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

net/publication/4141885

Relevance of actions and measurements in control performances

Conference Paper · August 2004

DOI: 10.1109/ASCC.2004.184859 · Source: IEEE Xplore

CITATION READS
1 34

1 author:

P. Albertos
Universitat Politècnica de València
384 PUBLICATIONS   3,805 CITATIONS   

SEE PROFILE

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

Profesion@l: Equilibrio de género en el espacio europeo View project

DGIP 23.15.26 - Control de un manipulador robótico continuo View project

All content following this page was uploaded by P. Albertos on 31 May 2014.

The user has requested enhancement of the downloaded file.

 Relevance of Actions and Measurements in Control Performances
Pedro Albertos

Universidad Politécnica de Valencia

Dept. of Systems Engineering and Control, P.O.Box. 22012
E-46071 Valencia, Spain. Fax: +34 96 3879579
e-mail: pedro@aii.upv.es

Abstract different from that of the uncontrolled system, the delay in

applying the control, or even the lost of control will not
In this paper, the relative relevance of input and output damage the control performances. On the other hand, if by
variables in achieving the control performance is analysed. applying the control, the expected closed loop behaviour is
In order to compare different control structures, the much faster than that of the open loop system, small
performances of the closed loop controlled system are changes in the control may result in large changes in the
expressed by the poles’ position. Attached to each input performances. This is even worst if the open-loop system is
(output), the control effort (control contribution) is defined unstable.
as the inverse of the maximum admissible delay in its path
without provoking instability. This delay, that can be The relation between control delays and control effort, in
roughly estimated by the shift of the open-to-closed loop the frame of single-input-single-output (SISO) systems was
poles under the variable connection, is defined as the already analysed in [2] and it was also pointed out that,
variable relevance in the control structure. As digital control under some specific circumstances, an additional delay may
implies open-loop control in the intersampling time improve the bandwidth of the closed loop system. These are
interval, there is always a time delay in the updating of the the so called phase conditionally stable systems [3].
control. Small time delays in the most relevant variables Time delays can be present in the process and/or in the
may result in unstable behaviour. actuators, and sometimes in the sensor instrumentation
Keywords: Time delays, control performance degrading, when the measurement requires some sort of transportation.
control effort, output control contribution, interaction, Some additional sources of unavoidable time delay appear
controller implementation. when digital controllers are used: analogue to digital
conversion, control algorithm computation, D/A conversion
and control signals interfacing as well as digital actuators
1 Introduction updating. Again, it is usual to consider the controller to be
designed for the ideal sampled data process, without taking
Dealing with multivariable control systems, the control is into account these delays.
composed by a number of elements leading to control
actions applied to a number of variables, either centralised In large-scale real-time systems design, all the mentioned
or decentralised, working together to achieve the global above delays plus the worst case execution time of each
control goals. The design is based on the availability of a control algorithm must not exceed the maximum allowed
predefined set of input and output variables and each delay to avoid loop instability.
variable contributes to the fulfilment of the control When digital controllers are implemented on
requirements. microprocessors and the processes to be controlled are very
The failure of an actuator or a measurement device, or even fast, the sampling period must be shortened and these
the temporal or permanent lost of communication between delays may degrade the controlled system performances if
some control and process variables, may result in a lost of they were not considered at the design stage. This is what
basic properties such as the global system stability. If this is happens, for instance, in the case of robot controllers where
the case, there is no control integrity in the design, [1]. some axis dynamics must be sampled every few
milliseconds and complex control algorithms must be
In the implementation of the control, in a digital system, implemented. A similar situation happens in the control of
there are unavoidable delays due to the physical constraints open loop unstable systems.
in the computation of the control actions or in the
communication network. These delays may be also the In multivariable control systems, the controllers applied to a
source of a degrading of performances and even provoke number of variables, either centralised or decentralised, are
the instability. implemented in the same computer. The priority in the
execution of a task highly influences the possible delay, [4].
The effect of these delays strongly depends on the strength Thus, the relevance of each control variable should be
of the control. If the purpose of the control is not very considered.
exigent, that is, if the controlled system dynamic is not very
Of course, control performance degradation is always That is, the control effort is proportional to the cut-off
present when an ideally designed controller is applied to a frequency and inversely proportional to the phase margin of
real plant. But the main problem is that in controlled the loop gain.
systems that seem to be closed-loop well damped and stable
Using a proportional controller, as far as the controller gain
may present stability problems if significant time delays
is increased, the cut-off frequency is increased and the
happen. The point here is that if a great control effort has
phase margin is decreased. That means, the control effort is
been used to get the nominal control, very short time delays
increasing and, as a consequence, the admissible delay in
become significant.
the control loop decreases.
The main purpose of this paper is to introduce the concept
of variable’s relevance, related to the control effort
Example 1. As an example, let us consider the plant
contribution of each variable in the control structure to
achieve the control performances. Also, the relation of this
relevance to the change in the process dynamics, expressed 20
G1 ( s) (7)
by the shift of the poles in closing the loop. To better isolate s ( s  3)( s 2  2s  6)
and characterise these effects, a pole placement by state
feedback control is initially assumed. Using a proportional controller with K=0.2, the Bode plot is
The paper is organised as follows. First, the problem is as in fig. 1, where the cut-off frequency is Z c =0.223 rad.s-1
stated for continuous time (CT) SISO systems. The basic and the phase margin is \ m = 81.47º = 1.42 rad.
result, that is, the relation between the control effort and the
maximum allowed time delay in the loop, is recalled. Then,
the same approach is developed for MIMO systems and the
measurement influence is also analysed. A number of
examples are included to illustrate the main result. The
conclusion section remarks this result.

