Chapter - 1

INTRODUCTION
The design of control systems is generally based on some mathematical model of
the system to be controlled. These mathematical models, analytical or
computational describing closely the physical system to be controlled are difficult
or rather impossible to precisely characterize and simulate. Further, a model, no
matter how detailed, is never a completely accurate representation of the physical
system. Both these aspects give rise to the concept of uncertainty which is always
likely to be present, may be in varying degree. There are several ways in which
the system can be uncertain, the major three are:

 The initial conditions of the system may not be accurately specified or

completely known.
 Systems experience disturbances from their environment. Also system
commands are typically not known a priori.
 The central source, however, is the uncertainty in the accuracy of a system
model itself or the system parameters may vary with time.

In control theory, robust control is an approach to controller design that explicitly

deals with uncertainty. Robust control methods are designed to function properly
provided that uncertain parameters or disturbances are found within some
(typically compact) set. Robust methods aim to achieve robust performance
and/or stability in the presence of bounded modelling errors. The early methods
of Bode and others were fairly robust; the state-space methods invented in the
1960s and 1970s were sometimes found to lack robustness, prompting research
to improve them. This was the start of the theory of robust control, which took
shape in the 1980s and 1990s and is still active today. In contrast with an adaptive
control policy, a robust control policy is static; rather than adapting to
measurements of variations, the controller is designed to work assuming that
certain variables will be unknown but bounded. Informally, a controller designed
for a particular set of parameters is said to be robust if it also works well under a
Page | 1
different set of assumptions. High-gain feedback is a simple example of a robust
control method; with sufficiently high gain, the effect of any parameter variations
will be negligible. From the closed loop transfer function perspective, high open
loop gain leads to substantial disturbance rejection in the face of system
parameter uncertainty. The major obstacle to achieving high loop gains is the
need to maintain system closed loop stability. Loop shaping which allows stable
closed loop operation can be a technical challenge. Robust control systems often
incorporate advanced topologies which include multiple feedback loops and feed-
forward paths. The control laws may be represented by high order transfer
functions required to simultaneously accomplish desired disturbance rejection
performance with robust closed loop operation. High-gain feedback is the
principle that allows simplified models of operational amplifiers and emitter-
degenerated bipolar transistors to be used in a variety of different settings. This
idea was already well understood by Bode and Black in 1927. The theory of
robust control began in the late 1970s and early 1980s and soon developed a
number of techniques for dealing with bounded system uncertainty. Probably the
most important example of a robust control technique is H-infinity loop-shaping,
which was developed by Duncan McFarlane and Keith Glover of Cambridge
University; this method minimizes the sensitivity of a system over its frequency
spectrum, and this guarantees that the system will not greatly deviate from
expected trajectories when disturbances enter the system.

The analysis of robustness deals with examining the effect of various

perturbations on system behaviour; the amount of perturbations that make the
system unstable and another aspect is to determine the worst case system
specifications under perturbations. Further, it is also of interest to develop, fixed
or varying, control schemes (controllers) depending upon system complexity and
performance requirements in presence of uncertainties. One of the approach is to
extend the existing established classical and modern control techniques. Another
approach is the development and utilization of special control techniques

Page | 2
incorporating robustness issues e.g. variable structure control, adaptive control
sliding mode control, quantitative feedback theory, neuro-fuzzy control etc.

In order to explain various techniques for the sake of continuity and

completeness, some of the definitions, terminologies and notations used in the
report are presented next.

1.1 UNCERTAINTY CLASSIFICATION

The uncertainties present in a system may be classified as structured or
unstructured based on their pattern of occurrence and as parametric, non-
parametric or mixed depending upon location of their existence. The uncertainly
tree is shown in Fig 1.1

Uncertainty

Parametric Non-Parametric

Structured Unstructured Structured Unstructured

Real Complex Complex Real Complex Complex

Fig. 1.1: Uncertainty Classification

The significant uncertainty is the values of system parameters under practical

operating conditions is termed as parametric uncertainty.

The non-parametric uncertainty refers to that aspect of system uncertainty which

is associated with un-modelled dynamics, truncation of high frequency modes,
non-linearities and effects of linearization and even time and randomness in the
system.

For structured uncertainly, the exact upper and lower limits for the variation are
known, while, for the unstructured uncertainty this information is available in
terms of some suitable norm only. Further these quantities may be expressed as
real and / or complex. The system representations and problem formulations of
various combinations of uncertainties are vastly different and covers a very wide
Page | 3
spectrum. In this work, the problem of structured parametric uncertainty has been
considered.

1.2 NEED OF DESIGN OF ROBUST CONTROL SYSTEM

Robust stability and performance are two indispensable virtues of any control
system. To stabilize most of the practical systems along with certain acceptable
performance levels, feedback control has to be exercised. Even such controlled
closed loop systems are prone to degradation in performance and sometimes the
loss of stability in the face of parametric perturbations. Controlling these twin
aspects of stability and performance with the help of suitable controllers, even in
presence of uncertainty, is termed as robust control or robust synthesis. The
literature is rich with various control methodologies and schemes to synthesize
robust controllers. Some of the schemes are presented here.

1.2.1 CONTROL DESIGN CLASSICAL WAY

Normally in the conventional control design for SISO system, the stability margin
is specified to ensure stability in the presence of model uncertainties. But, the
uncertainties or perturbations are not quantified, nor performance was taken into
account in terms of noise, disturbance etc. For MIMO systems, many of the SISO
methods cannot be scaled up.

