Modern Control Systems Analysis and Design Using MATLAB® Robert H. Bishop The University of Texas at Austin ADDISON-WESLEY PUBLISHING COMPANY Reading, Massachusetts * Menlo Park, California * New York Don Milis, Ontario + Wokingham, England + Amsterdam * Bonn Sydney + Singapore + Tokyo + Madrid + San Juan + Milan + ParisPreface Won to Profesor Itehard C. Dorf mont of this text. I also id P, Daniel Burkhart for their titude goes to my wife, Lynda, for assist- Manuscript and for her continuous support and RELB. Austin, Texas Contents Preface v 1 MATLAB Basics ba 1.1 Introduction. .......-+ 1 1.2 Statements and Variables... . 2 1.3 Matrices 10 14 Graphics eeu 16 15 Scripts ......- 21 2. Mathematical Modeling of Systems 25 2.1 Introduction. ...-.--++5 25 2.2 Spring-Mass-Damper System . . 26 2.3 Transfer Functions , 30 2.4 Block Diagram Models 35 2.5 Design Example. . . 45, 3 Control System Characteristics 3.1 Introduction... -. +--+ 3.2 Speed Tachometer System . . . 33 English Channel Boring Machines 4 Control System Performance 61 4.1, Introduction . . . . 5 61 4.2 Time-Domain Specifications . 61 4.3. Simplification of Linear Systems . 68 3 Control System Stability 71 5.1 Introduction... . 020.000 eee renee 7 viiviii Contents 5.2 Routh-Hurwitz Stability ©... ., Beer 5.3 Example: Tracked Vehicle Turning Control _ 16 6 Root Locus Method 81 6.1 Introduction... ..... 0... 81 62 Obtaining a Root Locus Plot. 82 6.3 Sensitivity and the Root Locus 88 7 Frequency Response Methods 91 71 Introduction... 2... 0... eee or 7.2 Bode Diagram eee 92 7.3 Specifications in the Frequency Domain... . 1... 95 74 Example: Engraving Machine System... ...... 9g 8 Stability in the Frequency Domain 103 8.1 Introduction... .. 0... = 103 82 Nyquist Plots 104 8.3 Nichols Charts 110 84 Examples 11 ® State-Space Methods 121 9.1. Introduction... .. . , - 121 9.2 Model Relationships - 122 9.3 Stability of Systems in the Time Domain |__| || | 195 94 Time Response . . 129 10 Control System Design 133 10.1 Introduction . . ee ae 133 10.2 Lead Compensation meet’ 10.3 Lag Compensators a oe ++ 136 10.4 Example: Rotor Winder Control System . . . 138 11 Robust Control Systems 151 ILL Introduction... 2.2... 2, eer iol 11.2 Robust PID Controlled Systems 152 Index/Glossary 159 Chapter 1 MATLAB Basics 1.1 Introduction MATLAB is an interactive program for scientific and engineering calculations. The MATLAB family of programs includes the base program plus a variety of toolbozes. The toolboxes are a collection of special files, called M-files, which extend the functionality of the base program. Together the base program plus the Control System Toolbox provide the capability to use MATLAB ‘or control system design and analysis. In the remainder of this book, whenever we refer to MATLAB, you can interpret that as meaning the base program plus the Control System Toolboz. Most of the statements, functions, and commands are computer platform independent. Regardless of what particular computer sys- tem you use, your interaction with MATLAB is basically the sai This book concentrates on this computer platform independent i teraction. A typical session will utilize a variety of objects that allow you to interact with the program. These objects are 1. statements and variables, 2. matrices, 3. graphics, and 4. scripts.2 Chapter 1. MATLAB Basics MATLAB interprets and acts on input in the form of one or more of these objects. Our goal in this chapter is to introduce each of the four objects in preparation for our ultimate goal of using MATLAB for control system design and analysis. ‘The manner in which MATLAB interacts with your computer system is computer platform dependent. Examples of computer de- pendent functions include installation, the file structure, generating hardcopies of the graphics, invoking and exiting a session, and mem- ory allocation. Questions related to platform dependent issues are not addressed here. This is not to imply that they are not impor- tant, but rather that there are better sources of information such as your Users Guide and your local resident expert. This book is not intended as a substitute for your Users Guide. Before proceeding, make sure that you can invoke a session and exit MATLAB . You need to be able to get to the command window and see the command prompt “ >>. To begin a session on a Macintosh you will probably double-click on the MATLAB program icon. On an IBM PC compatible you will probably type matlab at the DOS prompt. ‘The remainder of this chapter is organized as follows. There are four sections corresponding to the four objects listed above. In the first section we present the basics of statements and variables, Fol- lowing that is the subject of matrices. The third section presents an introduction to graphics, and the chapter concludes with a discussion on the important topic of scripts and Mfiles, 1.2 Statements and Variables Statements have the form shown in Figure 1.1. MATLAB uses the assignment so that equals (“=”) implies the assignment of the ex- pression to the variable. The command prompt is two right arrows, “>>”. A typical statement is shown in Figure 1.2, wherein we are entering a 2 x 2 matrix to which we-attach the variable name A. The statement is executed after the carriage return (or enter key) is pressed. The carriage return is not explicitly denoted in the remaining examples in this and subsequent chapters. 1.2, Statements and Variables 3 command prompt Wariable=expression Figure 1.1 MATLAB Statement Form. ‘The matrix A is automatically displayed afte: the statement is executed following the carriage return. If the statement is followed by a semicolon (;), the output matrix A is suppressed, as seen in Figure 1.3. The assignment of the variable A has been carried out even though the output is suppressed by the semicolon. It is often the case that your MATLAB sessions will include intermediate cal- culations for which the output is of little interest. You should use the semicolon whenever you have a need to reduce the amount of output. Output management has the added benefit of increasing the execution speed of the calculations, since displaying screen output, takes time. The usual mathematical operators can be used in expressions ‘The common operators are shown in Table 1.1. The order of the arithmetic operations can be altered by using parentheses. The example in Figure 1.4 illustrates that MATLAB can be used in a “calculator” mode, When the variable name and *=" are omit- ted from an expression, the result is assigned to the generic variable ans, MATLAB has available most of the trigonometric and elemen- tary math functions of a common scientific calculator. The Users >A=[1 2; 4 6] A= 1 2 carriage return 46 : Figure 1.2 Example Statement: Matrix Input4 Chapter 1. MATLAB Basics Figure 1.8 Using Semicolons to Suppress the Output. Guide has a complete list of available trigonometric and elementary math functions; the more common ones are summarized in Table 1.2. Variable names begin with a letter and are followed by any num- ber of letters and numbers (including underscores). Keep the name length to 19 characters since MATLAB remembers only the first 19 characters. It is a good practice to use variable names that describe the quantity they represent. For example, we might use the variable name vel to represent the quantity aircraft velocity. Generally, we do not use extremely long variable names even though they may be legal MATLAB names. Since MATLAB is case sensitive, the variables M and m are not the same variables. By case we mean upper and lower case. This is illustrated in Figure 1.5. The variables M and m are recognized as different quantities. ‘Table 1.1 Mathematical Operators. Addition , Subtraction Multiplication Division Power ~e + 1.2. Statements and Variables 5 >12.4/6.9 ans = 1.7971 Figure 1.4 Calculator Mode. MATLAB has several predefined variables, including Gi inf) (Nan¥i)and(j. JThree examples are shown in Figure 1.6. Nan stands for Not-a-Nuiber and results from undefined operations. Inf rep- resents -+co and pi represents x. The variable i = Y—T is used to represent complex numbers. ‘The variable j = Y/=I can be used for Table 1.2 Common Mathematical Functions. sin(X) | Sine of the elements of X cos(X) | Cosine of the elements of X asin(X) | Arcsine of the elements of X acos(X) | Arceosine of the elements of X tan(X) | Tangent of the elements of X atan(X) | Arctangent of the elements of X atan2(X,Y) | Four quadrant arctangent of the real elements of X and Y Absolute value of the elements of X Square root of X Imaginary part of X Real part of X Complex conjugate of X Natural logarithm of the elements of X logl0(X) | Logarithm base 10 of the elements of X (éxp(X) _ | Exponential of the elements of X6 Chapter 1. MATLAB Basics »M=[1 2]; om=[3 5 7]; Figure 1.5 Variables Are Case Sensitive. complex arithmetic by those who prefer it over i. These predefined variables can be inadvertently overwritten. Of course, they can also be purposely overwritten in order to free up the variable name for other uses. For instance, one might want to use i as an integer and reserve j for complex arithmetic. Be safe and leave these predefined variables alone, as there are plenty of alternative names that can be used. Predefined variables can be reset to their default values by using clear name (e.g,, clear pi). . The matrix A and the variable ans, in Figures 1.3 and 14, are stored in the workspace. Variables in the workspace are automati- pz=3444 Zz 3.0000 + 4.0000i aint ans = »0/0 Warning: Divide by zero ans = NaN Figure 1.6 Three Predefined Variables i, Inf, and Nan. 1.2. Statements and Variables 7 >who Your variables are: A M ans om z leaving 675516 bytes of memory free. Figure 1.7 Using the who Function to Display Variables. cally saved for later use in your session. The who function gives a lint of the variables in the workspace, as shown in Figure 1.7. MATLAB has a host of built-in functions. You can refer to the Users Guide for a complete list. We will describe each function we use as the need arises. The whos function lists the variables in the workspace and gives Additional information regarding variable dimension, type, and mem- ory allocation. Figure 1.8 gives an example of the whos function ‘The memory allocation information given by the whos function can be interpreted as follows. Each clement of the 2 x 2 matrix A Complex No 1 by2 No 1by1 No by 3 No y 1 Yes Grand total is (12 * 8) = 96 bytes, leaving 664912 bytes of memory free. Figure 1.8 Using the whos Function to Display Variables.Chapter 1. MATLAB Basics clear A »who Your variables are: M m z leaving 663780 bytes of memory free. Figure 1.9 Removing the Matrix A from the Workspace. pi ans = F 3.1416 + »format long; pi ans = oa 3.14159265358979 »format short e; pi ans = - 3.1416e+00 »format long e; pi ans = 3.141592653589793e+00 Figure 1.10 Outout Format Conte.) ™ 1.2, Statements and Variables 9 requires 8 bytes of memory for a total of 32 bytes, the 1 x 1 variable ans requires 8 bytes, and so forth. All the variables in the workspace are using a total of 96 bytes. The amount of remaining free memory depends upon the total memory available in the system. Computers with virtual memory will not display the remaining free memory. You can remove variables from the workspace with the clear function. Using the function clear, by itself, removes all items (vari- ables and functions) from the workspace; clear variables removes all variables from the workspace; clear namef name 2... removes the variables namel,; name2, and so forth. The procedure for re- moving the matrix A from the workspace is shown in Figure 1.9 A simple calculation shows that clearing the matrix A from memory freed up more than 32 bytes. In some cases, clearing a variable may not change the value of the displayed free memory at all. The who function displays the amount of contiguous remaining free memory. So, depending upon the “location” of the variable in the workspace, clearing the variable may or may not increase the displayed amount of remaining free memory. The point is that your available free memory may be more than displayed with the who or whos functions. All computations in MATLAB are performed in double precision. However, the screen output can be displayed in several formats This can be changed by using the format function shown in Figure 1.10. Once a particular format has been speci- fled, it remains in effect until altered by a different format input. » WHO. 77? Undefined function or variable ‘Symbol in question ==> WHO > Who 27? Undefined function or variable Symbol in question ==> Who Bioure 1.11 Fonction Names are Case Sensitive.10 Chapter 1. MATLAB Basics Remember that the output format does not affect the MATLAB computations — all computations are in double precision. On the other hand, the number of digits displayed does not nec- essarily reflect the number of significant digits of the number. This is problem dependent, and only you can know the true accuracy of the numbers that you input and that MATLAB displays. Since MATLAB is case sensitive, the functions who and WHO are not the same functions. The function who is a built-in func- tion, so typing who lists the variables in the workspace. On the other hand, typing the uppercase WHO results in the error mes- sage shown in Figure 1.11. Case sensitivity applies to all functions, 1.3 Matrices MATLAB is short for matrit laboratory. The Users Guide describes the program as a high-performance interactive software package de- signed to provide easy access to the LINPACK and EISPACK ma- trix software. Although we will not dwell on the matrix routines underlying our calculations, we will learn how to use the interactive capability to assist us in our control system design and analysis. We begin by introducing the basic concepts associated with manipulat- ing matrices and vectors. The basic computational unit is the matrix. Vectors and scalars can be viewed as special cases of matrices. A typical matrix expres- sion is enclosed in square brackets, [-]. ‘The column elements are separated by blanks or commas and the rows ate separated by semi- lons or carriage returns. Suppose we want to input the matrix A, where 1 aj va log(-1) —sin(x/2) —_cos(m/3) aresin(0.5) arccos(0.8) exp(0.8) A One way to input A is shown in Figure 1.12. The input style in Figure 1.12 is not unique. Matrices can be input across multiple lines by using a carriage return following the semicolon or in place of the semicolon. This is useful for entering large matrices. Different combinations of spaces Matrices u / v vA=[T, -4Y, sqrt(2);) ai] ee log(-1) sin(pi/2) cos(pi/3} Re asin(0.5), acos(0.8) exp(0.8)] A 1.0000, 0- 4.00001 1.4142 0+ 3.14161 1.0000 0.5000 0.5236 0.6435 2.2255, >A=l1 2:45] + —— od reat math] A= 12 45 Figure 1.12 Complex and Real Matrix Input with Automatic Di- mension and Type Adjustment. and commas can be used to separate the columns, and different com-” binations of semicolons and carriage returns can be used to separate the rows, as illustrated in Figure 1.12. No dimension statements or type statements are necessary when using matrices; memory is allocated automatically. Notice in the example in Figure 1.12 that the size of the matrix A is automatically adjusted when the input matrix is redefined. Also notice that the matrix elements can contain trigonometric and elementary math functions, as well as complex numbers. ‘The important basic matrix operations are addition and sub- traction, multiplication, transpose, powers, and the so-called array operations, which are element-to-element operations. The mathe- matical operators given in Table 1.1 apply to matrices. We will not discuss-matriz division, but be aware that MATLAB has a left- and right-matrix division capability. Matrix operations require that the matrix dimensions be com- patible. For matrix addition and subtraction this means that the matrices must have the same dimensions. If A is an n x m matrix12 Chapter 1. MATLAB Basics and Bis ap xr matrix, then A + B is permitted only if n = p and m =r. Matrix multiplication, given by A * B, is permitted only . Matrix-vector multiplication is a special case of matrix multiplication. Suppose b is a vector of length p. Multiplication of the vector 6 by the matrix A, where A is an nx m matrix, is allowed if m = p. Thus, y = AB is the nx 1 vector solution of Examples of three basic matrix-vector operations are given spose is formed with the apostrophe (’). We can pose and multiplication operation to create a vector inner product in the following manner. Suppose w and v ‘are m x 1 vectors. Then the inner product (also known as the dot »A=[1 3; 5 9]; Be[4 -7; 10 0}; i] ans = 5-4 15 9 »b=[1;5]; — — > + _ ea ans = 16 50 Figure 1.13 Three Basic Matrix Operati iti erations: Ad ‘iplica- tion, and Transpose. ares 1. Matrices 13 product) is given by w' + v. The inner product of two vectors is a wcalar, ‘The outer product of two vectors can similarly be computed ax w+ v', The outer product of two m x 1 vectors is an m x mi/ matrix of rank 1. Examples of inner and outer products are given In Figure 1.14. ‘The basic matrix operations can be modified for element-by- clement operations by preceding the operator with a period. The modified matrix operations are known as array operations. The com- monly used array operators are given in Table 1.3. Matrix addition and subtraction are already element-by-element operations and do not require the additional period preceding the operator. However, array multiplication, division, and power do require the preceding dot, as shown in Table 1.3. Suppose A and B are 2 x 2 matrices given by a-[2e]. a-[f 8]. a One ~ | ba ba "Then, using the array multiplication operator, we have auby ce | anba aaabr |” \ Act B= \ 2, »x=[Sipisin( awty <+—_____ ans = -27.9384 oxty! <———_f ans = 3.0327 -65.0000 49.3480 1.9055 -40.8407 31.0063 0.6065 -13.0000 9.8696 Figure 1.14 Inner and Outer Products. \€“4 Chapter 1. MATLAB Basics ‘Table 1.3 Mathematical Array Operators. + | Addition - | Subtraction + | Multiplication Division Power The elements of A. * B are the products of the corresponding ele- ments of A and B. A numerical example of two array operations is given in Figure 1.15. ___ Before proceeding to the important topic of graphics, we need to introduce the notion of subscripting using colon notation. The colon notation, shown in Figure 1.16, allows us to generate a row vector containing the numbers from a given starting value, xj, to a final value, x;, with a specified increment, dx. _ »A=[1;2;3];B=[-6;7;10]; »A*B + —_f Figure 1.15 Array Operations. 