Introduction To Process Control Analysis Mathematical Modeling Control and Optimization 1st Edition Victor A. Skormin (Auth.) Download PDF
Introduction To Process Control Analysis Mathematical Modeling Control and Optimization 1st Edition Victor A. Skormin (Auth.) Download PDF
com
https://textbookfull.com/product/introduction-to-control-engineering-
modeling-analysis-and-design-third-edition-mandal/
textbookfull.com
https://textbookfull.com/product/modeling-control-and-optimization-of-
natural-gas-processing-plants-1st-edition-william-a-poe/
textbookfull.com
https://textbookfull.com/product/managing-as-mission-nonprofit-
managing-for-sustainable-change-1st-edition-lori-heninger/
textbookfull.com
My Beautiful Monsters 03.0 - The Call of Monsters 1st
Edition J.B. Trepagnier
https://textbookfull.com/product/my-beautiful-monsters-03-0-the-call-
of-monsters-1st-edition-j-b-trepagnier/
textbookfull.com
https://textbookfull.com/product/sales-and-distribution-management-
decisions-strategies-and-cases-6th-edition-richard-r-still/
textbookfull.com
https://textbookfull.com/product/galliano-spectacular-fashion-1st-
edition-kerry-taylor/
textbookfull.com
https://textbookfull.com/product/elements-of-classical-and-quantum-
integrable-systems-gleb-arutyunov/
textbookfull.com
Springer Texts in Business and Economics
Victor A. Skormin
Introduction to
Process Control
Analysis, Mathematical Modeling,
Control and Optimization
Springer Texts in Business and Economics
Introduction to Process
Control
Analysis, Mathematical Modeling,
Control and Optimization
Victor A. Skormin
T.J. Watson School of Engineering
Binghamton University
Binghamton, NY, USA
This book is intended for graduate students and practitioners: it implies that readers
have a serious background in at least one of the four areas. A professional statis-
tician may discover how a regression model can facilitate control system design
and/or provide basis for process optimization. A control system engineer may
discover how to assess the accuracy of the traditional state-variable models. This
book may help industrial leaders recognize optimization as an additional resource
for improving process operation and motivate them to bring consultants. In the
Internet era promoting the concepts of recognizing the need for, knowing what is
available, and understanding how it works could be the most appealing feature of
this book. The book is fully consistent with the author’s motto, “I practice what I
teach and teach what I practice.”
vii
Preface
Modern manufacturing facilities can be utilized to their full potential only when
their operation is optimized. For the modern large-scale processes, even modest
steps toward optimum result in very significant monetary gains. The optimization in
Chap. 4 presents typical methods and problems of numerical optimization. These
methods are readily available from various sources, but only to those who are aware
of their existence and do recognize their applicability to particular situations. Once
an optimal regime of operation is found, it has to be implemented and maintained
over some time period. This task is hindered by adverse environmental effects and
undesirable dynamic behavior of the process. Both conditions could be alleviated
by (mostly discrete-time) control systems. Hence, Chap. 3 of this book features
typical control problems and their practical solutions. Mathematical models provide
a quantitative basis for the solution of optimization and control problems. In most
instances process optimization cannot be implemented experimentally. In numer-
ical optimization, mathematical models of the real processes are used as “guinea
pigs” for testing various operational regimes. Control system design is based on
mathematical models of the controlled processes in the form of transfer functions or
matrix–vector descriptions. Therefore, Chap. 2 shows how mathematical models
could be built and statistically validated. It includes cluster models that are not
commonly known but are very useful in some situations. Some statistical and
probabilistic concepts relevant to mathematical modeling are given in Chap. 1;
however, some of these techniques offer useful tools for process analysis. There-
fore, this book is not a book on optimization, control, mathematical modeling, or
statistical analysis—numerous books on these subjects already exist—this book is
intended to show how to apply these powerful disciplines as a set of tools to achieve
a very important goal: improving the operation of modern manufacturing facilities
(in chemistry, metallurgy, power generation, etc.)
ix
x Preface
This book reflects 40 years of the author’s career as a university educator and
industrial consultant, including professional experiences of his numerous former
students. The author is grateful to Christina Stracquodaine and Matthew Davis for
refining this text and specific numerical examples.
xi
xii Contents
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Introduction
xv
Process Control
It is a quantitative world
We bend the properties of matter
We are invincible and bold
In going after bread and butter
We learn in the eternal school
We have an unforgiving teacher
We use a computer as a tool
Discover feature after feature
We use a statistical technique
In space of numerous dimensions
Reveal relations, strong and weak
For the sake of our intentions
We are on an unbiased way
With all the knowledge, strength, and vigor
And shaping mother nature’s clay
Into magnificence of figure
We master rigors of control
Its theory and application
And never use a crystal ball
For our efforts’ validation
We want to minimize the loss
Of time and money being spent
And utilizing the resource
To fullest possible extent
xvii
Chapter 1
Statistical Methods and Their Applications
A random event is an event that may or may not occur as the result of a trial.
