cdes x (t) (8)

Figure 2. BoucWen model-based real-time force tracking scheme.

BoucWen model-based estimation of the current i. In order to check the force tracking accuracy in simulation, the actual MR damper force fact is then computed based on the estimated current i, the actual displacement and velocity. For the case under consideration where (1) does not depend on x due to the absence of an accumulator as in, e.g. [23], (1) is computed based on x and i only. The core of the force tracking method is depicted in gure 2 by the close-up. The same BoucWen model (15) is computed in real-time for the constant currents i = 0 A, 0.1imax , . . ., imax and for the actual x and x . The resulting forces f (0 A), f (0.1imax ), . . . , f (imax ) represent the estimated forces that would be generated by the MR damper if the applied MR damper currents were i = 0 A, 0.1imax , . . ., imax . Based on the forces f (0 A), f (0.1imax ), . . . , f (imax ) and the desired control force fdes , the interpolated current iint is derived by piecewise linear interpolationhere described for the case when |fdes | is larger than |f (0.2imax )| and smaller than or equal to |f (0.3imax )|as follows
(0.3imax 0.2imax )(|fdes | |f (0.2imax )|) 0.2imax + f (0.3imax ) f (0.2imax ) : |f (0.2imax )| < |fdes | |f (0.3imax )| = and sgn(fdes ) = sgn(f (0.3imax )) 0 : otherwise.

iint

(6)

where cdes is the desired viscous damping coefcient. The simulated force tracking accuracies are shown in gure 3 by the desired and actual force displacement trajectories and in gure 4 by the desired and actual force velocity trajectories. In addition to these trajectories, the trajectories resulting from constant current are also plotted to better visualize the sources of force tracking errors (FTEs). The simulated force tracking is based on the following simplications: undesirable current dynamics due to weak current drivers do not exist, which is fullled as shown in section 4 (gures 19 and 22), and MR uid dynamics due to the migration and aggregation of particles do not exist [39]. As can be seen from gures 3 and 4, fdes is not constrained by f (imax ) due to the choice of cdes = 4.85 104 N s m1 and the FTEs are mainly caused by: the residual force at 0 A of the MR damper, and the pre-yield stiffness of the MR damper.
3

The desired MR damper current ides takes the current constraints into account, yielding iint : 0 iint imax (7) ides = 0 : iint < 0 imax : iint > imax . The force tracking accuracy resulting from the proposed BoucWen model-based real-time force tracking scheme is validated by simulation and experiments, which are described in sections 3 and 4.

Smart Mater. Struct. 22 (2013) 045012

F Weber

Figure 3. Simulated tracking of viscous damping: force displacement trajectories.

Figure 6. Simulated tracking of viscous damping: close-up of force velocity trajectories.

Figure 4. Simulated tracking of viscous damping: force velocity trajectories.

Figure 7. Simulated tracking of viscous damping: forces (a) and desired current (b) versus time.

The close-ups in gures 5 and 6 and the gure 7 indicate how the BoucWen model-based real-time force tracking scheme works. (i) Within this section of the trajectory, fdes can be fully tracked because fdes is not constrained by f (0A), f (imax ) and the pre-yield stiffness. FTEs are not visible but exist because the interpolation (6) is piecewise linear but the hysteretic force part of the MR damper is a nonlinear function of current (4). However, the small force intervals used the interpolation (6), which corresponds to current intervals of 0.1imax , end up in negligibly small FTEs, i.e. |fdes fact |/|fdes | 0.0015. Due to the nonlinear relation between the hysteretic force of the MR damper and current (4), the estimated current i does not describe a pure half-sine as a function of time (gure 7(b)). (ii) Within this section of the trajectory, fdes < f (0 A) and sgn(fdes ) = sgn(f (0A)). The minimal FTE is achieved if i = 0 A is applied, which is realized by the real-time force
4

Figure 5. Simulated tracking of viscous damping: close-up of force displacement trajectories.

Smart Mater. Struct. 22 (2013) 045012

F Weber

Figure 8. Simulated tracking of viscous damping constrained by maximum MR damper force: force displacement trajectories.

