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

net/publication/4119270

Input-output analysis of decentralized relay systems

Conference Paper  in  Proceedings of the American Control Conference · January 2004

DOI: 10.1109/ACC.2004.183039 · Source: IEEE Xplore

CITATIONS READS
0 91

3 authors, including:

N. V. Faldin Igor Boiko

Tula State University Khalifa University of Science and Technology, (UAE)
12 PUBLICATIONS   28 CITATIONS    192 PUBLICATIONS   2,199 CITATIONS   

SEE PROFILE SEE PROFILE

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

PID Tuning View project

Dynamic Harmonic Balance View project

All content following this page was uploaded by Igor Boiko on 16 May 2014.

The user has requested enhancement of the downloaded file.

Proceeding of the 2004 American Control Conference WeP13.5
Boston, Massachusetts June 30 - July 2, 2004

Input-output Analysis of Decentralized Relay

Systems
Nikolay V. Faldin, Platon I. Fedorovski, Igor M. Boiko

The decentralized relay system, which is going to be

Abstract—A method of input-output analysis of
decentralized relay systems is proposed in the paper. The
proposed method furnishes a better accuracy than the
describing function method and does not require the
involvement of the filtering hypothesis. An example
illustrating application of the method and assessment of its
accuracy is given.
I. INTRODUCTION
analysed in the paper, can be represented by the block
diagram (Fig. 1). It consists of a plant given by the transfer
function W0 ( s ) , two identical hysteresis relays (R1 and R2)
and two linear filters given by the transfer functions W1 ( s )
and W2 (s ) . The filters given by W1 ( s ) and W2 ( s )
transform signal x into signals y1 and y 2 that have a phase
shift in respect to each other. As a result, relays R1 and R2

R ELAY control systems are extensively used in various

industries in many cases providing cheaper solutions
and better performance compared to other types of control
switch at different times and control U, which is the sum of
their outputs, is equivalent to the output of a certain dead-
zone relay (Fig. 2).
system. Some application examples are various on-off
U1
process control systems, DC motor, pneumatic and
hydraulic relay servo actuators, etc. Relays used in those γ 0T0
control systems normally have two-level ideal or hysteresis A
characteristic. Application of a decentralized relay may 2
provide some advantages over a single relay control. A
− t
Among those advantages the most important are: smaller 2
amplitude of self-excited oscillations and higher tracking
accuracy of servo control. A method of input-output (γ 0 + 1)T0
analysis of dead-zone relay servo systems was given in [1], U2
which was the extension of the methodology proposed in A
[2] and [3] for the hysteresis relay. The current paper
extends mentioned above methods to the case of two relays. 2
T0 2T0 t
The obtained results well agree with known exact [4] and A
−
approximate [5] methods of analysis of periodic motions in 2
relay systems as well as of periodic motions in
decentralized relay systems [6]. U
γ 0T0
II. PROBLEM FORMULATION AND MAIN
RESULTS A

W1(s) y1 U1
R1 T0 t
f U 2T0
x -A
W0(s)
R2 (γ 0 + 1)T0
y2 U2
W2(s)
Fig. 2. Resultant control.

Fig.1. Decentralized relay system with two relays. A. Analysis of the periodic motion
Let us denote the transfer functions:
Nikolay V. Faldin, Platon I. Fedorovski are with Automatic Control A
Systems Department Tula State University 92, Lenina Av., Tula 300600 W1* ( s ) = W0 ( s )W1 ( s ) ; W2* ( s ) = W0 ( s )W1 ( s ) ; being the
Russia Faldin@sau.tsu.tula.ru, Platon@sau.tsu.tula.ru,
2
Igor M. Boiko is with SNS-Lavalin SNC-Lavalin 909 5th avenue SW, amplitude of the relay outputs (R1 and R2); and 2κ being
Calgary, Alberta, T2P 3G5, Canada. i.boiko@ieee.org the hysteresis of relays R1 and R2. We can assume without
0-7803-8335-4/04/$17.00 ©2004 AACC 1749
loss of generality that in periodic motion, the signal
y 2 (t ) has a phase lag with respect to signal the y1 (t ) .The
resultant control U (t ) is the sum of the two relay outputs
[
as shown in Fig. 2. Let x = x1 , x 2 , ..., x n ]T be the state
vector of the system, 2T be the period of self-excited
oscillations and γ be the relative duration of the positive
pulse (Fig. 2). Let us denote the value of x in a periodic
motion at the time of the control U switch from zero to plus
B. Input-output analysis of type zero relay servo system
Suppose the transfer function doesn’t have zero poles.
Also suppose that in the autonomous mode there exists
symmetric oscillations with parameters T0 and γ 0 .
In the case of non-symmetric control (fig. 3) (it can occur
when a constant input signal is introduced) the parameters
of the oscillations can be defined by four function-vectors
x1 = x1 (τ 1, τ 2 , τ 3 , T ) , x 2 = x 2 (τ 1, τ 2 , τ 3 , T ) ,