1.3. Matricos starting value final value + 4 x=[xj: dx: x¢ ] + increment Figure 1.16 The Colon Notation. We can easily generate vectors using the colon notation, and as we shall soon see, this is quite useful for developing 2-y plots. Buppose our objective is to generate a plot of y = xsin(x) versus 1 for + =0,0.1,0.2,...,1.0. Our first step is to generate a table of y data. We can generate a vector containing the values of x at hich the values of y(z) are desired using the colon notation. This fp illustrated in Figure 1.17. Given the desired 2 vector, the vector la »x=[0:0.1:1]}4 yex.*sin(x); »ixy] ans = CO) 0 0.1000 0.0100 0.2000 0.0397 0.3000 0.0887 0.4000 0.1558 0.5000 0.2397 0.6000 0.3388 0.7000 0.4510 0.8000 0.5739 0.9000 0.7050 1.0000 0.8415 Figure 1.17 Generating Vectors Using the Colon Notation.14 Chapter 1. MATLAB Basics ‘Table 1.3 Mathematical Array Operators Addition Subtraction Multiplication Division Power ars The elements of A. « B are the products of the corresponding ele- ments of A and B. A numerical example of two array operations is given in Figure 1.15. Before proceeding to the important topic of graphics, we need to introduce the notion of subscripting using colon notation. The colon notation, shown in Figure 1.16, allows us to generate a row vector containing the numbers from a given starting value, xj, to a final value, x, with a specified increment, dx. »A=[1;2;3];B=[-6;7;10]; »A*B Figure 1.15 Array Operations. 14, Matrices = starting value final value 1 J x= [xii dx: xg] increment Figure 1.16 The Colon Notation. easily generate vectors using the colon notation, and au aa oe eet this is quite useful for developing 2-y plots. Buppose our objective is to generate a plot of y = esin(z) ans 8 for x = 0,0.1,0.2,...,1.0. Our first step is to generate a table of y data, We can generate a vector containing the values of z at hich the values of y(x) are desired using the colon notation. This illustrated in Figure 1.17. Given the desired x vector, the vector Figure 1.17 Generating Vectors Using the Colon Notation16 Chapter 1. MATLAB Basics u(2) is computed using the multiplication array operation. Creating a plot of y = zsin(z) versus z is a simple step once the table of ey data is generated, 1.4 Graphics Graphics play an important role in both the design and analysis of control systems. An important component of an interactive control system design and analysis tool is an effective graphical capability. A complete solution to the control system design and analysis will eventually require a detailed look at a multitude of data types in many formats. The important plot formats include root-locus plots, Bode plots, Nyquist plots, and time-response plots. The objective of this section is to acquaint the reader with the basic z-y plotting ca- Pability of MATLAB. More advanced graphics topics are addressed as the need arises. MATLAB uses a graph display to present plots. Some computer configurations allow both the command display and graph display to be viewed simultaneously. On computer configurations that allow only one to be viewed at a time, the command display will disappear when the graph display is up. The graph display is brought up automatically when a plot is generated using any function which generates a plot (e.g., the plot function). Switching from the graph display back to the command display is accomplished by pressing any the keyboard. The plot in the graph display is cleared gull junction at the command prompt. TheGhg function is used to switch to the graph display from the command display. There are two basic groups of graphics functions. ‘The first group of functions, shown in Table 1.4, specifies the type of plot. The list of available plot types includes 2-y plot, semilog and log plots. The second group of functions, shown in Table 1.5, allows us to customize the plots by adding titles, axis labels, and text to the plots and to change the scales and display multiple plots in subwindows, The standard z-y plot is created using the plot function. The y data in Figure 1.17 are plotted using the plot function as shown 14, Graphics 17 Table 1.4 Available Plot Formats. Plots the vector z versus the vector y. Plots the vector z versus the vector y. The z-axis is logio; the y axis is linear. Plots the vector z versus the vector y. The z-axis is linear; the y axis is logo. Plots the vector z versus the vector y. Creates a plot with logio scales on both axes. plot(x.y) semilogx(xy) semilogy(xy) loglog(x,y) igure 1.18. ‘The axis scales and line types are automatically ph ee Ts ac are labeled with the xlabel and ylabel commands; e title is applied with the title command. A grid can be placed p the plot by using the grid command. We see that a basic ey ot is generated with the combination of functions plot, xlabel, ! itle, and grid. ’ akg Tas cab placed on the graph by using the plot fune- jon with multiple arguments, as shown in Figure 1.19. The default Table 1.5 Functions for Customized Plots. Puts ‘text? at the top of the plot. Labels the 2-axis with ‘text’. Labels the y-axis with ‘text”, Puts ‘text? at (p1,p2) in screen coordinates where (0.0,0.0) is the lower left and > He (1.0,1.0) is the upper right of the screen. Subdivides the graphics window. Draws grid lines on the current plot. title(‘text’) H xlabel(‘text”) ylabel(‘text’) H text(pl,p2, text? 's¢3) subplot, P| grid18 Chapter 1. MATLAB Basics »x=[0: »y=x,*sin(x); »plot(x,y) »title(Plot of x sin(x) vs x ') »xlabel('x') »ylabel('y') grid Figure 1.18 A Basic 2-y Plot of zsin(z) versus 2. line types can also be altered. The available line types are shown in Table 1.6. The line types will be automatically chosen unless specified by the user. The use of the text function and changing the line types is illustrated in Figure 1.19. The other graphics functions loglog, semilogx, and semilogy are used in a similar fashion to plot. To obtain an 2-y plot where the z-axis is a linear scale and the y-axis is a logis scale, you would use the semilogy function in place of the plot function. ‘The cus- 14, Graphics yl and y2 19 by x=f0:0.1:175 me by ytmx.tsing); y2asin(x); _ [eakade fe > plot(x,y 1,2," » text(0.1 031 oan iy > text(0.1,0.85,'y2 = sin(x) ——' > xlabel('x'), ylabel('y1 and y2'), grid Figure 1.19 A Basic z-y Plot with Multiple Lines. Table 1.6 Line Types for Customized Plots. aS - | Solid line -- | Dashed line Dotted line Dashdot line20 Chapter 1. MATLAB Basics izing features listed in Table 1: ili it logo, semilog, and semilogy cet een 1e graph display can be subdivided into sma i The function subplot(mnp) subdivides the ak ae aa m x n grid of smaller subwindows, where m <2 andn <2. This means the graph display can be subdivided into two or four windows, The integer p specifies the window, where the windows are iwimbered left to right, top to bottom. This is illustrated in Figure 1.20, wh the graphics window is subdivided into four subwindows. etc Figure 1.20 Usi iti x exe ising subplot to Create a 2x2 Partition of the Graph 1.4. Scripts a 1.5 Scripts Up to this point, all of our interaction with MATLAB has been at . the command prompt. We enter statements and functions at the command prompt, and MATLAB interprets our inout and takes the appropriate action. ‘This is the preferable mode of operation when- fever your sessions are short and nonrepetitive. However, the real | power of MATLAB for control system design and analysis derives from its ability to execute a long sequence of commands stored in a file, These files are called M-files since the filename has the form file- name.m. A scriptis one type of M-file. The Control System Toolbox 4p a collection of M-files designed specifically for control applications. In addition to the pre-existing M-files delivered with MATLAB and the toolboxes, we can develop seripts for our applications, Scripts are ordinary ASCII text files and are created by using your own text editor. Creating and storing scripts are computer ‘platform dependent topics, which means that you need to seek out ithe appropriate expert at your location for more information. A script is just a sequence of ordinary statements and functions yhat you would use at the command prompt level. A script is invoked {at the command prompt level by simply typing in the filename (with- Fout the .m file type). Scripts can also invoke other scripts. When the script is invoked, MATLAB executes the statements and func- tions in the file without waiting for input at the command prompt The script operates on variables in the workspace. Suppose we want to plot the function y(t) = sinat, where a Js a variable that we want to vary. Using our favorite text editor, we write a script, which we will call plotdata.m. This is shown in Figure 1.21. We input a value of a at the command prompt, and in doing so we place a in the workspace. Then we execute the script by typing in plotdata at the command prompt. ‘The script plotdata.m will use the most recent value of a in the workspace. After executing the script we can enter another value of a at the command prompt and execute the script again. Your scripts should be well documented with comments. A com- ment begins with a %. If you put a header in your script comprised of several descriptive comments regarding the function of the script, yeChapter 1. MATLAB Basics % This is a script to plot the function y=sinalpha’ % % The value of alpha must exist in % to invoking the script. % the workspace prior t=[0:0.01:1]; y=sin(alpha*t); plot(t,y) xlabel("Time [sec]') ylabel('y(t) = sin( alpha * t )') grid Figure 1.21 A Simple Script to Plot the Function y(t) = sin at. then using the help function will display the header comments and describe the script to the user. This is illustrated in Figure 1.22. We use plotdata.m to develop an interactive capability with & as a variable, as shown in Figure 1.23. At the command prompt, we input a value of a = 10 followed by the script filename, which in this case is plotdata. The graph of y(t) = sinat is automatically »help plotdata Ht This is a script to plot the function y=sin(alpha*t), The value of alpha must exist in the workspace prior to invoking the script. Figure 1.22 Using the help Function. 23 Li, Scripts ‘command prompt palpha=10; plotdata sin ( alpha * 1) yo). ‘sin (alpha *t) y(t) igure 1.28 An Interactive Session Using a Script to Plot the Func- jon y(t) = sinat. a erated. We can now go back to the command prompt, enter a Value of a = 50, and run the script again to obtain the updated plot. ‘The graphics capability of MATLAB extends beyond the intro- luctory material presented here. We will investigate the issue of aphics further on an as-needed basis.24 Notes Chapter 1. MATLAB Basics Chapter 2 Mathematical Modeling of Systems 2.1 Introduction ‘The design and analysis of control systems is based on mathemati- tal models of complex physical systems. The mathematical models, which follow from the physical laws of the process, are generally highly coupled nonlinear differential equations. Fortunately, many jhysical systems behave Jinearly around an operating point within ‘tome range of the variables and it is possible to develop linear ap- [proximations to the physical systems. A Taylor series expansion is erally utilized in the linearization process. The linear approxima- Jon to the physical system is described by a linear, constant coeffi- ent ordinary differential equation. The Laplace transform method — is one way to compute the sohiition of the differential equation. The japlace transform can also be used to obtain an input-output de- jscription of the linear, time-invariant (LTT) system in the form of 8 transfer function. The application of the many “classical” and “modern” control system design and analysis tools are based on LTI mathematical models. MATLAB can be used with-LTI systems given in the form of transfer function descriptions or state-space descriptions (see Chapter 9, State-Space Methods). We begin this chapter by showing how to use MATLAB to assist, in the analysis of a typical spring-mass-damper mathematical model of a mechanical system. Using a MATLAB script, we will develop an interactive analysis capability to analyze the effects of natural 526 Chapter 2. Mathematical Modeling of Systems frequency and damping on the unforced response of the mass dis- placement. This analysis will utilize the fact that we have an ana- lytic solution that describes the unforced time response of the mass displacement, In the subsequent sections, we will discuss transfer functions and block diagrams. In particular, we are interested in how MATLAB can assist us in manipulating polynomials, computing poles and ze- 10s of transfer functions, computing closed-loop transfer functions, block diagram reduction, and computing the response of a system to a unit step input. The chapter concludes with the electric traction motor control design example found in MCS, pp. 79-81. The functions covered in this chapter are roots, roots1, se- ries, parallel, feedback, cloop, poly, conv, polyval, printsys, minreal, pamap, and step. 2.2 Spring-Mass-Damper System A spring-mass-damper mechanical aystem is shown in Figure 2.1. The motion of the mass, denoted by y(t), is described by the differ. ential equation Mij(t) + fe(t) + Ky(t) = r(2). (2.1) This system is described in MCS, pp. 96-41. The solution, y(t), of the differential equation describes the displacement, of the mass as 4 fanction of time. The forcing function is represented by r(t), The derivation of the spring-mass-damper mathematical model is based on the use of ideal springs and dampers, These ideal models for the spring and damper are based on lumped, linear, dynamic elements and only approximate the actual elements. The spring-mass-damper model, given in Eq. (2.1), is a linear, time-invariant approximation to the physical process; it is valid only in regions where the spring force is a linear function of the mass displacement and the damping due to friction is a linear function of the velocity. The mathematical model, given in Eq. (2.1), might represent an off-road vehicle shock absorber. Our objective could be to design an active control system to make the ride smoother when traversing 2.2. Spring-Mass-Damper System ar CC) Forcing function Figure 2.1 Spring-Mass-Damper System. paved roads. The control design and subsequent analysis would based on the vehicle shock absorber model in Eq. (2.1). Of course, true test of the control design is the road test. Only then can prove that the control design does in fact meet the objective of smoother ride on a bumpy road. We will soon see how to use [ATLAB to enhance our control design and analysis capability. Many physical processes are described by mathematical models nalogous to Eq. (2.1). AA typical electrical RLC circuit is described y an analogous mathematical model where the velocity, g(t), and he voltage, v(t), are analogous variables. This notion of analogous ystems is important in system modeling. Any experience gained In designing and analyzing control systems for mechanical systems scribed by Eq. (2-1) can be used in controlling analogous electrical, thermal, and fluid systems. EVES, ‘The unforced dynamic response, y(t), of the spring-mass-damper mechanical system is e-Srt sin(way/ ~ (2 t+ 8), where @ = cos"!