Professional Documents
Culture Documents
net/publication/3169028
Magnitude and symmetric optimum criterion for the design of linear control
systems: What is it and how does it compare with the others?
CITATIONS READS
102 337
2 authors, including:
M. Safiuddin
University at Buffalo, The State University of New York
64 PUBLICATIONS 175 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by M. Safiuddin on 26 January 2016.
Abstruct- The magnitude and symmetric optimization techniques are pole placement and the minimization of the integral time ab-
two related methods for designing “optimal” linear control systems in solute error (ITAE) are briefly evaluated. To determine the
the frequency domain. Both methods have been used in industry for worth of both the magnitude and symmetric optimum, the re-
some time, yet they do not appear to have been well-documented in the
English control engineering literature. The developments of the criterion
sponses of typical systems designed with these optima will
are examined, include general solutions for typical controllers. As a be compared to systems designed with pole placement and
basis for comparison, time domain response optimization techniques ITAE. Finally, control parameter solutions for common plant
such as minimizing the integral time absolute error, along with state and controller combinations are presented.
space techniques, i.e., pole placement, are briefly reviewed. Personal
computer simulation results of a normalized position control system are 11. DEVELOPMENT
OF MAGNITUDE [2]
OPTIMUM
given to illustrate these methods.
One possible objective in the design of a control system is
that the controlled system’s output should exactly and instan-
I. INTRODUCTION taneously reproduce its input. That is, the system’s transfer
function should be unity, i.e.,
HE DESIGN of industrial controllers is based on two
T tasks: determining the structure of the controller and ad-
justing the controller’s parameters to give an “optimal” sys-
C ( s ) / R ( s )= 1.
tem performance. As an aside, this design process is normally In other words, the system should be presentable on a Bode
done with “complete knowledge” of the plant. Furthermore, gain versus frequency diagram with a 0-dB gain of infinite
the plant is normally described by a linear time-invariant con- bandwidth and zero phase shift. In practice this is not possible
tinuous or discrete time model. The structure of the controller since every system will contain inductive- and capacitive-type
is chosen such that the system’s response can meet certain components that store energy in some form. It is these ele-
qualitative criteria; type I, 11, or 111 behavior, stable response, ments and their interconnections with energy dissipative com-
appropriate disturbance handling capabilities, etc. In the field ponents that produce the system’s dynamic response charac-
of control engineering, techniqueshave evolved to find an “op- teristics. Such systems reproduce some inputs almost exactly,
timal” set of controller parameters for a given plant and con- while other inputs are not reproduced at all. Signifying that
troller combination. These techniques include pole placement, the system’s bandwidth is less than infinite.
minimization of the integral of a function system error, and Once it is recognized that the system’s dynamics cannot
optimal control. Two related techniques that have not received be ignored, a new design objective is needed. One possible
much attention in the English control engineering literature, design objective is to maintain the magnitude response curve
with the exception of [ 2 ] , are the magnitude and symmetric as flat and as close to unity for a large a bandwidth as possible,
optimum. On the other hand, these techniques appear to be for a given plant and controller combination. This technique is
well-documented in the German literature [ 13. called the magnitude optimum. The magnitude optimum is an
This paper reviews the development and principles of both optimization technique, since it helps to determine the “best”
of these techniques. Furthermore, as a basis for comparison, controller coefficients for a given controller configuration
that previously had been determined to be best able to control
Paper IUSD 88-24, approved by the Industrial Control Committee of the a fixed plant. The fixed plant is assumed to have an mth-order
IEEE Industry Applications Society for presentation at the 1988 Industry Ap- transfer function of the form (Fig. 1)
plications Society Annual Meeting, Pittsburgh, PA, October 2-7. Manuscript
released for publication August 18, 1989.
J. W. Umland is with the Department of Mathematical and Aerospace
Engineering, 1012 Furnas Hall, State University of New York at Buffalo,
Buffalo, NY 14260. where P ( s ) is an mth-order polynomial n s,
M. Safiuddin is with the Department of Electrical and Computer Engineer- m
ing, 201 Bell Hall, State University of New York at Buffalo, Buffalo, NY
14260. P(s) = CbiS’.
IEEE Log Number 9034297. i =O
.OO
OO93-9994/90/05OO-0489$01 0 1990 IEEE
490 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS. VOL. 26, NO. 3. MAYIJUNE 1990
A
R L
GAS) G,(s)
I
Fig. 2. I controller for system with one delay.