Figure 9. Simulated tracking of viscous damping constrained by maximum MR damper force: force velocity trajectories.

tracking scheme, see gure 7. Consequently, fact tracks f (0A). The FTE here is given by the residual force at 0 A of the MR damper. (iii) Within this section of the trajectory, sgn(fdes ) = sgn(f (0A)) due to the pre-yield region. Due to sgn(fdes ) = sgn(f (0A)), i = 0 A minimizes the FTE, which is realized by the real-time force tracking scheme. (iv) This section starts where the pre-yield region ends and thereby sgn(fdes ) = sgn(f (0A)). From gures 5 and 6 it can be observed that the shortest trajectory to get back to fdes is to follow the trajectory resulting from imax . The BoucWen model-based real-time force tracking approach exactly does this: the estimated current steps from i = 0 A to imax (gure 7(b)) in order to come back to fdes as fast as possible (gure 7(a)) and thereby minimize the FTE. (v) As in section (i), the tracking of fdes is not constrained by f (0A), f (imax ) and the pre-yield stiffness and the resulting very small FTE of |fdes fact |/|fdes | 0.0015 is caused only by the linear piecewise interpolation (6). 3.2. Simulated tracking of viscous damping constrained by maximum MR damper force The BoucWen model-based real-time force tracking scheme is also checked for the case when fdes is constrained by f (imax ) due to a larger cdes , i.e. 1.5 4.85 104 N s m1 . Figures 8, 9 and 10(a) plot the resulting force tracking accuracies. In addition to the FTEs as described in section 3.1, FTEs due to the maximum MR damper force limitation f (imax ) occur. As can be seen from gure 10(b), the proposed real-time force tracking scheme outputs i = imax if |fdes | > |f (imax )| and sgn(fdes ) = sgn(f (imax )) due to (6) and (7). The simulations presented in sections 3.1 and 3.2 show that the sources of FTEs can be split into physical force constraints of MR dampers and inaccuracies of the real-time force tracking method under consideration. Neglecting the dynamics of the current driver and MR uid, the physical
5

Figure 10. Simulated tracking of viscous damping constrained by maximum MR damper force: forces (a) and desired current (b) versus time.

force constraints of MR dampers besides their semi-active restriction are: residual force at 0 A, maximum force at imax , and pre-yield stiffness. The FTEs due to the proposed BoucWen model-based real-time force tracking scheme: are caused by the piecewise linear interpolation in (6), which is a reasonable, control-oriented simplication of the fact that the relation between the hysteretic MR damper force and current is nonlinear (4), and are very small, i.e. |fdes fact |/|fdes | 0.0015, due to the small force intervals, which correspond to current intervals of 0.1imax , that are used in the interpolation scheme (6).

Smart Mater. Struct. 22 (2013) 045012

F Weber

Figure 11. Simulated tracking of clipped viscous damping with negative stiffness: force displacement trajectories.

Figure 13. Simulated tracking of clipped viscous damping with negative stiffness: forces (a) and desired current (b) versus time.

reduced compared to viscous damping, only that fdes is not constrained by |f (imax )|. As can be observed from gure 11, fdes describes an ellipse with negative inclination due to the negative stiffness term. The active desired control forces of viscous damping with negative stiffness that are plotted by the dashed blue line are clipped to zero, which evokes the horizontal and vertical trajectory parts of fdes . As indicated by the arrows in gure 12, the negative stiffness force leads to a force velocity trajectory that turns clockwise in the post-yield region. Looking at gures 1113, the following can be observed. (i) Within this section of the trajectory, fdes is not constrained by f (0A), f (imax ) and the pre-yield stiffness. The resulting FTE due to the proposed real-time force tracking scheme is caused by the linear piecewise interpolation in (6) and is |fdes fact |/|fdes | 0.0015. (ii) Within this section of the trajectory, fdes < f (0A) and sgn(fdes ) = sgn(f (0A)). The FTE is minimized by i = 0 A, which is applied by the real-time force tracking scheme. (iii) In this section the MR damper is operated in the pre-yield region and sgn(fdes ) = sgn(f (0A)). Hence, i = 0 A is the best choice to minimize the FTE. (iv) Within this section of the trajectory, the MR damper is still operated in the pre-yield region and sgn(fdes ) = sgn(f (0A))fdes due to the jump of fdes . As for section (iii), i = 0 A minimizes the FTE which is realized by the real-time force tracking scheme. (v) This section starts where the pre-yield region ends and thereby sgn(fdes ) = sgn(f (0A)). The shortest trajectory to get back to fdes is given by the trajectory resulting from imax . As visible in gure 13, the real-time force tracking scheme outputs the current step from i = 0 A to imax and reduces i very rapidly when fdes is reached in order to track fdes as in section (i). (vi) As in section (i), the FTE is only given by the real-time force tracking method under consideration.
6

Figure 12. Simulated tracking of clipped viscous damping with negative stiffness: force velocity trajectories.