¢. The initial displacement is y(0). ‘The tran- sient system response is underdamped when ¢ <1, overdamped when y(0) WO = 2)28 Chapter 2. Mathematical Modeling of Systems ¢ > 1, and critically damped when ¢ = 1. We Ay ar = 1. We can use MATLAB fe sieusine the unforced time response of the mass displacement following an initial displacement of y(0). Consi Cede (0). Consider the overdamped © Case 1: y(0)=0.15m, wn = V28d, G = as (4=2,5=3) © Case 2: y(0)=0.15m, w, = vist, G= aa (B=2,6=1) The MATLAB commands to generate the plot of the unforced re- sponse are shown in Figure 2.2. In the MATLAB setup, the variables »y0=0.15; wnesqrt(2);_ —<——____— »zetal=3/(2*sqrt(2)); zeta2=1/(2* 5 >t=[0:0.1:10]; ee »unforcedcommands t1=acos(zeta1 )*ones( I,Jength(t)); sa! t2=acos(zeta2)*ones(1,length(t)); a! o1=(y0/sqrt(1-zetatA2))ic2=(y0/sqrt(1-Becada2))s yl=cl *exp(-zetal *wn*t).*sin(wn*sqrt(1-zetal A2)*t+t1); y2nc2texp(-zetaZ*wntt) Psin(wntscrt(I-2etaZA2)*tvt2); buse2Mexp(-zeta2wntt}bi-bu; plot(ty1,"-t,y2,'~',t,bu,'",t,b1,"), grid a ae oe ylabel(’y(t) Displacement [m]') ie ['overdi = eles lamped zetal=',num2str(zetal),... text(0.2,0.80,[" amped y t pat 2 eon ‘underdar Zeta2=' num2str(zeta2),.. Figure 2.2 Script to Analyze the Spring Mass-Damper 22. Spring-Mass-Damper System u(0),Wn, t, Gy and C2 are input to the workspace at the command level. Then the script unforcedcommands.m is executed to gen- grate the desired plots. This creates an interactive analysis capa- bility to analyze the effects of natural frequency and damping on jhe unforced response of the mass displacement. You can investi- te the effects of the natural frequency and the damping on the ime response by simply entering new values of wa, (1, or (2 at the mmand prompt and running the script unforcedcommands.m ain. The time-response plot is in Figure 2.3. Notice that the ript automatically labels the plot with the values of the damping ficients. This avoids confusion when making many interactive wlations. The natural frequency value could also be automati- ly labeled on the plot. Utilizing scripts is an important aspect of veloping an effective interactive design and analysis capability in ATLAB . Since you can relate the natural frequency and damping the spring constant, K, and friction, f, you can also analyze the fects of K and f on the response. In the spring-mass-damper problem, the unforced solution to differential equation, given in Eq. (2.1), was readily available. general, when simulating closed-loop feedback control systems 0.05 8 (Displacement [rn] & ois” Time [sec] Figure 2.3 Spring-Mass-Damper Unforced Response.30 Chapter 2. Mathematical Modeling of Systems subject to a variety of inputs and initial conditions, it is not feasible to attempt to obtain the solution analytically. In these cases, we can use MATLAB to compute the solutions numerically and to display the solution graphically. The simulation capability of MATLAB will be discussed in detail in subsequent sections and chapters. 2.3. Transfer Functions The transfer function is an input-output description of an Ll sys- tem, as described in MCS, pp. 52-68. It relates the Laplace trans- form of the output variable to the Laplace transform of the input variable with zero initial conditions. Consider the LTI system de- scribed by the transfer function G(s), where X(9) Rs) nS + dyna} $ += tars + 0 S$ banis™ tb tas thy! OAs (2.3) where m n=conv(p.q) “*~———| multiply and @._| ne 314 9 i »value=polyval(n,-5) value= ~*——— Evaluate nis) at 9 0-5. ~66 Figure 2.5 Using conv and polyval to Multiply and Evaluate the Polynomials (35? + 2s + 1)(s +4). The function polyval is used to evaluate the value of a polynomial at the given value of the variable. ‘The polynomial n(s) has the value n(—5) = ~66, as shown in Figure 2.5. In the next example we will obtain a plot of the pole-zero lo- cations in the complex plane. This will be accomplished using the Pzmap function, shown in Figure 2.6. On the pole-zero map, ze- ros are denoted by an “o” and poles are denoted by an “x”. If the pzmap function is invoked without left-hand arguments, the plot is automatically generated. _ P: pole locations in column vector zer0 locations in column vector : a Gg) =num Oar 4 : IP.Zl-pzmap(num ) Figure 2.6 The pemap Function. 4. Transfer Functions 33, @ EXAMPLE 2.1 Transfer Functions Jonsider the transfer functions 6s? +1 (st 1)(s +2) Ms) = Sy34 ae FT 1 = Brae Be FH and tilizing a MATLAB script, we can compute the poles and zeros of 9), the characteristic equation of H(s), and divide G(s) by H(s) fe can also obtain a plot of the pole-zero map of G(s)/H(s) in the plex plane. The pole-zero map of the transfer functionG(s)/H(s) is shown Figure 2.7, and the associated MATLAB commands are shown in ure 2.8. The pole-zero map shows clearly the five zero locations, t it appears that there are only two poles. This cannot be the e since we know that the number of poles must be greater than equal to the number of zeros. Using the roots] function we can rtain that there are in fact four poles at s = -1. Hence, multiple ‘or multiple zeros at the same location cannot be discerned on pole-zero map. Imag Axis 4s Real Axis, Figure 2.7 Pole-Zero Map for G(s)/H(s).Chapter 2, Mathematical Modeling of Systems »numg=[6 0 1]; deng=[1 3 3 1]; »z=roots(numg) ze 0 + 0.40821 0 - 0.4082 »peroots1 (deng) <-____ Y anta[1 1]; n2=[1 2]; dl=[1 24%; d2=[1 -2%i]; a3. »numh=conv(n1,n2); denh=conv(d1 ,conv(d2,d3)); »printsys(numh,denh) num/den = SA243842 sacsesessasasineeitiesias SAB 43502445412 133; »num=conv(numg,denh); den=conv(deng,numh); »printsys(num,den) num/den = 6 SAS + 18 sA4 + 25 5A3 475 sA2 445412 S95 + 6544+ 14543 + 165A2 49542 »pzmap(num,den) Pole »title('Pole-Zero Map’) ae Figure 2.8 Transfer Function Example for G(s) and H(s). 24, Block Diagram Modolx 2.4 Block Diagram Models Suppose we have developed mathematical models in the form of transfer functions for the plant, represented by G(s), and the con- roller, represented by H(s), and possibly many other system com- nents such as sensors and actuators. Our objective is to inter- nect these components to form a control system. We will utilize [ATLAB functions to carry out the block diagram transformations. lock diagram models are described in MCS, pp. 54-69. ‘The process to be controlled is shown in Figure 2.9. A simple yen-loop control system can be obtained by interconnecting the lant and the controller in series as illustrated in Figure 2.10. We n use MATLAB to compute the transfer function from R(s) to (s), as illustrated in Example 2.2. EXAMPLE 2.2 Series Connection the process, represented by the transfer function G(s), be L 500s?" let the controller, represented by the transfer function G(s), be +t s+2° fe can use the series function to cascade two transfer functions j,(s) and G(s), as shown in Figure 2.11. The transfer function G,G(s) is computed using the series func- as shown in Figure 2.12. ‘The resulting transfer function, G.G(s), Qs) = G(s) = Process Gs) Ys) Us) ———>| Figure 2.9 Open-Loop System.36 Chapter 2. Mathematical Modeling of Systems 24, Block Diagram Models 37 Als) R(s) %s) »numg=[1]; deng=[500 0 0} pnumh=[1 1]; denh={1 2]; »[num,den]=series(numg,deng,numh,denh); »printsys(num,den) num/den = stl Figure 2.10 Open-Loop Control System. He G.G(s) = SE stl 500 sA3 + 1000 sA2 den ~ 500° + 10008?" Block diagrams quite often have transfer functions in parallel, In such cases, the function parallel can be quite useful. The parallel function is described in Figure 2.13. We can introduce a feedback signal into the control system by closing the loop with unity feedback, as shown in Figure 2.14. The signal E,(s) is an error signal; the signal A(s) is a reference input. In this control system, the controller is in the forward path and the closed-loop transfer function is Figure 2.12 Application of the series Function. re are two functions we can utilize to aid in the block diagram re tion process to compute closed-loop transfer functions for single- 4 multi-loop control systems. These functions are cloop and dback. | GeG(s) ‘The cloop function calculates the closed-loop transfer function 7) = EGG Figure 2.11 The series Function. Figure 2.13 The parallel Function.i 39 38 Chapter 2. Mathematical Modeling of Systems 4. Block Diagram Models Ea(s) [Controller] urs) [Process Rye ae] Ge(s) 4 G(s) yer) eet Figure 2.14 A Basic Control System with Unity Feedback. as shown in Figure 2.15 with the associated system configuration and assumes unity feedback with negative feedback as the default, ‘The feedback function is shown in Figure 2.16 with the associ ated system configuration, which includes Hs) in the feedback path. For both the eloop and feedback functions, if the input “sign” is omitted, then negative feedback is assumed. In Example 2.3 we show an application of the eloop function, and in Example 2.4 we show an application of the feedback function. @ EXAMPLE 2.3 The cloop Function Figure 2.16 The feedback Function. Another basic feedback control configuration is shown in Fig- 2.18. In this case, the controller is located in the feedback path. error signal, E4(s), is also utilized in this control system con- ration. The closed-loop transfer function is Let the process, G(s), and the controller, G(s), be as in Example 2.2 (sce Figure 2.12). To apply the cloop function we first uve the Peja series function to compute G.G(s), followed by the cloop function 1+ GH(s) to dose the loop. The command sequence is shown in Figure 2.17. The closed-loop transfer function, as shown in Figure 2.17, is > ¥(s) T(s) = —GeG(S)__ num _ stl ~T+G.G( den ~ 500s + 10005? +541" »numg~[1]; deng=[500 0 0]; »nume=[1 1]; denc=[1 2]; »{num1 ,den1 ]=series(numg,deng,nume,denc); »{num,den]=cloop(num1 dent,-1); »printsys(num,den) ss num/den = E ‘egative feedback (default) {num,den]=cloop(numi,dent,sign) s+1 +——_+ 500 s43 + 1000 sA2+ s+1 Figure 2.15 The cloop Function, Figure 2.17 Application of the cloop Function.40 Chapter 2. Mathematical Modeling of Systems Figure 2.18 A Basic Control System with the Controller in the Feedback Loop. WM EXAMPLE 2.4 The feedback Function Again, let the process, G(s), and the controller, H(s), be as in Ex: ample 2.2 (.¢., H(s) = G.(s)). To compute the closed-loop transfer function with the controller in the feedback loop we use the feed. back function. The command sequence is shown in Figure 2.19, ‘The closed-loop transfer function is T(s) = —G(s)_ num s+2 “T+ GH{(s) ~ den © 5008? + 10008? +s +1" The MATLAB functions series, cloop, and feedback can be used »numh=[1 1]; denh=[1 2}; [num,den}=feedback(numg,deng,numh,denh,-1); »printsys(num,den) num/den = $+2 —_— eee eer eeeee 500 83 + 1000 s*24 541 Figure 2.19 Application of the feedback Function, 2.4, Block Diagram Models 4 as aids in block diagram manipulations for multi-loop block die grams. This is illustrated in Example 2.5. M@ EXAMPLE 2.5 Multi-Loop Reduction is in Figure 2.20. This example Iti-loop feedback system is shown in Figure 2 dane found in MCS, pp.67-69. Our objective is to compute the closed-loop transfer function = Xb) Ts) = Re)" when 1 1 Gis) = a4" Gils) = s4r° etl stl P4544" 346" Gs(s) = Gas) = and atl aee For this example, a five-step procedure is followed: Ha) = Hp(s) = 2, Hs(s), Step 1: Input the system transfer functions into MATLAB . © Step 2: Move Hy behind Gi. EK 7 to>[om [ew > Y(e) Hs) Figure 2.20 Multi-Loop Feedback Control System.42 Chapter 2. Mathematical Modeling of Systems © Step 3: Eliminate the GsG4F loop. © Step 4: Eliminate the loop containing Hy. * Step 5: Eliminate the remaining loop and calculate T(s). ZThe five steps are illustrated in Figure 2.21, and the corresponding block diagram reduction is shown in Figure 2.22, ‘The result of executing the MATLAB commands is num _ 3° +4ot + 65° + 657+ 5842 den 123" + 20555 + 10669" + 2517s? + 31280? + 21060 FTE We must be careful in calling this the closed-loop transfer function, Recall that the transfer function is defined to the input-output rela- Hionship after pole-zero cancellations. If we compute the poles and zeros of T(s), we find that the numerator and denominator poly- nomials have (s +1) as a common factor. This must be canceled before we can claim we have the closed-loop transfer function. To ongt=[1]; dg1=[1 10}; ang2=[1]}; dg2=[1 1}; ong3=[1 0 1]; dg3=[1 4 4); ong4=[1 4=[1 6]; »nhl=[1 1}; dh1=[1 2]; onh2=[2}; dh2=[1]; »nh3=[1]; dh3=[1]; »nt=conv(nh2,dg4); d1=conv(dh2,ng4); »[n2a,d2a]=series(ng3,dg3,ng4,dg4); »[n2,d2]-feedback(n2a,d2a,nh1,dh1, +1); »[n3a,d3a]=series(ng2,dg2,n2,d2); »[n3,d3]=feedback(n3a,d3a,n1,d1); »[n4,d4]=series(ng1,dg1,n3,d3); »[num,den]=cloop(n4,d4,-1); Figure 2.21 Multi-Loop Block Reduction, 43 24. Block Diagram Models Ra ts) (a) Step 2 [=- z op - Eh Ge co fF (b) Step 3 Ris) cs) (c) Step 4 a we C(S) R(s) LO 3Gal t+ 205H2 + DGS (A) Step 5 ’ Figure 2.22 Block Diagram Reduction of Multi-Loop System (See Example 2.5 in MCS, pp. 67-69).45 4 : M4 Chapter 2. Mathematical Modeling of ‘Systems 45. Design Example 2.5 Design Example Ulectric traction motors are utilized on trains and transit vehicles. ‘The detailed block diagram model with the transfer functions of the Figure 2.23 The minreal Function. assist us in the pole-zero cancellation we wil it tion. The minreal function, shown in Fens ae pole-zero factors of a transfer function, The final step in the block reduction process is to cancel out the common factors, as shown in Figure 2.24. The closed-loop transfer function is given in Fig: lute 2.24 as Ts) = num/den. After the application of the minrecl function we find that the order of the denominator polynomial hex been reduced from six to five, implying one pole-nere cancellacio 1 pole-zeros cancelled 0.08333 SA4 + 0.25 sA3 + 0.25 sA2 + 0.25 s + 0.1667 SA5 + 16.08 8A4 + 72.75 sA3 + 137 sA2 4 123.7 s + 59.33 Figure 2.24 Application of the minreal Function, jower amplifier, armature controlled motor, and sensor, is shown in gure 2.25. This is Example 2.9 in MCS, pp. 79-81. Our objective js to compute the closed-loop transfer function and investigate the pponse of w to a commanded wz. The first step, as shown in Fig- re 2.26, is to compute the closed-loop transfer function w/w. The wed-loop characteristic equation is second-order with w, = 52 and = 0.012. Since the damping is low we might expect the response Jo be highly oscillatory. We can investigate the response w(t) to reference input, w(t), by utilizing the step function. The step nction, shown in Figure 2.27, calculates the unit step response of linear system. The step function is a very important function since control item performance specifications are often given in terms of the it step response. The state response, given by x(t), is an output the step function and will be discussed in detail in Chapter 9, itate-Space Methods. Include « in the left-hand argument list, but not be concerned with it for the time being. If the only objective is to plot the output, y(t), we can use the ep function without left-hand arguments and obtain the plot au- matically with axis labels. If we need y(t) for any reason other plotting, we must use the step function with left-hand argu- ts, followed by the plot function to plot y(t). We define t as a Figure 2.25 Electric Traction Motor Block Diagram.46 Chapter 2. Mathematical Modeling of Systems onumt=[10]; dent =[1 1]; num2~[1}; den2=[2 0.5] »num3=[540]; den3=[1]; num4=[0.1]; den4=[1]; »[nada]=series(num1,den1,num2,den2); »[nb,db]=feedback(na,da,num4,den4,-1); »[nc,dc]=series(num3,den3,nb,db); »[num,den}=cloop(no,de,-1); »printsys(num,den) num/den = 5400 —_—__ 2842 +2.5s +5402 Figure 2.26 Electric Traction Motor Block Reduction, Tow vector containing the times at which we wish the value of the output variable y(t), ‘The step response of the electric traction motor is shown in Fig- le 2.28. AAs expected, the wheel velocity response, given by y(t), is highly oscillatory. Figure 2.27 The step Function. 