Professional Documents
Culture Documents
Journal of Dynamic Systems, Measurement, and Control JANUARY 2013, Vol. 135 / 011004-1
C 2013 by ASME
Copyright V
In Eq. (1), Vgofr is the static volume of the front right suspen- € ufr þ Pfr Afr þ x17 þ Ktfr ðZufr yfr Þ ¼ 0
mufr Z
sion, c is the gas atomicity coefficient, and Pgofr is the static pres-
sure of the suspension which is equal to Pgofr ¼ mfr g=Afr . € ufl þ Pfl Afl þ x18 þ Ktfl ðZufl yfl Þ ¼ 0
mufl Z
Four first order differential equations are used as the dynamic € urr þ Prr Arr þ x19 þ Ktrr ðZurr yrr Þ ¼ 0
murr Z
model of the CVDTV. For instance, Eq. (2) represents the
dynamic equation of the front right suspension CVDTV € url þ Prl Arl þ x20 þ Ktrl ðZurl yrl Þ ¼ 0
murl Z
€ d þ Kd ðZd þ aP bR ZB Þ þ Cd Z_ d þ aP_ bR_ Z_ B ¼ 0
mD Z
2 1
x_ 17 ¼ x17 þ ufr (2) € B Pfr Afr Pfl Afl Prr Arr Prl Arl
mB Z
sfr sfr
Kd ðZd þ aP bR ZB Þ
Instantaneous response of the control valve is not possible in
Cd Z_ d þ aP_ bR_ Z_ B x17 x18 x19 x20 ¼ 0
the application. Thus, as in Eq. (2), a first order delay of sfr is sup-
posed in the dynamic model of the CVDTV. The values of the € þ Pfr Afr B2 Pfl Afl B1 þPrr Arr B2 Prl Arl B1
Iz R
system parameters are as in Refs. [21] and [27].
Kd ðZd þ aP bR ZB Þb Cd Z_ d þ aP_ bR_ Z_ B b
3.3 Nonlinear Mathematical Model of the Vehicle With þ x17 B2 x18 B1 þx19 B2 x20 B1 ¼ 0
Hydro-Active Suspension. Hydro-active suspension works under € þ Pfr Afr A1 þ Pfl Afl A1 Prr Arr A2 Prl Arl A2
Iy P
the adiabatic process [33]. Therefore, the gas spring behaves under
the ideal gas law. Nonlinear model of the front right suspension is þ Kd ðZd þ aP bR ZB Þa þ Cd Z_ d þ aP_ bR_ Z_ B a
derived in the following. Instantaneous pressure and volume of the þ x17 A1 þ x18 A1 x19 A2 x20 A2 ¼ 0 (6)
front right gas spring (Pgfr and Vgfr ) and static pressure and volume
of this gas spring satisfy the ideal gas equation
4 Sensors and Actuators
c Sixteen candidate measured variables (candidate sensors) are
Pgofr Vgofr
Pgfr ¼ c (3) supposed as in Fig. 6. These variables are y1 , y2 , y3 , and y4 as
Vgfr measures of the suspension displacements; y5 , y6 , y7 , and y8 as
Journal of Dynamic Systems, Measurement, and Control JANUARY 2013, Vol. 135 / 011004-3
measures of the axles accelerations; y9 , y10 , y11 , and y12 as meas- Like all types of semi-active suspensions, damping coefficients
ures of the vertical accelerations of the vehicle in the position of are the control inputs. Thus, selection of the manipulated variables
the wheels; y13 , y14 , and y15 as measures of the driver’s seat accel- (inputs) is not addressed here. These variables are ufr , ufl , urr , and
erations; and y16 as a measure of the vertical acceleration of the url as shown in Fig. 6.
body.
Other sensors can also be candidate, for instance, sensors meas- 5 Selection of Measured Variables
uring relative displacement between suspension and road surface.
However, in this study only the sensors given in Table 1 are used. Some tools of control structure selection depend on scaling such
This is done because of low costs, easy implementation, and suita- as most of the selection criteria based on input–output controllabil-
ble durability of theses sensors which make them proper for ity. However, incorrect scaling can result in improper selections.
application. On the other hand, scaling uncertain variables is not easy always.
