Control theory

This article is about control theory in engineering. For 1 Overview

control theory in linguistics, see control (linguistics). For
control theory in psychology and sociology, see control
theory (sociology) and Perceptual control theory.
Control theory is an interdisciplinary branch of engi-

Smooth nonlinear trajectory planning with linear quadratic

Gaussian feedback (LQR) control on a dual pendula system.

Control theory is
The concept of the feedback loop to control the dynamic behavior
of the system: this is negative feedback, because the sensed value
is subtracted from the desired value to create the error signal, • a theory that deals with influencing the behavior of
which is amplified by the controller. dynamical systems

• an interdisciplinary subfield of science, which orig-

inated in engineering and mathematics, and evolved
neering and mathematics that deals with the behavior of into use by the social sciences, such as economics,
dynamical systems with inputs, and how their behavior psychology, sociology, criminology and in the
is modified by feedback. The usual objective of con- financial system.
trol theory is to control a system, often called the plant,
so its output follows a desired control signal, called the Control systems may be thought of as having four func-
reference, which may be a fixed or changing value. To do tions: measure, compare, compute and correct. These
this a controller is designed, which monitors the output four functions are completed by five elements: detector,
and compares it with the reference. The difference be- transducer, transmitter, controller and final control ele-
tween actual and desired output, called the error signal, ment. The measuring function is completed by the de-
is applied as feedback to the input of the system, to bring tector, transducer and transmitter. In practical appli-
the actual output closer to the reference. Some topics cations these three elements are typically contained in
studied in control theory are stability (whether the output one unit. A standard example of a measuring unit is
will converge to the reference value or oscillate about it), a resistance thermometer. The compare and compute
controllability and observability. functions are completed within the controller, which may
Extensive use is usually made of a diagrammatic style be implemented electronically by proportional control, a
known as the block diagram. The transfer function, also PI controller, PID controller, bistable, hysteretic control
known as the system function or network function, is a or programmable logic controller. Older controller units
mathematical representation of the relation between the have been mechanical, as in a centrifugal governor or a
input and output based on the differential equations de- carburetor. The correct function is completed with a fi-
scribing the system. nal control element. The final control element changes
an input or output in the control system that affects the
Although a major application of control theory is in manipulated or controlled variable.
control systems engineering, which deals with the de-
sign of process control systems for industry, other
applications range far beyond this. As the general 1.1 An example
theory of feedback systems, control theory is useful
wherever feedback occurs. A few examples are in An example of a control system is a car’s cruise con-
physiology, electronics, climate modeling, machine de- trol, which is a device designed to maintain vehicle speed
sign, ecosystems, navigation, neural networks, predator- at a constant desired or reference speed provided by the
prey interaction, gene expression, and production the- driver. The controller is the cruise control, the plant is the
ory.[1] car, and the system is the car and the cruise control. The

1
2 2 CLASSIFICATION

system output is the car’s speed, and the control itself is • Nonlinear control theory - This covers a wider class
the engine’s throttle position which determines how much of systems that do not obey the superposition prin-
power the engine delivers. ciple, and applies to more real-world systems, be-
A primitive way to implement cruise control is simply to cause all real control systems are nonlinear. These
lock the throttle position when the driver engages cruise systems are often governed by nonlinear differential
control. However, if the cruise control is engaged on a equations. The few mathematical techniques which
stretch of flat road, then the car will travel slower going have been developed to handle them are more dif-
uphill and faster when going downhill. This type of con- ficult and much less general, often applying only
to narrow categories of systems. These include
troller is called an open-loop controller because there is no
feedback; no measurement of the system output (the car’s limit cycle theory, Poincaré maps, Lyapunov sta-
bility theorem, and describing functions. Nonlinear
speed) is used to alter the control (the throttle position.)
As a result, the controller cannot compensate for changes systems are often analyzed using numerical methods
on computers, for example by simulating their oper-
acting on the car, like a change in the slope of the road.
ation using a simulation language. If only solutions
In a closed-loop control system, data from a sensor mon- near a stable point are of interest, nonlinear systems
itoring the car’s speed (the system output) enters a con- can often be linearized by approximating them by a
troller which continuously subtracts the quantity repre- linear system using perturbation theory, and linear
senting the speed from the reference quantity represent- techniques can be used.[2]
ing the desired speed. The difference, called the error,
determines the throttle position (the control). The re-
sult is to match the car’s speed to the reference speed
2.2 Frequency domain versus time domain
(maintain the desired system output). Now, when the
car goes uphill, the difference between the input (the
Mathematical techniques for analyzing and designing
sensed speed) and the reference continuously determines
control systems fall into two different categories:
the throttle position. As the sensed speed drops below
the reference, the difference increases, the throttle opens,
and engine power increases, speeding up the vehicle. In • Frequency domain - In this type the values of the
this way, the controller dynamically counteracts changes state variables, the mathematical variables repre-
to the car’s speed. The central idea of these control sys- senting the system’s input, output and feedback are
tems is the feedback loop, the controller affects the sys- represented as functions of frequency. The in-
tem output, which in turn is measured and fed back to the put signal and the system’s transfer function are
controller. converted from time functions to functions of fre-
quency by a transform such as the Fourier transform,
Laplace transform, or Z transform. The advantage
2 Classification of this technique is that it results in a simplifica-
tion of the mathematics; the differential equations
that represent the system are replaced by algebraic
2.1 Linear versus nonlinear control theory
equations in the frequency domain which are much
simpler to solve. However, frequency domain tech-
The field of control theory can be divided into two
niques can only be used with linear systems, as men-
branches:
tioned above.
• Linear control theory - This applies to systems made
of devices which obey the superposition principle, • Time-domain state space representation - In this type
which means roughly that the output is propor- the values of the state variables are represented as
tional to the input. They are governed by linear functions of time. With this model the system being
differential equations. A major subclass is systems analyzed is represented by one or more differential
which in addition have parameters which do not equations. Since frequency domain techniques are
change with time, called linear time invariant (LTI) limited to linear systems, time domain is widely used
systems. These systems are amenable to powerful to analyze real-world nonlinear systems. Although
frequency domain mathematical techniques of great these are more difficult to solve, modern computer
generality, such as the Laplace transform, Fourier simulation techniques such as simulation languages
transform, Z transform, Bode plot, root locus, and have made their analysis routine.
Nyquist stability criterion. These lead to a de-
scription of the system using terms like bandwidth,
frequency response, eigenvalues, gain, resonant fre- 2.3 SISO vs MIMO
quencies, poles, and zeros, which give solutions for
system response and design techniques for most sys- Control systems can be divided into different categories
tems of interest. depending on the number of inputs and outputs.
 3