as x1 (T ,γ ) . Let us denote the value of x in a periodic x 3 = x 3 (τ 1, τ 2 , τ 3 , T ) , x 4 = x 4 (τ 1, τ 2 , τ 3 , T ) .

motion at the time of the control U switch from plus to zero Here x 1 corresponds to the control U switch from zero
as x 2 (T , γ ) . Define the phase locus of the relay system to plus, x 2 - from plus to zero, x 3 - from zero to minus,
with two relays as vectors x1 (T , γ ) and x 2 (T , γ ) . Let us x 4 - from minus to zero.
call the components of the phase locus R-characteristics. U
To determine the parameters of the self-excited
A
oscillations (period 2T and relative pulse duration γ) the
following set of equation should be solved:
~
y12 (T , γ ) = κ , Z1− (T , γ ) > 0 , τ1 τ2 τ 3 2T0 t
y12 (T , γ ) = −κ , Z 2− (T , γ ) < 0 . -A
where y12 (T , γ ) , y12 (T , γ ) are R-characteristics
pertaining Fig. 3. Control signal.
to the input signals of the relays R1 and R2 respectively;
~
Z1− (T , γ ) is the value of y&1 (t ) in periodic motion at the Let the control U be as shown in the Fig. 3. Introduce a
small constant input signal f. This will result in small
time of relay R1 switch from plus to minus; Z 2− (T ,γ ) is the deviations of the limit circle parameters ∆τ 1 , ∆τ 2 , ∆τ 3
value of y& 2 (t ) in periodic motion at the time of relay R2 and ∆τ 4 = ∆T .
switch from mines to plus. Let us write the conditions of proper relays switching
Consider the components of the phase locus. Let the with an accuracy of up to the first order of small quantity:
linear part of system be given by:
x& = Cx + DU , (1)  ∂ y 12 ∂ y 12
f W2 (0) −  y 12 + ∆τ 1 + ∆τ 2 +
 ∂ τ1 ∂τ2
where C ∈ R n× n , D ∈ R n ×1 , K ∈ R1× n , U is a scalar control. 
Find vector x value at the time of the control U switch from ∂ y 12 ∂ y 12 
zero to minus (this time corresponds to half of the period): + ∆τ 3 + ∆T  = κ
∂τ3 ∂T 

x(T ) = V (T )x(0) + V (T − γT )C −1 (V (γT ) − E ) DA , (2)
 ∂ y12 ∂ y12
where V (t ) is the solution of equation x& = Cx ; E is the f W1 (0) −  y12 + ∆τ 1 + ∆τ 2 +
identity matrix of respective dimension. Formula (2) is  ∂ τ1 ∂τ2

valid if vector x is continuous on t∈ [0, T ] . Since the 
∂ y12 ∂ y12
solution is symmetric on period 2T, for periodic motion the + ∆τ 3 + ∆T  = −κ
∂τ3 ∂T 
following holds: x(0) = − x(T ) and the first component of 
the phase locus x1 (T , γ ) is:  ∂ y 23 ∂ y 23
f W2 (0) −  y 23 + ∆τ 1 + ∆τ 2 +
x1 (T , γ ) = −( E + V (T )) −1V (T − γT )C −1 (V (γT ) − E) DA .  ∂ τ1 ∂τ2