3.3. Simulated tracking of clipped viscous damping with negative stiffness Viscous damping with negative stiffness is a preferable control law in structural control because the negative stiffness force increases the structural exibility at the control device position [9, 31, 32]. This enlarges the motion and thereby the energy dissipation in the control device, which nally leads to augmented structural damping. If this control law is to be emulated with MR dampers, the common method is to clip the active desired control forces to zero. This yields the desired, semi-active control force as follows (t) kdes x(t) cdes x : x (t)(cdes x (t) kdes x(t)) 0 fdes (t) = (9) 0 : x (t)(cdes x (t) kdes x(t)) < 0 where kdes is the desired stiffness coefcient. The chosen parameters for the numerical test are cdes = 2.909 104 N s m1 and kdes = 2 105 N m1 , where cdes is

Smart Mater. Struct. 22 (2013) 045012

F Weber

Figure 14. Cable damper setup with measurement and control hardware.

The force tracking analysis of clipped viscous damping with negative stiffness shows that: the pre-yield stiffness at imax limits how accurate the jump in fdes can be tracked by fact , and the active desired control forces are constrained, rst, by clipping and, second, by the residual force at 0 A.

4. Experimental results
4.1. Experimental setup The proposed real-time force tracking scheme is experimentally validated on a steel wire strand with a rotational MR damper positioned at 4% of the cable length from the left anchor (gure 14). Detailed information on the rotational MR damper under consideration can be found in [40]. The cable properties are: the tensile force is 22 kN, the mass per unit length including the additional clamped masses is 5.85 kg m1 and the length is 16.54 m. The displacement at the MR damper position is measured by a laser triangulation sensor and the cable acceleration is recorded by an accelerometer. The actual MR damper force is measured by a 500 kN load cell to determine the FTE. The desired MR damper current is produced by the current driver of type KEPCO R which almost fully compensates for the coil impedance. As a result, the time lag and difference between the desired and actual currents are negligibly small, see gures 19 and 22. The BoucWen model-based real-time force tracking scheme is implemented in the Matlab R /dSPACE R environment and runs at 2000 Hz sampling frequency in real time. The collocated velocity is estimated from the measured displacement and acceleration by adopting a kinematic Kalman lter [9]. Numerical instabilities of the BoucWen model-based force tracking scheme due to measurement noise in the input states are compensated by the following limitations: min(z ) = 2000, max(z ) = 2000, min(z) =
7

1, max(z) = 1. These limitations are required only when the cable is almost non-vibrating. If the cable amplitudes are larger than 0.2 mm at the MR damper position, the limitations of z and z do not constrain z and z, respectively. Notice that z is a non-dimensional variable with an operating range of [1, 1]. To make z less susceptible to measurement noise in the input states, the evolutionary variable z is ltered by a low pass of rst order (20 dB/decade) with corner frequency 300 Hz. 4.2. Validated BoucWen model of rotational MR damper First, the force responses of the rotational MR damper resulting from constant currents was measured using a hydraulic actuator of type Instron R that imposed a sinusoidal displacement of 8 mm amplitude at 1.5 Hz (gures 15 and 16). The measured trajectories show the following features. The pre-yield stiffness increases with increasing currents, as typically observed for many MR dampers and also seen in the simulated force displacement trajectories due to constant current in section 3 (gures 5 and 6). The viscous force part of the MR damper under consideration that can be read off from the force velocity trajectories between the Stribeck effect and x = 0 is very small [29]. The force response between the pre-yield and post-yield regions shows an almost linear and fairly slow increase as a function of the MR damper displacement and time, respectively. This slow force response results from the fact that, when the disc starts to rotate sinusoidally in one direction (starting from max(x)), rst, the particle chains at the outer diameter of the disc enter the post-yield regime due to the larger shear deformation at this position [40, 41]. Then, with increasing rotation angle and MR damper displacement, respectively, also particle chains that are located closer to the centre of the disc enter the post-yield regime. Finally, when the rotation of the disc corresponds

Smart Mater. Struct. 22 (2013) 045012

F Weber

Figure 15. Validated BoucWen model with Stribeck effect: force displacement trajectories.