• Single-input single-output (SISO) - This is the sim- tem may lead to overcompensation and unstable behav-
plest and most common type, in which one output is ior. This generated a flurry of interest in the topic, dur-
controlled by one control signal. Examples are the ing which Maxwell’s classmate, Edward John Routh, ab-
cruise control example above, or an audio system, in stracted Maxwell’s results for the general class of linear
which the control input is the input audio signal and systems.[4] Independently, Adolf Hurwitz analyzed sys-
the output is the sound waves from the speaker. tem stability using differential equations in 1877, result-
ing in what is now known as the Routh–Hurwitz theo-
• Multiple-input multiple-output (MIMO) - These are rem.[5][6]
found in more complicated systems. For exam-
ple, modern large telescopes such as the Keck and A notable application of dynamic control was in the area
MMT have mirrors composed of many separate seg- of manned flight. The Wright brothers made their first
ments each controlled by an actuator. The shape of successful test flights on December 17, 1903 and were
the entire mirror is constantly adjusted by a MIMO distinguished by their ability to control their flights for
active optics control system using input from mul- substantial periods (more so than the ability to produce
tiple sensors at the focal plane, to compensate for lift from an airfoil, which was known). Continuous, reli-
changes in the mirror shape due to thermal expan- able control of the airplane was necessary for flights last-
sion, contraction, stresses as it is rotated and distor- ing longer than a few seconds.
tion of the wavefront due to turbulence in the atmo- By World War II, control theory was an important part of
sphere. Complicated systems such as nuclear reac- fire-control systems, guidance systems and electronics.
tors and human cells are simulated by computer as
Sometimes, mechanical methods are used to improve the
large MIMO control systems.
stability of systems. For example, ship stabilizers are fins
mounted beneath the waterline and emerging laterally. In
contemporary vessels, they may be gyroscopically con-
3 History trolled active fins, which have the capacity to change their
angle of attack to counteract roll caused by wind or waves
acting on the ship.
The Sidewinder missile uses small control surfaces placed
at the rear of the missile with spinning disks on their outer
surfaces and these are known as rollerons. Airflow over
the disks spins them to a high speed. If the missile starts
to roll, the gyroscopic force of the disks drives the control
surface into the airflow, cancelling the motion. Thus, the
Sidewinder team replaced a potentially complex control
system with a simple mechanical solution.
The Space Race also depended on accurate spacecraft
control, and control theory has also seen an increasing
use in fields such as economics.

4 People in systems and control

Main article: People in systems and control

Many active and historical figures made significant con-

tribution to control theory including:

• Pierre-Simon Laplace (1749-1827) invented the Z-

transform in his work on probability theory, now
Centrifugal governor in a Boulton & Watt engine of 1788 used to solve discrete-time control theory problems.
The Z-transform is a discrete-time equivalent of the
Although control systems of various types date back to Laplace transform which is named after him.
antiquity, a more formal analysis of the field began with a • Alexander Lyapunov (1857–1918) in the 1890s
dynamics analysis of the centrifugal governor, conducted marks the beginning of stability theory.
by the physicist James Clerk Maxwell in 1868, entitled
On Governors.[3] This described and analyzed the phe- • Harold S. Black (1898–1983), invented the concept
nomenon of self-oscillation, in which lags in the sys- of negative feedback amplifiers in 1927. He man-
4 5 CLASSICAL CONTROL THEORY