1.2.2 MODEL BASED CONTROL DESIGN ISSUES

Analytical or computational models cannot truly characterize and emulate the
phenomenon. A model no matter how detailed is never a completely accurate
representation of a real physical system. To overcome such issues, we introduce
the concept of robustness in control system. Feedback in a closed loop control
system reduces the effect of disturbances, errors and parameter changes on the
performance of a control system. In the presence of disturbances and sensor
noises, if we wish to design high performance control systems (Robust Control
System), we must include the following considerations in the design steps:

Page | 4
 a) Tracking performance (keep the tracking error small)
b) Disturbance rejection (keep the output y small for disturbance input)
c) Sensitivity to modelling errors (make sensitivity small)
d) Stability margin (make stability robust)
e) Sensitivity to sensor noise (make sensitivity small)
1.3 ROBUST CONTROL
Given a family of uncertain system P{p(s,q) : q ϵ Q} to be controlled along with
a set of design goals, in terms of desired performance measures, the problem of
robust control is to find a suitable controller K that ensures the stability of P with
acceptable performance over the entire uncertainty set Q or determine that none
exists.

Like all other engineering designs, the controller design too involves suitable
trade-offs i.e. a satisfactory comprise amongst various desired performance
specifications. Determination of whether or not there is any possible controller
that can provide suitable trade-offs amongst the performance specifications is as
important as aspect as synthesizing an appropriate controller when one exists. If
no controller can achieve suitable trade-offs then the desired specifications may
have to be relaxed or respelled and/or the system to be controlled may have to be
redesigned with some alterations. In practice, existing controller design methods
are often successful at finding a suitable controller, when one exists. However, if
no controller is able to achieve desired performance it is very difficult to comment
that no suitable controller exists as another design approach or designer may be
able to find a suitable controller.

The controller, thus, obtained may be a dynamic or static controller. The elements
of the dynamic controller are variables which tend to change, or may be selected,
in response to the changes in system parameters or environment e.g. gain
scheduling, adaptive, neuro-fuzzy controllers etc. Whereas, the elements of a
static controller are fixed and this constant set is able to provide desirable
performance over the entire uncertainty set. Compared to dynamic controllers,
static controllers are economical, less complex, and easy to implement but
Page | 5
involve greater degree of trade-offs. In this work emphasis is on designing a static
controller

1.4 WHEN IS A CONTROL METHOD SAID TO BE

ROBUST?
Informally, a controller designed for a particular set of parameters is said to be
robust if it also works well under a different set of assumptions. High-gain
feedback is a simple example of a robust control method; with sufficiently high
gain, the effect of any parameter variations will be negligible. From the closed
loop transfer function perspective, high open loop gain leads to substantial
disturbance rejection in the face of system parameter uncertainty. Other examples
on robust control include sliding mode and terminal sliding mode control.

The major obstacle to achieving high loop gains is the need to maintain system
closed loop stability. Loop shaping which allows stable closed loop operation can
be a technical challenge.

Robust control systems often incorporate advanced topologies which include

multiple feedback loops and feed-forward paths. The control laws may be
represented by high order transfer functions required to simultaneously
accomplish desired disturbance rejection performance with robust closed loop
operation.

1.5 APPLICATIONS OF ROBUST CONTROL

Almost all the practical physical systems encountered in day-to-day life are
uncertain in some sense or the other. Thus, the significance of robust control
theory can said to be universal. This fact has been demonstrated by many
researchers by applying the concepts of robustness to various divergent fields
ranging from engineering systems such as electrical machines, power systems,
engine control flexible structures, flight control, mechanical systems, process
control and robotics etc. , to many non-engineering systems such as economical,
biological, agricultural, socio-technical etc. However, limited applications are
reported for non-engineering systems, though they are more prone to existence of
Page | 6
uncertainties owing to vague mathematical representations. Even the engineering
systems which can be modelled with reasonable precision, it is almost impossible
to eliminate the uncertainty in totality. Simultaneously, efforts have been made
to develop suitable control techniques to handle the uncertainties and provide
acceptable good performance over wide range of parameters.

1.6 SCOPE OF REPORT

The field of robust control is very vast with wide spectrum of tools and techniques
which have been developed over the past three decades. The study and the
treatment of the present thesis considers linear continuous time-invariant
deterministic systems, real parametric structured uncertainties and extensions of
classical and modern control methods.

The study covers the aspects of robust stability, as well as, robust performance
both in time and frequency-domains. The constant controller framework has been
considered for both single-point, single-output (SISO), as well as, multi-input
(MIMO) systems.

1.7 ORGANISATION OF THE WORK

This chapter presents introduction to the robust control problem with reference to
its relevance and significance. System representation under uncertainty and few
mathematical preliminaries essential for understanding the problem are also
presented.

The Chapter-two details the historical development of the research in the area of
robust control in the form of Literature Survey and critical review.

In Chapter-three the problems of robust stability is studied with the help of

suitable examples, emphasising the extension of existing methods for the same.
The robustness measured for stability have also been investigated.

The Chapter-four is devoted to the design of controller satisfying the problem of

robust stability and performance using tuning of controller and linearization of
plant model.
Page | 7
The Chapter-five provides the conclusion of the whole work of the project along
with proposals for future work.

1.8 SUMMARY
Uncertainty in system model is unavoidable. It can be parametric or non-
parametric, structured or non-structured. Ability of a control system to remain
stable and have acceptable level of performance even in presence of such
uncertainties gives rises to the concept of robustness. When dealing with
uncertain systems, analysis has to be carried out for entire family of systems
generated owing to the parametric variations. Here such family of systems is
considered to have invariant degree, independent uncertainty structure and affine
linear dependency on uncertainly vector. The significance of the robust control
analysis and design is evident from the extensiveness of research in the area.

Page | 8
 Chapter - 2
LITERATURE REVIEW
The robust control problem, i.e., the problem of designing accurate control
systems in the presence of significant plant uncertainties, is classical. However,
over the last few decades significant new theory has been developed for the
solution of this problem and the term robust control for this classical problem is
only of recent vintage (1973). An extensive set of approaches towards robustness
analysis and control using a wide variety of mathematical techniques inclusive of
classical, modern and specialized approaches are available. In this, we confine
the term robust control to the design of fixed controllers. The concept of robust
control theory has found divergent application to areas such as aerospace system,
chemical processes, power networks etc.