Figure 17. Measured tracking of viscous damping: force displacement trajectories.

Figure 16. Validated BoucWen model with Stribeck effect: force velocity trajectories. Figure 18. Measured tracking of viscous damping: force velocity trajectories.

to approximately 2 mm of MR damper displacement for the case under consideration, all particle chains are operating in the post-yield regime and consequently the full post-yield force is produced. The parameters = 10 000 1 m1 , = 10 000 1 m1 , n = 2 and A = 15 000 1 m1 are selected by trial and error to obtain the same slope of the simulated force displacement trajectories in the pre-yield region as measured. The nonlinear relation between the hysteretic force part and current (4) is modelled by a piecewise linear curve t and implemented in Matlab R /dSPACE R by a one-dimensional look-up table. The viscous force part is estimated from the measured force velocity trajectories in the post-yield region. The resulting parameters are ca = 34 N s m1 and cb = 8 N s m1 A1 . The parameters of the second order lter to model the Stribeck effect are chosen to be 0 = 2 12 rad s1 and = 0.78 to obtain the same force overshoot during the same displacement range as seen in the measurements. Since the approach (15) does not take into account the effect that the post-yield
8

regime is a function of the rotation angle as described above, the simulated and measured trajectories show a discrepancy where the MR damper force leaves the pre-yield regime and enters the post-yield regime (gures 15 and 16). In contrast, the model shows a very good accuracy in the pre-yield region and in the part of the post-yield region where all particle chains are operated in the sliding regime. 4.3. Measured tracking of viscous damping The proposed real-time force tracking scheme is rst tested by tracking fdes resulting from viscous damping (8) with cdes = 2000 N s m1 . This choice of cdes guarantees that the desired force is not constrained by |f (imax )|. The measured force tracking is plotted in gures 17 and 18 during two periods of vibration of a free decay response of the cable. It is observed that max(fdes ) is smaller than max(fdes ). The reason for this

Smart Mater. Struct. 22 (2013) 045012

F Weber

Figure 19. Measured tracking of viscous damping: desired and actual currents versus time.

Figure 20. Measured tracking of clipped viscous damping with negative stiffness: force displacement trajectories.

is that the cable peak velocity at the MR damper position is greater when the cable swings downwards than upwards. FTEs are mainly visible in sections (i) and (ii). (i) Section (i) denotes the location in the trajectory where the MR damper force gets out of the pre-yield regime and enters the post-yield regime, where the desired force is not constrained by |f (0A)| and |f (imax )|. Due to |fdes | < |f (0A)| in the pre-yield region, the MR damper locks the cable. When the desired force can be tracked in the post-yield region, the MR damper starts to rotate. The transition between the operating condition locked and rotating induces higher frequency components in the collocated cable displacement, which are then also present in the collocated velocity x and consequently in fdes , due to the control law fdes = cdes x . (ii) The oscillations visible in section (ii) represent the transient closed-loop response of the initially triggered higher frequency components in section (i). The mean FTE within the trajectory part from point a to point b, where the force tracking is not constrained by the pre-yield stiffness, f (0A) and f (imax ), is 0.0780. This value is of the same order as the measured FTEs presented in [23, 42] and therefore can be seen as acceptably small, considering that the MR damper force is controlled in real-time by a model-based feed-forward without force feedback from a sensor. The desired and actual currents are depicted in gure 19, which demonstrate that the current driver of type KEPCO R is capable to almost fully compensate for the coil impedance of the MR damper under consideration. Hence, the modelling approach in section 2 that does not take into account current dynamics due to insufciently powerful current drivers is justied. The current time history shows the typical peaks to imax when the BoucWen model-based real-time force tracking scheme targets to reach fdes as fast as possible after the pre-yield region, see (iv) in section 3.1. After this current
9