aged to develop stable negative feedback amplifiers 5.1 Closed-loop transfer function
in the 1930s.
For more details on this topic, see closed-loop transfer
• Harry Nyquist (1889–1976) developed the Nyquist
function.
stability criterion for feedback systems in the 1930s.
• Richard Bellman (1920–1984) developed dynamic
The output of the system y(t) is fed back through a sensor
programming since the 1940s.[7]
measurement F to the reference value r(t). The controller
• Andrey Kolmogorov (1903–1987) co-developed the C then takes the error e (difference) between the refer-
Wiener–Kolmogorov filter in 1941. ence and the output to change the inputs u to the system
under control P. This is shown in the figure. This kind
• Norbert Wiener (1894–1964) co-developed the
of controller is a closed-loop controller or feedback con-
Wiener–Kolmogorov filter and coined the term
troller.
cybernetics in the 1940s.
This is called a single-input-single-output (SISO) control
• John R. Ragazzini (1912–1988) introduced digital system; MIMO (i.e., Multi-Input-Multi-Output) systems,
control and the use of Z-transform in control theory with more than one input/output, are common. In such
(invented by Laplace) in the 1950s. cases variables are represented through vectors instead of
• Lev Pontryagin (1908–1988) introduced the simple scalar values. For some distributed parameter sys-
maximum principle and the bang-bang principle. tems the vectors may be infinite-dimensional (typically
functions).
• Pierre-Louis Lions (1956) developed viscosity so-
lutions into stochastic control and optimal control
methods. r + e u y
C P
–
5 Classical control theory F

To overcome the limitations of the open-loop controller,

control theory introduces feedback. A closed-loop con- A simple feedback control loop
troller uses feedback to control states or outputs of a
dynamical system. Its name comes from the information If we assume the controller C, the plant P, and the sen-
path in the system: process inputs (e.g., voltage applied sor F are linear and time-invariant (i.e., elements of
to an electric motor) have an effect on the process outputs their transfer function C(s), P(s), and F(s) do not depend
(e.g., speed or torque of the motor), which is measured on time), the systems above can be analysed using the
with sensors and processed by the controller; the result Laplace transform on the variables. This gives the fol-
(the control signal) is “fed back” as input to the process, lowing relations:
closing the loop.
Closed-loop controllers have the following advantages
over open-loop controllers: Y (s) = P (s)U (s)

• disturbance rejection (such as hills in the cruise con- U (s) = C(s)E(s)

trol example above)
• guaranteed performance even with model uncertain- E(s) = R(s) − F (s)Y (s).
ties, when the model structure does not match per- Solving for Y(s) in terms of R(s) gives:
fectly the real process and the model parameters are
not exact
( )
• unstable processes can be stabilized P (s)C(s)
Y (s) = R(s) = H(s)R(s).
• reduced sensitivity to parameter variations 1 + F (s)P (s)C(s)

• improved reference tracking performance The expression H(s) = 1+FP(s)P (s)C(s)

(s)C(s) is referred to
as the closed-loop transfer function of the system. The
In some systems, closed-loop and open-loop control are numerator is the forward (open-loop) gain from r to y, and
used simultaneously. In such systems, the open-loop con- the denominator is one plus the gain in going around the
trol is termed feedforward and serves to further improve feedback loop, the so-called loop gain. If |P (s)C(s)| ≫
reference tracking performance. 1 , i.e., it has a large norm with each value of s, and if
A common closed-loop controller architecture is the PID |F (s)| ≈ 1 , then Y(s) is approximately equal to R(s) and
controller. the output closely tracks the reference input.
 5

5.2 PID controller

A
For more details on this topic, see PID controller. P (s) =
1 + sTp
and insert all this into expression for closed-loop transfer
The PID controller is probably the most-used feedback function H(s), then tuning is very easy: simply put
control design. PID is an initialism for Proportional-
Integral-Derivative, referring to the three terms operating
on the error signal to produce a control signal. If u(t) is 1
K= , Ti = Tf , Td = Tp
the control signal sent to the system, y(t) is the measured A
output and r(t) is the desired output, and tracking error and get H(s) = 1 identically.
e(t) = r(t) − y(t) , a PID controller has the general
form For practical PID controllers, a pure differentiator is nei-
ther physically realisable nor desirable[8] due to amplifi-
cation of noise and resonant modes in the system. There-
∫ fore, a phase-lead compensator type approach is used in-
d
u(t) = KP e(t) + KI e(t)dt + KD e(t). stead, or a differentiator with low-pass roll-off.
dt
The desired closed loop dynamics is obtained by adjust-
ing the three parameters KP , KI and KD , often it- 6 Modern control theory
eratively by “tuning” and without specific knowledge of a
plant model. Stability can often be ensured using only the
In contrast to the frequency domain analysis of the clas-
proportional term. The integral term permits the rejec-
sical control theory, modern control theory utilizes the
tion of a step disturbance (often a striking specification in
time-domain state space representation, a mathematical
process control). The derivative term is used to provide
model of a physical system as a set of input, output
damping or shaping of the response. PID controllers are
and state variables related by first-order differential equa-
the most well established class of control systems: how-
tions. To abstract from the number of inputs, outputs
ever, they cannot be used in several more complicated and states, the variables are expressed as vectors and the
cases, especially if MIMO systems are considered. differential and algebraic equations are written in matrix
Applying Laplace transformation results in the trans- form (the latter only being possible when the dynami-
formed PID controller equation cal system is linear). The state space representation (also
known as the “time-domain approach”) provides a conve-
nient and compact way to model and analyze systems with
1 multiple inputs and outputs. With inputs and outputs, we
u(s) = KP e(s) + KI e(s) + KD se(s)
s would otherwise have to write down Laplace transforms
( ) to encode all the information about a system. Unlike the
1 frequency domain approach, the use of the state-space
u(s) = KP + KI + KD s e(s)
s representation is not limited to systems with linear com-
with the PID controller transfer function ponents and zero initial conditions. “State space” refers
to the space whose axes are the state variables. The state
of the system can be represented as a vector within that
( ) space.[9][10]
1
C(s) = KP + KI + KD s .
s

There exists a nice example of the closed-loop system dis- 7 Topics in control theory
cussed above. If we take:
PID controller transfer function in series form 7.1 Stability

( ) The stability of a general dynamical system with no input

1 can be described with Lyapunov stability criteria.
C(s) = K 1 + (1 + sTd )
sTi
• A linear system is called bounded-input bounded-
1st order filter in feedback loop output (BIBO) stable if its output will stay bounded
for any bounded input.