2.1 HISTORICAL REVIEW

A historical review of recent results in robust control is presented. Starting with

the classical theory and then moving on to modern and specialized approaches.

2.1.1 CLASSICAL SENSITIVITY DESIGN PERIOD

The earlier proposed solution to the “robust control problem” appears in the
patent of H.S Black in 1927. In this now classical patent, Black first proposed
feedback and large-loop gains for the design of an accurate system given
significant plant uncertainties. Unfortunately, most “accurate” system designed
this way were dynamically unstable. It was not until the results of Nyquist in 1932
that the trade-off between dynamic stability large-loop gain was analytically
understood. The Nyquist frequency domain stability criterion and Black’s
concept of large loop gain for system accuracy formed the basis of robust control
design developed in the classical book by Bode published in 1945. Bode also
introduced the differential sensitivity function to provide an analytical measure
for system accuracy improvement, at least for sufficiently small plant variations.
The Bode approach to the design of robust system was extended to finite plant

Page | 9
variations by Horowitz. A number of research papers have appeared based on the
classical theories of Nyquist, Bode and Horowitz. The focus during this period of
time was on loop shaping of single input single output (SISO) systems for
stability, sensitivity reduction, noise suppression etc.

2.1.2 STATE-VARIABLE PERIOD

The next major period in control system theory is state-variable period. In the
early 60’s, R.E. Kalman introduced a number of key variable concepts, i.e.
controllability, observability, optimal linear-quadratic state feedback (LQSF),
optimal state estimation (Kalman filtering), etc. A thorough exposition of the
major results associated with this period may be found in the text of Anderson
and Moore. Unfortunately, with a few notable exceptions, the problem of plant
certainty was largely ignored during this period. One such notable exception was
the introduction of the sensitivity comparison matrix in 1964 by Cruz and Perkins
for analysis of multiple input multiple output (MIMO) systems. The sensitivity
comparison matrix provided an analytic tool for the comparison of closed-loop
vs. open-loop accuracy improvement. It was an early attempt to extend SISO
sensitivity results to MIMO systems. Siljak advocated the use of parameter plane
and optimization techniques for control system analysis and design under
parameter uncertainity. The major contribution of this period, commonly referred
to as the sensitivity design problem were summarized by Cruz and later presented
in textbook form by Frank. In general, the effort during this period evolved
around problems of trajectory insensitivity, eigenvalue/eigenvector insensitivity
etc.

2.1.3 MODERN ROBUST CONTROL PERIOD

In the late 70’s and early 80’s, a renowned interest appeared in the problem of
plant uncertainity. At about the same time, some significant results were being
reported on the analysis of multivariable systems in the frequency domain. In
particular, the concept of coprime matrix fraction description of multivariable
systems was designed as a design tool by Youla and Desoer in 1976 and 1981.
Page | 10
Around the same time Rosenbrock and MacFarlane and Postlethwaite generalized
the Nyquist stability criterion for multivariable systems. In this parameterization
of all stabilizing compensators were introduced. This parameterization has come
to play a key role in the robust stabilization of multivariable systems and is often
referred to as the Yaula parameterization. This confluence of interest in
uncertainity and multivariable systems led to the current period. For the first time
the term robust control, appeared in the title of a conference paper by Davison
in1973 and in a journal article by Pearson and Stats in 1974.

However the seeds for the modern robust control were planted in the two papers
written in early 60’s. One was by James in 1963 which introduced the concept of
‘small-gain’ principle, which plays such a key role in robust stability criterion.
The other was by Kalman in 1964, which demonstrated for SISO systems that
optional LQ state-feedback control laws had some very strong robustness
properties i.e., infinite gain margins and 60-degree phase margins. In 1977,
Safonov and Athans demonstrated that these gains and phase margins extended
to MIMO systems for gain and phase variations in each input channel to the plant.
Unfortunately, when state-estimate feedback is used instead of state-feedback
these desirable robustness properties vanish. Many researchers Neimark,
Aizeman, Mitrovic, Meerov, Thaler, Ackerman etc contributed significantly
towards development of the parametric approach.

The widespread interest in robust design of control systems subjected to

structured perturbations enlarged the scope of parameter space approach by
inclusion of lyapunov method, as well as, frequency domain concepts. The
frequency domain response description is found to be a more natural way to
describe the unstructured uncertainties. Many research efforts in robust design
under structured and unstructured perturbations via frequency domain
characterization are reported. The first major contribution was by Doyle based on
singular value analysis. Doyle and Stein were able to show, however, that the
desirable loop return-difference properties of the optimal LQSF control law can
be recovered by suitable design of the Kalman filter in the feedback loop. Because
Page | 11
of the considerable design experience with optimal LQ and LQG design during
the state-variable period, there was strong interest in extending robustness results
to this class of problems, and a multivariable robust design philosophy emerged,
which was identified as the LQG/LTR(Linear quadratic guassian/loop transfer
recovery). A more sophisticated result was derived by Chen and Desor.
Unfortunately, all of the robustness results based on singular values require the
plant uncertainty to be unstructured and for structured uncertainty the results
become conservative. To address this shortcoming, Doyle proposed the concept
of structured singular value (SVV), however, their computation and optimization
still remains a difficult task.

Kharitonov’s Theorem published in 1978 in Russian technical literature, provided

spark to the research in this field. However, the theorem began to receive attention
from 1983 onwards only, which provided strong motivation for a decade of
furious work in real parametric uncertain system. The work of Bialas and Barmish
are credited for exposing the power of Kharitnov’s theorem. The first notable
result following Kharitnov’s theorem is due to Soh, Berger and Dabke, who gave
a way to compute the radius of the largest stability ball in the space of polynomial
coefficients around the nominal stable system. The next significant development
was the Edge theorem given by Barlett, Hallot and Huang, which added fuel to
the fire. They proved that the root space of the entire family of uncertain system
is bounded by the root loci of the exposed edges i.e. the entire family is stable, if
and only if, all the edges are stable. In 1987 Biernacki, Hwang and Bhattacharya
extended the results by calculating the largest stability ball in the space of
parameters of the plant transfer function itself.

