Beruflich Dokumente
Kultur Dokumente
Kani1
Advanced Control Systems Laboratory and
Cognitive Robotics Laboratory,
Control and Intelligent Processing
Center of Excellence,
School of Electrical and Computer Engineering,
College of Engineering,
University of Tehran,
Office No. 110,
North Kargar Street,
Tehran 1439957131, Iran
e-mail: h.kani@ut.ac.ir
1
Corresponding author.
Manuscript received April 27, 2015; final manuscript received November 25,
2015; published online January 6, 2016. Assoc. Editor: Robert J. Wood.
2 3
BJ Btot tsinat 2JJ sinat
BJ Btot tt 6 cosat 7
Ur t; 0 e 2JJ 6
4 D D 7 (16)
2Ktot tsinat BJ Btot tsinat 5
cosat
D D
q
1 1=2
where a and D are defined as 0:5JJ 1 D and
q l lt 2sin2 atc D2 2 sin2 atc sin2 atc D2
D
4JJ Ktot BJ Btot 2 , respectively. It is supposed that the
(18)
expression 4JJ Ktot BJ Btot 2 is always positive in our system.
This assumption is imposed by the available and prevalent ranges where c is defined as JJ Ktot 2 BJ Btot 2 . On the other
for spring and damper coefficients and also by the practical hand, although the parameters JJ and BJ are constant, the value of
ranges of the joints moment of inertia and its damping constant. Btot t and Ktot t is varying in time when the joint impedance
By considering the fact that the Euclidean norm of the state transi- changes. So, the value of the common stability margin, k0 , and l
tion matrix can be depicted by the square of its greatest singular depend on the impedance of the joint. Consequently, to find a
value and using some mathematical operations, it can be function envelope for the Euclidean norm of all the state transition
shown that matrices, kUr t; 0k, it is sufficient to choose the maximum and
minimum value of the parameters l and k0 , respectively. The
maximum value of l corresponds to the time that sinat 1. It is
BJ Btot tt noticeable that the maximum value of l and the minimum value
kUr t; 0k le 2JJ
(17) of k0 can occur for two different joint impedance values.
nl
X Xnr _ Dt1 at at Dt is applied to the terms of q_ i and
of at
r=l
W Fli dli :d qli hr Fri dri :dqri hr (22) _h r=l in Eq. (23). Since all the parameters at the time instance t Dt
i1 i1 i
are known, the final representation of the cost function is
l=r
where the function di specifies the active pairs. In each time
l=r
instance, the value of inputs, qi , should be calculated. As a X nl
1 l 2bli l 2 2bl
Ct ki qi t htr i qli t htr
result, their values are specified in order to minimize the partial i1
2 Dt Dt
work, DW, at each time instance. By replacing Eq. (4) in Eq. (22)
and also using dat atdt,
_ DW is calculated as qli t Dt ht Dtr dli
X nr
Xnl
1 l l 2 1 1 r 2bri r 2 2br
DW t ki qi t htr kil qli t Dt ki qi t htr i qri t htr
2 2 i1
2 Dt Dt
i1
2
_ qri t Dt ht Dtr dri (24)
ht Dtr2 bli q_ li hr Dt dli
Moreover, we assume that ht is also known. This assumption is
X nr
1 r r 2 1 logical, because it is common to use a position sensor in any imple-
ki qi t htr kir qri t Dt mented practical joint. As Eq. (24) declares, the cost function has a
i1
2 2 l=r
quadratic form and the only unknown parameters are qi . On the
2 2
other hand, the following constraints must be met simultaneously:
ht Dtr bri q_ ri hr _ Dt dri (23)
!
Amp X nl
l l l l
l X nr
r r r r
r
ut r ki qi tbi q_ i t di ki qi tbi q_ i t di
Although by minimization of DW we do not guarantee to mini- Ktot i1 i1
mize the entire work, it results in an acceptable suboptimal solu-
tion. To have a more simple expression for DW, the approximation T ht (25)
(
qli hr > Dxin ; dli 1
(26)
qri hr > Dxin ; dri 1
(
qli hr Dxout ; dli 0
(27)
qri hr Dxout ; dri 0
(
kil qli hr Flmax; ; dli 1
(28)
kir qri hr Frmax ; dri 1
(
_ > 0 ; dl 1
kil qli hr bli q_ li hr i (29)
r r r _r
ki qi hr bi q i hr _ > 0 ; dr 1
i
be provided in different ways. One of the simplest and most appli- displacement. For the sake of simplicity and without loss of gener-
cable methods is using DC motors and pulleys to convert rotary to ality, these motors are assumed to be identical and also have the
linear motion which is then used in our proposed actuator. A sim- same pulleys with equal diameters. Moreover, PID controllers are
ple PID controller is also chosen to control the angular position of considered to be the same and their transfer functions are in the
each DC motor such a way that the end sides of springs and damp- parallel form.