1 • Stability for nonlinear systems that take an input is

F (s) = input-to-state stability (ISS), which combines Lya-
1 + sTf
punov stability and a notion similar to BIBO stabil-
linear actuator with filtered input ity.
6 7 TOPICS IN CONTROL THEORY

For simplicity, the following descriptions focus on which has a pole at z = 1.5 and is not BIBO stable since
continuous-time and discrete-time linear systems. the pole has a modulus strictly greater than one.
Mathematically, this means that for a causal linear sys- Numerous tools exist for the analysis of the poles of a sys-
tem to be stable all of the poles of its transfer function tem. These include graphical systems like the root locus,
must have negative-real values, i.e. the real part of each Bode plots or the Nyquist plots.
pole must be less than zero. Practically speaking, stabilityMechanical changes can make equipment (and control
requires that the transfer function complex poles reside:
systems) more stable. Sailors add ballast to improve the
stability of ships. Cruise ships use antiroll fins that extend
• in the open left half of the complex plane for con- transversely from the side of the ship for perhaps 30 feet
tinuous time, when the Laplace transform is used to (10 m) and are continuously rotated about their axes to
obtain the transfer function. develop forces that oppose the roll.
• inside the unit circle for discrete time, when the Z-
transform is used.
7.2 Controllability and observability
The difference between the two cases is simply due to Main articles: Controllability and Observability
the traditional method of plotting continuous time versus
discrete time transfer functions. The continuous Laplace
transform is in Cartesian coordinates where the x axis is Controllability and observability are main issues in the
the real axis and the discrete Z-transform is in circular analysis of a system before deciding the best control strat-
coordinates where the ρ axis is the real axis. egy to be applied, or whether it is even possible to control
or stabilize the system. Controllability is related to the
When the appropriate conditions above are satisfied a sys- possibility of forcing the system into a particular state by
tem is said to be asymptotically stable: the variables of using an appropriate control signal. If a state is not con-
an asymptotically stable control system always decrease trollable, then no signal will ever be able to control the
from their initial value and do not show permanent os- state. If a state is not controllable, but its dynamics are
cillations. Permanent oscillations occur when a pole has stable, then the state is termed stabilizable. Observability
a real part exactly equal to zero (in the continuous time instead is related to the possibility of observing, through
case) or a modulus equal to one (in the discrete time case). output measurements, the state of a system. If a state is
If a simply stable system response neither decays nor not observable, the controller will never be able to deter-
grows over time, and has no oscillations, it is marginally mine the behaviour of an unobservable state and hence
stable: in this case the system transfer function has non- cannot use it to stabilize the system. However, similar
repeated poles at complex plane origin (i.e. their real and to the stabilizability condition above, if a state cannot be
complex component is zero in the continuous time case). observed it might still be detectable.
Oscillations are present when poles with real part equal
to zero have an imaginary part not equal to zero. From a geometrical point of view, looking at the states of
each variable of the system to be controlled, every “bad”
If a system in question has an impulse response of state of these variables must be controllable and observ-
able to ensure a good behaviour in the closed-loop system.
That is, if one of the eigenvalues of the system is not both
x[n] = 0.5n u[n] controllable and observable, this part of the dynamics will
then the Z-transform (see this example), is given by remain untouched in the closed-loop system. If such an
eigenvalue is not stable, the dynamics of this eigenvalue
will be present in the closed-loop system which therefore
1 will be unstable. Unobservable poles are not present in
X(z) =
1 − 0.5z −1 the transfer function realization of a state-space represen-
tation, which is why sometimes the latter is preferred in
which has a pole in z = 0.5 (zero imaginary part). This
dynamical systems analysis.
system is BIBO (asymptotically) stable since the pole is
inside the unit circle. Solutions to problems of uncontrollable or unobservable
system include adding actuators and sensors.
However, if the impulse response was

7.3 Control specification

x[n] = 1.5n u[n]
Several different control strategies have been devised in
then the Z-transform is
the past years. These vary from extremely general ones
(PID controller), to others devoted to very particular
1 classes of systems (especially robotics or aircraft cruise
X(z) = control).
1 − 1.5z −1
 7

