ISSN 1068-798X, Russian Engineering Research, 2008, Vol. 28, No. 1, pp. 83–87. © Allerton Press, Inc., 2008.

Original Russian Text © A.G. Lyutov, 2007, published in STIN, 2007, No. 10, pp. 26–30.

Synthesis of Control Systems

for Complex Mechatronic Equipment
A. G. Lyutov
DOI: 10.3103/S1068798X0801019X

Ensuring optimal properties of systems regulating time synthesis of optimal control actions, largely
complex mechatronic equipment (various technologi- because there is no direct relation between the parame-
cal processes, robot systems, electromechanical ters of the control system being synthesized and the
devices, etc.), when the characteristics of the equip- variable weighting factors of the functional being min-
ment and the external perturbations are indeterminate, imized [1, 3].
demands the real-time synthesis of optimal control The synthesis of an adaptive control system usually
actions in the course of control-system operation. This consists of two stages [4]: synthesis of the basic loop;
may be described as integrated synthesis, since it and synthesis of the adaptation loop. Whereas there are
involves simultaneous formulation of control algo- numerous methods of solution for the second stage, the
rithms (laws) and control actions [1]. selection of the adjustable parameters of the basic loops
In constructing such control systems, indeterminacy (which is one of the problems in the first stage) has not
is a major problem. Thus, in machining, the incom- been adequately formalized, and the solution mainly
pleteness of a priori and current information on the depends on the designerís knowledge and experience.
object of control is due to factors such as change in the
operating conditions of the equipment, unstable charac- In addition, traditional synthesis of optimal control
teristics of the material being machined and the tool actions in designing an adaptive control system is only
material, unstable characteristics and wear of the equip- possible on the basis of an identification approach to
ment, and wear of the tool [2]. determine the structure and parameters of the basic
loop, since data on the structure and parameters of the
The action of these factors leads to uncontrollable object of control are required as the initial information
variation in the forces and thermal load on the mecha- for synthesis. However, this approach to adaptive con-
tronic equipment, which requires considerably less trol considerably increases the computational expendi-
intense operation of the equipment. Existing design tures in integrated synthesis and significantly reduces
methods for machining-control systems are unable to the speed of the adaptation loop.
consistently ensure their required operational charac-
teristics and reliability. Accordingly, we need new approaches considerably
simplifying the real-time synthesis of optimal control
actions and improving the operation of the control sys-
EXISTING APPROACHES tem in conditions of indeterminacy. Such an approach
In the control of dynamic objects, optimal-control [5, 6] may provide the basis for constructing the control
methods give the best results. The traditional approach system of mechatronic equipment using search-free
here reduces to the synthesis of optimal control algo- adaptation methods.
rithms (laws) at the stage of control-system design.
However, when the object (in the presence case, mecha-
SELECTION OF THE OPTIMIZATION
tronic equipment) operates with indeterminate charac- CRITERION
teristics and with indeterminate properties of the sur-
roundings, real-time synthesis of optimal control The most important aspect of the optimization prob-
actions is required in the course of control-system oper- lem in control-system synthesis is selection of the opti-
ation, i.e., adaptive optimization of the control system mization criterion reflecting the goal and conditions of
is essentially required. control. The optimization criterion is usually specified
Obviously, the implementation of such algorithms in the form of functionals or target functions to be min-
assumes that the procedures of synthesis may be for- imized.
malized and are computationally simple. Traditional Usually, the structures and approximate parameter
approaches (for example, on the basis of classical qua- (weighting-factor) values of the functional are prelimi-
dratic functionals) do not lend themselves to the real- narily specified at the stage of control-system design. If