Random event is a concept that reflects the qualitative side of random phenomena.
Typically, random events are labeled as A, B, C,. . . and are characterized by
probability. Probability of a random event A, P[A], is a positive number that does
not exceed 1, i.e. 0 P[A] 1. Event A is impossible if P[A] ¼ 0; event A is certain
if P[A] ¼ 1.
Frequency of a random event A represents the likelihood of the occurrence of
A
this event as the result of a trial. Unsurprisingly it is defined as the ratio NN where N is
A
the total number of trials, and N is the number of trials where event A has occurred.
N BA1 N A1 N BA2 N A2
P½B=A1 P½A1 þ P½B=A2 P½A2 þ P½B=A3 P½A3 þ A2
N A1 N N N
N BA3 N A3 N BA1 þ N BA2 þ N BA3
þ ¼
N A3 N N
P½B=A1 P½A1
P½A1 =B ¼
P½B=A1 P½A1 þ P½B=A2 P½A2 þ P½B=A3 P½A3
Now, one can compute the probability of poor quality product subject to the
short-time electrical failure reported at 10 am:
Now re-estimate the probability of M1, M2, and M3 subject to the failure of the
first converter:
P½A=M1 P½M1
P½M1 =A ¼
P½A=M1 P½M1 þ P½A=M2 P½M2 þ P½A=M3 P½M3
:17 :3333
¼ ¼ :447
:17 :3333 þ :13 :3333 þ :08 :3333
:13 :3333
P½M2 =A ¼ ¼ :342
:17 :3333 þ :13 :3333 þ :08 :3333
:08 :3333
P½M2 =A ¼ ¼ :211
:17 :3333 þ :13 :3333 þ :08 :3333
Re-estimate the probabilities of M1, M2, and M3 subject to the failure of the
second converter:
:17 :447
P½M1 =A ¼ ¼ :555
:17 :447 þ :13 :342 þ :08 :211
1.1 Random Events 5
:13 :342
P½M2 =A ¼ ¼ :321
:17 :447 þ :13 :342 þ :08 :211
:08 :211
P½M3 =A ¼ ¼ :124;
:17 :447 þ :13 :342 þ :08 :211
67 211 108
P BSS =A ¼ ¼ :298, P BSS =B ¼ ¼ :413, P BSS =C ¼ ¼ :266,
225 511 406
4 3
P BSS =D ¼ ¼ :125, P BSS =F ¼ ¼ :115
32 26
P B=BSS ¼
P BSS =B P½B
SS SS SS
P B =A P½A þ P B =B P½B þ P B =C P½C þ P BSS =D P½D þ P BSS =F P½F
:413 :426
¼
:298 :187 þ :413 :426 þ :266 :338 þ :125 :027 þ :115 :022
:1759
¼ ¼ :537
:0557 þ :1759 þ :0899 þ :0034 þ :0025
P C=BSS
P BSS =C P½C
¼ SS SS SS
P B =A P½A þ P B =B P½B þ P B =C P½C þ P BSS =D P½D þ P BSS =F P½F
:266 :338
¼
:298 :187 þ :413 :426 þ :266 :338 þ :125 :027 þ :115 :022
:0899
¼ ¼ :274
:0557 þ :1759 þ :0899 þ :0034 þ :0025
A random variable can have any numerical value, x(k), at each trial, k ¼ 1,2,. . . is
the trial (realization) index. How can the general properties of a random variable be
extracted? Consider an array: x(k), k ¼ 1,2,. . .,N.
Detect minimum and maximum values of the variable within the array, XMIN and
XMAX. Divide interval [XMIN, XMAX] into M divisions
Define step Dx ¼ (XMAXXMIN)/M.