A control problem can have several specifications. Sta- (called “nominal parameters”) are never known with ab-
bility, of course, is always present: the controller must solute precision; the control system will have to behave
ensure that the closed-loop system is stable, regardless of correctly even when connected to physical system with
the open-loop stability. A poor choice of controller can true parameter values away from nominal.
even worsen the stability of the open-loop system, which Some advanced control techniques include an “on-line”
must normally be avoided. Sometimes it would be desired identification process (see later). The parameters of the
to obtain particular dynamics in the closed loop: i.e. that model are calculated (“identified”) while the controller it-
the poles have Re[λ] < −λ , where λ is a fixed value self is running: in this way, if a drastic variation of the pa-
strictly greater than zero, instead of simply asking that
rameters ensues (for example, if the robot’s arm releases
Re[λ] < 0 . a weight), the controller will adjust itself consequently in
Another typical specification is the rejection of a step dis- order to ensure the correct performance.
turbance; including an integrator in the open-loop chain
(i.e. directly before the system under control) easily Analysis
achieves this. Other classes of disturbances need differ-
ent types of sub-systems to be included.
Analysis of the robustness of a SISO (single input sin-
Other “classical” control theory specifications regard the gle output) control system can be performed in the fre-
time-response of the closed-loop system: these include quency domain, considering the system’s transfer func-
the rise time (the time needed by the control system tion and using Nyquist and Bode diagrams. Topics in-
to reach the desired value after a perturbation), peak clude gain and phase margin and amplitude margin. For
overshoot (the highest value reached by the response be- MIMO (multi input multi output) and, in general, more
fore reaching the desired value) and others (settling time, complicated control systems one must consider the the-
quarter-decay). Frequency domain specifications are usu- oretical results devised for each control technique (see
ally related to robustness (see after). next section): i.e., if particular robustness qualities are
Modern performance assessments use some variation of needed, the engineer must shift his attention to a control
integrated tracking error (IAE,ISA,CQI). technique by including them in its properties.

Constraints
7.4 Model identification and robustness
A particular robustness issue is the requirement for a con-
A control system must always have some robustness prop- trol system to perform properly in the presence of input
erty. A robust controller is such that its properties do not and state constraints. In the physical world every signal is
change much if applied to a system slightly different from limited. It could happen that a controller will send control
the mathematical one used for its synthesis. This specifi- signals that cannot be followed by the physical system: for
cation is important: no real physical system truly behaves example, trying to rotate a valve at excessive speed. This
like the series of differential equations used to represent it can produce undesired behavior of the closed-loop sys-
mathematically. Typically a simpler mathematical model tem, or even damage or break actuators or other subsys-
is chosen in order to simplify calculations, otherwise the tems. Specific control techniques are available to solve
true system dynamics can be so complicated that a com- the problem: model predictive control (see later), and
plete model is impossible. anti-wind up systems. The latter consists of an additional
control block that ensures that the control signal never ex-
System identification ceeds a given threshold.

For more details on this topic, see System identification.

8 System classifications
The process of determining the equations that govern the
model’s dynamics is called system identification. This 8.1 Linear systems control
can be done off-line: for example, executing a series
of measures from which to calculate an approximated Main article: State space (controls)
mathematical model, typically its transfer function or ma-
trix. Such identification from the output, however, cannot For MIMO systems, pole placement can be performed
take account of unobservable dynamics. Sometimes the mathematically using a state space representation of the
model is built directly starting from known physical equa- open-loop system and calculating a feedback matrix as-
tions: for example, in the case of a mass-spring-damper signing poles in the desired positions. In complicated sys-
system we know that mẍ(t) = −Kx(t) − Bẋ(t) . Even tems this can require computer-assisted calculation capa-
assuming that a “complete” model is used in designing the bilities, and cannot always ensure robustness. Further-
controller, all the parameters included in these equations more, all system states are not in general measured and
8 9 MAIN CONTROL STRATEGIES

so observers must be included and incorporated in pole List of the main control techniques
placement design.
• Adaptive control uses on-line identification of the
process parameters, or modification of controller
8.2 Nonlinear systems control gains, thereby obtaining strong robustness proper-
ties. Adaptive controls were applied for the first
Main article: Nonlinear control time in the aerospace industry in the 1950s, and have
found particular success in that field.
Processes in industries like robotics and the aerospace • A hierarchical control system is a type of control sys-
industry typically have strong nonlinear dynamics. In tem in which a set of devices and governing software
control theory it is sometimes possible to linearize such is arranged in a hierarchical tree. When the links in
classes of systems and apply linear techniques, but in the tree are implemented by a computer network,
many cases it can be necessary to devise from scratch the- then that hierarchical control system is also a form
ories permitting control of nonlinear systems. These, e.g., of networked control system.
feedback linearization, backstepping, sliding mode con-
trol, trajectory linearization control normally take advan- • Intelligent control uses various AI computing ap-
tage of results based on Lyapunov’s theory. Differential proaches like neural networks, Bayesian probability,
geometry has been widely used as a tool for generaliz- fuzzy logic,[11] machine learning, evolutionary com-
ing well-known linear control concepts to the non-linear putation and genetic algorithms to control a dynamic
case, as well as showing the subtleties that make it a more system.
challenging problem.
• Optimal control is a particular control technique in
which the control signal optimizes a certain “cost in-
8.3 Decentralized systems control dex": for example, in the case of a satellite, the jet
thrusts needed to bring it to desired trajectory that
Main article: Distributed control system consume the least amount of fuel. Two optimal con-
trol design methods have been widely used in indus-
When the system is controlled by multiple controllers, the trial applications, as it has been shown they can guar-
problem is one of decentralized control. Decentraliza- antee closed-loop stability. These are Model Predic-
tion is helpful in many ways, for instance, it helps control tive Control (MPC) and linear-quadratic-Gaussian
systems to operate over a larger geographical area. The control (LQG). The first can more explicitly take
agents in decentralized control systems can interact using into account constraints on the signals in the system,
communication channels and coordinate their actions. which is an important feature in many industrial pro-
cesses. However, the “optimal control” structure in
MPC is only a means to achieve such a result, as it
8.4 Deterministic and stochastic systems does not optimize a true performance index of the
closed-loop control system. Together with PID con-
control trollers, MPC systems are the most widely used con-
trol technique in process control.
Main article: Stochastic control
• Robust control deals explicitly with uncertainty in
A stochastic control problem is one in which the evolution its approach to controller design. Controllers de-
of the state variables is subjected to random shocks from signed using robust control methods tend to be able
outside the system. A deterministic control problem is to cope with small differences between the true sys-
not subject to external random shocks. tem and the nominal model used for design. The
early methods of Bode and others were fairly ro-
bust; the state-space methods invented in the 1960s
and 1970s were sometimes found to lack robustness.
9 Main control strategies Examples of modern robust control techniques in-
clude H-infinity loop-shaping developed by Duncan
Every control system must guarantee first the stability of McFarlane and Keith Glover of Cambridge Uni-
the closed-loop behavior. For linear systems, this can versity, United Kingdom and Sliding mode control
be obtained by directly placing the poles. Non-linear (SMC) developed by Vadim Utkin. Robust methods
control systems use specific theories (normally based on aim to achieve robust performance and/or stability
Aleksandr Lyapunov's Theory) to ensure stability without in the presence of small modeling errors.
regard to the inner dynamics of the system. The possibil-
ity to fulfill different specifications varies from the model • Stochastic control deals with control design with un-
considered and the control strategy chosen. certainty in the model. In typical stochastic control
 9