Another noteworthy extension of Kharitnov’s theorem is the Generalized

Kharitnov Theorem (GKT) given by Chapellat and Bhattacharya providing
necessary and sufficient conditions for robust stability of a closed loop system
containing an interval plant in the forward, through the concept of extremal
sets. The extremal set concept is useful in characterizing robust stability under
parametric uncertainty, worst case parametric stability margins, frequency
Page | 12
domain plots such as Bode and Nyquist envelopes, stability margins
and mixed uncertainty stability and performance margins. In
combination with the highly efficient computational techniques of
determining stability margins, such as the Tsypkin-Polyak locus and
those based on the Mapping theorem of Zadeh and Desoer, the GKT
provides the control engineer with a set of powerful techniques for analysis
and design.

Lyapunov function theory is another popular approach for robust control

analysis and design owing to its applicability to even non-linear and time-
varying system. Patel, Toda and Sirdhar, and Yedavalli and Colleagues
exploited Lyapunov function theory for computing stability bounds for a
state-variable system representation. These bounds gives a quantitative
measure of robustness for the system under consideration. Garofalo,
Celentano and Glielmo, Shi and GAO and Boyd and Barratt have
contributed richly towards this end.

Barmish and DeMarco, Leal and Gibson, Mansour and Anderson and
Haddad and Bernstein have utilized the approach through p ar a me tri c
Lya punov fun ction . Th e mo nograph o f Sa fonov contains an
extensive discussion of the application of Lyapunov function approach
to robust stabilization.

One more approach utilized for robust stability analysis is state-variable

minimax approach to the robust control problem. Here the dynamics is
given in state-space form and the performance measure is taken to be the
maximum of an appropriate norm over all admissible parameter variations.
This approach leads to a guaranteed level of performance for all possible
parameter variations. Chang and Peng, represented an early attempt to design
for both robust performance and robust stabilization.

Hurwitz and his colleagues evolved in the early 80’s, an approach to the
control of uncertain systems, which is based on loop-gain shaping and
Page | 13
plant uncertainty represented by ‘templets’ of possible plant transfer
function values at a given frequency. This approach is referred to as
qualitative-feedback-theory (QFT) approach. The QFT approach for
SISO system is summarized by Hurwitz. Extensions of the QFT approach
to multivariable systems are given in a paper by Yaniv and Hurwitz.

Recently Hurwitz stability conditions have been used to design robustly

system. This approach is referred to as Hurwitz-condition approach, has
the advantage that structured parameter variations can be dealt with
directly.

Then there was an approach by Carlucci and Donati in 1975, the so called
norm-uncertainty approach. In this approach, the uncertainty in
characterized by a bound on a branch space norm of the error between
nominal system output and actual system output. Based on the nominal
model, controllers are then designed to yield guaranteed-cost
performance. Some significant new theory has been developed for the
design of robust multivariable system in the frequency domain. Recently,
new mathematical results have appeared, which expedite the computation
of optimal sensitivity solutions. In particular, Ball and Helton, used a
Beurling-Lax theorem to solve a general optimal interpolation problems
and Glover developed a theory of model reduction, which has important
computational implications for the optimal interpolation problem.
Interpolation with bounded analytic matrices play a key role in
multivariable sensitivity optimization and robust optimization.

A number of papers on the robust control of multivariable system appeared

in that issue. The main focus of these papers was the use of singular values
in the design of robust multivariable systems in the frequency domain.
Basically, this represented an extension of . The classical Bode design
approach to multivariable systems. In November 1982, the special issue
of the IEE Proceedings, part D on Sensitivity and Robustness was

Page | 14
published. This special issue included a number of important papers
on robust control which contributed immensely towards development
of this field. The IEEE special issue of 19S I and the 1FEE special issue
in 1982 provided the fundamental analysis tools for robust control which
are applicable to diverse areas. In January 1993, the IFAC came out with a
special issue of Automatic on robust control, to provide some of the latest
directions and perspective of developments in robust control techniques.

Dorato has presented an extensive review of robust contro l. Dorato,

Tempo and Muscat have compiled a detailed bibliography for the
same. Siljak have reviewed the contributions to the parameter space
methods. Dahleh, Tesi and Vicino have presented an Overview of extremal
properties for robust control of interval plants. banuaish and Kang have surveyed
a subset of the body of research sparked by Kharitonov theorem.

This area is establishing to such a stage that forms the part of course studies
and quite a good number of text and reference books on robust control are available
in context with the present work specific review in focussed sub-areas is presented
separately in subsequent sections.

2.2 ROBUST STABILITY ANALYSIS

The problem of ensuring the stability of a closed loop system subjected to

perturbations has been of considerable Interest to researchers for quite some time.
Stability robustness analysis can be viewed from two perspectives, namely, time
domain and frequency domain analysis. The analysis in the frequency domain is
called out using the singular value decomposition where the non-singularity
matrix is the criterion in developing the robustness conditions. Barnett presents
a useful summary and comparison of various robustness tests available, with
respect to their conservatism. Bounds are obtained by Kantor and Andres in the
frequency domain using eigenvalue and M-matrix analysis. On the other hand,
the time domain stability robustness analysis using Lyapunov stability concepts
is illustrated by Bernett and Storey, Bellman, Davison, in the context of robust
Page | 15
controller design, and Desoer, Keel and Bhattacharyya have extended classical
control techniques to systems having uncertain parameters for obtaining bounds
on extremal gain and phase margins. Patel, Toda and Sridhar have presented
explicit hounds on the perturbation of a linear system to maintain stability. Patel
and Toda have given bounds for 'highly structured perturbations' as well as
'weakly structured perturbations', While Lee treats only 'weakly structured
perturbations'. Using Lyapunov approach Yedavalli developed bounds both for
structured and unstructured perturbations for an asymptotically stable system and
addressed the issue of ‘conservatism’ in the time domain stability robustness
bounds. Martin derived a stability robustness measure for a perturbed linear
system. Zhou and Khargonekar used some new robust stability bounds for
structured uncertainty and proved them less conservative in comparison to the
existing ones, Juang presented a new technique to calculate element bounds of
allowable linear time-invariant perturbations such that the poles of a system are
maintained in any pre-determinate region of the complex plane providing useful
quantitative measures for both stability and performance robustness. Liu and
Daley have shown system robustness improvement using polynomial pole
assignment and have combined time domain and frequency domain performance
specifications to realize joint optimum robustness. Dickman investigated a
singular value-based approach to specify the robustness of a multivariable linear
feedback system in state-space representation and has shown that a lower bound
for the robustness measure is maximized when the forbenius norm of the closed
loop system matrix is minimized.