2 The control effort in SISO systems

Consider the time delay effect in SISO CT systems. If the

process model is expressed by the transfer function
y (s)
G p ( s) (1)
u ( s)
and the two degree of freedom controller is represented by
u ( s ) f ( s )r ( s)  k ( s) y ( s) (2)
where r(s) is the external reference, the loop-gain is given Figure 1: Bode plot of the G1(s) P(K = 0.2)-controlled
by plant
G ( s) G p ( s)k ( s ) (3) The control effort is:
Let us assume that the loop phase margin is \ m . This 0.223
CE 0.157 ; o ' 6.37 s
means that the maximum admissible time delay, ', to keep 1.42
the system stable is such that
This is the maximum delay without reaching instability. If
'.Zc \ m (4) the gain is increased, for instance, K = 0.6, the control effort
becomes CE= 1.6, and the maximum delays is reduced to
where Z c is the cut-off frequency, that is, the maximum
0.67 s.
frequency such as
2.1 State feedback control
G ( jZ c ) 1 (5) Now, consider the same issue if the controller is designed
For monotonically damped systems (no resonance modes) and implemented by state feedback. Assume that the
this frequency is usually smaller than the system bandwidth. process is modelled by the tern (A,b,c), where A is the
system matrix dim nxn, b is the matrix input, dim nx1, and
Definition 1. The control effort in a loop is defined as the c is the output matrix, dim 1xn:
inverse of the maximum admissible time delay:
x( t ) Ax( t )  bu( t ) ; y (t ) cx(t ) (8)
1 Zc
CE ' (6) Without any loss of generality, a controllable canonical
\m
form, [5], is chosen for the tern (A,b,c). Denote by A(s), the
characteristic polynomial of A,
 n n K ( s ) P( s )  A( s )
n (14)
A( s ) sI  A – ( s  ai ) s n  ¦ A1 .s i 1 ; An ¦ ai (9)
i 1 i 1 i 1 are the entries of the feedback matrix -k, in (10), with
n
where ai , i 1, n are the open-loop poles of the system, Kn  Pn  An ¦ ai  pi .
and An is the s n 1 coefficient. i 1

The loop frequency response is given by

If, as a result of a pole assignment or any optimisation
approach, a linear state feedback control law is designed, K ( jZ )
such as: G ( jZ )
A( jZ )
(15)
u (t ) kx(t )  r (t ) (10) K n ( jZ ) n 1    K1
G ( jZ )
the controlled system will be, nominally: ( jZ ) n  An ( jZ ) n 1    A1
x (t ) ( A  bk ) x(t )  br (t ) (11)
Denote by This representation will give some insight in the relation-
ship between the control effort and the poles shifting.
n n
2.2 Control effort and poles’ shifting
P( s ) sI  A  bk – (s  pi ) ; Pn  ¦ pi
(12) If a delay of ' units of time in the control action, (10), is
i 1 i 1
assumed, the complete system will be
the closed-loop characteristic polynomial, where x (t ) Ax(t )  bkx(t  ')  br (t  ') (16)
pi , i 1,  n are the closed-loop poles of the controlled
system, and Pn is the s n 1 coefficient. The loop transfer function is shown in fig. 3, being