problems, it is assumed that there exist random noise 12 Further reading

and disturbances in the model and the controller, and
the control design must take into account these ran- • Levine, William S., ed. (1996). The Control Hand-
dom deviations. book. New York: CRC Press. ISBN 978-0-8493-
8570-4.
• Energy-shaping control view the plant and the con-
troller as energy-transformation devices. The con- • Karl J. Åström and Richard M. Murray (2008).
trol strategy is formulated in terms of interconnec- Feedback Systems: An Introduction for Scientists
tion (in a power-preserving manner) in order to and Engineers. (PDF). Princeton University Press.
achieve a desired behavior. ISBN 0-691-13576-2.

• Self-organized criticality control may be defined as • Christopher Kilian (2005). Modern Control Tech-
attempts to interfere in the processes by which the nology. Thompson Delmar Learning. ISBN 1-
self-organized system dissipates energy. 4018-5806-6.

• Vannevar Bush (1929). Operational Circuit Analy-

sis. John Wiley and Sons, Inc.
10 See also
• Robert F. Stengel (1994). Optimal Control and Esti-
mation. Dover Publications. ISBN 0-486-68200-5.
11 References
• Franklin et al. (2002). Feedback Control of Dy-
[1] Antunes, Ricardo; Gonzalez, Vicente (3 March 2015). namic Systems (4 ed.). New Jersey: Prentice Hall.
“A Production Model for Construction: A Theo- ISBN 0-13-032393-4.
retical Framework”. Buildings 5 (1): 209–228.
doi:10.3390/buildings5010209. • Joseph L. Hellerstein, Dawn M. Tilbury, and Sujay
Parekh (2004). Feedback Control of Computing Sys-
[2] trim point
tems. John Wiley and Sons. ISBN 0-471-26637-X.
[3] Maxwell, J.C. (1868). “On Governors”. Proceed-
ings of the Royal Society of London 16: 270–283. • Diederich Hinrichsen and Anthony J. Pritchard
doi:10.1098/rspl.1867.0055. JSTOR 112510. (2005). Mathematical Systems Theory I - Mod-
elling, State Space Analysis, Stability and Robustness.
[4] Routh, E.J.; Fuller, A.T. (1975). Stability of motion. Tay-
Springer. ISBN 3-540-44125-5.
lor & Francis.

[5] Routh, E.J. (1877). A Treatise on the Stability of a Given • Andrei, Neculai (2005). “Modern Control Theory
State of Motion, Particularly Steady Motion: Particularly - A historical Perspective” (PDF). Retrieved 2007-
Steady Motion. Macmillan and co. 10-10.
[6] Hurwitz, A. (1964). “On The Conditions Under Which • Sontag, Eduardo (1998). Mathematical Control
An Equation Has Only Roots With Negative Real Parts”. Theory: Deterministic Finite Dimensional Systems.
Selected Papers on Mathematical Trends in Control The-
Second Edition (PDF). Springer. ISBN 0-387-
ory.
98489-5.
[7] Richard Bellman (1964) Control Theory, Scientific Amer-
ican 211(3):186–200 • Goodwin, Graham (2001). Control System Design.
Prentice Hall. ISBN 0-13-958653-9.
[8] Ang, K.H., Chong, G.C.Y., and Li, Y. (2005). PID con-
trol system analysis, design, and technology, IEEE Trans • Christophe Basso (2012). Designing Control Loops
Control Systems Tech, 13(4), pp.559-576. for Linear and Switching Power Supplies: A Tutorial
[9] Donald M Wiberg. State space & linear systems. Schaum’s Guide. Artech House. ISBN 978-1608075577.
outline series. McGraw Hill. ISBN 0-07-070096-6.
• Briat, Corentin (2015). Linear Parameter-Varying
[10] Terrell, William (1999). “Some fundamental control and Time-Delay Systems. Analysis, Observation, Fil-
theory I: Controllability, observability, and duality — tering & Control. Springer Verlag Heidelberg. ISBN
AND— Some fundamental control Theory II: Feed- 978-3-662-44049-0.
back linearization of single input nonlinear systems”.
Amer. Math. Monthly 106: 705–719 and 812–828.
doi:10.2307/2589614. For Chemical Engineering
[11] Liu, Jie; Wang, Golnaraghi, Kubica (2010). “A
novel fuzzy framework for nonlinear system con- • Luyben, William (1989). Process Modeling, Simula-
trol”. Fuzzy Sets and Systems 161 (21): 2746–2759. tion, and Control for Chemical Engineers. Mc Graw
doi:10.1016/j.fss.2010.04.009. Hill. ISBN 0-07-039159-9.
10 13 EXTERNAL LINKS