peak, the commanded current generates approximately the correct actual force and becomes zero when |fdes | < |f (0A)|. The higher frequency components in the desired current, which mainly take place immediately after the current peak, are caused by the higher frequency components in fdes due to the transition from the pre-yield region to the post-yield region. The different current maxima when fdes is tracked in the post-yield region correspond to the fact that max(fdes ) is smaller than max(fdes ) due to the larger peak velocity of the cable when it swings downwards than upwards. 4.4. Measured tracking of clipped viscous damping with negative stiffness In order to guarantee that the desired force is not constrained by the maximum MR damper force at 3.5 A, i.e. |fdes | < |f (imax )|, cdes is reduced compared to viscous damping only because the stiffness term in (9) increases the maximum desired force signicantly. The desired parameters in (9) for the test under consideration are chosen as cdes = 1150 N s m1 and kdes = 28 000 N m1 . The resulting force displacement and force velocity trajectories during two periods of vibration during the free decay response of the cable are plotted in gures 20 and 21 and the corresponding desired and actual currents are displayed in gure 22. It is seen that: (i) The current spike within the section (i) forces fact to follow the trajectory due to imax and thereby minimizes the FTE. The resulting FTE is given by the nite pre-yield stiffness of the force displacement trajectory at imax . Only the hypothetical case of an innite large pre-yield stiffness due to innite large imax would allow tracking the jump in fdes by fact without FTEs. (ii) During the section (ii), fact < fdes despite the current is still i = imax . Here, fdes cannot be precisely tracked due to the effect that the full development of the post-yield force requires a rotation in the disc that corresponds

Smart Mater. Struct. 22 (2013) 045012

F Weber

Figure 21. Measured tracking of clipped viscous damping with negative stiffness: force velocity trajectories.

Figure 22. Measured tracking of clipped viscous damping with negative stiffness: desired and actual currents versus time.

to approximately 2 mm damper displacement. This characteristic of the MR damper under consideration is described in section 4.1, where the full force response in the post-yield region also requires approximately 2 mm damper displacement. (iii) In this section, the force tracking is not constrained by f (0A), f (imax ), the post-yield region and the dependency of the post-yield force on the rotation angle as in section (ii). The mean FTE due to the real-time force tracking scheme within the trajectory part from point a to point b is 0.0569. Compared to the measured FTEs obtained in [23, 42], this value represents a fairly small value for the real-time controlled MR damper force without force feedback from a sensor. (iv) In this section, rst fact > fdes due to f (0A) and then fact > fdes because of the clipping. The real-time force tracking scheme commands the correct current for minimum FTE, which is i = 0 A. The reason why fact does not track f (0A) from the very beginning of this section although i = 0 A is applied are remanent magnetization effects in the particles due to the magnetization in the previous sections (i)(iii) [23]. The remanent magnetization evokes a slightly larger force at i = 0 A than f (0A) that results from the application of i = 0 A for a longer time as, e.g., when measuring the MR damper characteristics at constant current [23], see section 4.2, gure 15.

5. Summary and conclusions


A new approach to track the desired control force in real-time with MR dampers is presented. The approach is based on a BoucWen model-based feed-forward without feedback from a force sensor. The BoucWen model is adopted to estimate the MR damper force in real-time for the actual MR damper displacement and velocity, respectively, and as if the constant currents of i = 0 A, 0.1imax , . . ., imax were applied. The resulting estimated force that would be generated by the MR
10

damper if the constant currents of i = 0 A, 0.1imax , . . ., imax were applied together with the desired control force are then used to estimate the desired current in real-time by a linear piecewise interpolation scheme. The model-based feed-forward control scheme is tested by simulation to track linear viscous damping and clipped viscous damping with negative stiffness. These two control laws are selected because the working behaviour of the proposed force tracking scheme can be shown well. However, it is emphasized that the proposed force tracking scheme can be adopted to track any desired control force resulting from, e.g., clipped optimal control, sliding mode control, fuzzy logic, neural network-based schemes and others. It is demonstrated that the proposed real-time force tracking scheme is able to track the desired control force in both cases with a tracking error 0.0015 if the desired control force is not constrained by the physical force constraints of MR dampers, which are pre-yield stiffness, residual force at 0 A and maximum force at imax . It is explained why a current peak minimizes the force tracking error when the MR damper is operated in the pre-yield region and the signs of the desired control force and the pre-yield MR damper force are equal. The force tracking scheme is experimentally validated on a rotational MR damper that is connected to a vibrating steal wire strand. It is explained how the BoucWen model is run in real time in the Matlab R /dSPACE R environment without numerical instabilities. The experimental results show that the model-based feed-forward force tracking scheme is able to track the two control laws under consideration without any numerical instability and with an average force tracking error of 0.0570.078. It is therefore concluded that the proposed control scheme is a reliable and efcient force tracking tool for the real-time control of MR dampers.