2.5. Design Example Wheel velocity aT ‘motoresponse.m % This script computes the step % response of the Traction Motor % Wheel Velocity {y.x,t]=step(num,den,t); plot(ty),grid xlabel('Time [sec]") ylabel('Wheel velocity") Figure 2.28 Traction Motor Wheel Velocity Step Response.48 Chapter 2. Mathematical Modeling of Systems ——__________ Notes Chapter 3 Control System Characteristics 3.1 Introduction We introduce feedback to 1. decrease the sensitivity of the system to plant variations, 2. enable adjustment of the system transient response, 3, reject disturbances, and 4, reduce steady-state tracking errors. ‘The advantages of feedback (listed above) come at the cost of in- creasing the number of components and system complexity, reduc- tion in the closed-loop system gain, and the introduction of possi- ble instabilities. However, the advantages of feedback outweigh the disadvantages to such an extent that feedback control systems are found everywhere. In this chapter, the advantages of feedback are illustrated with two examples. Our objective is to illustrate the use of MATLAB in the control system analysis. In the first example, we introduce feedback control to a speed tachometer system in an effort to reject disturbances. The tachome- ter speed control system example can be found in MCS, pp. 125-128. The reduction in system sensitivity to plant variations, adjust- ment of the transient response, and reduction in steady-state error are demonstrated in a second example. ‘This is the English Channel boring machine example found in MCS, pp. 134-197. 4950 Chapter 3. Control System Characteristics 3.2 Speed Tachometer System ‘The open-loop block diagram description of the armature controlled de-motor with a load torque disturbance, T,(s), is shown in Fig- ure 3.1. The values for the various parameters, taken from Example 2.9 in MCS, pp. 79-81, are given in Table 3.1. We have two in- puts to our system, V,(s) and Ty(s). Relying on the principal of superposition, which applies to our LTI system, we consider each input separately. To investigate the effects of disturbances on the system, we let V,(s) = 0 and consider only the disturbance 7,(s). Conversely, to investigate the response of the system to a reference input, we let Ty(s) = 0 and consider only the input Vq(s) The closed-loop speed tachometer control system block diagram is shown in Figure 3.2. The values for K, and K, are given in Table 3. Tour system displays good disturbance rejection, then we expect the disturbance Tu(s) to have a small effect on the output .(s). Consider the open-loop system shown in Figure 3.1 first. We can use MATLAB to compute the transfer function from T,(s) to w(s) and evaluate the output response to a unit step disturbance (i.e., T(s) = 1/s). The time response to a unit step disturbance is shown in Figure 3.3. The script opentach.m, shown in Figure 3.3, is used to analyze the open-loop speed tachometer system. ‘The open-loop transfer function is sls) _ nam Tals) ~ den ~ 26515" Tals) Vals) + LY lal) Kn Lim) V- w(s) Figure 3.1 Open-Loop Speed Tachometer Control System where K, is the Back Electromotive-Force Constant. 42. Speed Tachometer Syste 51 Table 3.1 Tachometer Control System Parameters. Ry\ Km | J) f | Ks | Ka| Ke 1 | 10 |2}05/o1] 54) 1 Since our desired value of w(t) is zero (remember that Va(s) = 0), the steady-state error is just the final value of w(t}, which ve denote by wo(t) to indicate open-loop. The steady-state error, s hoa on the plot in Figure 3.3, is approximately the value of the spe a = T seconds, We can obtain an approximate value ofthe steady- Alate error by looking at the last value in the output vector vos ti We generated in the process of making the plot in Figure 3.3. The approximate steady-state value of w, is w,(00) FY we(7) = —0.6632 rad/see. verifies that we have in fact reached steady-state, Se ai fashion, we begin the closed-loop system analysis by computing the closed-loop transfer function from Ty) t0 (3), an then generating the time response of w(t) to aunt step disturbance input. The output response and the seriptclosedtach.m are shown in Figure 3.4. The closed-loop transfer function from the disturbance 5) gus Rs) + rr] 98) Plis+f ea Figure 3.2 Closed-Loop Speed Tachometer Control System.a Open-Loop Disturbance Stop Response %Speed Tachometer Example % Ra=1; Km=10; J=2; f=0.5; Kb=0.1; num1=[1]; dent =[J,f}; num2=[Km*Kb/Ra]; den2=[1]; [num,den}=feedback(num1 den! num2,den2); % Chaige Sign oF Wanctar trction wie Bhs nume-num; iveutbanes has pegeten an nea sgatlve sign in the diagram. printsys(num,den) ae % Compute response to [yo,xst}=step(numden); plot(tyo) <1 step detubance, title(‘Open-loop Disturbance Step Response") —— Habel( time [sec]'), ylabel('speed’), grid yotlength(t)) —<—{__Steadstate error Figure 3.3 Open-Loop Analysis of the Tachometer System. Chapter 3. Control System Characteristics 32. Speed Tachometer System 53 Closed. Loop Disturbance Step Respense speed [rad/sec] 9 TOOT TOF 0006 OOO COT OIE OUIF OUTS OOS Ot time [sec] closedtach.m % Speed Tachometer Example % Raw; Km=10; J=2; f=0.5; Kb=0.1; Ka=54; Kt=1; num1=[1 num2=[Ka*Kt]; den2=[1]; num3=[Kb]; den3=[1]; num4=[Km/Ra]; den4=[1]; [numa,dena]=parallel(num2,den2,num3,den3); [numb,denb]=series(numa,dena,num4,den4); [num,den]=feedback(num1 ,den1 ,numb,denb); % J, 7 age aA OTT NUM=-NUMT ~« ——| as negative signin the block printsys(num,den) - % [Compute response te [yc,x,t]=step(num,den); plot(t,yc) ~+— ster disturbance. title(‘Closed-loop Disturbance Step Response’ xlabel('time [sec]'), ylabel('speed [rad/sec}’), grid % ye(length(t)) Figure 3.4 Closed-Loop Analysis of the Tachometer System.54 Chapter 3. Control System Characteristics input is ~ den 2s + 541.5" As before, the steady-state error is just the final value of w(t), which we denote by w.(t) to indicate closed-loop. The steady-state error is shown on the plot in Figure 3.4. We can obtain an approximate value of the steady-state error by looking at the last value in the output vector y., which we generated in the process of malking the plot in Figure 3.4, The approximate steady-state value of w is We(00) & we(0.02) = -0.0018 rad/sec. We generally expect that w.(00)/ws(00) < 0.02. ‘The ratio of closed. loop to open-loop steady-state speed output due to a unit step dis. turbance input, in this example, is we(c0) we(00) We have achieved a remarkable improvement in disturbance rejec- tion. It is clear that the addition of the negative feedback loop reduced the effect of the disturbance on the output. This demon. strates the disturbance rejection property of closed-loop feedback systems. = 0.0027. 3.3 English Channel Boring Machines ‘The block diagram description of the English Channel boring ma- chines is shown in Figure 3.5. The transfer function of the output due to the two inputs is K L Oe) = ey (= ayia RO) + app ROU) The effect of the control gain K’ on the transicut response is shown in Figure 3.6 along with the script english1.m used to generate the plots. Comparing the two plots in Figure 3.6, it can be seen that decreasing K decreases the overshoot. Although it is not ae obvious 1.3. English Channel Boring Machines 55 disturbance D(s) boring Figure 3.5 Boring Machine Control System Block Diagram. from the plots in Figure 3.6, it is also true that decreasing K de- creases the selling time. This can be verified by taking a closer look (at the command level) at the data used to generate the plots. This example demonstrates how the transient response can be altered by feedback control gain K. Based on our analysis thus far, we would prefer to use K = 50. However, there are other considerations that must be taken into account before we can establish the final design. Before making the final choice of K, it is important to consider the system response to a unit step disturbance. This is shown in Fig- ure 3.7. We see that increasing K reduces the steady-state response of e(t) to the step disturbance. The steady-state value of o(t) is 0.02 and 0.01 for K = 50 and 100, respectively, The steady-state errors, percent overshoot, and settling times are summarized in Table 3.2. ‘The steady-state values are predicted from the final value theorem as follows: Jiged = tgs s(e+ 12) +K If our only design consideration is disturbance rejection, we would prefer to use K = 100. ‘We have just experienced a very common imde-off situation in control system design. In this particular example, increasing K leads to better disturbance rejection, while decreasing K leads to better performance (via less overshoot and quicker settling time). The final56 Chapter 3. Control System Characteristics 15, __Sjep response for K=100_ 4 —[orerthcok 7] ‘settling tine | os| 0 002 04 06 08 i123 ta ie 1 time [sec] Ls ‘Shep response for K= time [sec} english1.m oe to a Unit Step Input R(s)=1/s for K=50, 100 numge{1]; deng=[1 12 0}; K1=100; K2=50; num1=K1*numg; num2=K2*numg % [numa,dena]=cloop(num1,deng {numb,denb]=cloop(num2,deng); ~ % t=[0:0. 5 i -_ (yl x,t]=step(numa,dena,t); [y2,x,t]-step(namb,denb,t); subplot(211),plot(t.y1), title('Step Response for K=100") xlabel(‘time [Sec]'),ylabel(‘c(t)") subplot(212),plot(t,y2), title('Step Response for K=50') xlabel('time [sec]'),ylabel(‘c(t)") ‘Create a x and. loplats with” is labels, Figure 3.6 The Response to a Step Input with K=100 and K=50. 4.3, English Channel Boring Machines Disturbance Response for K=100. ois; oof e 0.005} 02 04 06 08 1 12 14 16 18 2 time [sec] 0.005, Disturbance Response for K=100 (2 04 0605112 a 2 time [sec] cenglsh2.m % Response to a Disturbance D(s)=1/s for K=50, 100 % humge[1]; deng=(1 12 0]; K1=100; K2=50; % (oedipal [ruma,dena]=feedbacl{numg,dengk1,1); [numb,denb}=feedback(numg,deng, % t=[0:0.05:2.0]; BE ead ya labels. [yl xt]=step(numadena,t); [y2.x.t}=step(nurib,denb,t); subplot(211),plot(ty1) * title(‘Disturbance Response for K=100') xlabel‘time [sec]'),ylabel('c(t)") subplot(212),plot(tiy2) title('Disturbance Response for K=50') xlabel(‘time [sec]'),ylabel(‘e(t)") Figure 3.7 The Response to a Step Disturbance with K=100 and K=50.Chapter 3. Control System Characteristics Table 3.2 Response of the Boring Machine Control System for K = 50 and K = 100. K =50| K = 100 P.O.| 0 10 T | 1a 13 ems | 2% 1% decision on how to choose K rests with the designer. So you see that while MATLAB can certainly assist you in the control system design, it cannot replace your decision-making capability and intuition. The final step in the analysis is to look at the system sensitivity ‘to changes in the plant. The system sensitivity is given by (Eq. 3.62, MCS, p. 137) 1 Mel KG(s) o(s +12) +1 +k" We can compute the values of $(s) for different values of s and generate a plot of the system sensitivity. For low frequencies, we can approximate the system sensitivity by 12s Ko ing the gain K reduces the system sensitivity. ‘The system sensitivity plots are shown in Figure 3.8 for K = 50. The sensitivity approximation is also shown in Figure 3.8. (5) = ahah Fuglish Channel Boring Machines System sensitivity to plant variations 0608 02 04 O65 Se s(s+12) | SO ~s eta) 4K 12 ‘System sensitivity to plant variations, pe ike of - ~ I 3 0 ca w [rad/sec} % System Sensitivity Plot % K=50; num=[1 12 0}; fen=[1 12 ne .A2 + 12%; d nx 1245; d2=K; S2=n2./d2; % plot(w,abs(S),w,abs(S2)) /=[0.1:0.05:20]; s=w*i; 1 Lay S.A2 +12*84K; S=n/ cg approximate sensitivity. | Jot(real(S),imag(S)) ce fitle'System Sensitivity to Plant Variations ') xlabel(‘Real(S)"), ylabel("Imag(S)), arid a pause “Sate xlabel('w [rad/sec]'), ylabel(Abs(S)'), grid 1 s(s) = £66412) 3 (5412) +K [Setupvecworar | K];| 8 =# @to evaluate Figure 3.8 System Sensitivity to Plant Variations.60 Chapter 3. Control System Characteristics Notes Chapter 4 Control System Performance 4.1 Introduction Primary concerns in control system design are stability and perfor- mance. Performance is an issue for stable systems and is the topic of this chapter. In order to design and analyze control systems, We must first establish adequate performance specifications. Per- formance specifications can be presented in the time domain or the frequency domain. Time-domain specifications generally take the rm of settling time, percent overshoot, rise time, and steady-state ror specifications. Stability and frequency-domain specifications addressed in the next chapters. ‘This chapter is organized as follows. In the next section we Investigate time-domain performance specifications given in terms of transient, response to a given input signal and the resulting steady- slate tracking errors. ‘The chapter concludes with a discussion of simplification of linear systems. The MATLAB functions introduced in this chapter are impulse ‘and Isim, These functions are used to simulate linear systems. 4.2 Time-Domain Specifications "Lime-domain performance specifications are generally given in terms of the transient response of a system to a given input signal. Since the actual input signals are generally unknown, astandard fest input 6162 Chapter 4. Control System Performance Figure 4.1 Step Response of a Second-Order System. signal is used. The test signals are of the general form x(t) =t, and the corresponding Laplace transform is When n = 1,2, and 3 we have the n 22) an step, ramp, and parabolic ir repectivaly ‘An impulse function is also used as a test eines fits 1¢ standard performance measures are usually defined in terms : 1e step response and the impulse response. The most common Sep opens performance measures are percent overshoot (P.O.), er tims (B), peak time (T,), and settling time (T,), as shown in Consider the second-order system shown in Figure 4.2. The 4.2, Time-Domain Spe _ Gh Cc | 5(5+an) ia Figure 4.2 Single-Loop Second-Order Feedback System. closed-loop output is oe Rs) CO) = Sy mas bak We have already discussed the use of the step function to com- pute the step response of a system. Now we address another im- portant test signal: the impulse. The impulse response is the time derivative of the step response, We compute the impulse response with the impulse function shown in Figure 4.3. We can obtain a plot similar to Figure 4.5(a) in MCS, p, 162 with the step function, as shown in Figure 4.4. Using the impulse function, we can obtain a plot similar to Figure 4.6 in MCS, p. 163. The response of a second-order system for an impulse function input is shown in Figure 4.5. In the script, we set w, = 1, which is 7] | SYSTEM 7” T Figure 4.3 The impulse Function.64 Chapter 4. Control System Performance zeta = 0.1, 0.2, 0.4,0.7, 1.0,2.0 % Compute step response for a second-ordel % Duplicate Figure 4.5(a) inMCS, p. 162 % t=[0:0.1:12]; num=[1]; zetal=0.1; den1=[1 2*zetal 1); zeta2=0.2; den2=[1 2*zeta2 1]; ~ zeta3=0.4; den3=[1 2*zeta3 1]; zetad~0.7; den4=[1 2*zetad 1]; zetaS=1.0; denS=[1 2*zetaS 1]: Zeta6-2.0; den6=[1 2+zeta6 1]; (y1.x,t]-step(num,dent ,t); [y2,x,t]=step(num,den2,); {y3,x,t]~step(num,den3,t); [y4,x,t]=step(num,dendt): USext-step(rum,den5,t); [y6,x,t]=step(num,den6,t); plot(ty1,t.y2,t,y3,ty4,t,y5,t,y6) xabel(' wn*t'), ylabel(‘c(t)') <— title(‘zeta = 0.1, 0.2, 0.4, 0.7, 1.0, 2.0") Figure 4.4 Reananca af + Cn 65, 42 Time-Domain Specilic 1 | 0.6} 04} 02| (tiiwn o} 02 4 05 a a a Impulseresponse.m % Compute impulse response for a second-order system % Duplicate Figure 4.6 in MCS, p. 163 % t=[0:0.1:10]; num=[1]; zetal=0.1; denl=[1 2*zetal 1]; “*] zeta2=0.25; den2=[1 2*zeta2 1]; zeta3=0. n3=[1 2*zeta3 1]; zeta4=1.0; den4=[1 2*zeta4 1]; % [y1.x,t]-impulse(num,den1 ,t); [y2,x,t]=impulse(num,den2,t); [y3,x,t]=impulse(num,den3,t); [y4.x,t]=impulse(num,den4,t); % ; der ); der plot(ty1,ty2,ty3.ty4) | Gee xlabel(' wn*t’), ylabel('g(t)/wn') title(‘zeta = 0.1, 0.25, 0.5,1.0"), grid Figure 4.5 Response of a Second-Order System to an Impulse.66 Chapter 4. Control System Performance SYSTEM rc) GS) | capa Figure 4.6 The Isim Function. cauivalent to computing the step response versus wnt. This gives us @ mote general plot valid for any w, > 0. Ih many eases, you may need to simulate the syatem response to an arbitrary but known input. In these cases, you can use the Isine function. The Isim function is shown in Figure 4.6. An example of the use of Isim is given in Example 4.1. @ EXAMPLE 4.1 Mobile Robot Steering Control The block diagram for a steering control system for a mobile robot is shown in Figure 4.7 (see MCS, pp. 174-176). Controller Vehicle dynamics R(s) Cs) Tee F ee ie 770 | Figure 4.7 Block Diagram of a. ‘Steering Control System for a Mobile Robot. 