83
84 LYUTOV

the modeling results for the synthesized control system In particular, the classical functional with an addi-
are unsatisfactory (as is generally the case), the struc- tive control-expenditure function may be reduced to a
ture or parameters of the functional are iteratively cor- functional with a multiplicative control-expenditure
rected until acceptable results are obtained. function
In general, existing approaches to iterative correc- t2 t2
tion of the functional to be minimized are complex and
difficult to formalize [2, 3]. This significantly limits ∫ ∫
I = V [ x ( t 2 ) ] + Q [ x ( t ), t ] dt + U [ u ( t ), t ] dt
their applicability in the real-time synthesis of optimal t1 t1
θ2
(3)
control actions.
Within the framework of the given approach [5, 6],
formulation of an optimization criterion on the basis of
∫
= V [ x̃ ( θ 2 ) ] + Q [ x̃ ( θ ), θ ] dθ,
θ1
functionals with a multiplicative control-cost function
may be proposed [5]. Such a functional takes the form where
t2 dθ = { 1 + U [ u ( t ), t ]/Q [ x ( t ), t ] }dt. (4)
∫
I = V [ x ( t 2 ) ] + Q [ x ( t ), t ]U [ u ( t ), t ] dt, (1)
t1 SOLUTION OF THE OPTIMIZATION PROBLEM
where V is the final-state function of the process being A method of linearized substitution has been devel-
controlled; x(t2) is the final-state vector of the process; oped for solution of the optimization problem using the
Q and U are scalar functions of the state vector of the proposed functionals [6]. In this method, switching to a
process x(t) and the control vector u(t), respectively, as new independent variable in the functional reduces the
well as the time t. initial differential equations of the closed control sys-
Functionals of the form in Eq. (1) permit the deriva- tem to equivalent (in general, nonlinear) differential
tion of analytical relations between the weighting fac- equations [7].
tors of the functional and the coefficients of the equiva- After linearization of these equations, the coeffi-
lent equations of the synthesized control system. In this cients of the polynomials of the corresponding equiva-
case, the complex traditional optimal-synthesis proce- lent transfer functions of the closed control system are
dure is used only once at the stage of control-system functions of the weighting factors of the functional.
design, in order to determine the structure and initial This eliminates the need for traditional numerical para-
parameters of the regulator. metric optimization. The weighting factors may be
All that remains is a simple procedure of purposeful regarded as tunable parameters of the optimal regulator.
formulation of optimal characteristics of control, by The proposed approximate method considerably
correcting the regulator coefficients on the basis of the reduces the computational complexity of the algo-
relations already obtained between these coefficients rithms in the synthesis of optimal control actions and
and the weighting factors of the functional. permits automatic selection of the weighting factors of
The functional in Eq. (1) may be determined by sub- the functional to be minimized, by adaptive adjustment
stitution, i.e., by replacing the independent variable t in the course of control-system operation.
with the new variable θ, related to t by the differential In the general case, for linearized control systems of
equation dθ = U[u(t), t]dt; thus mechatronic equipment based on functionals with a
θ2 multiplicative expenditure function, optimization con-
sists of the following steps.
∫
I = V [ x̃ ( θ 2 ) ] + Q [ x̃ ( θ ), θ ] dθ, (2)
1. Selection of the functional to be minimized,
θ1
reflecting the quality of the control processes and the
where x̃ (θ2) and x̃ (θ) are, respectively, the vectors of corresponding constraints, from the class of traditional
the final and current states of the controlled process, functionals with an additive control-expenditure func-
corresponding to the new equations of state. tion.
The functional in Eq. (2) is equivalent to the initial 2. Determination of the structure and initial param-
functional in Eq. (1) in terms of the optimization prob- eters of the closed control system and regulator by the
lem but does not explicitly contain the function U[u(t), t]. appropriate optimization method for the selected func-
As a result, for some classes of dynamic control sys- tional with weighting factors whose values take limited
tems, analytical relations are obtained between the account of the control expenditures and the specified
weighting factors of the functional being minimized differential equation of the mechatronic equipment.
and the coefficients of the equation of the control sys- 3. Reduction of the functional with an additive con-
tem being optimized. This significantly simplifies inte- trol-expenditure function to a functional with a multipli-
grated synthesis. cative expenditure function, in accordance with Eq. (3).