G' ( s ) k ( sI  A) 1 b.e  's (17)

A1 r s + us  1 x1 s 
 's
e
An As 
-
-
r s  u s  -
1 1 1
+ +
s s s
-
x n s 
Ks 
x2 s  x1 s 

+ Figure 3: State feedback delayed control

Kn
+ Assume that Pn !! An , that is, the closed loop poles have
K1
been shifted well on the left, to improve the time response
a) of the system. In order to estimate the phase margin of the
controlled system (11), taken into account (5) and (15), it
r s + u s  1 x1 s  yields
As 
- Kn
G ( jZc ) 1, o 1 # o Zc # K n  Pn  An (18)
jZ c
Thus, the control effort is proportional to the poles’ shifting,
Ks  from the open to the closed loop system, that is,
b) Kn An  Pn
CE | (19)
Figure 2: a) State feedback control; b) Control loop Mm Mm

Remark 1: As the phase margin is usually desirable to be

As seen in fig. 2.a, the state feedback control may be easily around 1 rad, the maximum admissible time delay is around
expressed by a loop transfer function, as shown in fig. 2.b, the inverse of the poles’ shifting.
K( s )
G( s ) k ( sI  A ) 1 b (13) Example 2. Let us consider the state feedback design for
A( s )
the plant G1(s) in the Example 1. If the closed loop poles
where the coefficients of the polynomial
are desired to be assigned at p = [- 4+2j, -4-2j, -8, -10], the Assume a dynamic feedback through the system:
state feedback law for the canonical controllable state
variables is x̂( t ) ( A  lc ) x̂( t )  bu( t )  ly( t )(20)
where the design matrix (row vector) l is computed to
k = [1600 982 232 21] assign the observer’s poles in order to get a good tracking
of the observed state, x̂( t ) [5]. If the control law (10) is
The free response of the system output, from x0T=[1 0 0 0] now:
is depicted in fig. 4, curve a. Note that
u( t )  kx̂( t )
K4 P4  A4 26  5 21
the feedback controller transfer function is:

Thus, the total poles’ shifting is 21. The Bode plot for the u( s )
k( s ) k ( sI  A  lc  bk ) 1 l (21)
open-loop transfer function (13), [G(s)= k ( s  IA) 1 b ], y( s )
presents a cut-off frequency Z c = 21.6 rad.s-1 and the phase Observe than, in this case, there are two design parameters:
margin is \ m = 74º = 1.3 rad. That means a maximum time k to assign the closed loop poles and l to define the
observer’s poles.
delay of 60 ms, close to the inverse of the pole shifting. The
same response, if a delay '=65 ms is included, is also If either the measurement or the control action are delayed,
plotted (curve b), showing an unstable behaviour. to evaluate their relevance, the maximum admissible delay
in the loop transfer function should be computed. Taking
1 into account (21) and (13), this transfer function is:
0.8
a
a: state feedback G( s) k ( s).G ( s) k ( sI  A  lc  bk ) 1l ˜ c( sI  A) 1b (22)
b
b: id. with delay
0.6
c: o utput feedback
0.4 d: id. with delay Example 3. Let us consider again the system in the previous
example. A state observer is included such that its poles are
free res ponse

0.2
located at: po = [-9,-10,-15,-20]. The observer gain, l (such
0 that eig(A-lc)= po), is:
b
c
-0.2
l = [-49 -798 -4254 4728]T
-0.4

-0.6 d
The output response, starting from the same initial
conditions is as depicted in fig. 4.c. The observer initial
-0.8
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
state is null, and due to the initial error in the state, there is a
tim e worse transient.