67 42. Time-Domain Specifications mobilerobot.m % Compute the response of the Mobile Robot % System to a triangular wave input % | numg=[10 20]; deng=[1 10 0]? [num,den]=cloop(numg,deng); .1:8.2. [V1 v2iv3]; settee danas eee t(try, tu), : tian [sec]’), ylabel(‘theta [rad Figure 4.8 Transient Response of the Mobile Robot Steering Con- trol System to a Ramp Input.68 Chapter 4. Control System Performance Suppose the steering controller, G1(s), is K ls) = K+. When the input is a ramp, the steady-state error is A w= (4.1) where Ky = WK. The effect of the controller constant, Kz, on the steady-state error is evident from Eq, (4.1). Whenever Ky is large, the steady-state error is small, and vice versa, We can simulate the closed-loop system response to a ramp input using the Isim function. The controller gains K, Ky and the system gain K can be represented symbolically in the script so that various values can be selected and simulated. The results are shown in Figure 4.8 for Ky = =1, Ky =2, and r = 1/10, 4.3 Simplification of Linear Systems In practice, it may be necessary to approximate a higher-order trans- fer function model with a lower-order model, For example, it may be impractical to implement a high-order controller in a control sys- jem. However, it may be possible to develop a lower-order appror. imate controller that closely matches the input-output response of the high-order controller. A procedure for approximating transfer functions is given in MCS, pp. 185-187. We can use MATLAB ta compare the approximate model to the actual model, as illustrated in the following example. M EXAMPLE 4.2 A Simplified Model Consider the third-order system 6 "0) = sya Ts +o 13. Simplification of Linear Systems 69 ‘A necond-order approximation (see MCS, pp. 187-188) is 1.60 1) = Fy 354s +100" A compatison of their respective step responses is given in Figure 4.9. % Compare step response for sega % num1=[6]; dent=[1 6 11 67) num2=[1.6]; den2=[1 2.584 1.6 t=[02 is [y1,x,t]=step(num1 dent ,t); [y2,x,t]=step(num2,den2,t); plot(ty1,t.y2), grid zi xlabel('time [sec]'), ylabel('step response") Figure 4.9 Step Response Comparison for an Approximate Transfer Vunction Versus the Actual Transfer Function.70 Chapter 4. Control System Performance ————_______._ Notes Chapter 5 Control System Stability ’ 1 Introduction he stability of a closed-loop control system is a fundamental issue Jn controls. Generally speaking, an unstable closed-loop control sys- Jorn is of little practical value. For linear systems, a necessary and {ficient condition for a feedback system to be stable is that all the les of the system transfer function have negative teal parts. In her words, the poles must lie in the left-half plane for the system be stable, The Routh-Hurwitz stability method provides a struc- sed mechanism for determining the number of unstable poles of «¢ closed-loop characteristic equation. This allows us to obtain a cs” or “no” answer to stability without explicitly calculating the es. ‘This chapter begins with a discussion of the Routh-Hurwitz sta- llity method, We will see how MATLAB can assist us in the stabil- lly analysis by providing an easy and accurate method for comput- ling the poles of the characteristic equation. For the case where the ‘characteristic equation is a function of a single parameter, it will be possible to generate a plot displaying the movement of the poles as the parameter varies. The chapter concludes with an example. ‘The function introduced in this chapter is the function for, which In used to repeat a number of statements a specific number of times, 772 Chapter 5. Control System Stability 5.2 Routh-Hurwitz Stability ‘The Routh-Hurwitz criterion is a necessary and sufficient criterion for stability. Given a characteristic equation with fixed coefficients, we can use Routh-Hurwitz to determine the number of roots in the right-half plane. For example, consider the characteristic equation qs) =? +s? +2s+24=0 associated with the closed-loop control system shown in Figure 5.1. ‘The corresponding Routh-Hurwitz array is shown in Figure 5.2. The two sign changes in the first column indicate that there are two roots of the characteristic polynomial in the right-half plane; hence the closed-loop system is unstable. Using MATLAB we can verify the Ronth-Hurwitz result. by directly computing the roots of the characteristic equation, as shown in Figure 5.3, using the root: function. Recall that the roots function computes the roots of a polynomial. Whenever the characteristic equation is a function of a singlel parameter, the Routh-Hurwitz method can be utilized to determine the range of values that the parameter may take while maintaining stability. Consider the closed-loop feedback system in Figure 5.4. The characteristic equation is q(s) = 3° +s? 448+ K =0. Using a Routh-Hurwitz approach we find that we require 0 < K <8 for stability (see MCS, p. 215). We can use MATLAB to verify this G(s) = ——__1____ (9 Syeda 0B Figure 5.1 Closed-Loop Control System with T(s) = C(s)/R(s) 1/(s? +s? +.2s +24). ‘2. Routh-Hurwitz Stability 73 nee at sign change: gure 5.2 Routh-Hurwitz Array for the Closed-Loop Control Sys- mn with T(s) = C(s)/R(s) = 1/(s? + 5? +28 +24). sult graphically. As shown in Figure 5.5, we establish a vector of malues for K at which we wish to compute the roots of the charac- ristic equation, Then using the roots function we calculate and Jot the roots of the characteristic equation, as shown in Figure 5.5. It can be seen that as K increases, the roots of the characteristic ation move toward the right-half plane as the gain tends toward = 8, and eventually into the right-half plane when K > 8. This is ‘graphical verification of the Routh-Hurwitz result obtained above. the next chapter we will discover a compact method of obtaining pnumg=[1]; deng=[1 1 2 23]; »[num,den]=cloop(numg,deng); »roots(den) ans = 3.0000 1.0000 + 2.6458: 1.0000 - 2.6458: Figure 5.3 Using the roots Function to Compute the Closed-Loop Control System Poles of the System Shown in Figure 5.1.4 Chapter 5. Control System Stability Figure 5.4 Closed-Loop Control System with 7(s) = C' =| K/(s° +28? +454 K). ees : f 8 le 3 ' peommcsumetioscncoo HX 9696-36303 ‘Script computes the roots of the characteristic % equation (s) = 543 + 2502 +454 K for O0. % the stability region for the two track vehicle % control problem % a=[0.1:0.01:3.0]; Km[20:1:1 20]; = X= 0*K; ya"; «ina Hat eon a nelength(K); m=length(a); | ¥#eters of sppropriate length for i=1:n least one root of the characteristic equation in the right-half plane, The process is repeated until the entire selected range of a and K ig exhausted. Then, the plot of the (a, K) pairs defines the separation between the stable and unstable regions, for j=1:m Gel, 8, 17, K(D#10, KO*aD: = ts(q); net een ae ifmax(Fealp)) > 0, ()=K(0s yl=al + E(s) [eva K end cera : ‘i nade ait) i ° plot(xy). arid, abel), ylabel‘a') | ; Figure 5.8 Stability Region for a and K for Two-Track Vehicle Figure 5.7 Turning Control for a Two-Track Vehicle. ‘Turning Control.8 Chapter 5. Control System Stabilit 63. Example: Tracked Vehicle Turning Control The region to the left of the plot of a versus K in Figure 5.8 i the stable region, since that corresponds to K < 126. If we assume that r(t) = At,t > 0, then the steady-state error i Pee s(s+1)(s+2)(s+5) A 10A te 20" s(5 + 1)(s+2)(s+5)+K(sta) 2 aK” where we have used the fact that ay s(s+1)(s +2)(s +5) 1 BO) = Taga = THEE TSF KETO Aol Given the steady-state specification, e,, < 0.24A, we find that tl specification is satisfied when 10A aK < 0-244, or ak > 41.67. (6. Any values of a and K that lie in the stable region in Figure 5 and satisfy Eq. (5.1) will lead to an acceptable design. Por examp! K = 70 and.a = 0.6 will satisfy all the design requirements. T' closed-loop transfer function (with a = 0.6 and K = 70) is % problem with a-0.6 and K=70 % This script computes the ramp response % for the two-track vehicle turning control % t=[0:0.01:10]; u=ty—{_¥ «unit ramp input 70s +42 —— SSS numge=[1 0.6]; denge=[1 1]; s+ 859 + 175? + 80s +42 ile numg=I70]; denge[1 7 10 0}; The associated closed-loop poles are [num,den}=cloop(numa,dena): [numa,dena]=series(numgc,dengc,numg,deng); 2s = -10767, {y,x]=Isim(num,den,u,t); «tin 6 = 0.5781, plot(ty,tu), grid f 8 —0.1726 + 3.1995%, and xlabel(‘time [sec]'), ylabel('c(t)') 5 = ~0.1726 — 3.1995i, j The corresponding unit ramp input response is shown in Figure 5. The steady-state error is less than 0.25, as desired. jicle Turning Control. re 5.9 Ramp Response for a = 0.6 and K = 70 for ‘Two-TrackChapter 5. Control System Stability’ Notes Chapter 6 oot Locus Method .1 Introduction 1c relative stability of a control system is related to the location the roots of the closed-loop characteristic equation. ‘The tran- nit response (ie., settling time, overshoot, etc.) of a linear control tem is also related to the location of the poles and zeros of the ed-loop transfer function. The closed-loop system relative sta- lity and performance can sometimes be adjusted by changing a ameter, such as a control gain. The root locus method provides raphical representation of the locus of roots of the characteristic ation as one parameter is varied. The graphical representation called the root locus plot. ‘An approximate root locus sketch can be obtained by applying orderly procedure outlined in MCS, pp. 241-253. Alternatively, can use MATLAB to obtain an accurate root locus plot. How- r, do not be tempted to rely solely on MATLAB for obtaining ot locus plots while neglecting the manual steps in developing an proximate root locus. The fundamental concepts behind the root, sus method are buried in the manual steps and it is essential to lly understand their application. ‘The chapter begins with a discussion on obtaining a root locus Jot with MATLAB. This is followed by a discussion of the connec- ns between the partial fraction expansion, dominant poles, and 81a2 Chapter 6. Root Locus Method 6.2. Obtaining a Root Locuy Plot 83 the dlosed-loop system response. Root sensitivity is covered in the final section. expansions of rational functions, 6.2 Obtaining a Root Locus Plot Figure 6.2 The rlocus Function. Consider the closed-loop control system in Figure 6.1. ‘The closed. loop transfer function is the root locus plot associated with Eq. (6.1) are shown in Figure 6.3 along with the associated root locus plot. Invoking the rlocus fune- tion without left-hand arguments results in an automatic generation of the root locus plot. When invoked with left-hand arguments, the Flocus function returns a matrix of root locations and the associated sin vector. The steps to obtain a root locus plot with MATLAB are as fol- lows: 1. Obtain the characteristic equation in the form given in Ea. (6.2) where & is the parameter of interest, and 2. use the rlocus function to generate the plots. T(s) = (9) ____—K(s +1(s +3) Ris) s(8+2)(3+3) + K(s+ 1)" ‘The characteristic equation can be written as sti 14K ary =o Teh The form of the characteristic equation in Eq. (6.1) is necessary to Referring to Figure 6.3, we can see that as K increases, two branches of the root locus break away from the real axis. ‘This means that for some values of K, the closed-loop system character istic equation will have two complex roots. Suppose we want to find the value of K corresponding to a pair of complex roots. We can use the rlocfind function to do this, but only after a root locus has been obtained with the rlocus function. Executing the rlocfind function Will result in a cross-hair marker appearing on the root-locus plot. You move the cross-hair marker to the location on the locus of in. terest and hit the enter key. The value of the parameter K and the selected point will then be displayed in the command display. The use of the rlocfind function is illustrated in Figure 64. Continuing our third-order root locus example, we find that when K = 20.5775, the closed-loop transfer function has three poles and , (6.2) Where k is the parameter of interest to be varied from 0 Ska. The rlocus function is shown in Figure 6.2. ‘The steps to obtaining Als) — Figure 6.1 Closed-Loop Control System with Unspecified Gain K.Oo) Chapter 6, Root Locus Method Figure 6.3 The Root Locus for the Characteristic Eg. (6:1). two zeros at ~2.0505 + 4.3297¢ poles:s = | —2.0505 — 4.32275 } , zeros: —.8989 Considering the closed-loop pole locations only, we would expect that the real pole at s = ~.8989 would be the dominant pole (see MCS, p. 166). To verify this, we can study the closed-loop system Fesponse to a step input, R(s) = 1/s. For a step input we have Os) = 20.5775(s + 1)(s +3) s(s + 2)(s +3) + 20.5775(s +1)” Generally, the first step in computing e(t) is to expand Hq. (6.3) in @ partial fraction expansion (see MCS, pp. 45-52). The residue (6.3) 6.2, Obtaining a Root Locus Plot 85 »p=[1 1]; q=[1 5 6 0]; rlocus(p,q) »rlocfind(p,q) —*——~[ aigeind foiows the Haebe Select a point in the graphics window selected_point = -2.0509 + 4.3228: ans = 20.5775 «—_{ a tl Waled (Rw selected pone] Figure 6.4 Using the rlocfind Function. function can be used to expand Eq, (6.3), as shown in Figure 6.5. ‘The residue function is described in Figure 6.6. The partial fraction expansion of Eq. (6.3) is =1.3786 + 1.70101 C8) = 55 2.0505 + 4.32081 Comparing the residues we see that the coefficient of the term cor- responding to the pole at s = -.8989 is considerably smaller than the coefficient of the terms corresponding to the complex-conjugate poles at s = —2,0505 + 4.3227%. From this we expect that the in- fluence of the pole at s = —.8989 on the output response c(t) is 0.24296,2. Obtaining a Root Locus Plot 87 hot dominant. The settling time is then predicted by considering the complex-conjugate poles. The poles at s = -2.0505 + 4.32271 respond to a damping of ¢ = 0.4286 and a natural frequency of = 4.7844. Thus, the settling time is predicted to be K-25: »lrp,k]=resi re 1.3786 - 1.70101 0.2429 3.0000 sing the step function, as shown in Figure 6.7, we find that T, = 6 seconds. So our approximation of settling time T, © 1.95 is a irly good good approximation. The percent overshoot is predicted be p= P.O. = 100 exp-$*/V1-@ = 22.5%, scan be seen in Figure 6.7, the actual overshoot is very nearly 1%. Clearly, the prediction of overshoot is too low. In this example the role of the system zeros on the transient sponse is illustrated. The proximity of the zero at s = —1 to the le at s = —0.8989 reduces the impact of the pole on the transient, -2,0505 + 4.3228: -0.8989 ol 0 0s a ‘Time (sec) »K=20.5775;nume=K*[1 4 3]; den=[1 5 6+K K]; »step(num,den), grid Figure 6.7 Step Response for the Closed-Loop System in Figure 6.1 with K = 20.5775." Figure 6.6 The residue Function,88 Chapter 6. Root Locus Method Figure 6.8 Converting a Partial Fraction Expansion Back to a Ra- tional Function. response. The main contributors to the transient response are the complex-conjugate poles at s = —2.0505 + —4.3228i, One final point regarding the residue function. You can convert, the partial fraction expansion back to the polynomials num/den, given the residues (r), the pole locations (p), and the direct terms (X), with the command shown in Figure 6.8. 6.3 Sensitivity and the Root Locus ‘The roots of the characteristic equation play an important role in defining the closed-loop system transient response. The effect off parameter variations on the roots of the characteristic equation is a useful measure of sensitivity. ‘The root sensitivity can be defined to be Ors a 4 Rk i) We can utilize Eq. (6.4) to investigate the sensitivity of the roots of the characteristic equation to variations in the parameter k. If we change k by a small finite amount Ak, and evaluate the modified root rj + Ary, it follows from Eq. (6.4) that ane (65) i The quantity Sf is a complex number, Referring back to the third- order example in the previous section, if we change Ka factor of 5%, 6.4. Sensitivity and the Root Locus 89 we find that the dominant complex-conjugate pole at s = —2.0505+ 1.3298i changes by Ar; = —0.0025 — 0.168% when K changes from K = 20.5775 to K = 21.6064. From Eq. (6.5), it follows that Sha 0.0494 — 2.3355%. ‘Vhe sensitivity Sf‘ can also be written in the form Sp! = 2.3360 £268.7872°. ‘The magnitude and direction of Si! provides a measure of the root sensitivity. The script used to perform the above sensitivity calcu- lations is shown in Figure 6.9. d The root sensitivity measure may be useful for comparing the sensitivity for various system parameters at different root locations. However, the root sensitivity measure may not be that useful when tilized in the design process. It is primarily an analysis measure. pfsensitivity.m ‘% Compute the system sensitivity to a parameter % variation % K=20.5775; den=[1 5 6+K K]; r1=roots(den); % dk=1.0289; ~t % Km=K+dk; denm=[1 5 6+Km Km]; r2=roots(denm); dr=r1-r2; % ‘S=dr/(dk/K); <«———_ sehaitivity forma 7] Figure 6.9 Sensitivity Calculations for the Root Locus for a 5% Change in K = 20.5775.Chapter 7 equency Response Methods -1 Introduction the frequency response of a system is the steady-state output re- jonse due to a sinusoidal input signal. In the previous chapters have discussed the system response to various other test signals cluding steps, ramps, parabolas, and impulses. In this chapter, we ill investigate the response of systems to sinusoidal inputs. ‘The frequency response methods are based on considering the ;ponse of linear systems to sinusoidal input test signals as the fre- juency of the sinusoidal test signal varies. A linear, time-invariant system has the characteristic that, in the steady-state, the output response due to a sinusoidal input differs from the input only in magnitude and phase. The transfer function describing the sim soidal behavior of the system is obtained by replacing s with jw in the system transfer function G(s). Then, for a fixed w, G(jw) is a complex number with a magnitude and phase. The magnitude and phase of G(jw) can be represented graphically as w varies. This type of graphical representation is known as a Bode diagram. It is possible to develop control system performance specifications in the frequency domain so that an effective control system design method- ology using the Bode diagram can be used. The chapter begins with an introduction to the Bode diagram. Subsequently, the connection between the frequency response and performance specifications in the time-domain is discussed. ‘The o192 Chapter 7. Frequency Response Methods chapter concludes with an illustrative i 2 0 ve example to gain experi dlesigning a control system in the frequency domain dion it functions covered are bode and logspace. The bode func- ion is used to generate a Bode diagram, and the logspace function generates a logarithmically spaced vect: i generates lgacitl spaced vector of frequencies utilized by 7.2 Bode Diagram Suppose we have the transfer function (see MCS, p. $21 ) = 5(1 + 0.18) 30+ 05s) + S$ hat (ray ie oe diagram corresponding to Eq. (7.1) is shown in Figure 7.1. iagram consists of the logarithmic gain in dB versus w in one plot and the phase (1) versus w in a second plot, ‘The manual steps for sketching an approximate Bode diagram are given in MCS, pp. 308-317. As with the root locus plots, it be temptin; to rely exclusively on MATLAB to obtain your Bode diagrams. ‘Treat MATLAB as one tool in your tool kit that you can use to design and s ocr of 8 3 © -t00}.. 