Page | 16
 Chapter - 3
MATERIALS AND METHODS

3.1 Introduction

Due to the presence of energy storing elements in any plant the desired output
may be less than or greater than the reference input (desired value). An automatic
controller compares the actual value of the plant with the reference input,
determines the deviation, and produces the control signal that will reduce the
deviation to zero value or to a small value as depicted in Figure 3.1.

Figure 3.1. Complete Block Diagram of Automatic Controller Used In

Plant

3.2 Classifications of Industrial Controllers

1. Proportional controllers
2. Integral controllers
3. Proportional-plus-derivative controllers
4. Proportional-plus-integral-plus-derivative controllers

Page | 17
3.2.1 Proportional Controller:

In the case of a proportional controller, the output of the controller u(t) is

proportional to the actuating error signal e(t) as shown in figure 3.2. The output
response of the controller is given by the equation

u(t) = Kp*e(t)

In Laplace-transformed quantities:

U(s)
= Kp (Where Kp is the proportional gain.)
E(s)

The output of the controller is as shown in Figure 3.3.

Fig. 3.2. Actuating Error Signal Fig. 3.3. Controller Output Signal

3.2.2 Integral Controller:

For an integral controller, the output u(t) is changed at a rate proportional to the
actuating error signal as shown in figure 3.4. Its output depends on the magnitude
of error as well as the time up to which the error persists.

Page | 18
The output response of the controller is given by equation:

du
=Ki*e(t)
dt

U(s)
In Laplace-transformed quantities: = Ki/s
E(s)

The output of the controller is as shown in Figure 3.5.

Figure 3.4 Actuating Error Signal Figure 3.5 Controller output signal

This controller is used for reducing the steady state error. Output signal increases
continuously if steady state error does not decrease with time. This control action
will try to decrease error more efficiently as the factor of time is also introduced
in it.

3.2.3 Proportional-plus-derivative controllers:

Derivative control action, when added to a proportional controller, provides a

means of obtaining a controller with high sensitivity. In derivative type of
controller the magnitude of output signal is proportional to the rate of change of
actuating error signal. It anticipates Derivative control action is always used along
with proportional control because if the constant error persists, its rate of change
is zero and derivative controller will not work in this condition. Proportional
controller will work in this condition also.
Page | 19
 de
u(t) = KD
dt

3.2.4 Proportional-plus-integral-plus-derivative controllers:

PID control is a feedback mechanism which is used in control system. This type
of control is also termed as three term control. By controlling the three parameters
- proportional, integral and derivative we can achieve different control actions for
specific work. PID is considered to be the best controller in the control system
family. For PID control the actuating signal consists of proportional error signal
added with derivative and integral of the error signal. Therefore, the actuating
signal for PID control is: –

𝑑𝑒(𝑡)
𝑒𝑎 (𝑡) = 𝑒(𝑡) + 𝑇𝑑 + 𝐾𝑖 ∫ 𝑒(𝑡)𝑑𝑡
𝑑𝑡

PID controller combines advantages of proportional, derivative and integral

control actions.

3.3 Kharitonov's Theorem

Kharitonov's theorem is a result used in control theory to assess the stability of
a dynamical system when the physical parameters of the system are used to check
if the system is stable (i.e. if all roots have negative real parts). Kharitonov's
theorem can be used in the case where the coefficients are only known to be
within specified ranges. It provides a test of stability for a so-called interval
polynomial, while Routh-Hurwitz is concerned with an ordinary polynomial.

An interval polynomial is the family of all polynomials

p(s) = a0 + a1s + a2s2 + a3s3 + …. + ansn

Where each coefficient ai belongs to real numbers which can take any value in the
specified intervals.

An interval polynomial is stable (i.e. all members of the family are stable) if and
only if the four so-called Kharitonov polynomials are stable.

k1(s) = l0 + l1s1 + u2s2 + u3s3 + l4s4 + l5s5 + …

Page | 20
 k2(s) = u0 + u1s1 + l2s2 + l3s3 + u4s4 + u5s5 + …

k3(s) = l0 + u1s1 + u2s2 + l3s3 + l4s4 + u5s5 + …

k4(s) = u0 + l1s1 + l2s2 + u3s3 + u4s4 + l5s5 + …

These Kharitonov polynomials are formed by using the maximum and minimum
values of the system parameters in the characteristic equation. Basically, for any
particular frequency wo these four polynomials represent four vertices of a
rectangle termed as Kharitonov Rectangle. If the solution of these four
polynomials lie inside this Kharitonov Rectangle, then the system is said to be
stable.

Fig. 3.6 Kharitonov Rectangle

The advantage of using Kharitonov's theorem is that although in principle we are

testing an infinite number of polynomials for stability, we need to test only four.
This we can do using Routh-Hurwitz or any other method. So it only takes four
times more work to be informed about the stability of an interval polynomial than
it takes to test one ordinary polynomial for stability.

Kharitonov's theorem is useful in the field of robust control, which seeks to

design systems that will work well despite uncertainties in component behaviour
due to measurement errors, changes in operating conditions, equipment wear and
so on.