13 External links
• Control Tutorials for Matlab, a set of worked-
through control examples solved by several different
methods.

• Control Tuning and Best Practices

• Advanced control structures, free on-line simulators

explaining the control theory
 11

14 Text and image sources, contributors, and licenses

14.1 Text
• Control theory Source: https://en.wikipedia.org/wiki/Control_theory?oldid=674292191 Contributors: The Anome, Taral, Ap, Seb, Pier-
reAbbat, ChangChienFu, Heron, Ubiquity, Michael Hardy, Lexor, Kku, Glenn, Poor Yorick, Lommer, GRAHAMUK, Barak~enwiki,
Revolver, Daniel Sperry, Ralesk, Robbot, LuckyWizard, Gandalf61, Ancheta Wis, Matt Gies, Giftlite, Rs2, BenFrantzDale, Dratman,
Guanaco, Siroxo, Wmahan, JanHRichter, Bact, Onco p53, Mrdarrett, Simoneau, PeR, Netlad, Frenchwhale, Mike Rosoft, D6, Jwdi-
etrich2, Jiy, Discospinster, Eb.hoop, Rich Farmbrough, TedPavlic, Guanabot, Pak21, Style, Ignignot, Fenice, Gauge, Rgdboer, Euyyn,
Spalding, Billymac00, Mdd, HasharBot~enwiki, Sponge, Hammertime, Mononoke~enwiki, Wtshymanski, Cburnett, Suruena, Aschwarm,
Hazard, Ceyockey, Oleg Alexandrov, Simetrical, Mindmatrix, Benbest, Jeff3000, Julien Tuerlinckx, GregorB, Graham87, Yuriybrisk, Tail-
boom22, Rjwilmsi, Ikelos, Yamamoto Ichiro, Titoxd, SystemBuilder, Neonil~enwiki, Mathbot, Vonkje, Jpkotta, Gwernol, Wavelength,
Pmg, Sasuke Sarutobi, Gaius Cornelius, Jugander, Kdcarver, Spike Wilbury, Welsh, Jpbowen, Gnusbiz, Sahodaran, Wknight94, Bullhad-
dha, Scott Dial, XieChengnuo, Sardanaphalus, Attilios, SmackBot, Kurochka, Royalguard11, Sattyam, Ignacioerrico, Ppntori, GeneralAl-
gernon, Kurykh, Frederick12, Oli Filth, Cvgittings, Complexica, Nbarth, Torzsmokus, SundarBot, Jmdaly, Khukri, S Roper, DavidEll,
StephenReed, Jbergquist, Tilin, Dankonikolic, Bn, Khazar, Copeland.James.H, EmreDuran, Chandu15, Tawkerbot2, Chetvorno, Ro8269,
Adam Keller, CRGreathouse, BeenAroundAWhile, Rawling, Anakata, Engelec, Drinibot, ShelfSkewed, Myasuda, Pulkitgrover, Perroboy,
Briantw, Holnet, Hazmat2, Marek69, OrenBochman, Seh444, Daniel Kellis, JAnDbot, MER-C, The Transhumanist, DCwom, Magiola-
ditis, Riceplaytexas, Tupeliano~enwiki, First Harmonic, Americanhero, User A1, Mastlab, Sigmundg, Absurdburger, R'n'B, Erkan Yil-
maz, Pharaoh of the Wizards, Jiuguang Wang, Rod57, Mbbradford, Sigmundur, Jamesontai, Martial75, Dedmonds, Swrkoab, VolkovBot,
ICE77, JohnBlackburne, Jimmaths, Constant314, Erik the Red 2, Serrano24, IPSOS, Someguy1221, Mickraus, Magmi, Martin451, Ph-
moreno, EverGreg, Mathmanta, John43ny, DrJunge, Dan Polansky, Chimin 07, Lightmouse, Canglesea, A Sengupta, Tuxa, Mani excel,
Sfan00 IMG, ClueBot, Bob12794, Niceguyedc, Dja25, DragonBot, CrazyChemGuy, Poree, Boden Rodenberger, Ykhwong, Brews ohare,
Ricmagno, D1ma5ad, Marklar2007, Control.optimization, DOI bot, Spradlig, Cha'yu, Download, MrVanBot, CarsracBot, Numbo3-bot,
Tide rolls, Lightbot, Jarble, ‫سعی‬, Legobot, Luckas-bot, Yobot, RHB100, Tamtamar, AnakngAraw, AnomieBOT, Coronafire, Daniele
Pugliesi, Zander the Mander, Citation bot, Raffamaiden, Chongkian, Shadowjams, Dudered, FrescoBot, Dinhxuanduyet, Citation bot 1,
Didier wkpd, Puzl bustr, SchreyP, David Pardy, EUvin, Brian the Editor, Markeditwiki, DARTH SIDIOUS 2, Regancy42, Wikitanvir-
Bot, Tommy2010, Mmeijeri, Vsh3r, Zndr.k.94, ChristianSchluchter, Suslindisambiguator, Arman Cagle, Cblambert, Vorziblix, Capricorn
umair, RoyKok, ChuispastonBot, ClueBot NG, MelbourneStar, Thelctrclengnr, Cm578707, Konuts, Theraft, MerlIwBot, Helpful Pixie Bot,
Wojciech.ziniewicz, Snow Blizzard, Brad7777, Gibbja, Jcmg1964, JYBot, Aseemkrishna, Ropongiyoyogi, Limit-theorem, Smilesgiles89,
Phmresearch, Mbord, Samu72~enwiki, Jodosma, FunnyPika, Otherocketman, Writers Bond, Monkbot, HMSLavender, Jorge Guerra Pires,
Loraof, 4x8rot, KasparBot and Anonymous: 279