150 L 107 ior Tor 10r 0s ior Frequency (rad/sec) 10 Figure 7.1 The Bode Plot Associated with Eq. (7.1). 7.2. Bode Diagram 93 analyze control systen to develop the capability to anually obtain approximate Bode diagrams. There is no substitute for a clear understanding of the underlying theory. ‘A Bode diagram is obtained with the bode function shown in Figure 7.2. ‘The Bode diagram is automatically generated if the hode function is invoked without left-hand arguments. Otherwise, the magnitude and phase characteristics are placed in the workspace Unrough the variables mag and phase. A Bode diagram is obtained with the plot function using mag, phase, and w. ‘The vector w contains the values of the frequency in radians/sec at which the Bode diagram will be calculated. If w is not specified, MATLAB will mitomatically choose the frequency values by placing more points in rogions where the frequency response is changing quickly. Since the Bode diagram is a log scale, if you choose to specify the frequencies explicitly, itis desirable to generate the vector w using the logspace function. ‘The logspace function is shown in Figure 7.3. ‘The Bode diagram in Figure 7.1 is generated with the commands shown in Figure 7.4. The bode function automatically selected the Figure 7.2 The bode Function.* Chapter 7. Frequency Response Methods 7,3. Specifications in the Frequency Domain 95 bodescript.m ‘n points between 10" and 10? % Bode plot script for Figure 7.21 in MCS, p. 324 | 96 f] num=5*[0.1 1]; fl={1 0}; f2=[0.5 1}; f8=[1/2500 .6/50 1]; denccorv(f conv({2,f3)); = E 3 Bode(num,den) Figure 7.4 The Script for the Bode Diagram in Figure 7.1. frequency range as w = 0.1 to 1000 rad/sec. This range is user- nelectable with the logspace function. .3 Specifications in the Frequency Domain Bxeeping in mind our goal of designing control systems that sat- certain performance specifications given in the time-domain, we must establish a connection between the frequency response and the transient time response of a system. The relationship between spec- ifications given in the time domain to those given in the frequency domain depend upon approximation of the system by a second-order system with the poles being the system dominant roots, This ap- proximation is discussed in MCS, pp. 241-255. Consider the second-order system shown in Figure 7.5. The s[mag,phase,w}=bode(num,den,w); »semilogx(w,20"logt0(mag)), grid »xlabel(Frequency [rad/sec}), ylabel(20*log(mag) [dBy') ot 10° 10! 102 10° Frequency [rad/sec] Figure 7.3 The logspace Function. Figure 7.5 Single-Loop Second-Order Feedback System.96 Chapter 7. Frequency Response Methods closed-loop transfer function is e cs Fons FOR Ts) (72) ‘The Bode diagram magnitude characteristics associated with the closed-loop transfer function in Eq. (7.2) are shown in Figure 7.6. The relationship between the resonant frequency, «s,, the meximuny oF the frequency response, M,,, and the damping ratio, C. and the yalural frequency, wis shown in Figure 7.17 (and in Figure 7.10 in MCS, p-810). The information in Figure 7.7 will be quite helpful in designing control systems in the frequency domain while satisfying time-domain specifications, We have seen that we can relate frequency-domain specifications ‘o time-domain specifications by using the information contained in |__Performance specifications |___een domain damping ratio © Mp, Tisetime > wp overshoot <> Mp, 10 g 0, ? -10 ae g 30 OG Figure 7.6 Second-Order Closed-Loop System Characteristics. in 7 1. Specifications in the Mequeney Domain 9 welogspace(-1,1,400); «—p—— num=wn2; for i=1:110 zeta=zeta+0.005; > den=[1 2*zeta*wn wnA2]; [mag,phase,w]=bode(num,den,w); 2(i)=zeta; [mp(i),! end : subplot(211),plot(z,mp),grid xlabel(‘zeta'), ylabel(‘Mpw') subplot(212),plot(z,wr),grid xlabel('zeta’), ylabel('wr/wn") : )[]emax(mag) stores the index ofthe minima magn evr. Figure 7.7 The Relationship Between (M;_,w,) and (¢,w,) for a Second-Order System.98 Chapter 7. Frequency Response Methods the closed-loop Bode diagram. Stability is an important issue that can be addressed in the frequency domain by considering the open- loop transfer function. This topic will be addressed in the next chapter. 7.4 Example: Engraving Machine System Consider the block diagram model in Figure 7.8. This example can be found in MCS, pp. 932-335. Our objective is to design K so that the closed-loop system has an acceptable time response to a step command. A functional block diagram describing the frequency domain design process is shown in Figure 7.9. First we choose K = 2 and subsequently iterate on K if the performance is unacceptable. A script, shown in Figure 7.10, is used in the design. The value of K is defined at, the command level. Then the script is executed and the closed-loop Bode diagram is generated. The values of My, and w, are determined by inspection from the Bode diagram. Those values are used in conjunction with Figure 7.7 to determine the corresponding values of ¢ and wa. Given the damping ratio, ¢, and the natural frequency, w,, the- settling time and percent overshoot are estimated using the formulas * Gn 4 Tix 7) P.O. 100exp If the time-domain specifications are not satisfied, then we adjust K and iterate. ‘The values for ¢ and w, corresponding to K _ = 0.88. This leads to a prediction of P.O. are ¢ = 0.29 and 38% and T, = 16 Motor, screw, and scribe holder C(s) Figure 7.8 Engraving Machine Block Diagram Model. 7.4, Example: Engraving Machine System 99 initial gain K 20*logiOmag) [4B] nbS85 05 a If satisfied, then exit and continue analysis. Determine mpo and Gy. Establish relationship between ‘specs and time domain specs Figure 7.9 Frequency Design Functional Block Diagram for the Eneraving Machine.100 Chapter 7. Frequency Response Methods engravescript 1m num=[K]; den=[1 3 2 K];~<«! welogspace(-1,1,400); [mag,phase,w]=bode(num,den,w); [mp,!]=max(mag);wr=w(|); mp.wr »K=2} engravescript! 15.6740 po= 38.5979 Figure 7.10 Frequency Design Script for the Engraving Machine. 74. Example: Engraving Machine System 101 sconds. The step response, shown in Figure 7.11, is a verification Lhat the performance predictions are quite accurate and the closed- loop system performs adequately. In this example, the second-order system approximation is rea- wonable and leads to an acceptable design. However, the second- order approximation may not always lead directly to a good design. Fortunately, with MATLAB we have the possibility to construct an interactive design facility that can assist us in the design process by reducing the manual computational loads while providing easy access to a host of classical and modern control tools. on) Ke2; num=[K]; den=[1 32K]; t=[0:0.01:20]; u=1,02*ones(length(t),1); =0.98*ones(length(t),1); : 1n1.38*ones(length(t),1);._ ———[ggge y.xJ=step(num,den,t); C plot(ty,tustlu,tl), grid xlabel('Time [sec]'), ylabel(‘c(t)") Figure 7.11 Engraving Machine Step Response for K = 2.8.1 Introduction tability of a control system can be determined with frequency- nse _methods.. The: basis. for the frequency-domain stability alysis is the Nyquist: stability criterion. Issues of absolute sta- ility as well as telative stability can be addressed in the frequency domain. Graphical methods play an important role in the frequency- domain design and analysis of control systems. We will utilize sev- eral frequency-domain plots in our stability investigations, and, of course, we will use MATLAB to aid in obtaining our plots. ‘The chapter begins with a discussion of the Nyquist stability criterion and the Nyquist diagram and Nichols chart. We will also revisit the Bode diagram in our discussions on relative stability. Two examples are given which illustrate the frequency-domain design ap- proach. We will make use of the frequency response of the closed- loop transfer function, T(jw), as well as the loop transfer function GH (ju). We present an illustrative example that shows how to deal with a time delay in the system by utilizing a Padé approximation. The functions covered in this chapter are nyquist, nichols, margin, pade, and ngrid. 103104 Chapter & Stability in the Frequency Domain 8.2 Nyquist Plots ‘The Nyquist stability criterion is based on Cauchy's theorem, which is concerned with mapping contours in the complex s-plane. Con- sider the system in Figure 8.1. The closed-loop transfer function ° G(s) ‘ 7) = Ty Gay and the characteristic equation is F(s) =1+GH(s) =0. Allof the zeroes of F'(s) must lie in the left-hand s-plane for stability. We choose a contour, I, in the s-plane which encloses the entire right-hand s-plane, and plot I’r in the F(s)-plane and determine the number of encirclements of the origin. Equivalently, we can plot Tp in the P(s)-plane and determine the number of encirclements of the =I point, where P(s) = F(s) —1. The Nyquist stability criterion can be stated as follows: A feedback control system is stable if and only if, for the contours I'p, the number of counterclockwise encir- elements of the (~1,0) point is equal to the number of poles of P(s) with positive real parts (see MCS, p. 362). The plot of Ip is the Nyquist plot. It is generally more difficult to 8.2. Nyquist Plots 105 generate the Nyquist plot manually than the Bode diagram. How- ever, we can use MATLAB to generate the Nyquist plot rather easily. ‘The Nyquist plot is generated with the nyquist function, as shown in Figure 8.2. When nyquist is used without left-hand argu- ments, the Nyquist plot is automatically generated; otherwise, you must use the plot function to generate the plot using the vectors re and im. One cautionary remark regarding Nyquist plots: Some time in the course of using the nyquist function you may find that your Nyquist plot looks strange or that some information appears to be missing. It may be necessary in these cases to use the axis func- tion to override the automatic scaling and use the nyquist function with left-hand arguments in conjunction with the plot function. In this way you can focus in on the —1 point region for your stability analysis, as illustrated in Figure 8.3. Figure 8.1 Single-Loop Feedback Control System. Figure 8.2 The nyquist Function,106 Chapter 8. Stability in the Frequency Domain »num=(0.5]; den=[1 2 10.5 }; »axis([-1.0,.1,-0.1,0.1]); »[re,im]=nyquist(num,den); »plot(reim),grid —*—f Figure 8.3 The nyquist Function with Manual Scaling. Up to this point we have been considering absolute stability only. In other words, our concern has been whether a system is stable or, not. However, relative stability measures of gain and phase margins. can be determined from both the Nyquist plot and the Bode dia- gram. The gain margin is a measure of how much the system gain would have to be increased for the GH (jw) locus to pass through the (—1,0) point, thus resulting in an unstable system. ‘The phase margin is a measure of the additional phase lag required before the system becomes unstable. Gain and phase margins can be deter- mined from both the Nyquist plot and the Bode diagram. Consider the system shown in Figure 8.4. Relative stability can | Figure 84 A Closed-Loop Control System Example for Nyquist and Bode with Relative Stability. 82. Nyquist Plots 107 100%] Fag. 108 num=(0.5}; den=[1 2 10.5]; —s [mag,phase,w]=bode(num,den); ‘Frequency (rad/sec) margin(mag,phase,w); oT oo [Pies aia 100 101 Fretcncy atic) Figure 8.5 The margin Function. be determined from the Bode diagram using the margin function. ‘The margin function is invoked in conjunction with the bode func- tion to compute the gain and phase margins. The margin function is shown in Figure 8.5. If the margin function is invoked without left-hand arguments, the Bode diagram is automatically generated with the gain and phase margins labeled on the diagram. This is illustrated in Figure 8.6 for the system that is shown in Figure 8.4. The script to generate the Nyquist plot for the system in Fig- ure 8.4 is shown in Figure 8,7. In this case, the number of poles of GH(s) with positive real parts is zero and the number of counter- clockwise encirclements of —1 is zero; hence the closed-loop system is stable. We can also determine the gain and phase margins, as indicated in Figure 8.7.108 Chapter 8. Stability in the Frequency Domain 8.2. Nyquist Plots 109 8 1s Gain Margin = 3.017 Phase Margin = 49.41 num=[0.5]; den=[1 2 10.5 ]; % % Plot Nyquist and compute Gain and Phase % Margins for GH(s) = 0.5/sA3+2sA2+5+0.05 % num=[0.5}; den=[1 2 1 0.5 J; hase ma % K [mag,phase,w]=bode(num,dert); [Gm,Pm,Wcg,Wcp ]=margin(mag,phase,w); % w-logspace(-1,1,200); ~«—f5 % [mag,phase,w]=bode(num,den,w); % margin(mag,phase,w); nyquist(num,den) | title({'Gain Margin = ‘jnum2str(Gm) ' Phase Margin = ‘,num2str(Pm)]) < 0.215. Referring to Figure 7.7 (or MOS, p.316) we find that My, < 2.45.4 Chapter 8. Stability in the Frequency Domain % Liquid Control System Analysis % num=K*[0.0743 -0.4460 0.8920]; d1=[1 1]; d2=[30 1]; d3=[1/9 1/3 1]; d4=[ 0.0743 0.4460 0.8920]; den=conv(d1 ,conv(d2,conv(d3,d4))); % welogspace(-2,1,400); [ Compote gain ane {mag,phase,w]=bode(num,den,w); opener {Gm Pm Weg Mep]amargin(mag, phase.) bode(num,den) <«——_____| title(['Gain Margin = ‘num2str(Gm), ... " Phase Margin = ';num2str(Pm)]) “*—{" ‘abet plot Figure 8.12 Bode Diagram for the Liquid Level Control System. 44, Example 5 Gels) mf G(s) ROS) + K(+2) |__aee desired -D sel speed speed Figure 8.13 Battlefield Vehicle Speed Control System. We must keep in mind that the information in Figure 7.7 is for wecond-order systems only and can be used here only as a guideline. We now compute the closed-loop Bode diagram and check the values of Mp,. Any gain K for which Mp, < 2.45 may be a valid gain for our design, but we will have to investigate further to include step responses to check the actual overshoot. The script in Figure 8.14 aids us in this task. In keeping with the spirit of the design steps in MCS, pp. 992-402, we investigate further the gains K = 20,10, and 4.44 (even though M,, > 2.45 for K = 20). We can plot the step responses to quantify the overshoot, as shown in Figure 8.15. Alternately, we could have used a Nichols chart to aid the design process. This is shown in Figure 8.16. ‘The results of the analysis are summarized in Table 8.1 for K = 20,10, and 4.44. Suppose we choose K = 10 as our design gain. ‘Then we obtain the Nyquist plot and check relative stability. This is shown in Figure 8.17. The gain margin is GM = 49.56 and the phase margin is PM = 26.11°. ‘Table 8.1 Actual Response for Selected Gains. K 4.44] 10 | 20 Percent overshoot | 5% | 30% | 50% Settling time 35 | 5 | 6 Peak time 14 | 1.0 | 07 Ca 31% | 17% | 9%6 Chapter 8 Stability in the Frequency Domain 4, Examples ut battlestep.m t=[0:0.1:10]; K=20; % welogspace(0,1,200); K=20; “~ % forie1:3, <+———_______ numge=K+[1 2]; dengo=[1 1]; numg=[1]; deng=[1 2 4]; [nums,dens]=series(numgc,denge,numg,deng); [num,den]=cloop(nums,dens); [mag,phase,w]=bode(num,den,w); if i==1, magi =mag; phase1=phase; K=10; end =2, mag2=mag; phase2=phase; K=4.44; end =3, mag3=mag; phase3=phase; end for i-1:3 <—_—_______| numgo=K*[1 2]; dengo=(1 1]; numg=[1]; deng=[1 2 4]; [nums,dens]=series(numgc,dengc,numg,deng); [num,den]=cloop(nums,dens); Ly.x]=step(num,den,t); if in=1, yl=y; K=10; end , y3=y; end % end loglog(w,mag1,'-',w,mag2,'-',w,mag3,'~"), grid xlabel('frequency [rad/sec]'), ylabel(’mag') plot(ty1,'-",ty2,"-',ty3,'-"),grid xlabel(‘time [sec]'), ylabel(‘c(t)') Figure 8.14 Remotely Controlled Battlefield Vehicle Closed-Loop System Bode Diagram Script. Figure 8.15 Remotely Controlled Battlefield Vehicle Step Re- sponse.us Chapter 8. Stability in the Frequency Domain battlenichote.m '% Remotely Controlled Battlefield Vehicle % numge=[1 2]; dengc=[1 1]; eS numg=[1]; deng=[1 24]; “= {mum,den]-series(numge,denge,numg,deng); welogspace(-1,2,200); axis({-210,0,-24,36]) ~*~} % K1=20; K2=10; K3=4.44; ——___| [mag1,ph1 ,w]=nichols(K 1 *num,den,w); [mag2,ph2,w}=nichols(k2*num,den,w); [mag3,ph3,w}=nichols(k3*num,den,w); plot(ph1 ,20*iog1 0(mag ph3,20*log1 0(mag3), Figure 8.16 Remotely Controlled Battlefield Vehicle Nichols Chart. 