The second component x (T , γ ) can be determined as
2
∂ y 23 ∂ y 23 
+ ∆τ 3 + ∆T  = −κ
a solution of (2) at t=γT subject to control is U = A and ∂τ3 ∂T 

initial state is x (T , γ ) :
1
 ∂ y14 ∂ y14
f W1 (0) −  y14 + ∆τ 1 + ∆τ 2 +
x 2 (T , γ ) == V (γT )x 1 (T , γ ) + C −1 (V (γT ) − E ) DA .  ∂ τ1 ∂τ2

∂ y14 ∂ y14 
+ ∆τ 3 + ∆T  = κ . (3)
∂τ3 ∂T 


1750
 Here where δx ij - R-characteristic corresponding to periodic
y 12 = y12 (τ 1 , τ 2 , τ 3 , T ) τ = γ T ,τ = T , = − y 23 ,
1 0 0 2 0
τ 3 = (1+ γ 0 )T0 , T = T0
signal δU ij (t ) . According to (4):

y12 = y12 (τ 1 , τ 2 , τ 3 , T ) τ = − y14 , ∂x i δx ij

1 = γ 0 T0 ,τ 2 = T0 , = lim . (5)
τ 3 = (1+ γ 0 )T0 , T = T0 ∂τ j ∆τ j → 0 ∆τ j
∂yik yik (τ 1, τ 2 , τ 3 , T ) The analysis of increment δU ij (t ) proves the following
= , k = 1,4 .
∂τ j ∂τ j τ 1 =γ 0T0 ,τ 2 = T0 ,
τ 3 = (1+ γ 0 )T0 ,T = T0
formulas to be valid if the plant is linear:

Consider symmetric periodic functions U i (t ) ∂x1 ∂x 3 ∂x1 ∂x 2 ∂x 4 ∂x1 ∂x 3

=− , =− = , =− ,
∂τ 1 ∂τ 3 ∂τ 2 ∂τ 3 ∂τ 1 ∂τ 3 ∂τ 1
corresponding to R-characteristics where time t = 0 stands
for respective relay switching. For U i (t ) t = 0 stands for ∂x 2 ∂x1 ∂x 4 ∂x 2 ∂x 3 ∂x1 ∂x1 ∂x1
= + − , = + + ,
relay R2 switch from minus to plus. Asymmetry of control ∂τ 1 ∂τ 2 ∂τ 2 ∂τ 2 ∂τ 1 ∂τ 1 ∂τ 2 ∂τ 3
U can be caused by a variation of the time of any switch of ∂x 4 ∂x 2 ∂x1 ∂x 4 ∂x1 ∂x 3 ∂x 2 ∂x 4
either of the relays. Denote asymmetric control as follows: = − − , =− , =− . (6)
~ ∂τ 3 ∂τ 2 ∂τ 2 ∂τ 2 ∂T ∂T ∂T ∂T
U ij (t ) = U i (t ) + δU ij (t ) , Derive from (3) the following equalities with the use of
where U i (t ) is a symmetric periodic function; δU ij (t ) is a equations (6):
∆T = 0 , ∆τ 1 = ∆τ 2 − ∆τ 3 , (7)
control increment, which too is a 2T-periodic function and
If the input signal is not introduced the following
caused by the increment τ j . The typical control increment
equalities are valid:
is shown in Fig. 4
− y12 = κ , − y12 = −κ , − y23 = −κ , − y14 = κ .
U 1 (t )
Taking into consideration (7), rewrite equations (3) as:
A
 ∂ y 12 ∂ y 12   1 1 
 −  ∆τ 1 +  ∂ y 2 + ∂ y 2  ∆τ 2 = f W 2 ( 0 ) ,
t  ∂ τ1 ∂ τ 3   ∂τ2 ∂τ3 
-A    
 ∂ y12 ∂ y12   2 2
 −  ∆τ 1 +  ∂ y1 + ∂ y1  ∆τ 2 = f W1 (0) . (8)
δ U 11 ( t )  ∂ τ1 ∂ τ 3   ∂τ2 ∂ τ 3 
  