Acknowledgments
The author gratefully acknowledges the nancial support of Empa, Swiss Federal Laboratories for Materials Science

Smart Mater. Struct. 22 (2013) 045012

F Weber

and Technology, D ubendorf, Switzerland, and the technical support of the industrial partner Maurer S ohne GmbH and Co. KG, Munich, Germany.

[19]

References
[1] Aguirre N, Ikhouane F, Rodellar J, Wagg D J and Neild S A 2010 Viscous and Dahl model for MR dampers characterization: a real time hybrid test (RTHT) validation Proc. 14th European Conf. on Earthquake Engineering (Ohrid, Aug.Sept. 2010) [2] Christenson R E, Spencer B F Jr and Johnson E A 2006 Experimental verication of smart cable damping J. Eng. Mech. 132 26878 [3] Gattulli V, Lepidi M and Potenza F 2009 Seismic protection of frame structures via semi-active control: modeling and implementation issues Earthq. Eng. Eng. Vib. 8 62745 [4] Lee H-J, Jung H-J, Moon S-J, Lee S-K, Park E-C and Min K-W 2010 Experimental investigation of MR damper-based semiactive control algorithms for full-scale ve-story steel frame building J. Intell. Mater. Syst. Struct. 21 102537 [5] Lin Y Z and Christenson R 2011 Real-time hybrid test validation of a MR damper controlled building with shake table tests Adv. Struct. Eng. 14 7992 [6] Ma slanka M, Sapinski B and Snamina J 2007 Experimental study of vibration control of a cable with an attached MR damper J. Theor. Appl. Mech. 45 893917 [7] Metered H, Bonello P and Oyadiji S O 2009 The experimental identication of magnetorheological dampers and evaluation of their controllers J. Mech. Syst. Signal Process. 24 97694 [8] Weber F, Distl H, Feltrin G and Motavalli M 2009 Cycle energy control of MR dampers on cables Smart Mater. Struct. 18 015005 [9] Weber F and Boston C 2011 Clipped viscous damping with negative stiffness for semi-active cable damping Smart Mater. Struct. 20 045007 [10] Wu W J and Cai C S 2010 Cable vibration control with a semiactive MR damper-numerical simulation and experimental verication Struct. Eng. Mech. 34 61123 [11] Zapateiro M, Karimi H R, Luo N and Spencer B F Jr 2009 Frequency domain control based on quantitative feedback theory for vibration suppression in structures equipped with magnetorheological dampers Smart Mater. Struct. 18 095041 [12] Spencer B F Jr and Nagarajaiah S 2003 State of the art of structural control J. Struct. Eng. 129 84556 [13] Ou J and Li H 2010 Analysis of capability for semi-active or passive damping systems to achieve the performance of active control systems Struct. Control Health Monit. 17 77894 [14] Weber F, Feltrin G and Motavalli M 2005 Measured LQG controlled damping Smart Mater. Struct. 14 117283 [15] Yang F, Sedaghati R and Esmailzadeh E 2009 Development of LuGre friction model for large-scale magneto-rheological uid damper J. Intell. Mater. Syst. Struct. 20 92337 [16] Jim enez R and Alvarez-Icaza L 2005 LuGre friction model for a magnetorheological damper J. Struct. Control Health Monit. 12 91116 [17] Karamodin A and Kazemi H 2010 Semi-active control of structures using neuro-predictive algorithm for MR dampers Struct. Control Health Monit. 17 23753 [18] Zapateiro M, Luo N, Karimi H R and Veh J 2009 Vibration control of a class of semiactive suspension system using
11

[20]

[21]

[22]

[23]

[24]

[25]

[26]

[27]

[28]

[29]

[30]

[31]

[32]

[33]

[34]

[35]