44, Examples 119 Gain Margin = 49.56 Phase Margin = 26.11 Imag. Axis battlenyquist.n % Remotely Controlled Battlefield Vehicle % Nyquist plot for K=10 % numge=10*[1 2]; dengo=[1 1]; numg=[1]; deng=[1 2 4]; [num,den]=series(numgc,dengc,numg,deng); [mag,phase,w]-bode(num,den); [Gm,Pm,Wcg,Wep]=margin(mag,phase,w); % nyquist(num,den); title({'Gain Margin = ',pum2str(Gm), ... "Phase Margin = ‘,num2str(Pm)]) Figure 8.17 Nyquist Chart for the Remotely Controlled Battlefield Vehicle with K = 10.120 Chapter 8. Stability in the Frequency Domain Notes Chapter 9 State-Space Methods 9.1 Introduction In the previous chapters we considered control system design and analysis in the frequency domain. We utilized the Laplace trans- form to transform the linear, constant coefficient differential equa- tion model into an algebraic expression in terms of the complex variable s. ‘Then we operated on our system in input-output (or transfer function) form bas + SF ag + this + bo Fast ols) = ape) = GU) RIs) In this chapter we begin to look at control system design and analysis in the time domain. In contrast to the frequency-domain approach, the time-domain method utilizes a state-space tepresen- tation of the system model, given by Az + Bu De+Hu (9.1) ce The vector & is the state of the system, A is the constant n xn system matrix, B is the constant n x m input matrix, D is the constant p xn output matrix and HT is a constant p x m matrix. ‘The number of inputs, m, and the number of outputs, p, are taken to be one since we are considering only single-input, single-output problems. Therefore c and u are not bold variables. 121,122 Chapter 9, State-Space Methods Figure 9.1 The State-Space Representation bode Function. ‘The main elements of the state-space representation in Eq. (9-1) are the state vector 2 and the constant matrices (A,B, D,H). Since the main computational unit in MATLAB is the matrix, the state-space representation lends itself well to the MATLAB envi- ronment. In fact, MATLAB covers so many aspects of state-space methods that we will not be able to discuss them all here. The new functions covered in this chapter are tf2ss and ss2tf. Most of the functions covered in the previous chapters also apply here, For example, the the bode function can be utilized with a state-space model, as shown in Figure 9.1. ‘The same idea ap- plies to series, parallel, feedback, cloop, printsys, minreal, step, pzmap, impulse, Isim, rlocus, rlocfind, residue, bode, nyquist, and nichols. 9.2 Model Relationships Given a transfer function we can obtain an equivalent state-space representation, and vice versa, MATLAB has two functions that convert systems from transfer function to state space and back, ‘The function tf2ss converts a transfer-function representation to a state-space representation; the function ss2tf converts a state-space representation to a transfer function. These functions are shown in Figure 9.2. 123 9.2, Model Relationship c 7 mum@) Cis) = Gis) Us) = “ary UO) Figure 9.2 Linear System Model Conversion. For instance, consider the third-order system Ole) __2s? +8846 (02) Ts) = Rie) = PRP + 1s FO We can obtain a state-space representation using the tf2ss function as shown in Figure 9.3. The state-space representation of Eq. (9.2) is given by Eq. (9.1) where 8 -16 -6 1 A=| 1 0 0], B=|O], 0 1 0 0 pD=([28 6], H=(0) Notice that the printeys function lists the system matrices as a, b,¢,d. ‘The conversion to our notation is as follows: and an A, bo B, cD, do.124 Chapter 9. State-Space Methods % Convert G(s) = (2sA2+85+6)/(sA3+8542+165+6) % to a state-space representation % num-=[2 8 6]; den=[1 8 16 6]; [A,B,D,H}=tf2ss(num,den); printsys(A,B,D,H) »convert a= xl x2 x3 xl -8.00000 -16.00000 -6.00000 x2 1.00000 0 0 x3 0 1.00000 0 b= ul x1 1.00000 x2 0 x3 0 c= x1 x2 x3 yl 2.00000 8.00000 6.00000 d= ul yl O Figure 9.3 Conversion of Eq. (9.2) to a State-Space Representa- tion, 9.3, Stability of Systeme in the Time Domain 125, 9.3 Stability of Systems in the Time Domain Suppose we have a system in state-space form as in Eq. (9.1). The stability of the system can be evaluated with the characteristic egua- tion associated with the system matrix A. The characteristic equa- tion is det(sI — A) =0. (93) "Phe characteristic equation is a polynomial in s. If all of the roots of the characteristic equation have negative real parts (j.e., Re(si) < 0,¥i), then the system is stable. ‘When the system model is given in the state-space form we must, calculate the characteristic polynomial associated with the A ma- trix. In this regard we have several options. We can calculate the characteristic equation directly from Eq. (9.3) by manually comput- ing the determinant of (sf — A). ‘Then we can compute the roots using the roots function to check for stability, or alternatively, we can utilize the Routh-Hurwitz method to detect any unstable roots. Unfortunately, the manual computations can become lengthy, espe- cially if the dimension of A is large. We would like to avoid this manual computation if possible, As it turns out, MATLAB can assist in this endeavor. ‘The poly function described in Chapter 2 can be used to com- pute the characteristic equation associated with A. Recall that poly is used to form a polynomial from a vector of roots. It can also be used to compute the characteristic equation of A, as illustrated in Figure 9.4, wherein input matrix, A, is 8 -16 -6 A=|1 0 0 o 1 0 and the associated characteristic polynomial is +85? + 163 +6. If A is an nx n matrix, poly(A) is the characteristic equation represented by the n-+1 element row vector whose elements are the coefficients of the characteristic equation.126 Chapter 9. State-Space Methods [ CoetFicients of characteristic —] polynomial in descending order >A=[-8 -16 ~ >p=poly(A) p- 1.0000 8.0000 16.0000 6.0000) »roots(p) ans = ~5.0861 -2.4280 coms -0.4859 Figure 9.4 Computing the Characteristic Equation of A with the poly Function. ™@ EXAMPLE 9.1 Automatic Test System The state-space representation for the automatic test system (see MCS, pp. 462-465) is &=Az+ Bu (4) o 10 0 o-1 1], B=|0 0.0 -5 K Our design specifications are (i) step response with a settling time less than two seconds, and (ii) overshoot less than 4%, We assume that the state variables are available for feedback so that the control where 9.3, Stability of Systems in the Time Domain 127 is given by u= (-Ky,-Ka,—Ko) 2. (9.8) We must select the gains K, K,,K and Ks to meet the performance specifications. Using the design approximation TR a <2 and P.O. x 100exp"/VI@ < 4, On we find that (> 0.72 and w, > 28. ‘This defines a region in the complex plane in which our dominant roots must lie to have any chance of meeting the design specifica- tions. Substituting Eq, (9.5) into Eq, (9.4) yields 0 1 0 = 1 e=A‘e, (9.6) -KK, -KK, -(6+KKs) where A" is the revised A matrix. The characteristic equation asso- ciated with Eq, (9.6) can be obtained by evaluating det(sI—A") = 0. ‘This results in Kytka | K motkK If we view KK as a parameter and let K, = 1, then we can write Eq, (9.7) as follows: 4(3-+1)(8-+5) + KKa(s? + (9.7) ot + Ragas + 3(5 +1)(s +5) We place the zeros at s = —4 + 2 in order to pull the locus to the left in the s-plane. Thus our desired numerator polynomial is s? +8s-+20. Comparing corresponding coefficients leads to Kot Ka _ ve moo 8 and Tatil 20. Therefore Kz 0.35 and K3 = 0.05. We can now plot a root locus with KK, as the parameter, as shown in Figure 9.5. The 1+KKs128 Chapter 9. State-Space Methods Imag. Axis Valid region to meet NV performance specs =28 =0.72 qutolocus m % Root locus script for the Automatic Test System % including performance specs regions num=[1 8 20}; den=[1 6 5 0]; cs: rlocus(num,den); hold on <<} Figure 9.5 Root Locus for the Automatic Test System. 9.4, Time Respense 129 Figure 9.6 Step Response for the Automatic Test System. characteristic equation, Eq, (9.7), is SF +83+20 1+ KK +5) > ‘The selected gain, KKs = 12, lies in the performance region, as shown in Figure 9.5. The rlocfind function is used to determine the value of K Kg at the selected point. The final gains are K = 240.00 Ky, = 1.00 Ky, = 0.35 Ks = 0.05 The controller design results in a settling time of about 1.8 seconds and an overshoot of 3%, as shown in Figure 9.6. 9.4 Time Response ‘The time response of the system in Eq. (9.1) is given by the solution to the vector differential equation a(t) =exp(Adja(0)+ ['explA@-A]Burdr. (98)130 Chapter 9. State-Space Methods The matrix exponential function in Eq, (9.8) is the state transition matrix, (0), where H(t) = exp(At). We can use the function exxpm to compute the transition matrix for a given delta time, as illustrated in Figure 9.7. The expm(A) func- tion computes the matrix exponential whereas the exp(A) function returns e*% for each of the elements aj; € A. —t The time response of the system in Eq. (9.1) can be obtained by using the Isim function. The Isim function can accept as input nonzero initial conditions as well as an input function. ‘This is il lustrated in Figure 9.8 for the RLC network (see MCS, Chapter 9) described by the state-space representation a-[} 3] 7 a-[§] , D=[1 0] ,and H=0. The initial conditions are 2,(0) = 22(0) = 1 and the input u(t) = 0. If we can compare the results obtained by the Isim function and by multiplying the initial condition state vector by the state transition matrix, we find exactly the identical results. At 1 = 0.2 the state transition matrix is given in Figure 9.7. The state at t = 0.2 is predicted by the state transition methods to be a — | 0.9671 —0.2968 cay _ f 0.6703 2 Jeg | 01484 0.5219} \ za J, \ 0.6703 J * The state at 0.2 is also predicted with the Isim function to be (0.2) = 22(0.2) = 0.6703. f>A=[0 -2; 1-3]; dt=0.2; Phixexpm(Arat) Phi = 0.9671 -0.2968 «| sate transi 0.1484 0.5219 esse ‘i Figure 9.7 Computing the State Transition Matrix for a Given 9.4. Time Response 131 1 os o.| xl os 07 a8 054 A=[0-2;1 -3]; [zare i Ly x]elsim(A,B,D,H,u,t,x0) subplot(211), plot(t,x(:,1)) xlabel(‘time [sec]’), ylabel('x1') subplot(212), plot(t,x(:,2)) xlabel(‘time [sec]'), ylabel('x2") Figure 9.8 Computing the Time Response for Nonzero Initial Con- ditions and Zero Input.132 Chapter 9. State-Space Methods Notes Chapter 10 Control System Design 10.1 Introduction It is often possible to achieve stability and meet all the control sys- tem performance specifications by adjusting one or two parameters. We introduced many examples in the previous chapters illustrating design by adjusting a few parameters. However, in many cases it is necessary to add a dynamic compensator into the system. Altering a control system to meet relative stability and performance spec- ifications is called compensation. We say that our compensators are dynamic in the sense that the compensator is itself a system described by a transfer function or state-space representation with internal states. A compensator is shown in the control system in Figure 10.1. The compensator in Figure 10.1 is a cascade or series ‘Spacecraft rotational motion Compensator model 8 desired Figure 10.1 A Compensated Control System for Spacecraft Rota- tional Motion. 133134 Chapter 10. Control System Design compensator since it is placed in the feedforward path. A come Pensator placed in the feedback path is known as a feedback com, pensator. Compensators can be placed in other paths (e.g., inner feedback loops) in the control system as well. ‘The main topic of this chapter is compensation of feedback con- trol systems. ‘There are many approaches to compensation. We will consider lead and lag compensators and present a design exam, ple that uses both the root locus method and the Bode frequency. domain method to design the compensators. One of the most com, monly used compensators is the proportional plus derivative plug integral (PID). The PID compensator is covered in Chapter 11. 10.2 Lead Compensation Consider the series compensator G(s) = Aete, (10.1) “The selection of the variables Kz, and p is based on satisfying the design performance specifications. Whenever lel < pl the compensator in Eq. (10.1) is a lead compensator, The pole-zero 1. The maximum value of the phase lead occurs at a frequency Wm, where 1 om = VB = ae The maximum phase angle at Wm is dm, where zl 1 sin bm = T° Consider, for example, the lead compensator 10(s +1) +10" ‘The associated Bode diagram is shown in Figure 10.3. The maxi- mum value of the phase lead occurs at on = VB= VO. ‘The maximum phase lead is rm = arcsin GAs) = where a = 10. i: i The phase-lead compensator is a differentiator type compen- sator. This can be seen by considering the case when |p| >> |2|. Then it follows that ue Glo) #8 FF We can design lead compensators with frequency-domain design techniques utilizing Bode diagrams as well as with root locus design methods. The lead compensator increases the phase margin, thus providing additional stability, and increases the system bandwidth to provide speedier dynamic response.136 Chapter 10. Control System Design bode(num,den) Figure 10.8 Bode Diagram of the Lead Compensator. 10.3 Lag Compensators Again consider the series compensator K(s+2) G.(s) = £2) te Se Whenever bi <|2l the compensator G(s) is a lag compensator. The pole-zero diagram of the lag compensator is shown in Figure 104. The lag compensator can also be written as in Eq. (10.2) where a < 1. The maximum value of the phase lag occurs at un = Vp = V0. 10.3. Lag Compensators 137 Figure 10.4 Pole-Zero Diagram of the Lag Compensator. Consider, for example, the lag compensator (e410) Gels) = FF ‘The associated Bode diagram is shown in Figure 10.5. We see that the lag compensator is an integration type compen- sator by considering |2| >> |p|. Then G(s) ¥ K+ Ke, ‘This has the same form as the widely used lag compensator Gs) = Ky + AE. ‘This is known as a proportional plus integral (PI) compensator. ‘The lag compensator is applicable when high steady-state ac- curacy is required. Although it is possible to increase steady-state accuracy by simply increasing the system gain, this often leads to unacceptable transient response and sometimes instability. This problem is overcome with the addition of a lag compensator and properly chosen values of K,p, and 2. The lag compensator de- creases the system bandwidth thus suppressing high frequency noise and slows down the transient response.138 Chapter 10. Control System Design Pe bas Figure 10.5 Bode Diagram of the Lag Compensator. 10.4 Example: Rotor Winder Control System ‘The rotor winder control system is shown in Figure 10.6 (see MCS, pp. 542-545). The design objective is to achieve high steady-state accuracy to a ramp input. The steady-state error to a unit ramp Figure 10.6 Rotor Winder Control System. 10.4, Example: Hotor Winder Control System 139 1/s?, is input, R(s) ’ c= I where a fm C268 Kem ty Of course, the performance specification of overshoot and settling time must be considered as well as steady-state tracking error. In all likelihood, a simple gain will not be satisfactory. So we consider active compensation utilizing lead and lag compensators using both Bode diagrams and root locus plots. Our approach is to develop a series of scripts to aid in the compensator designs. Consider first a simple gain controller, G.