RUSSIAN ENGINEERING RESEARCH Vol. 28 No. 1 2008

 SYNTHESIS OF CONTROL SYSTEMS FOR COMPLEX MECHATRONIC EQUIPMENT 85

4. Replacement of the independent variable accord- where ε(t) and u(t) are the control error and the control
ing to Eq. (4) in the differential equations of the optimal action in the closed one-dimensional control system;
closed control system, in accordance with the rules of the first term is the quadratic estimate of the quality of
variable substitution in differential equations. control; the second characterizes the energy expendi-
5. Linearization of the equivalent nonlinear differ- tures of the regulator executive mechanisms and
ential equation of the closed control system. The coef- expresses the constraint in optimization.
ficients of the linearized equation are then direct analyt-
ical functions of the weighting factors of the functional The primary synthesis problem here is to determine
to be minimized, which are also adjustable parameters the optimal transfer function Φ*(s) = H(s)R*(s)[1 +
of the basic control loop. H(s)R*(s)]–1 of the closed control system from the con-
6. Correction of the initial parameters of the optimal dition J min. Here H(s) and R*(s) are the transfer
regulator (see step 2), by purposeful change in the tun- functions of the object of control (the mechatronic
able parameters in the equivalent linearized differential equipment) and the optimal regulator, in which the
equation of the control system, until the required qual- weighting coefficient α—or some function β = β(α)—
ity of control is obtained. is present as a direct parameter; s is the Laplacian oper-
Steps 1–5 are executed once at the stage of control- ator.
system design. Step 6 may be formalized, automated,
and executed in real time in the course of control-sys- In accordance with the foregoing approach, an
tem operation, on the basis of a particular adaptation equivalent differential equation of the control system
method where necessary. may be written, in terms of the control error ε(t); it is
obtained from the initial differential equation of the
closed control system by replacement of the indepen-
EXAMPLE
dent variable. (The resulting equation is not given here,
As an example, consider the practical use of the pro- on account of its unwieldiness.) After linearizing this
posed method for optimization on the basis of the qua- equation in the vicinity of the steady state— ε̇ (∞) =
dratic quality functional J of control, with a single
weighting coefficient α (α > 0) ε̇˙ (∞) = … = 0—and returning to real time, we may
∞ ∞
write the corresponding optimal equivalent transfer
function of the control system (with zero boundary con-
∫ ε ( t ) dt + α ∫ u̇ ( t ) dt,
2 2
J = (5) ditions), which is the solution of the problem of mini-
0 0 mizing Eq. (5), in the general form

Φ *equ ( s, β )
m
am s + … + a1 s + a0 (6)
= -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-,
b °n ( 1 + β ) [ 1 + ( 2n + 1 )β ]s + … + b °3 ( 1 + β ) ( 1 + 7β )s + b °2 ( 1 + β ) ( 1 + 5β )s + b °1 ( 1 + 3β )s + b °0
n–1 n 2 3 2

where β = αk2; k = const; a0, a1, …, am are parameter and parametric optimization simply involves varying β.
values in the numerator of the transfer function; b °0 , This not only considerably increases the effectiveness
of parametric optimization but also permits adaptive
b 1° , …, b n° are the initial parameter values in the optimization of the control system using search-free
denominator of the transfer function, determined for methods [4].
small α.
Thus, the equivalent transfer function in Eq. (6) of CONSTRUCTING THE ADAPTATION LOOP
the closed control system is the approximate optimal OF AN OPTIMAL CONTROL SYSTEM
solution of the problem of minimizing Eq. (5) and
defines the single-parameter set {Φ*(s, β)} of stable Construction of the self-tuning loops of control sys-
(when 0 < β < ∞) transfer functions of the control sys- tems for linearized continuous objects of control is
tem that ensure physical feasibility of the regulator often based on the Lyapunov-function method or a gra-
transfer functions R*(s, β). The dynamic properties of dient method employing sensitivity functions [4].
the control system with transfer function Φ*(s, β) are The direct Lyapunov method permits the determina-
completely determined with known transfer function tion of the structure of the self-tuning (adaptive) system
by H(s), by specifying a single weighting factor β. with a standard model known to satisfy the stability
On the basis of Eq. (6), factorization of piecewise- condition [8]. However, its practical use is relatively
rational expressions (associated with numerical solu- difficult, primarily on account of difficulty in selecting
tion of a system of nonlinear equations) is unnecessary, the appropriate Lyapunov function.