A The equivalent gains of the relays can be found as
quotients of averaged over the period control component U
γ 0 T0 t and averaged input signals of the relays y1 and y2 .
U1 U2
Fig. 4. Increment δ U 11 (t ) . K1 = , K2 = . (9)
y1 y2
Let us denote For U1 , U 2 , y1, y2 the following formulas are obviously
~
x 1j = x1 (τ 1,..., τ j + ∆τ j , τ j +1,..., T ) , true:
~ A( ∆τ1 − ∆τ 3 ) A∆τ 2
x 2j = x 2 (τ1,..., τ j + ∆τ j , τ j +1,..., T ) , U1 = , U2 = , (10)
2T0 2T0
~
x 3j = x 3 (τ 1,..., τ j + ∆τ j , τ j +1,..., T ) , y1 = f W1 (0) − U W1 (0) W (0) ,
~
x 4j = x 4 (τ1,..., τ j + ∆τ j , τ j +1,..., T ) . y 2 = f W 2 ( 0) − U W 2 ( 0 ) W ( 0 ) . (11)
By definition A∆τ 1
where U = .
~
x ij − x i T0
∂x i
= lim . (4) Considering (7), (9) – (11) the following formulas of the
∂τ j ∆τ j →0 ∆τ j
equivalent gains of the relays can be derived:
Since the superposition principle is valid for the linear A( 2∆τ 1 − ∆τ 2 )
K1 = ,
part of the system, the following can be set forth for R- f W1 (0) 2T0 − 2 A∆τ 1 W1 (0) W (0)
characteristics ~
xi j A∆τ 2
K2 = . (12)
~ f W2 (0) 2T0 − 2 A∆τ 1 W2 (0) W (0)
x ij = x i + δx ij ,

1751
 Variables ∆τ1 and ∆τ 2 can be obtained from (8). As a of up to the first order of small quantity – with the use of
result, the following final formulas of the equivalent gains equations (7):
 ∂ y12 ∂ y12   1 1
∆τ 1 +  ∂ y 2 + ∂ y 2  ∆τ 2 = b0 − ν 2 , (15)
can be obtained:
 −
A( 2d1 − d 2 )  ∂ τ1 ∂ τ 3   ∂τ2 ∂τ3 
K1 = ,    
d W1 (0) 2T0 − 2 Ad1 W1 (0) W (0) Consider propagation of signal f through the filter W1 ( s )
Ad 2 with the constant and ramp components being of interest,
K2 = . (13)
d W2 (0) 2T0 − 2 Ad1 W2 (0) W (0) with the use of Laplace transform.
where: g g
L−1 (( 1 + 0 )W1 ( s )) =
 ∂ y 1 ∂ y 12   ∂ y12 ∂ y12  2 s
s
d = 2 −  + − g g g
 ∂ τ 1 ∂ τ 3   ∂ τ 2 ∂ τ 3  = L−1 ( 1 W1 (0) + 1 C1 + 0 W1 (0) + Z ( s ))
2 s s
s
 ∂ y 2 ∂ y12   ∂ y 12 ∂ y 12  Here, Z(s) is the Laplace transform of the motion
− 1 −  + ,
 ∂ τ 1 ∂ τ 3   ∂ τ 2 ∂ τ 3  component vanishing with time. It will not be taken into
account below. Factor C1 can be determined through the
 ∂ y 2 ∂ y12   ∂ y1 ∂ y12 
d1 = W2 (0)  1 +  − W1 (0)  2 + , g1
representation of W1 ( s ) as partial fractions and taking
 ∂ τ 2 ∂ τ 3   ∂ τ 2 ∂ τ 3  s2
 ∂ y 1 ∂ y 12   ∂ y 2 ∂ y12  g1
d 2 = W1 (0)  2 −  − W 2 ( 0)  1 − . the factor at . Obviously:
s
 ∂ τ 1 ∂ τ 3   ∂ τ 1 ∂ τ 3 
g g
L−1 (( 21 + 0 )W1 ( s )) = W1 (0) g1t
s s
C. Input-output analysis of type one relay servo system +C1 g1 + W1 (0) g 0 + Z (t ) . (16)
Let us consider input-output analysis of the system if the Reasoning along similar lines, consider the input signal
plant contains one zero pole. Let the system be as depicted f(t) propagation through the filter W2 ( s ) . Therefore, the
in Fig. 5
following equations are valid with (15) and (16) taken into
account:
y1 U1 b0 = W2 (0) g 0 + C2 g1 ,
W1(s)
a0 = W1 (0) g 0 + C1 g1 . (17)
R1 U ~
f x g1
x W0(s) Here, C2 is the factor at in the partial fraction
s
R2 representation (similar to C1 ). In accordance with (14) and
y2 U2 k x1 (16),
W2(s)
s W1 (0) g1 = k1U ,
W2 (0) g1 = k2U , (18)
Fig. 5. Type one control system.
where k1 and k 2 are the gains of the integrators in partial
Let the signal U be k
fraction representations of (W0 ( s ) + )W1 ( s ) and
0 s
U (t ) = U (t ) + U ,
k
where U 0 (t ) is a symmetric 2T-periodic function (Fig. 2). (W0 ( s ) + )W2 (s ) respectively.
s
If control U is asymmetric the following formula is valid:
A∆τ 1
x1 (t ) = kU t + ν + x10 (t ), (14) Since U = , then
T0
where ν is a constant value determined by the initial
k1 A∆τ 1
conditions, x10 (t ) is a 2T-periodic function with zero g1 = ,
2T0W1 (0)
average over the period value. Asymmetric periodic motion
k2 A∆τ1
may exist in the system if the input signal f (t ) = g1t + g 0 g1 = . (19)
is introduced. Let in the steady oscillatory process the ramp 2T0W2 (0)
input applied to the relays be: Considering (17), (19), and the following equations:
L−1 ( f ( s )W1 ( s )) = a1t + a0 , L−1 ( f ( s )W2 ( s )) = b1t + b0 . ν 1 = W1 (0)ν , ν 2 = W2 (0)ν ,
Write the conditions of relay switches with an accuracy the set of equations (15) can be rewritten as:

1752
  ∂ y12 ∂ y12   ∂ y1 ∂ y12  transfer functions W1* ( s ) and W2* ( s ) are represented as
 −  ∆τ 1 +  2 +  ∆τ 2 −
 ∂ τ 1 ∂ τ 3   ∂ τ 2 ∂ τ 3  partial fractions. Obviously the following statements are
n ∂ xi i
∂ y1i l , where ∂ x l
C k A∆ τ 1 i m ∂ xi
l , ∂ y2
− 2 2 = W2 ( 0)( g 0 − ν ), true: =∑ =∑ is the
2T0W2 ( 0) ∂ τ j l =1 ∂ τ j ∂ τ j l =1 ∂ τ j ∂τ j
 ∂ y12 ∂ y12   ∂ y 2 ∂ y12  derivative corresponding to a particular fraction; n , m the
 −  ∆τ 1 +  1 +  ∆τ 2 −
 ∂ τ 1 ∂ τ 3   ∂ τ 2 ∂ τ 3  number of partial fractions in transfer functions W1* ( s ) and
C k A∆τ 1 ∂ x il
− 11 = W1 (0)( g 0 − ν ). (20) W2* ( s ) respectively. Derivatives can be obtained
2T0W1 (0) ∂τ j
from a variational equation pertaining to a particular partial
Let us denote q = g 0 − ν . The equivalent gains of the fraction.
relays are determined as above (formulas (9)) but y1 and Let us obtain the formula of ∂ x1 ∂ τ 1 for transfer
y2 are now equal to: k
~ function W ( s ) = . Variational equation for this
y1 = qW1 (0) − W1* (0)U + C1 g1 , s+a
~ element is:
y 2 = qW2 (0) − W 2* (0)U + C 2 g1 .
d δx11
~ ~ + a δx11 = k δU 11 (22)
Here, W1* ( s ) and W 2* ( s ) are the partial fraction dt
k Increment δ U 11 (t ) is depicted in Fig. 4. Control
representation of (W0 ( s ) + )W1 ( s ) and
s
variation δ U 11 (t ) is approximately equal to δ-function:
k
(W0 ( s ) + )W2 ( s ) respectively with the integrators being
s δU11 (t ) = A ∆τ1 δ (t − τ1 ) . Equation (22) has the following
not considered. Now we can write the formulas of the solution:
equivalent gains of the relays: δx (t ) = δx (0) e − a t + kAe − a ( t −τ 1 ) ∆τ 1 , 0 ≤ t ≤ 2T . (23)
A( 2d1 − d 2 )W1 (0) In a periodic motion δx ( 2T ) = δx (0) . Then according to
K1 = ~ ,
2( d (W1 (0)) T0 − Ad1 W1* (0)W1 (0) + C1k1 Ad1 )
2 (23), the following formula of derivative of R-characteristic
can be obtained:
Ad 2W2 (0)
K2 = ~ . (21) 0 0
Ad1 W2* (0) C2 k2 Ad1 ∂ x1 kAeT aγ
2
2W2 (0) ( d T0 − + ) = .
(W2 (0)) 2 (W2 (0)) 2 ∂ τ 1 e 2T 0 a − 1
In formulas (21): To obtain the formula of the derivative of R-
characteristic for the element with transfer function
 ∂ y 1 ∂ y 12 C 2 k 2 A   ∂ y12 ∂ y12 
d= 2 − −  + − Cs + D
W ( s) = , this transfer function can be
 ∂ τ 1 ∂ τ 3 2T0W2 (0)   ∂ τ 2 ∂ τ 3 
s + 2αs + α 2 + β 2
2