(s), where Gs) = K. ‘Then, Clearly, the larger we make K, the smaller the steady-state error éxy. However, we must consider the effects of increasing K on the transient response. This is shown in Figure 10.7. When K = 500, our steady-state error for a ramp is 10% but the overshoot is 70% and the settling time is around 8 seconds for a step input. We consider this to be unacceptable performance and turn to compensation. The two important compensator types that we consider are lead and lag compensators. First we try a lead compensator K(s+2) G(s any (= Gea) where |2| < |p|. ‘The lead compensator will give us the capability to improve the transient response. We will use a frequency domain approach to design the lead compensator. ‘Suppose we desire a steady-state error of less than 10% to a ramp input. Then we desire Ky = 10.140 y Chapter 19. Control System Design et) rotorgain.m K(50 100 200 500 }; «| numg=[1]; deng=[1 15 50 0}; oie deng=[ Oo}; % for i=1:4 [nums,dens]=series(K(i),1,nu , | numg,der [num,den}=cloop(nums,dens); _ [y,x]J=step(num,den,t); Ys(z,i)=y; end % Plot(t, Y¥s(:,1),"',t,Ys(:,2),""1,¥: xlabel(‘time [sec]'), ylabel(‘c(t)') ~———_f it, ¥s(:,4),'-") Figure 10.7 Transient Response for Simple Gain Controller 10.4. Example: Rotor Winder Control System M4 In addition to the stcady-state specifications, suppose also that we desire to meet certain performance specifications: (i) settling time T, <3 seconds, and (ii) percent overshoot for a step input < 10%, Solving the approximate formulas e aoe 10 and Te 7- P.O. % 100 exp"/V1-? for ¢ and w, yields ¢ 0.5912 and wp, = 2.2555. The phase margin requirement becomes Os Com © Gop = 60 degrees. ‘The steps leading to the final design are as follows: 1. Obtain the uncompensated system Bode diagram with K = 500 and compute the phase margin. 2. Determine the amount of necessary phase lead. 3. Evaluate a where sing, = 224. 4, Compute 10loga and find the frequency wm on the uncom- pensated Bode diagram where the magnitude curve is equal to =10loga. 5. In the neighborhood around w,, on the uncompensated Bode, draw a line through the 0-dB point at w,, with slope equal to the current slope plus 20 dB/dec. Locate the intersection of the line with the uncompensated Bode to determine the lead compensation zero location. Then calculate the lead compen- sator pole location as p = az. 6. Draw the compensated Bode and check the phase margin. Re- peat any steps if necessary.M42 Chapter 10. Control System Design 7. Raise the gain to account for attenuation (1/a). 8. Verify the final design with simulation using step functions, and repeat any steps if necessary. We utilize three scripts in the design. The design scripts are shown in Figures 10.8, 10.9, and 10.10. The first script is for the uncom- pensated Bode, the next is for the compensated Bode, and the final script is for the step response analysis. The final lead compensator design is 100¢ 5) (s +3. Gels)= 25) The settling time and overshoot specifications are satisfied, but K, = 5, resulting in a 20% steady-state error to a ramp input. It is possible to continue the design iteration and refine the compensator somewhat, although it should be clear that the lead compensator has added phase margin and improved the transient response as anticipated. ‘To improve the steady-state errors we can consider the lag com- pensator. The lag compensator has the form K(s+2) (stp)? where |p| < |z|. We will use a root locus approach to design the lag compensator, although it can be done using Bode as well. The desired root location region of the dominant roots are specified by G(s) = $= 0.5912 and w, = 2.2555, The steps in the design are as follows: 1. Obtain the root locus of the uncompensated system. 2. Locate suitable root locations on the uncompensated system which lie in the region defined by ( = 0.5012 and w, = 2.2555. 3, Caleulate the loop gain at the desired root location and the system error constant, Kusncomp 10.4, Example: Rotor Winder Control System 143 mag [dB] frequency [rad/sec] rotoriead.m K=500; numg=[1]; deng=[1 15 50 0}; Inum,den]=series(K,1,numg,deng); resaaee phase welogspace(-1,2,200); ieee [mag,phase,w}=bode(num,den,w); [Gm,Pm,Wcg,Wcp]=margin(mag,phase,w); % Phi=(60-Pm)*pi/180; <——f % alpha=(1+sin(Phi))/(1-sin(Phi)) | ——«——{ % M--10*log1 O(alpha)*ones(length(w),1); % [mag,phase,w]=bode(num,den,w); semilogx(w,20*log1 0(mag),W.M), grid : xlabel( "frequency [rad/sec]'), ylabel(’mag [d8]") Figure 10.8 Lead Compensator: Uncompensated Bode.144 oft) rumg={1] deng=[1 15 50 numge=K*[1 3.5]; dengc=[1 25 ]}ne—| '8 [num,den]=series(numgc,denge,numg,deng); welogspace(-1,2,200); [mag,phase,w}-bode(num,den,w); [Gm,Pm,Wcg,Wep ]=margin(mag,phase,w); % bode(num,den) title({"Gain margin = ',num2str(Gm), .. "Phase margin = ',num2str(Pm)]) Figure 10.9 Lead Compensator: Compensated Bode. Chapter 10. Control System Design BM 10.4. Example: lotor Winder Coutrol System K=1800; % numge[1]; deng=[1 15 50 0}; numge=K*[1 3.5]; dengc=[1 25]; % [nums,dens]=series(numge,denge,numg,deng); [num,den]=cloop(nums,dens); % t=[0:0.01:2]; step(num,den,t) Figure 10.10 Lead Compensator: Step Response. M5146 5 ur Chapter 10. Control System Desig 10.4, Example: Rotor Winder Control System 4. Compute a = ee where K,,. 10. 5. With @ known, determine suitable locations of the comps sator pole and zero so that the compensated root locus sti passes through desired location. 6. Verify with simulation and repeat any steps if necessary. The design methodology is shown in Figures 10.11, 10.12, and 10.13; Using the rlocfind function, we can compute the gain K associat: with the roots of our choice on the uncompensated root locus that lie in the performance region. We then compute a to ensure that we achieve the desired K,. We place the lag compensator pole an zero in order not to impact the uncompensated root locus. In Fig} ure 10.12, the lag compensator pole and zero are very neat the origi at z= —0.1 and p = ~0.01. The settling time and overshoot specifications are nearly satisfi and K, = 10 as desired. It is possible to continue the design iterati and refine the compensator somewhat, although it should be ch that the lag compensator has improved the steady-state errors to’ Tamp input relative to the lead compensator design, The final k compensator design is Imag, Axis rotorlag.m_ numg=[1]; deng=[1 15 50 0}; axis([-15,1,-10,10]); clg; rlocus(numg,deng); hold on 100(s + 0.1) Ode) = ET oa)” ‘The resulting performance is summarized in Table 10.1. Table 10.1 Compensator Design Results. % zeta=0.5912; wn=2.2555; % xe[-10:0.1:-zeta*wn); y=-(sqrt(1-zetaA2)/zeta)*xs -xc.A2); Controller Gain, K | Lead | Lag xc=[-10:0.1:-zeta*wn];cesqrt(wnA2-xc.42); % Step overshoot 70% | 8% | 13% plot(xiys's Xs"? Xie, Settling time (sec) 8 1] 4 Steady-state error for ramp | 10% | 20% | 10%. K 10 5 | 10 Figure 10.11 Lag Compensator: Uncompensated Root Locus.148 Chapter 10. Control System Design rotorlagl.m rumg=[1}; deng-[1 15 50 0]; numge={1 0.1]; dengo=[1 0.01}; [num,den]=series(numge,denge,numg,deng); axis([-15,1,-10,10]); cg; rlocus(num,den); hold on % zeta=0.5912; wn=2.2555; -zeta*wn]; y=-(sqrt(1-zeta42)/zeta)*x; Figure 10.12 Lag Compensator: Compensated Root Locus. 10.4. Example: Rotor Winder Control System 149 rototlag2.m K=100; % numge[1]; deng-[1 15 50 0}; numge=K*[1 0.1]; denge=[1 0.01}; % [nums,dens]=series(numgc,dengc,numg,deng); [num,den]=cloop(nums,dens); % step(num,den) Figure 10.13 Lag Compensator: Step Response.150 Chapter 10. Control System Design Notes Chapter 11 Robust Control Systems 11.1 Introduction Designing a highly accurate control system in the presence of plant uncertainty is a classical design problem. In the previous chapters, we have generally assumed that the plant parameters are well known and designed our control system accordingly. In practice, the plant parameters are never precisely known and may vary slowly over time. It is desirable to design a control system that performs adequately over a range of plant parameters. A control system is robust when it maintains a satisfactory level of stability and performance over a range of plant parameters and disturbances. Tn this chapter, we begin to investigate robust control systems. In particular we consider the commonly used proportional plus deriva- Dis) Figure 11.1 Feedback Control System with Reference and Distur- bance Inputs and a Prefilter. 151152, Chapter 11. Robust Control $ shee Live plus integral (PID) controller, Our feedback control system has the form shown in Figure 11.1. Notice that the system has a pro: filler G(s). The role of the prefilter in contributing to optimum performance is discussed in MCS, pp. 594-596. 11.2 Robust PID Controlled Systems ‘The PID controller has the form Ks? + Kys + Ky GAs) : Notice that the PID controller is not a rational function (i.e., the degree of the numerator polynomial is greater than the degree of the denominator polynomial). You will experience difficulty if yo attempt to input the PID controller into MATLAB in the standa numerator and denominator fashion. Generally speaking, the prol Jem can be resolved by utilizing the conv function rather than th series function in your manipulations. The objective is to choose the parameters Ki,Kz, and Ky meet the performance specifications and have desirable robustness properties. Unfortunately, it is not. immediately clear how to choose the parameters in the PID controller to obtain certain robustness characteristics. We will show by an illustrative example that it i possible to choose the parameters iteratively and verify the robust. ness by simulation. Using MATLAB helps in this process since the entire design and simulation can be mechanized utilizing scripts an easily executed again and again. A complete exposition on the sub- ject of robust control analysis and design is beyond the scope of thi book. @ EXAMPLE 11.1 Robust Control of Temperature Consider the feedback control system in Figure 11.1, where 11.2, Robust PID Controlled Systems 153 and the nominal value of ¢o is @=1 We will design a compensator based on co = 1 and check robustness by simulation. Our design specifications are as follows: (i) settling time T, < 0.5 seconds, and (a) optimum ITAE performance for a step input (see MCS, pp. 176-185). In our design, we will not utilize a prefilter to meet specicn tion (ii) but will instead show that acceptable performance ({.e,, low overshoot) can be obtained by increasing the system gain. ‘The closed-loop transfer function is (Kas? + Kis + Ka : (ia) T8)= SF G+ Kes Ot hst ‘The associated characteristic equation is 2 +as + iets) where Kt = Ke +2, 14h 24K Ky 2+ Ks ling time requirement T, < 4 leads us to choose the roots ey Me fo the left of the ¢ = Cur = 8 line in the s-plane, as shown in Figure 11.2, to ensure ‘that the locus travels into the required performance region. We have chosen a= 16 andb=10 ce cheure the locus travels past the s = —8 line. We select a point tom the root locus in the performance region, and using the rlocfind function, we find the associated gain K* and the associated value of154 Chapter 11, Robust Control Systers »tlocfind(num,den) Figure 11.2 Root oe ot Locus for the PID Compensated Temperature q- For the point we have chosen we find that ‘K* =118, Then, with K*, a, and 6 we can solve for the PID coefficients as 11.2. Robust PID Controlled Systems 155 follows: Ky = K*-2=116, Ky = a(2 + Ks) —1 = 1187, Kz = (2 + K3) = 8260. To meet the overshoot performance requirements for a step input we will utilize a cascade gain K that will be chosen by iterative methods using the step function. This is illustrated in Figure 11.3. The step response corresponding to K = 5 has an acceptable overshoot of 2%, With the addition of the gain K = 5, the final PID controller is Gls) = K Kgs? + Kut Ka _ 1163? + en (112) We did not used the prefilter as is done in the design Example 11.5 in MCS, pp. 594-595. Instead we increased the system gain to obtain satisfactory transient response. Now we can consider the question of robustness to changes in the plant parameter co. Our investigation into the robustness of our design consists of a step response analysis using the PID controller given in Eq. (11.2) for a range of plant parameter variations of 9 € (0.1, 10]. The results are displayed in Figure 11.4. The script is written to compute the step response for a given co. It might be a good idea to place the input of op at the command prompt level to make the script more interactive. The simulation results indicate that the PID design is robust with respect to changes in cy. The differences in the step responses fot co € [0.1,10] are barely discernible on the plot. If the results showed otherwise, it would be possible to iterate on the design until an acceptable performance was achieved. ‘There exist various control design methods that incorporate ro- bustness directly into the design process, but their presentation here is outside the scope of this text. The interactive capability of MATLAB allows us to check our robustness by simulation, although this is clearly not the most desirable approach to design.156 Chapter 11. Robust Control Systems 0 0.05 Or O15 02 025 03 ‘Time (sec) Gah Wa Ureorpansated root locus. K=5; <——_| Increase syatem gain to redice overshoot. K3=Ks-2, K1=a*(2+K3)-1, K2=b*(2+K3) = ‘PID gang numge=K+[K3 K1 K2]; denge=[1 0]; numg=[1]; dengf1 2 1]; 1% nums=conv(numge,numg); caamaretaiios ; ce denseconv(denge,deng); «| PRPS tne 11.2, Robust PID Controlled Systems 157 3 | & E 4b 4 ~ ~ 02} + he i. aaa % 0.05 On 0.15 02 0.25 03 como J; deng=(1 2*cO cONZI numge=5*[116 1187 8260]; denge=[1 0]; % numa=conv(numg,numgc); dena=conv(deng,dengc); % [num,den}=cloop(numa,dena); % step(num,den) [num,den]=cloop(nums,dens); step(num,den) Figure 11.3 Step Response for the PID Temperature Controller. Figure 11.4 Robust PID Controller Analysis with Variations in co158 Chapter 11. Robust Control Systems Notes Index/Glossary axis Controls the manual axis sealing on plots; 105 bode Computes a Bode frequency response plot; 92-93, 107, 110, 122 clear Removes variables and functions from memory; 6, 9 clg Clears plots from the graph window; 16 cloop Computes the closed-loop system with unity feedback; 26, 37-40, 122 conv Multiplies two polynomi- als via convolution; 26, 31-32, 152 ‘end Terminates a for function; ie expm Computes the matrix ex- ponential function; 130 feedback Computes the feed- back interconnection of two sys- tems; 26, 37-40, 49, 122 for Repeats a group of state- ments a specific number of times; 1,75 159 format Controls the output for- mat; 9 grid Draws grid lines on the current plot; 17 help Invokes the help facility; 22 impulse Computes the unit im- pulse response of a system; 61, 63, 122 loglog Generates an z-y plot using log-log scales; 17-18, 20 logspace Generates a logarith- mically spaced frequency vec- tor for frequency response anal- ysis; 92-95 Isim Computes the time response of a system to an arbitrary in- put and initial conditions; 61, 66, 68, 122, 130 margin Computes the gain mar- gin, phase margin, and associ- ated crossover frequencies from frequency response data; 103, 107-110160 minreal Transfer function pole- zero cancellation; 26, 44, 122 ngrid Draws grid lines on a Nichols chart; 103, 110 nichols Computes a Nichols fre- quency response plot; 103, 110, 122 nyquist Computes a Nyquist frequency response plot; 103, 105-106, 122 pade Computes an n-th order Padé approximation to a time delay; 103, 112 parallel Computes a parallel system connection; 26, 36-37, 122 plot Generates a linear z-y plot; 16-19, 45, 93, 105, 110 poly Returns the characteris- tic equation when the input is a matrix and returns a polyno- mial when the input is a vec- tor containing the roots of the polynomial; 26, 31, 125-126 polyval Polynomial evaluation; 26, 32 printsys Prints state-space and transfer function representations of linear systems in a readable form; 26, 34, 39, 40, 44, 46, 122-124 pzmap Plots the pole-zero map of a linear system; 26, 32, 122 residue Computes the residues, Index/Glossary poles, and direct terms of a partials fraction expansion; 82, 84-86, 88, 122 rlocus Computes the root lo- cus of a linear system; 82-83, 122 rlocfind Finds the gain asso- ciated with a given set of roots on a root locus plot; 82-83, 122, 129, 146, 153 roots Computes the roots of a polynomial; 26, 31, 72-73, 125 roots1 Same as the roots fune- tion, but gives more accurate answers when there are repeated roots; 26, 31, 33 shg Shows the graph window; 16 series Computes a series sys- tem connection; 26, 35-38, 40, 122, 152 semilogx Generates an z-y plot using semilog scales with the 2- axis logio and the y-axis linear; 17-18, 20 semilogy Generates an 2-y plot using semilog scales with the wraxis logio and the z-axis lin- ear; 17-18, 20 ss2tf Converts a state-space sys- tem representation to a trans- fer function representation; 122- 123 step Computes the unit step response of a system; 26, 45- 46, 63, 87, 122, 155 Index/Glossary subplot Subdivides the graph display into sub-windows; 17, 20 tfss Converts a transfer func- tion system representation toa state-space representation; 122- 123 title Puts a title on the cur- rent plot; 17 who Lists the variables in the workspace; 7, 9-10 whos Lists the variables in the workspace including their size and type; 7,9 xlabel Puts an z-axis label on the current plot; 17 ylabel Puts @ yaxis label on the current plot; 17 161