RUSSIAN ENGINEERING RESEARCH Vol. 28 No. 1 2008

86 LYUTOV

g(t) ε(t)
R*(s, β0, β1, ..., βµ)
u(t)
H(s)
x(t) R* ( s, β 0, β 1, β 2, …, β µ )
1 Φ °equ ( s, β 0, β 1, β 2, …, β µ )
= ------------ -------------------------------------------------------------------------
-.
βµ β1 β0 H ( s ) [ 1 – Φ °equ ( s, β 0, β 1, β 2, …, β µ ) ]
λµ λ1 λ0
------ ... ------
s s
------
s Thus, the parameters β0, β1, β2, …, βµ are the tun-
...
able parameters of the regulator in the basic loop.
β0 (s, β , β , ..., β )
Wsm 0 1 µ
If the quadratic function of the mismatch error ε a (t) =
2
...
β1 (s, β , β , ..., β )
[x(t) – xM(t)]2 is selected as the measure of the mis-
Wsm 0 1 µ
match of the output coordinates of the self-tuning sys-
... ... ... ... ... ... ...
... tem x(t) and the model xM(t), the figure shows the cor-
βµ (s, β , β , ..., β )
responding structure of the self-tuning system with a
Wsm 0 1 µ standard model synthesized by the gradient method [9].
εa(t) Here λ0, λ1, λ2, …, λµ are coefficients determining the
Φst(s) self-tuning rate; Wsm is the sensitivity model; Φst(s) is
the transfer function of the standard model; g(t) is the
Structure of self-tuning system.
master signal.
The proposed approach to optimization simplifies
the real-time synthesis of optimal control actions and
Gradient methods of synthesis yield the structure of also improves the operation of the control system in
the self-adjusting system, but it is fairly difficult to conditions of indeterminacy. The effectiveness of this
ensure the stability of this system and the required approach is due to the simplification of the algorithms
dynamic characteristics. In addition, there are difficul- in the synthesis of optimal control actions and to auto-
ties in determining the sensitivity function of the output mated selection of the weighting factors of the func-
coordinate of the self-tuning system for each of the tun- tional to be minimized, by adaptive tuning in the course
able coefficients. When using each method, an indepen- of control-system operation.
dent problem is the selection of tunable parameters of
the basic loop of the self-tuning system. These findings are of practical importance. Despite
extensive research on the subject, the synthesis of adap-
The proposed approach to optimization eliminates tive control systems for mechatronic equipment
the difficulties in the application of gradient methods remains a pressing concern, because accurate methods
and permits their use in the construction of self-tuning are inapplicable in many practical cases on account of
loops and in adaptive optimization of the control sys-
incompleteness and indeterminacy of the a priori and
tem. This may be based on the following consider-
ations. current information on the object to be controlled and
its operating conditions.
1. Adjustment of the self-tuning system is based on
a single-parameter {Φ*(s, β)} or multiparameter In this context, there is great practical interest in the
{Φ*(s, β0, β1, β2, …, βµ)} set of stable transfer func- development of effective approximate methods of syn-
tions, which, in this case, removes the constraints on the thesis, characterized by simple adaptive-control algo-
applicability of gradient methods due to the need for rithms and control equipment, flexibility (robustness)
stability of the self-tuning system. and resistance to interference of the control systems,
and reduced requirements on the measured informa-
2. The parameters β0, β1, β2, …, βµ appear as tunable tion.
parameters of the basic loop of the self-tuning system
and completely determine the dynamic properties of
the transfer function Φ*. REFERENCES
3. Difficulties in determining the sensitivity func- 1. Spravochnik po teorii avtomaticheskogo regulirovaniya
tions of the output coordinate of the self-tuning system (Handbook on Automatic Control Theory), Krasov-
skii, A.A., Ed., Moscow: Nauka, 1987.
are eliminated, since the dependence of the coefficients
of transfer function Φ* on each of the tunable parame- 2. Zoriktuev, V.Ts. and Lyutov, A.G., Upravlenie protses-
ters is known. This permits the determination of the sami mechanoobrabotki detalei aviatsionnykh dvigatelei
v usloviyakh neopredelennosti (Control of the Machin-
corresponding models of sensitivity. ing of Airplane-Engine Components in Conditions of
The optimal transfer function R*(s, β0, β1, β2, …, Indeterminacy), Moscow: Izd. MAI, 2003.
βµ) of the regulator is determined from the equivalent 3. Krut’ko, P.D., Maksimov, A.I., and Skvortsov, L.M.,
Algoritmy i programmy proektirovaniya avtomati-
° (s, β0, β1, β2, …, βµ) correspond-
transfer function Φ equ cheskikh sistem (Design Algorithms and Programs for
ing to Eq. (6), by the inverse-operator method Automatic Systems), Moscow: Radio i Svyaz, 1988.