 ∂ y 2 ∂ y12 C1k1 A   ∂ y 12 ∂ y 12  represented as a sum of two transfer functions with

− 1 − −  + , complex conjugate coefficients.
 ∂ τ 1 ∂ τ 3 2T0W1 (0)   ∂ τ 2 ∂ τ 3 
Formulas of the derivatives R-characteristics for various
 ∂ y 2 ∂ y12   ∂ y 1 ∂ y 12  elements are presented in Table 1. Note that the table
d 1 = W 2 ( 0)  1 +  − W1 (0)  2 + , provides not all the derivatives but only those, which are
 ∂ τ 2 ∂ τ 3   ∂ τ 2 ∂ τ 3  necessary for calculation of the equivalent gains of the
 ∂ y1 ∂ y12 C2 k 2 A  relays.
d 2 = W1 (0)  2 − − − Table 1 has the following notation
 ∂ τ 1 ∂ τ 3 2T0W2 (0) 
. L1 = e Tα ( 2+γ ) sin Tβ ( 2 − γ ) + e Tαγ sin Tβγ ,
 ∂ y 2 ∂ y12 C1k1 A 
− W2 (0)  1 − −  L2 = e Tα ( 2+γ ) cos Tβ ( 2 − γ ) − e Tαγ cos Tβγ ,
 ∂ τ 1 ∂ τ 3 2T0W1 (0) 
M 1 = e Tα ( 3+γ ) sin Tβ (1 − γ ) + e Tα (1+γ ) sin Tβ (1 + γ ) ,
M 2 = e Tα ( 3+γ ) cos Tβ (1 − γ ) − e Tα (1+γ ) cos Tβ (1 + γ ) ,
D. Derivatives of R-characteristics
To calculate the equivalent gains of the relays it is N 1 = e Tα ( 3−γ ) sin Tβ (1 + γ ) + e Tα (1−γ ) sin Tβ (1 − γ ) ,
necessary to know the derivatives ∂ x i ∂ τ j . Suppose that N 2 = e Tα ( 3−γ ) cos Tβ (1 + γ ) − e Tα (1−γ ) cos Tβ (1 − γ ) ,