Figure 4: Free response of state and output feedback The loop transfer function is now given by (22). Again, if a
control of the process in Example 1 delay of '=60 ms is introduced, the response starts to
oscillate, as shown in fig. 4.d.
Remark 2: Although the system was defined as a SISO
system, (7), the feedback includes not only the output, and Remark 3: In this case, the delay may be considered in the
the full state is assumed to be measured (see figure 2.a). In data acquisition system (delay in the output), in the control
order to consider the relevance of each measurement, if the updating (delay in the control) or in the communication
control schema is open at any measurement point, the loop channels (any of them or both). The effect is always the
gain should be computed and the maximum admissible same.
delay should be determined. Note that the poles’ shifting is Remark 4: In the general case, the concept of control effort
null except in the case the x4 state variable feedback is cannot be only connected with the poles shifting. In fact, for
open. a proportional controller, k(s) = k, as shown in the Example
The delay in the measurement may be analysed in a more 1, there is no pole shifting but it is clear that as far as the
general framework. gain increases the allowed time delay decreases and, thus,
the control effort should increase. See also Remark 2.
2.3 Output feedback control
Let us now consider a dynamic output feedback controller,
k(s) being its transfer function. It is assumed that this 3 MIMO systems
controller has been designed to produce a similar effect to
that of the state feedback pole assignment, through an In the control of MIMO systems, the pole placement is
implicit observer or filter. Thus, an extra-degrading of the achieved by the feedback matrix K, measuring the state and
performances should be expected. applying simultaneously the vector of control actions. As
before, a state estimator or observer can be used if a
reduced number of variables is measured. The whole sensors and actuators being active and without any delay.
control structure is as depicted in fig. 5. To evaluate this relevance, assume a delay is introduced in
between zi and yi (fig. 5). The maximum admissible delay is
computed from the open-loop transfer function between yi
Process and zi. This transfer function is expressed by:
uj yi
yi ( s )
i c ( sI  A  BK i ) 1 ki (27)
.... Hj .... .... Si .... zi ( s )
where ic represents the i-row of the output matrix, C.
vj Control zi
Computing the loop phase margin \ m as well as the cut-off
frequency, Z c , the measurement relevance is evaluated by
ref (6).

Figure 5: MIMO controlled plant In a similar way, for a generic input, like uj, its relevance is
characterised by the control effort, that is, the inverse of the
There are a number (p) of measurements, zi, gathered from maximum admissible delay in updating the control signal
the outputs through the respective measurement device and without instabilising the system. In order to evaluate the
sampler, Si, as well as (m) control actions vj updated by relevance of each control input, the remaining control
means of the hold devices, Hj. actions are assumed to be closed and the path related to this
The system model is given by: variable is open. In this case, the transfer function to be
computed is:
x (t ) Ax (t )  Bu (t ) ; y (t ) Cx (t ) (23)
v j( s )
Let us assume state feedback, that is, C = I, the control law j k ( sI  A  B . j K )1b j (28)
being: u j( s )
And, again, by (6), the variable relevance is computed.
u (t )  Kx (t ) (24)
The closed loop control system is defined by the tern Example 4. Let us consider a typical exothermic reactor as
( A  BK , B , C ). Adopt the following notation: described, for instance, in [6], [7]. The layout, as depicted
in fig. 6 includes two actuators controlling the reactant
flow, F(t), and the cooling water flow, Fj(t). It also includes
ª 1k º three sensors providing the measurement of the reactor and
K >k1 k2 ... kn @ «  » (25) the cooling jacket temperatures, T and Tj respectively, as
« »
¬ m k¼ well as the main product concentration, Ca. These variables
are chosen as state variables. Thus, the input and state
where ki is the i-column of K, and jk is the j-row of K.
vector are, respectively:
Similarly, bi is the i-column of the input matrix B, and so
on. Also denote by
x >C a T Tj @T ; u >F Fj @T (29)
Kj >k1 k 2  k j 1 0 k j 1  k n @ (26) A linearised model of the reactor, as described in the
to represent the K matrix with null j-column and references, is given by (23), where:

ª 1k º ª- 1.705 - 0.2519 0 º ª 2.918 0 º

«  » «
A « 23.088 - 28.71 » «
20.9 » ; B « - 28.6 0 »»
« » «¬ 0 200.3 - 216.89 »¼ «¬ 0 - 415.29»¼
iK « 0 »
«  » and C is the unity matrix. The poles of the open-loop
« » system (23) are:
¬mk¼
^a i ` ^eig ( A)` ^2.5878,7.73,236.987` ; An 247.3
to represent the K matrix with null i-row.
The relevance of each variable is defined in the following Assume the requirement of assigning the closed loop poles
way: at p = {-320, -340, -360}. That is: Pn 1020 .