RUSSIAN ENGINEERING RESEARCH Vol. 28 No. 1 2008

 SYNTHESIS OF CONTROL SYSTEMS FOR COMPLEX MECHATRONIC EQUIPMENT 87

4. Fradkov, A.L., Adaptivnoe upravlenie v slozhnykh siste- 7. Lyutov, A.G., Optimizing Control Systems by Linear-
makh: bespoiskovye metody (Adaptive Control in Com- ized Functional Substitution, Optimizatsiya i upravlenie
plex Systems: Search-Free Methods), Moscow: Nauka, protsessom rezaniya, mekhatronnye stanochnye sistemy:
1990. Sb. trudov mezhdunar. nauch.-tekhn,. konf. (Optimiza-
5. Lyutov, A.G., Optimization Criteria for Dynamic Pro- tion and Control of Cutting and Mechatronic Machine-
cesses with Multiplicative Control-Expenditure Func- Tool Systems: Proc. Intern. Conf.), Ufa: BashGU, 2004,
tions, Optimal’noe upravlenie mekhatronnymi stanoch- pp. 148–154.
nymi sistemami (Optimal Control of Mechatronic 8. Petrov, B.N., Rutkovskii, V.Yu., and Zemlyakov, S.D.,
Machine-Tool Systems), Ufa: UGATU, 1999, pp. 159– Adaptivnoe koordinatno-parametricheskoe upravlenie
162. nestatsionarnymi ob”ektami (Adaptive Coordinate–
6. Lyutov, A.G., Optimization of Control Systems by Parametric Control of Nonsteady Objects), Moscow:
Means of Functionals with a Multiplicative Expenditure Nauka, 1980.
Function, Optimal’noe upravlenie mekhatronnymi 9. Lyutov, A.G., Self-Tuning Algorithms of Optimal Con-
stanochnymi sistemami (Optimal Control of Mecha- trol System for Mechatronic Objects, Mekhatronika,
tronic Machine-Tool Systems), Ufa: UGATU, 1999, avtomatizatsiya, upravlenie (Mechatronics, Automation,
pp. 163–165. and Control), Ufa: UGATU, 2005, pp. 149–154.

RUSSIAN ENGINEERING RESEARCH Vol. 28 No. 1 2008