Compute the number of realizations within each interval, i.e. number of reali-
zations nj such that XMINþ(j1)Dx < X(k) XMIN þ jDx, j ¼ 1,2,. . .,M
Compute frequencies fj ¼ nj/N, j ¼ 1,2,. . .,M
Build a histogram showing frequencies fj vs x(k) values like the one shown
below in Fig. 1.1 (it is said that a histogram relates values of a random variable to
the frequency of occurrence of these values).
Then assume that N ! 1, M ! 1, and Dx ! 0. It can be seen that the histogram
turns into a continuous line that represents the distribution law of the random
variable x(k), this is called the probability density P(x).
The probability density function can be used to define the probability of random
variable x(k) which satisfies the condition x1 < x(k) x2 as follows:
xð2 xð2 xð1
Frequencies
f3
f2 fj
f1
fM
XMIN DX XMAX
ð
1 ð
1 ð
1
One can realize that P½1 < xðkÞ 1 ¼ PðxÞdx PðxÞdx ¼ PðxÞdxþ
0 0 0
ð0
PðxÞdx ¼ 1:
1
It is known that there are several “typical” distribution laws, however, the most
common is called normal distribution, which provides the best description of most
random phenomena that can be observed in real life.
Normal Distribution has the following probability density:
( )
1 ðx μÞ2
PðxÞ ¼ pffiffiffiffiffi exp ¼ Pðx; μ; σ Þ;
σ 2π 2σ 2
where μ and σ are parameters of the normal distribution law: μ — mean value and
σ — standard deviation. It is said that the normal distribution law reflects funda-
mental properties of nature. In terms of manufacturing, the standard deviation of
any variable characterizing the product represents effects of “forces of nature” on a
manufacturing process, and the mean value represents effects of operators’ efforts
and adjustment of equipment.
The following is the definition of the probability of a normally distributed
random variable x(k), satisfying the condition x1 < x(k) x2:
xð2 xð2 xð1
P½x1 < xðkÞ x2 ¼ Pðx; μ; σ Þdx ¼ Pðx; μ; σ Þdx Pðx; μ; σ Þdx ¼ Φðz2 Þ Φðz1 Þ
x1 0 0
Visit https://textbookfull.com
now to explore a rich
collection of eBooks, textbook
and enjoy exciting offers!
8 1 Statistical Methods and Their Applications
where
zð1 2
xμ x1 μ x2 μ 1 z
z¼ , z1 ¼ , z2 ¼ , Φðz1 Þ ¼ pffiffiffiffiffi exp dz and
σ σ σ 2π 2
0
zð2 2
1 z
Φðz2 Þ ¼ pffiffiffiffiffi exp dz :
2π 2
0
Note that function Φ(z)is an odd function, i.e. ΦðzÞ ¼ ΦðzÞ. This function is
tabulated and is readily available. One should realize that this function continuously
increases between zero and infinity, as shown below.
ð0 2 ð
1 2
1 z 1 z
Φð0Þ ¼ pffiffiffiffiffi exp dz ¼ 0 and Φð1Þ ¼ pffiffiffiffiffi exp dz ¼ :5
2π 2 2π 2
0 0
1
Mx ½N ¼ Mx ½N 1 þ ½xðNÞ Mx ½N 1
N
1 pffiffiffiffiffiffiffiffiffiffiffiffi
Vx ½N ¼ Vx ½N 1 þ ½½xðNÞ Mx ðNÞ2 Vx ½N 1, and Sx ½N ¼ Vx ½N
N1
Similar derivations could result in the formula for the estimation of variance.
Confidence Intervals for Mean Values and Variances One should realize that as
with any statistical estimate, estimates of mean value and standard deviation
extracted from N observations of random variable x(k) are different from
corresponding parameters of the distribution law. How is the discrepancy between
1.2 Random Variables 11
the “true” and the estimated characteristics estimated? It is known that the estima-
tion error is a random variable with its own statistical distribution. It is also known
that estimation according to the above formulae results in unbiased estimates,
i.e. the mean value of the estimation error is zero. The knowledge of the distribution
law results in the following expression for the confidence interval for the mean
value:
P Mx ðNÞ ΔM ðN; αÞ μ Mx ðNÞ þ ΔM ðN; αÞ ¼ 1 2α
Sx ðNÞ
where ΔM ðN; αÞ ¼ tðN; αÞ pffiffiffiffi
N