Page | 21
Example:

Considering an interval polynomial

P(s,q) = [0.25,1.25] s3 + [2.50,3.50] s2 + [0.80,1.20] s + [0.35,1.15]

The four Kharitonov polynomials are formed as

K1(s) = 0.35 + 0.80 s + 3.50 s2 +1.25 s3

K2(s) = 1.15 + 1.20 s + 2.50 s2 +0.25 s3

K3(s) = 1.15 + 0.80 s + 2.50 s2 +1.25 s3

K4(s) = 0.35 + 1.20 s + 3.50 s2 +0.25 s3

Using the classical Hurwitz criterion, it is verified that all four Kharitonov
polynomials are stable, thus, the interval polynomial family is robustly stable.

3.4 Ziegler–Nichols method

The Ziegler–Nichols tuning method is a method of tuning a PID controller using
experience to learn and improve. It was developed by John G.
Ziegler and Nathaniel B. Nichols. It is performed by setting the I (integral)
and D (derivative) gains to zero. The "P" (proportional) gain, Kp is then increased
(from zero) until it reaches the ultimate gain Ku, at which the output of the control
loop has stable and consistent oscillations. Ku and the oscillation period Tu are
used to set the P, I, and D gains depending on the type of controller used:

Control Type Kp Ti Td
P 0.5Ku - -
PI 0.45Ku Tu/1.2 -
PD 0.8Ku - Tu/8
Classic PID 0.6Ku Tu/2 Tu/8

Table 3.1 Multiplication Factors for Ziegler-Nichols Method

Page | 22
These 3 parameters are used to establish the correction u(t) from the error e(t) via
the equation:
𝑡
1 𝑑𝑒(𝑡)
𝑢(𝑡) = 𝐾𝑝 (𝑒(𝑡) + ∫ 𝑒(𝜏)𝑑𝜏 + 𝑇𝑑 )
𝑇𝑖 𝑑𝑡
0

Which has the following transfer function relationship between error and
controller output:

1 𝑇𝑑 𝑇𝑖 𝑠 2 + 𝑇𝑖 𝑠 + 1
𝑢(𝑠) = 𝐾𝑝 (1 + + 𝑇𝑑 𝑠) 𝑒(𝑠) = 𝐾𝑝 ( )𝑒(𝑠)
𝑇𝑖 𝑠 𝑇𝑖 𝑠

Example:

Let us take for example the process:

1
𝐺𝑝 (𝑠) =
(𝑠 + 1)(5𝑠 + 1)(0.2𝑠 + 1)

The ultimate gain will be:

1
𝐺𝑝 (𝑠) =
(𝑠 + 1)(5𝑠 + 1)(0.2𝑠 + 1)

And the frequency of oscillation will be:

1 + 5 + 0.2
𝜔𝐶𝑂 = √ = 2.48998
1 ∗ 5 ∗ 0.2

So the period of oscillation at the ultimate gain is:

2𝜋
𝑇𝑢 = = 2.52339
𝜔𝐶𝑂

The following will be the Ziegler-Nichols controller settings:

 P control:
𝐾𝑢
𝐾𝑐 = = 18.72
2
 PI control:
𝐾𝑢
𝐾𝑐 = = 17.02
2.2
Page | 23
 𝑃𝑢
𝜏1 = = 2.10
1.2
 PID control:
𝐾𝑢
𝐾𝑐 = = 22.02
1.7
𝑃𝑢
𝜏1 = = 1.26
2
𝑃𝑢
𝜏𝐷 = = 0.32
8
3.5 Ackermann's formula
Ackermann's formula is a control system design method for solving the pole
allocation problem. One of the primary problems in control system design is the
creation of controllers that will alter the dynamics of a system and alter the poles
to a more suitable, and sometimes more stable, state. Such a problem can be
tackled by many different methods; one such solution is the addition of a feedback
loop in such a way that a gain is added to the input with which one can change
the poles of the original system. If the system is controllable, an efficient method
for pole placement is Ackermann's formula, which allows one to choose arbitrary
poles within the system.

Computing the characteristic polynomial and choosing a suitable feedback matrix

can be a challenging task, especially in larger systems. One way to make
computations easier is through Ackermann's formula. For simplicity's sake,
consider a single input vector with no reference parameter r, such as

𝑢(𝑡) = −𝑘𝑥(𝑡)

𝑥̇ (𝑡) = 𝐴𝑥(𝑡) − 𝐵𝑘 𝑇 𝑥(𝑡),

where 𝑘 𝑇 is a feedback vector of compatible dimensions. Given that the system

is still controllable, Ackermann's method states that the design process can be
simplified by the following equation:

𝑘 𝑇 = [0 0 … 0 1]𝐶 −1 Δ𝑛𝑒𝑤 (𝐴)

Page | 24
In which Δ𝑛ew (A) is the desired characteristic polynomial evaluated at matrix A.

Example:

Consider

We know from the characteristic polynomial of A that the system is unstable

since det(𝑠𝐼 − 𝐴) = (𝑠 − 1)(𝑠 − 2) − 1 = (𝑠 − 1)2 , the matrix A will only
have positive eigenvalues. Thus, to stabilize the system we shall put a feedback
gain 𝐾 = [ 𝑘1 𝑘2 ].

From Ackermann's formula, we can find a matrix k that will change the system
so that its characteristic equation will be equal to a desired polynomial. Suppose
we want

Thus,

Computing the controllability matrix yields

Page | 25
 Chapter - 4

RESULT AND DISCUSSION

In this project work, parameters used to run the model are gains of different
controllers, coupling co-efficient/constants of linearized model, input and output
of model. The obtained values and other parameters required to run simulation in
the work are taken from literature.