These are the main reasons for implementation of such input–-
output selection criteria, which are independent from scaling.
Table 1 Candidate sensors description State space description of the vehicle with linear hydro-active
suspension is as Eq. (7). The matrices are described in Appendix
Output variable Description Output variable Description A of Ref. [21]
y1 Zufr ðZB A1 P B2 RÞ y9 € B A1 P
Z € B2 R
€ 8
y2 Zufl ðZB A1 P þ B1 RÞ y10 € B A1 P
Z € þ B1 R
€ < x_ ¼ AH x þ BH u u þ BH w w
y3 Zurr ðZB þ A2 P B2 RÞ y11 € B þ A2 P
Z € B2 R
€ y ¼ CH x þ DH u u þ DHw w (7)
y4 Zurl ðZB þ A2 P þ B1 RÞ y12 € B þ A2 P
Z € þ B1 R
€ :
z ¼ EH x þ FHu u þ FH w w
y5 € ufr
Z y13 €D
Z
y6 € ufl
Z y14 aP€
y7 € urr
Z y15 bR€ where x is the state vector, y is the measured variables vector, and
y8 € url
Z y16 €B
Z z is the controlled variables vector (see general control system
setup in Fig. 1).
Fig. 7 Evaluation of four candidate sensor types based on the JRSNP criterion
Journal of Dynamic Systems, Measurement, and Control JANUARY 2013, Vol. 135 / 011004-5
where “:” shows the Schur or Hadamard multiplication. For a Set number 100 rad=s > RHP zeros
nonsquare system, matrix inversion in Eq. (9) is substituted with
pseudo-inverse. For the first time, RGA has been presented in Ref. Set0001 0.0084
Set0002 0.0000
[34] as a steady state measure to assess interactions in decentral- Set0003 0.0000, 0.0361 6 34.1j, 0.0000 6 0.0000j
ized control systems. In most of the studies in the field of control Set0004 0.0074
structure selection, RGA has been used as a method for control
configuration selection. However, it can be shown that RGA is
also suitable for input–output selection. The law is not to select
input–output sets which cause big RGA elements. Because their However, if it is not possible to totally avoid unstable transmis-
related system P is substantially difficult to control and by using a sion zeros, the input–output sets must be chosen such that they
specific class of controllers, the system will be nonrobust [13]. So, result in the least number of RHP zeros and these zeros should be
it is suitable to K have small elements. Also to have diagonal away from the imaginary axis as far as possible [35]. Table 2
dominancy, K I should be small. These two goals can found a shows the RHP zeros slower than 100 rad=s for the remaining
criterion, which is RGA number candidate sets.
X Based on Table 2, set 0001 and set 0004 are more viable for
X
RGA number : ¼kK I ksum ¼ 1 kij þ kij (10) control. Next, the quantified criteria from the state controllability
i¼j i6¼j and observability are used to tackle the final selection.
Smaller RGA numbers cause easier controls. The RGA num- 5.4 Hankel Singular Values. In control theory, eigenvalues
bers for four remaining candidate sets are plotted in Fig. 9. determine the system stability. However, Hankel singular values
Whereas RGA numbers of these sets are too close to each other, represent the energy of each system state variable. For a system in
the confident selection is impossible. With respect to this crite- state space description, Hankel singular values are
rion, it seems that set 0004 and set 0003 have worse results than
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
that of set 0002. However, in the sense of this criterion, set 0001 rH ¼ ki ðLc Lo Þ (11)
is the best set in low frequencies.
Lc and Lo are the controllability and observability gramians,
5.3 Right Half Plane Zeros. Due to the serious limitations respectively, and ki are the system eigenvalues. A candidate set,
imposed by RHP transmission zeros in multivariable plants, it is which causes higher Hankel singular values, benefits more joint
desirable to avoid them in any sensor selection. It is well known controllability and observability. That is why such a candidate set
that the location of transmission zeros is related to the sensors. is more viable for control.