l=r
ers track their set points, qid (Fig. 3(b)). The most important operational index of the proposed actuator
is set-point tracking of the joint angle and its impedance, simulta-
neously. It is so important that we can change the impedance of
5 Simulation Results the joint independent of its angular position. Figure 4(a) shows
In this section, the variable impedance actuator is simulated. the joint position and its desired angular set-point signal. The joint
We assume that four real springs are located in each side of the impedance and its corresponding desired value are displayed in
joint and their spring constants are chosen as Fig. 4(b). These experiments were done in presence of an external
f1 103 ; 2 103 ; 2 103 ; 5 103 gN=m. Using this set of known torque of 0:1 sinh and the control mechanism compen-
spring constants, we are able to have a range of spring constants sates it efficiently. These two figures obviously describe that the
between 1 103 N=m and 10 103 N=m in each side of the proposed control mechanism is completely able to control both
joint by combination of active springs. Consequently, according the angular position and impedance in presence of a known exter-
to Sec. 4.2, the joint impedance, Ktot , has a discrete value in the nal torque. Figure 5 shows the deflections of all the spring sets.
range of 2 103 r 2 N=m and 20 103 r 2 N=m. On the Being more accurate, it can be seen that the deflection of each
other hand, corresponding to each spring, there exists an energy inactive spring and damper set is kept in the range of Dout when
dissipating element whose damping constant is chosen in propor- the joint tracks its set points.
tion to its related spring constant. So, in each side of the joint, In addition, the allowed switching time must be considered for
there is a set of four dampers with damping constants assumed as any changes in the joint impedance. According to Eq. (15) and the
f10; 20; 20; 50gN s=m. The phenomenon that the damping con- values of system parameters (Table 1), and also assuming
stant varies in proportion to the variation of its corresponding k 0:5k0 ; c 2l, nine switches are admissible in order to
stiffness occurs in some artificial muscles such as pneumatic types maintain the desired exponential stability in a time interval of
[37]. Other parameters of the variable impedance actuator are 20 s. If an application requires a faster switching sequence,
illustrated in Table 1. Using a total of eight real springs for the Eq. (21) can be used to determine the switching times. It should
joint requires eight corresponding DC motors to control their be noticed that each change in the impedance of the joint can be
Fig. 7 Power consumption of the joint versus the joints impedance values. For each impedance, the power consumption for
l r
all of the possible cases of choosing Ktot ; Ktot =r 2 is plotted with circles. The curve connects the minimum power consumption
l r
points and the value of Ktot ; Ktot =r 2 is also written for these minimum points.
taken into account as only one switch between subsystems if just impedances. Thus, there exists a degree-of-freedom to choose the
the control of each springs end side is tight enough. values of the left and right side impedances in order to reach a
Besides set-point tracking, the other important issue is the determined joint impedance value. Investigating this fact, we
energy consumption. Consequently, the power consumption of the measure the joint power consumption at the frequency of 2 Hz for
actuator is calculated for set-point tracking of 0:5p sin2pft in all the possible joint impedance values. Moreover, for each joint
l r
five different frequencies. In addition, in each frequency, the sim- impedance, all of the possible combinations of Ktot and Ktot are
ulation is done for all of the possible joint impedance values and considered. Figure 7 shows that the least power consumption in
l r
the result is illustrated in Fig. 6. The diagram shows that we each specified impedance is occurred when Ktot and Ktot gain pos-
should expect more power consumption for higher joint fre- sible nearest values to each other.
quency. On the other hand, in each frequency, maximum energy The proposed control mechanism is capable of rejecting exter-
consumption occurs in a specific impedance and this impedance nal disturbances. Thus, in the last simulation test, the performance
increases with the operating frequency. of the actuator is studied in presence of a constant torque exerted
Moreover, as we declared in Sec. 4.2, the joint impedance is to the joint as an external unknown disturbance. As illustrated in
calculated by summation of the joints left and right side Fig. 8, a constant torque of 0:5N m is exerted at t 2 for 0:3 s
Fig. 10 Set-point tracking performance of the designed actuator: (a) the joint angular position and (b) the joint impedance
value