A possible control law is:

Definition 2. The relevance of a variable is the inverse of
ª 858.5 68.5676 4.6683 º
the maximum admissible time delay in its transmission path u (t ) Kx (t ); K «  40.463  4.2238  0.5505 » (30)
without reaching the instability. ¬ ¼
If the u1 input is closed, the poles of the loop gain for the
For the measurement yi, its relevance is determined by the
second input are:
inverse of the maximum admissible delay in the sensor Si
without making unstable the whole system, while the other ^eig( A  b1 .1 k )` ^ 405.7 ,336.1,49.5`;
 S2 405.7  336.1  49.5 791.4 3.1 Real-time control of MIMO systems
The implementation of a MIMO control system in a single
Thus, the pole shifting in closing the u2 path is: computer requires the adequate scheduling of the different
S2 – Pn= 228.7 tasks. A priority is assigned to each task. According to their
priority, different delays appear in the execution of each
task, [8]. In order to minimise the degrading of control
performances, the highest priorities should be assigned to
the most relevant tasks. Otherwise, under similar delays, the
degrading of the control loops with higher control effort
will be more noticeable.

4 Conclusions

The concept of relevance for control attached to a variable

of the control system has been introduced and discussed. A
direct measure of the relevance of a variable has been
established in connection with the maximum admissible
delay in the transmission of this signal, either from the
Figure 6: Exothermic reactor process to the control (measurement) or from the control to
the process (control action).
In order to evaluate the relevance of a measurement, in this
case one of the state variables, the corresponding column of A close connection of the relevance of a variable to the
K should be considered null to compute the open-loop poles’ shifting due to its existence has been suggested and
poles. According to (26), denoting by verified as valid for special cases, like the state feedback
control.
K1 >K @k 1 0 >0 k2 k3 @
The concept is applicable in SISO and MIMO systems,
and the open-loop poles for x1 are: leading to some guidelines in designing digital control
systems implemented on a single CPU and in the real time
^eig ( A  BK )` ^0.7
1 1942 .9  0.4585`;
scheduling of the tasks.
Thus, the pole shifting in closing this path is: References
1S – Pn= 2963.6
[1] Grossdidier, P., Morari, M., and Holt B.. “Closed-loop
Proceeding in this way, the result is summarised in Table 1. properties from steady state gain formulation”. IEC
Fund. 24 221-235. 1985
In this Table 1, in order to compute the relevance of each
variable, the open-loop poles are computed (o-l poles), and [2] Albertos, P. and Olivares, M. “Time Delay
their sum evaluated (S). The total shift if the loop is closed Limitations in Control Implementation.” European
at this variable is computed as the difference between S and Control Conference. 1999.
the sum of the closed-loop poles. This figure indicates the
[3] Albertos, P. “Phase conditionally stable systems”
relevance of each variable. The computation of the
Asian Control Conference. Singapore, 2002.
maximum admissible delay for each loop will give the
precise measurement of the variable’s relevance. [4] P. Albertos and A. Crespo. Real-time control of non-
uniformly sampled data systems. Control Engineering
Table 1. Relevance of variables in MIMO control system
Practice, 7:445-458, 1999.
Variable u2 u1 x1 x2 x3
[5] Ogata, K. Modern Control Engineering. Prentice-Hall,
o-l poles -405.7 0.2 0.7 -2513.3 -340.9 4th edition, 2002.
-336.1 -135.6 1492.9 –340.5 –225.2
-49.5 -340.6 –0.4585 –127.2 ±179.5i [6] Marlin, T.E. Process Control. Chem. Eng. McGraw
Hill, Singapore, 2nd edition, 2000
S=Sum of
poles -719.4 -475.9 1485.1 -2981 -719.4 [7] Perez M. and Albertos. P. “Self oscillating and chaotic
Relevance behaviour of a PI controlled CSTR with control valve
228.7 544 2963.6 -1961 228.7 saturation”. Journal of Process Control, Vol. 14.pp
= S-Pn
Thus, it appears clearly that the measurement of the state 51-59. 2004.
variable x1 is the most relevant activity, followed by the [8] Albertos P., Crespo A., Ripoll I., Vallés M., Balbastre
updating of the first control variable. Their failure will P., “RT control Scheduling to reduce control
provoke the system instability. On the other hand, the performance degrading”. 39th IEEE Conference on
integrity is assured with respect to the rest of variables, one Decision and Control 2000. December 12-15. Sydney.
by one.

View publication stats