The output sets, namely set 0001 and set 0004, are different
only in y1 and y4 . In Fig. 10, Hankel singular values of two trans-
fer functions between the four inputs and every remaining candi-
date measured variable (y1 and y4 ) are shown.
The clear result of Fig. 10 is to select set 0001 as the system
measured variable set. Fig. 11 Interaction evaluation of the system by means of
Gershgorin’s bands
6 Control Configuration Selection
The main subject of control configuration selection is to mini- restricted information flow exists. This will cause decreases in
mize the loop interactions. A proper control configuration selec- decentralized systems performance. However a trade-off should
tion gives a configuration which has the maximum possible be taken among the performance decreases and other benefits of
diagonal dominancy. Based on Ref. [13], if there is no limitation decentralized control systems [13]. A decentralized PID controller
for using steady state RGA criterion, it can be used as a powerful is used to control the system. Figure 12 shows the implemented
means for selecting control configurations. The steady state RGA control system setup. The PID controller gains are tuned using
matrix of the system is presented as below ultimate sensitivity method [37] besides doing some trial and
errors.
ufr ufl urr url The vehicle speed is assumed to be constant at 60 km/h. The
2 3 vehicle is excited by a bump road input. The results are presented
y1 1:124 0:124 0:098 0:098
6 7 in time domain.
y2 6
6 0:091 1:025 0:065 0:131 7
7 The bounce, roll, and pitch accelerations of the driver’s seat are
Kð0Þ ¼ 6 7 presented in Fig. 13 for both passive and hydro-active suspen-
y3 6
4 0:098 0:098 1:097 0:098 7
5 sions. In hydro-active suspension, these accelerations are
y8 0:002 0:003 0:001 1:006 decreased well.
The magnitude of the force between the road and the tire is pro-
portional to the displacement between the tire axis and the road.
Pairing such that the system RGA is close to unitary matrix
This displacement is called tire dynamic deflection. The contact
guarantees having the maximum possible diagonal dominancy
between the tire and the road depends on this vertical force. Con-
and integrity. As seen here, the diagonal and the off-diagonal ele-
sequently, keeping the tire dynamic deflection to small amounts is
ments of the system RGA are close to one and are little, respec-
necessary to have an acceptable contact between the tire and the
tively. Also no diagonal element is negative. So, the pairs (ufr ; y1 ),
road. The variations of tire dynamic deflection for passive and
(ufl ; y2 ), (urr ; y3 ), and (url ; y8 ) are chosen in order to have the mini-
hydro-active suspension systems are presented in Fig. 14. The
mum possible loop interaction.
decrease in tire dynamic deflection in hydro-active system shows
Evaluating the interaction is done in Fig. 11 using column
having a proper tire–road contact.
Gershgorin’s bands. In Ref. [36], a comprehensive study on
Nyquist arrays and Gershgorin’s bands has been presented.
Nyquist array of a transfer function matrix PðsÞ is an array of the
graphs in which the graph ijth is the location of the Nyquist pij ðsÞ.
The column Gershgorin’s bands P of a transfer
function
matrix PðsÞ
m
are circles with the radius of i¼1 pij ðjxÞ and the center of
i6¼j
pii ðjxÞ. If the Gershgorin’s band related to pii ðsÞ does not include
the origin, system PðsÞ has diagonal dominancy in that channel.
As shown in Fig. 11, Gershgorin’s bands include the origin,
whereas the diameters of Gershgorin’s circles are fairly small, that
is, small interactions. Regarding this, decentralized PID controller
design is addressed in the following.
7 Controller Design
In a centralized control system, each input is determined by the
feedback from all the measured variables. That is, full information
exchange exists, whereas in a decentralized control system only a Fig. 12 Control system setup of the vehicle suspension
Journal of Dynamic Systems, Measurement, and Control JANUARY 2013, Vol. 135 / 011004-7
Journal of Dynamic Systems, Measurement, and Control JANUARY 2013, Vol. 135 / 011004-9