14.2 Images
• File:Boulton_and_Watt_centrifugal_governor-MJ.jpg Source: https://upload.wikimedia.org/wikipedia/commons/c/c7/Boulton_and_
Watt_centrifugal_governor-MJ.jpg License: CC BY 3.0 Contributors: Own work Original artist: Dr. Mirko Junge
• File:Commons-logo.svg Source: https://upload.wikimedia.org/wikipedia/en/4/4a/Commons-logo.svg License: ? Contributors: ? Original
artist: ?
• File:Complex-adaptive-system.jpg Source: https://upload.wikimedia.org/wikipedia/commons/0/00/Complex-adaptive-system.jpg Li-
cense: Public domain Contributors: Own work by Acadac : Taken from en.wikipedia.org, where Acadac was inspired to create this graphic
after reading: Original artist: Acadac
• File:Feedback_loop_with_descriptions.svg Source: https://upload.wikimedia.org/wikipedia/commons/2/24/Feedback_loop_with_
descriptions.svg License: GFDL Contributors: Own work Original artist: Myself
• File:Folder_Hexagonal_Icon.svg Source: https://upload.wikimedia.org/wikipedia/en/4/48/Folder_Hexagonal_Icon.svg License: Cc-by-
sa-3.0 Contributors: ? Original artist: ?
• File:People_icon.svg Source: https://upload.wikimedia.org/wikipedia/commons/3/37/People_icon.svg License: CC0 Contributors: Open-
Clipart Original artist: OpenClipart
• File:Portal-puzzle.svg Source: https://upload.wikimedia.org/wikipedia/en/f/fd/Portal-puzzle.svg License: Public domain Contributors: ?
Original artist: ?
• File:Simple_feedback_control_loop2.svg Source: https://upload.wikimedia.org/wikipedia/commons/9/90/Simple_feedback_
control_loop2.svg License: CC BY-SA 3.0 Contributors: This file was derived from: Simple feedback control loop2.png: <a
href='//commons.wikimedia.org/wiki/File:Simple_feedback_control_loop2.png' class='image'><img alt='Simple feedback control
loop2.png' src='https://upload.wikimedia.org/wikipedia/commons/thumb/4/45/Simple_feedback_control_loop2.png/50px-Simple_
feedback_control_loop2.png' width='50' height='17' srcset='https://upload.wikimedia.org/wikipedia/commons/thumb/4/45/Simple_
feedback_control_loop2.png/75px-Simple_feedback_control_loop2.png 1.5x, https://upload.wikimedia.org/wikipedia/commons/thumb/
4/45/Simple_feedback_control_loop2.png/100px-Simple_feedback_control_loop2.png 2x' data-file-width='439' data-file-height='150'
/></a>
Original artist: Simple_feedback_control_loop2.png: Corona
• File:Smooth_nonlinear_trajectory_planning_control_on_a_dual_pendula_system.png Source: https://upload.wikimedia.org/
wikipedia/commons/a/a1/Smooth_nonlinear_trajectory_planning_control_on_a_dual_pendula_system.png License: Public domain
Contributors: http://www.youtube.com/watch?v=LaDd0gqwCCs Original artist: DaveSzeto
• File:Text_document_with_red_question_mark.svg Source: https://upload.wikimedia.org/wikipedia/commons/a/a4/Text_document_
with_red_question_mark.svg License: Public domain Contributors: Created by bdesham with Inkscape; based upon Text-x-generic.svg
from the Tango project. Original artist: Benjamin D. Esham (bdesham)
• File:Wikibooks-logo-en-noslogan.svg Source: https://upload.wikimedia.org/wikipedia/commons/d/df/Wikibooks-logo-en-noslogan.
svg License: CC BY-SA 3.0 Contributors: Own work Original artist: User:Bastique, User:Ramac et al.
12 14 TEXT AND IMAGE SOURCES, CONTRIBUTORS, AND LICENSES

14.3 Content license

• Creative Commons Attribution-Share Alike 3.0