4.1 Operating Conditions of The Proposed Model Of Throttle And PID

Controller (Numerical Values)
The dynamical equations of the throttle model are:

𝐽𝜙̈(𝑡) = Σ 𝑀(𝑡) = 𝑀𝑚𝑜𝑡𝑜𝑟 (𝑡) + 𝑀𝑠𝑝𝑟𝑖𝑛𝑔 (𝑡) + 𝑀𝑑𝑎𝑚𝑝𝑖𝑛𝑔 (𝑡)

𝑀𝑚𝑜𝑡𝑜𝑟 (𝑡) = (𝑑𝑢𝑡𝑦)𝐶𝑠

𝑀𝑠𝑝𝑟𝑖𝑛𝑔 (𝑡) = −𝐾𝑠 (𝜑(𝑡) − 𝜑𝑒𝑞 )

𝑀𝑑𝑎𝑚𝑝𝑖𝑛𝑔 (𝑡) = −𝐾𝑑𝜑̇ (𝑡)

𝐽𝜙̈(𝑡) = −𝐾𝑠 (𝜑(𝑡) − 𝜑𝑒𝑞 ) − 𝐾𝑑𝜑̇ (𝑡) + (𝑑𝑢𝑡𝑦)𝐶𝑠

𝜑(𝑡 = 0) = 0;

𝜑̇ (𝑡 = 0) = 0;

The following operating conditions are obtained for a throttle-

Inertia of the throttle plate : J = 2.5e-4 kg-m2

Spring constant : Cs = 0.2 N-m/rad

Viscous friction constant : Cd = 0.04 N-m-s/rad

Torque constant : Ct = 4.0 N-m

1
𝜙̈(𝑡) = [−𝐾𝑠 (𝜑(𝑡) − 𝜑𝑒𝑞 ) − 𝐾𝑑𝜑̇ (𝑡) + (𝑑𝑢𝑡𝑦)𝐶𝑠 ]
𝐽

Page | 26
 Fig.4.1. shows the complete block diagram of throttle control model of an isolated
system with constant parameters.

Fig.4.1 Block Diagram of a Throttle Model

The block diagram of throttle model with PID controller as a single subsystem with
single input and single output is:

Fig.4.2 Block Diagram of Throttle Model with PID Controller as a Single

Subsystem

Page | 27
• In this throttle model, a PID controller (standard for linear controls) is first added
to create a control loop.
• A signal builder block is used for flexibility, grouping of multiple signals, and
simulating multiple scenarios.
• The simulation results can then be viewed with the Scope feature, even if there are
multiple signals.
A unit step signal is given to the signal builder block as the input.

Fig.4.3 Input given to the system via signal builder block

4.2 Plant Response

4.2.1 Case 1: Without tuning of PID controller

The values of controller parameters used in PID controller are given in table 4.1.

P 1

I 1

D 0

Table 4.1 Values of Controller Parameters without Tuning

Page | 28
 Fig. 4.4 Steady state response of throttle speed control model and PID
controller of an isolated system.

• The above plot shows the output obtained with a unit step input and the
initial values of P, I and D as 1, 1 and 0 respectively.
• This plot shows the irregularities and high values of steady state error. The
irregularities include high overshoot and vibrations and we need to remove
these irregularities.
4.2.2 Case 2: With tuning of PID controller

In order to improve the irregularities to obtain the desired system behavior, we

influence our system behavior by using PID controller and tuning its parameters.
The values of controller parameters used in PID controller are defined as
variables and are given in table 4.2.

P p

I i

D d

Table 4.2 Values of Controller Parameters

Page | 29
The values of p, i and d are defined in the workspace window of MATLAB as the
predefined values 1, 1 and 0 as earlier.

The response of the system for the predefined values of the PID controller is:

Fig.4.5 System Response before Tuning

After tuning the PID controller and defining the values of p, i and d by adjusting
the response time and transient behaviour of controller we get the response as:

Fig 4.6 System Response after Tuning

Page | 30
The following table shows the controller parameters before and after tuning:

Tuned Block
P 0.18067 1
I 1.7522 1
D -0.00048953 0

Table 4.3 Values of Controller parameters before and after Tuning

The following table shows the performance and robustness of the PID controller
before and after tuning:

Tuned Block
Rise Time 0.0717 seconds 0.0157 seconds
Setting Time 0.338 seconds NaN seconds
Overshoot 8.58% 4.41%
Peak 1.09 1.04
Gain margin Inf dB @ Inf rad/s Inf dB @ Inf rad/s
Phase margin 69 dB @ 19.8 rad/s 62.7 dB @ 89.4 rad/s
Closed-loop stability Stable Stable

Table 4.4 Performance Parameters before and after Tuning

The above performance parameters in time domain guarantees the robustness of

the control system.

Page | 31
Observing the frequency response parameters of our PID controller before and
after tuning we can conclude that our system is stable because of positive values
of gain margin and phase margin. This is shown by the following Bode plot:

Fig. 4.7 Bode Plot for Tuned and Block Response

Page | 32
The following plot shows the output of the plant after tuning the parameters of
PID controller:

Fig. 4.8 Output of the Plant after Tuning

When we look at the values of the scope what we see is a much smoother
behaviour and response is reaching at 90 degrees a lot faster.

4.3 Linearization of Non-linear System

Due to natural causes such as accumulation of dirt and dust, the friction in the
throttle system increases which changes the block parameters such as spring
constant, torque constant and the viscous friction constant. All these uncertainties
can effect control qualities. To deal with these uncertainties we undergo
linearization of our plant by giving the tolerance level to all the uncertain
parameters. So in order to get the exact range of values we provide 20 percent
tolerance to the uncertain parameters. For this we apply linear control theory and
specify the linearization of Simulink block. For linearization, we define the
analysis points, by putting input perturbations at controller input response and
output measurement at plant output response and then observe the plant behaviour
between these two points.

Page | 33
The following figure shows the step response after linearization of our plant
model which includes uncertainties in spring constant:

Fig. 4.9 Step Response after Linearization of Plant Model

By seeing the step response we cannot directly find the worst case or the boundary
conditions of the uncertain parameters. To do the worst case analysis of our
linearized model we use the following inbuilt functions in MATLAB.