1753
 P1 = e Tα ( 4−γ ) sin Tβγ + e Tα ( 2−γ ) sin Tβ ( 2 − γ ) , W2 ( s ) =
2
, A = 2 , κ = 0.02 .
0.8s + 1
P2 = e Tα ( 4 −γ ) cos Tβγ − e Tα ( 2 −γ ) cos Tβ ( 2 − γ ) . In the autonomous mode, stable self-excited oscillations
of period T0 = 0.7 and parameter γ 0 = 0.58 exist in the
TABLE 1.
W (s ) k system. The equivalent gains of the relays calculated with
k
the use of the above formulas are:
s s+a
K1 = 0.24 , K 2 = 0.636
∂x 1 kA γ − 1)
( kAeTaγ Simulation of the system via integration of the original
∂τ 1 2
e 2Ta − 1 differential equation (DE) with successive Fourier analysis
was done for the proposed method accuracy assessment.
∂x1 0 kAeTa Both frequency responses: calculated on the basis of the
∂τ 2 developed approach and through DE solution are presented
e 2Ta − 1
in Table 2.
∂x1 −
kAγ kAeTa (1+ γ ) TABLE 2
∂τ 3 2 e 2Ta − 1 Frequency, Proposed approach DE solution
∂x 2 kA kA[eTa + eTa ( 2 −γ ) − eTa (1−γ ) ]
s −1
φ ,o А φ ,o А
∂τ 1 2 e 2Ta − 1
0.1 -1.9 0.751 -1.879 0.752
∂x 2
−
kAγ kAeTa (1−γ ) 0.2 -3.92 0.75357 -3.85 0.7538
∂τ 2 2 e 2Ta − 1 0.3 -5.891 0.75631 -5.803 0.7562
∂x 2 kAeTa 0.4 -7.718 0.76015 -7.879 0.7594
0 −
∂τ 3 e 2Ta − 1
W (s ) The proposed method does not require the filter
Cs + D
hypothesis to be valid, unlike the describing functions
s 2 + 2αs + α 2 + β 2 method. Moreover, the actual shape of the oscillations is
A[( D − cα ) L1 + сβL2 ] taken into account when deriving the formulas of the
∂x 1
equivalent gains.
∂τ 1 β ( e 4Tα − 2e 2Tα cos 2Tβ + 1)
∂x 1 A[( D − cα )( e3Tα + eTα ) sin Tβ IV. CONCLUSION
4Tα 2Tα
+
∂τ 2 β (e − 2e cos 2Tβ + 1) A method of input-output analysis of decentralized relay
systems is proposed above. Although the proposed
+ сβ ( e3Tα − eTα ) cos Tβ )] approach results in more complex final formulas and the
β ( e 4Tα − 2e 2Tα cos 2Tβ + 1) analysis procedure compared to one relay case, it still
∂x1 A[( D − Cα ) M 1 + сβM 2 ] remains reasonably simple and provides many advantages
over the describing function method.
∂τ 3 β ( e 4Tα − 2e 2Tα cos 2Tβ + 1)
∂x 2 ∂x1 ∂x 4 ∂x 2 REFERENCES
+ −
∂τ 1 ∂τ 2 ∂τ 2 ∂τ 2 [1] N. Faldin, N. Panferov, I. Boiko, “Input-output Analysis of Dead-
Zone Relay Control Systems,” Proc. of 2001 American Control
∂x 2 A[( D − cα ) N 1 + CβN 2 ] Conference, Arlington, VA, USA, pp. 1507-1512, 2001.
[2] S. A. Rudnev, N. V. Faldin. "Linearization of Relay Servo System
∂τ 2 β ( e 4Tα − 2e 2Tα cos 2Tβ + 1) with Respect to Control Signal". Russian Ac. Sc. Transactions on
Control and Control Systems, № 2, 1998, pp. 36 - 43 (in Russian).
∂x 2 ∂x1 [3] I. Boiko. "Input-output Analysis of Limit Cycling Relay Feedback
∂τ 3 −
∂τ 2 Control Systems". Proc. of 1999 American Control Conference, San
Diego, California, USA, pp. 542 - 546, 1999.
∂x 4 A[( D − cα ) P1 + сβP2 ] [4] Ya. Z. Tsypkin. Relay Control Systems. Cambridge, England, 1984.
4Tα 2Tα [5] D. P. Atherton. Nonlinear Control Engineering - Describing Function
∂τ 2 β (e − 2e cos 2Tβ + 1) Analysis and Design. Workingham, Berks, UK: Van Nostrand
Company Limit
[6] Z.J. Palmor, Y. Halevi, T. Efrati. “A General and Exact Method for
III. EXAMPLE OF INPUT-OUTPUT ANALYSIS Determining Limit Cycles in Decentralized Relay Systems”.
Let the transfer functions of system be: Automatica, 31 (9), pp. 1333-1339, 1995.
2
W0 ( s ) = , W1 ( s ) = 1 ,
(0.5s + 1)(0.04 s 2 + 0.2 s + 1)

1754

View publication stats