A general form for the controller transfer function is This function may now be expanded in a Taylor series about
w =o,
G C ( s )= a ( ~ ) / 2 ~ (2)
where a(s) is an nth-order polynomial in s,
n
a ( s )= C a ; s ' .
i =O
Normally, the order of the controller is less than that of the p4(jo>lz,= o = 1.
plant, i.e., m > n. The standard controller structures can be
found via proper choice of the polynomial coefficients: The first 2n - 1 odd derivatives of IM(jw)l 2 evaluated at w = 0
are set equal to zero. Furthermore, it is possible to set the
proportional (P) al # 0 , a; = 0 first 2n even derivatives to zero, eliminating the effect of low-
i E [0, n],i # 1 frequency components in the magnitude response function.
integral (I) a0 # 0, a; = 0 Defining the magnitude response function as approximately
equal to zero for low frequencies, i.e.,
i E [ l , n3
proportional plus integral (PI) ao, a I# 0, a; =0
i E [2,n] and satisfied exactly for w = 0. It is proven from calculus
proportional plus integral ao, a1 , a2 # 0 , a; = 0 that, for a curve to be horizontal at a point, its first derivative
plus derivative (PID) i E [3, n ] . must be zero at that point. Furthermore, for a curve to be flat,
or to have zero curvature at a point, its second derivative must
The closed-loop unity feedback transfer function is given equal zero at that point. Therefore, the magnitude optimum
bY technique seeks to eliminate the sensitivity of the magnitude
response function to low frequencies by equating to zero the
C(s) - Gc(s)G,(s)
(3) coefficients of the Taylor series expansion that would be most
R(s) 1 +Gc(s)Gp(s)' significant for low frequencies. The final general result that
For the general plant and controller transfer functions this may can be used to find the controller coefficients is
be rearranged to be
(4)
which is a simple first-order delay. In this case, giving a PI controller of the form
TABLE I
OPTIMAL CONTROLLER SETTINGS FOR SPECIFIC PLANT CONFIGURATIONS
~
Magnitude TI = 0 7, = 0
TI = 0 I 0
72 = 4.7 a 8.4 a 4.3 5.5 a 0.63 KpD
Ty = 2Kp0
TI > a TI = TI
T2 = 0 PI 0
72 = 4.7 a 8.4 a 4.3 5 . 5 m 1 .6(KPa/Tl)D
Ty = 2KPa
TI = TI
PID 72 = T2 4.7 a 8.4 a 4.3 NIA 0.75(KPa/m)D
T, = 2KPa
TI S a 71 = TI
T2 > 0 P 72 = 0 4.7 a 8.4 a 4.3 10 a NIA
Ty = 2KPa
Symmetric TI 3 a 21 = 4 0 3.1 a 16.5 a 4.3
T2 = 0 PI 72 = 0 17 a 1.6 (KPa/Ti)D
Ty = 8Kpu2/Ti (7.6 a) (13.3 a) (8.1)
TI = 40 3.1 a 16.5 a 43
PID 72 = T2 17 a 1.6(Kpa/T,)D
T , = 8Kpa2/T, (7.6 a) (13.3 a) (8.1)
Fig. 4 . PI controller for system with one free integrator and one delay.
bance inputs. In effect, the secondary or imbedded system The large time constants in the fixed plant may be treated
determined by the transfer function as being free integrators, as long as Tis >> 1 is true. Except
at low frequencies, this is a true statement. Furthermore, this
T ( s )= 1/[1 + G c ( s ) G 1 ( ~ ) G 2 ( ~ ) 1 (12) assumption takes on importance when considering at what fre-
quency the open-loop magnitude response curve crosses over
will be optimized in a manner similar to the magnitude opti-
the 0-dB axis on a Bode magnitude plot. Thus the transfer
mum. To accommodate free integrators, an approximation of
function for the fixed plant may be approximated as
the open-loop transfer function will be used.
The fixed-plant transfer function is written in a factored
form, i.e.,
1 1 or
rather than as a polynomial, as was PreviouslY used. It is The unity feedback open-loop transfer function G ( s ) is
assumed that the plant contains one small delay and a number
of large delays, Ti >> 0 . The controller is written in a similar
factored form. G ( s )=
nrXl(1 +Tis) 1
(17)
n = p +q
TlYS (1 + 0s)n:Tis I=I
G c ( s )= = r=l (’ 7 i S ) 1 / T > s . (14) This transfer function can be multiplied by one in the form
UMLAND AND SAFIUDDIN: DESIGN OF LINEAR CONTROL SYSTEMS 493
G
(1 +is)
This expression may be further simplified by considering the % (22)
product +G 1 + -sn7 + TLns72 ( 1 +as)
fi(%)
i=l =Q(l+$).