[36]

neural network and backstepping techniques Mech. Syst. Signal Process. 23 194653 Won J and Sunwoo M 2009 Fuzzy modelling approach to magnetorheological dampers: forward and inverse model J. Syst. Control Eng. 223 105565 Dyke S J, Spencer B F Jr, Sain M K and Carlson J D 1996 Modeling and control of magnetorheological dampers for seismic response reduction Smart Mater. Struct. 5 56575 Chen Z H and Ni Y Q 2011 On-board identication and control performance verication of an MR damper incorporated with structure J. Intell. Mater. Syst. Struct. 22 155165 Russo R and Terzo M 2011 Design of an adaptive control for a magnetorheological uid brake with model parameters depending on temperature and speed Smart Mater. Struct. 20 115003 Weber F and Ma slanka M 2012 Frequency and damping adaptation of a TMD with controlled MR damper Smart Mater. Struct. 21 055011 Jung H-J, Spencer B F Jr and Lee I-W 2003 Control of seismically excited cable-stayed bridge employing magnetorheological uid dampers J. Struct. Eng. 129 87383 Yang G, Spencer B F Jr, Jung H-J and Carlson J D 2004 Dynamic modeling of large-scale magnetorheological damper systems for civil engineering applications J. Eng. Mech. 130 110714 Wang D H and Liao W H 2011 Magnetorheological uid dampers: a review of parametric modelling Smart Mater. Struct. 20 023001 Stribeck R 1902 Die wesentlichen eigenschaften der gleit- und rollenlagerthe key qualities of sliding and roller bearings Z. Ver. Dtsch. Ingen. 46 13428 Stribeck R 1902 Die wesentlichen eigenschaften der gleit- und rollenlagerthe key qualities of sliding and roller bearings Z. Ver. Dtsch. Ingen. 46 14327 Weiss K D, Carlson J D and Nixon D A 1994 Viscoelastic properties of magneto- and electro-rheological uids J. Intell. Mater. Syst. Struct. 5 7725 Sims N D, Holmes N J and Stanway R 2004 A unied modelling and model updating procedure for electrorheological and magnetorheological vibration dampers Smart Mater. Struct. 13 10021 Marathe S S, Wang K W and Gandhi F 2004 Feedback linearization control of magnetorheological uid damper based systems with model uncertainties Smart Mater. Struct. 13 100616 Hgsberg J 2009 The role of negative stiffness in semi-active control of magneto-rheological dampers Struct. Control Health Monit. 18 289304 Iemura H and Pradono M H 2009 Advances in the development of pseudo-negative-stiffness dampers for seismic response control Struct. Control Health Monit. 16 78499 Lu K C, Loh C H, Yang J N and Lin P Y 2008 Decentralized sliding mode control of a building using MR dampers Smart Mater. Struct. 17 055006 Neelakantan V A and Washington G N 2008 Vibration control of structural systems using MR dampers and a modied sliding mode control technique J. Intell. Mater. Syst. Struct. 19 21124 Xu Z-D and Guo Y-Q 2006 Fuzzy control method for earthquake mitigation structures with magnetorheological dampers J. Intell. Mater. Syst. Struct. 17 87181 Lin P Y, Roschke P N and Loh C H 2007 Hybrid base-isolation with magnetorheological damper and fuzzy control J. Struct. Control Health Monit. 14 384405

Smart Mater. Struct. 22 (2013) 045012

F Weber

[37] Wang D H and Liao W H 2005 Modeling and control of magnetorheological uid dampers using neural networks Smart Mater. Struct. 14 11126 [38] Wang W and Song Y L 2012 Nonlinear vibration semi-active control of automotive steering using magneto-rheological damper Meccanica 47 202739 [39] Laun M H, Schmidt G, Gabriel C and Kieburg C 2008 Reliable plate-plate MRF magnetorheometry based on validated radial magnetic ux density prole simulations Rheol. Acta 47 104959

[40] Boston C, Weber F and Guzzella L 2010 Modeling of a disc-type magnetorheological damper Smart Mater. Struct. 19 045005 [41] Han Y-M, Choi S-B and Wereley N M 2007 Hysteretic behavior of magnetorheological uid and identication using preisach model J. Intell. Mater. Syst. Struct. 18 97381 [42] Russo R and Terzo M 2011 Modelling, parameter identication, and control of a shear mode magnetorheological device Proc. Inst. Mech. Eng. I 225 54962

12

Anda mungkin juga menyukai