1. [mgain,wcvalue] = wcgain(linsys1)
Input parameter = Name of linearized plant model (linsys1 in this case)
Output parameters = Worst case value of uncertain parameters and
maximum gain at that point
As per the case model, the values obtained are
mgain=
LowerBound: 1.1133
UpperBound: 1.1134
Critical Frequency: 9.5955
wcvalue=
Cd = 0.0480
Page | 34
2. sys_worst = usubs(linsys1,wcvalue)
Input parameter = Name of linearized plant model (linsys1 in this case) and
worst case value

Output= Model of worst case value

Fig. 4.10 Output of Worst Case Model in MATLAB

Page | 35
 3. step (linsys1,sys_worst)
This function gives the plot of worst case uncertain parameters in our plant
model.

Fig. 4.11 Step Response of Worst Case Uncertain Parameters

4.4 SUMMARY
This chapter formulates the problem of robust stability. Various methods and
their applications to determine robust stability are detailed with their help of
throttle model. A number of quantitative measures are discussed. IT has been
observed that robustness measure provides an upper bound value of the
perturbation for maintaining the stability. This upper bound may be maximised
by modifying the criteria for robustness measure.

Page | 36
 Chapter - 5
CONCLUSION AND FUTURE SCOPE
To know the physical systems for the purpose of modelling and control have been
the long effort of interest to researchers since long. Linear system theory provides
powerful tools in the applications of modelling and control. It has been observed
that modelling inaccuracies due to uncertainties result in inferior performance and
control. The aim of this project is to study the types of uncertainties and their
effects on model representation as well as on system response. Further, an
overview of robust control has been presented along with a proposed design
method. The conclusion of the work with respect to each chapter is summarized
here.

5.1 SUMMARY

A classification of uncertainties has been presented as structured/unstructured,

parametric/non-parametric and real/complex. The polynomial and matrix
representations if uncertainties in system models has been discussed. Some
mathematical preliminaries are included for the sake of completeness and
reference. The significance of robust control is given with the help of various
application areas. The scope of the work and organization of the thesis have been
presented.

The literature on robust control has been reviewed exhaustively. The references
of classical ad modern control applied to robust study have been discussed with
reference to their concepts, techniques and applications. The publications on,
robustness analysis have been searched with respect to stability and performance.

The stability robustness study is a prime issue in system analysis. Various

methods and their applications to determine robust stability are detailed with the
help of suitable examples. The classical methods namely, root locus ad Nyquist
criterion provide some insight to robustness analysis and are in general not
suitable for rigorous study. Some modifications of these techniques have been

Page | 37
proposed in the literature. Small gain and Kharitonov’s theorems are directly
applicable to robustness analysis. Explicit bounds on perturbation of a linear
system to maintain the analysis are used as robustness measures. Few Lyapunov
equation based robustness measures have been studies with numerical examples.
The issues of conservativeness have also been included.

The analysis of performance in time domain is mainly expressed in terms of

locations of poles/eigenvalues in complex plane. Usual specifications such as rise
time, peak overshoot, settling time etc. can be defined by specific regions in the
complex plane. The performance in presence of uncertainties can be evaluated by
restricting the pole locations to allowable region(s). The allowable region(s) is
(are) the intersection of desired specification constraints. Cost functions
formulation for various time domain specifications has been presented.

The classical frequency domain methods such as Bode plot, Nyquist criterion and
Nichol’s chart are used to analyse the robust performance of uncertain system in
terms of family of curves. The curves bounding the external values form an
envelope. The worst case performance (gain margin and phase margin) can be
evaluated from these envelopes. Suitable examples are included for illustrations.
In case of multi constraints, numerical optimization is used to evaluate the
performance. Few techniques of optimization used in control theory are
presented.

An introduction has been presented to few robust control design methods. The
pole assignment of control is simple and promising. Pole assignment with
additional constraints on performance provides the robustness. A state feedback
control design has been proposed based on pole assignment and simultaneously
optimizing cost functions for performance. It has been observed that this control
results in acceptably good robustness. The algorithm of the proposed design is
given and computational aspects have been discussed. The effectiveness of
proposed algorithm has been demonstrated through a case study.

Page | 38
5.2 FUTURE SCOPE

Efforts may be made to develop new measures of stability robustness reducing

the conservatism. New cost functions may be investigated for performance.
Application of optimization increases the computational burden therefore, search
for simple and easy methods is demanding. The possible extension of proposed
robust design method to present work may be possible in terms of considering
different uncertainty structures, different models structures and different
combinations of specifications etc.

Page | 39
 REFERENCES

1. Swami A.K. October 2003. ‘Robustness Analysis and Control for

Parametric Uncertainties’, Doctor of Philosophy in Electrical Engineering,
G.B.P.U.A.T, Pantnagar. (2003)
2. MATLAB Documentaion
(https://www.mathworks.com/discovery/pid-control.html)
3. Gopal M. 1997. “Control Systems: Principles And Design”, McGraw Hill
Inc., Volume II (1997)
4. Ogata K. 2015. “Modern Control Engineering”, Pearson Inc., Volume V
(2015)
5. Ackermann J. 1993. Robust Control-Systems with Uncertain Physical
Parameters, Springer-Verlag, London, UK. (1993)

Page | 40
 BIO-DATA OF STUDENTS

1. Name: Utsav Goyal

Date of Birth: 22-July-1996

Father’s Name: Mr. Sanjay Goyal

Address: Dehradun, Uttarakhand

Contact No.: +91-9045164277

2. Name: Vivek Sharma

Date of Birth: 18-June-1996

Father’s Name: Mr. V.K. Sharma

Address: Rishikesh, Uttarakhand

Contact No.: +91-9897731487

3. Name: Shubham Mittal

Date of Birth: 24-July-1994

Father’s Name: Mr. Ramdhan Mittal

Address: Rudrapur, Uttarakhand

Contact No.: +91-8979012349

4. Name: Saloni Garg

Date of Birth: 06-February-1997

Father’s Name: Mr. Virendra Garg

Address: Dehradun, Uttarakhand

Contact No.: +91-9690720951

Page | 41