and the imbedded system to be optimized is
1
Upon multiplying this product out, one would have a series
U s )= Ty7 (23)
of the form
1 + -sn + -s2(1
n
+ as)
The unknowns in this equation are the equivalent integration
time constant Ty and the control parameter 7. The optimiza-
tion then proceeds by applying the same principles used for
Assuming that 7;s >> 1, this series may be truncated after the the magnitude optimum. This includes substituting j w for s,
first-order terms, i.e., and setting the first 2n derivatives of the modulus, squared of
the frequency response function, evaluated at w = 0, equal to
zero. The resulting optimum criteria is
7 =4na (24)
Furthermore, to minimize the magnitude of the higher order T y = 2o (25)
terms, neglected here, the following requirement is made:
and the open-loop transfer function is
~i =7 i = [ l , n]. 1 +4as
G ( s )=
Thus 8a2s2(1 os) ' +
The 0-dB crossover frequency for such a system is wc = 1/2a,
and as can be seen from the Bode magnitude plot for this trans-
fer function (Fig. 7), the curve displays twofold symmetry
494 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 26, NO. 3. MAYIJUNE 1990
Fig. 8. PI controller with input filter for system with one free integrator and one delay
about the (1/2a, 0 dB) point. Thus explaining the symmet- IV. REVIEW
OF CLASSICAL
DESIGN
METHODS
ric optimum designation. Furthermore, the knowledge of the Two techniques used in the past to determine the “best”
crossover frequency helps to validate the assumptions made controller coefficients for a given controller and plant config-
earlier. For the controller at crossover uration are the Butterworth pole assignment method and the
minimization of the integral time absolute error (ITAE) [3].
7;s = 4na/2a = 2n > I Upon first inspection, these two techniques appear quite dif-
and becomes much greater than one for large n, implying ferent, yet the method of application for both is the same.
that it is appropriate to neglect the higher order terms in the The end result of both methods is a set of prototype transfer
expanded product. As for the large time delays in the plant, functions that meet their respective criteria. Once the order of
it is given that Ti >> U so that, at crossover, the system is known, the controller coefficients are determined
by fitting the controller and plant to the appropriate prototype
Tis = T i / 2 ~>> 1, transfer function.
The Butterworth pole pattern is a simple-pole configuration
thus allowing the large time delays to be approximated by where the n poles of the system are equally spaced on a circle
integrators. of radius WO in the left half-plane. 00 denotes the bandwidth
Note that, to return to the original form of the controller, of the system. Higher order systems with such pole patterns
the integration time constant TI must be evaluated in terms of display somewhat oscillatory step responses. In an effort to
the equivalent time constant T, and the plant parameters Ti: add damping to such systems the Magarineworth pole pattern
has been proposed [ 5 ] .
A second set of prototype transfer functions is found from
the minimization of the ITAE for a step input to the system.
The error e ( t ) is defined as
As an example, again consider the design of a PI controller
for a plant that consists of one free integrator and one time e(t)= r(t)- c(t) = 1 - c ( t )
delay (Fig. 4). This is the same problem, that when analyzed for a unit step input. The ITAE is then
via the magnitude optimum the resulting controller was forced
back to a proportional configuration. Here the solution is
ITAE = J, t’le(t’)ldt’.
0
7 =4a
The step response of systems designed according to this prin-
8K, u2 ciple display substantially less overshoot than is designed to
T’ = - fit the Butterworth forms.
To
To 1 U US VI. RESULTS
AND COMPANSONS
G,(s) = -~.
8K,a2 s Both the magnitude and symmetric optimum can be applied
to a variety of plant and controller configurations, some of
Unfortunately, this system displays a 43% overshoot to a which may be represented by a general block diagram shown
step input at the reference, with a rise time of 3 . 1 ~ On . the in Fig. 9.The specific plant configurations along with the ap-
other hand, the maximum deflectionto a step load disturbance, propriate “optimum” controller settings are given in Table I.
Z , is l d ( K , a / T o ) Z . To improve the system’s input response, Furthermore, several performance measures are given for each
an input filter is used to cancel the zero of the controller (Fig. system. These performance measures are for the system’s re-
8). Furthermore, this filter is positioned outside the loop, so sponse due to either a reference unit step input r ( t ) or a 1/4
as not to affect the load response. The input step response unit step load disturbance d ( t ) . The rise time t , is defined as
overshoot is then found to be 8.l%, with a rise time of 7 . 6 ~ . the time for the response c ( t ) to reach the required value be-
While this extra input filter is an ad hoc remedy for a pre- fore the first overshoot. Settling time ts is defined as the time
viously optimized system, it should be remembered that only required for the variation of the response to fall within f2%
a system related to the original was optimized. In actuality, of the required value. The percentage of maximum overshoot,
no conditions were placed on the original system. Therefore, or deviation, of the controlled variable from its required value
it may have been anticipated that the actual system might re- is denoted %OS. Note that, to have a dynamic system, it is
spond in an unacceptable fashion. assumed that U # 0, K, # 0. For the symmetric optimum, the
UMLAND AND SAFIUDDIN: DESIGN OF LINEAR CONTROL SYSTEMS 495
I I
Fig. 9. General block diagram to be applied to Table 1.
coefficients are given in Table 11. The input step responses Time (scc)
are shown in Fig. 11, and the disturbance step responses are Fig. 12. Unit step disturbance response.
shown in Fig. 12. These responses were obtained from sim-
ulations done on a personal computer using the TUTSIM [7] worth designed systems. For the input step response, the ITAE
continuous dynamic system simulation software. The value of system responds faster with less overshoot than either of the
CY is 0.025. other two systems. The symmetric optimum system has the
As Figs. 1 1 and 12 show, the responses of the ITAE de- slowest rise time shown, while its overshoot and settling time
signed system is superior to both the symmetric and Butter- are both less than that for the Butterworth designed system.
496 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 26, NO. 3, MAYIJUNE 1990
ITAE -‘
Symmetric and
0.8 - Butterwor t h
Time (6s~)
Fig. 13. Unit step input response.
ACKNOWLEDGMENT
The authors acknowledge the contributions of the late Darl
C. Washburn, Jr., of Westinghouse Electric Corporation,
whose seminar notes provided the basis for this work. His
- biography follows.
tems,” in Conf. Rec. IEEE Ind. Appl. Soc. 1985 Annu. Meeting, Mohammed Safiuddin (M’60-SM’79) received
85CH2207-9, pp. 1130-1 140. the B.E. degree from Osmania University, Hyder-
[6] L. M . Boychuk and V. S . Elsukov, “Structural-parametric synthesis of abad, India, and the M.S. degree from the Univer-
regulation of reduced order,” Sov. J . Automat. Inform. Sci., vol. sity of Illinois, Urbana, both in electrical engineer-
19, no. I , pp. 89-92, 1986. ing, and the M.B.A. and Ph.D. degrees from the
171 W. E. Reynolds and J . Wolf, TUTSIM Users Manual, Applied i, State University of New York at Buffalo in 1959,
Palo Alto, CA, 1987. 1960, 1971, and 1982, respectively.
He worked as a Junior Engineer in Andhra
Pradesh State Electricity Board (India) for over a
year before arriving in the U.S. He joined the Sys-
tems Control Department of Westinghouse Electric
Corporation, Buffalo, in June 1960 as an Associate Engineer. He progressed
through the Engineer, Senior Engineer, and Fellow Engineer positions to be-
come Manager of Product/Strategic Planning in 1982 in the Power Electronics
and Drive Systems Division and was later appointed Technical Advisor in the
Marketing Department of the same Division. His interest in continuing edu-
cation has kept him in close contact with the State University of New York at
Buffalo where he has done part-time teaching since the early 1960’s and has
been an adjunctlpart-time Associate Professor since 1977. He is President
of STS International, a technology service firm he established in September
1985. His areas of technical interest cover static power conversion and opti-
Jeffrey W. Umland (S’88) received the B.S.M.E. mal control systems as applied to coordinated ac and dc motor drive systems
degree in 1985 from the State University of New for industrial processes. He has been awarded ten patents in this field and has
York (SUNY) at Buffalo where he is currently published numerous technical papers.
working toward the Ph.D. degree in mechanical en- Dr. Safiuddin has served as Chairman of Industrial Controls Committee
gineering. of IAS (1985-1987), Chairman of Education Committee (1978-1988), and
He has been a Research Assistant, Lecturer, and a Director of the IEEE Buffalo Section (1983-1986). He is an Associate
Teaching Assistant at SUNY-Buffalo. His main re- Member of the New York State Society of Professional Engineers. He was
search interests are vibration control in large space awarded the Roscoe Allen Gold Medal in 1957 by Osmania University for
structures and friction behavior under unsteady op- excellence in the subject of hydraulics and was nominated for the prestigious
erating conditions. B. G. Lamme scholarship of Westinghouse by his Division in 1968 and 1980.
Mr. Umland was awarded a SUNY-Buffalo Fe- He is a member of the Pi Mu Epsilon (mathematics) and Beta Gamma Sigma
lowship in 1985. He is a member of Tau Beta Pi and Pi Tau Sigma. (business) honor societies.