You are on page 1of 39

Fundamentals of Automatic Process Control

Basically, automatic control is relatively easy to comprehend. In many ways it is much like manual control. However, the
automatic controller does not necessarily duplicate what the human operator does by hand. Automatic equipment gives
continuous, minute attention to the control application. Automatic controllers can compute and remember, but they
cannot reason from new conditions, nor can they forecast beyond the data which are built into them. Thus, these are the
basic differences between human and automatic-control means—factors that govern automatic controller design and
use.

Typical Process

The term process, as employed here, signifies the function and operations utilized in treatment of material. In Fig. 1, for
example, the operation of adding heat energy to water is a process. The steam coils, tank, pipes, and valves comprise the
plant within which the heating process is accom¬ plished. The parts and characteristic behavior of this typical process are
analyzed in the following paragraphs to portray what effect these factors have on the controllability of the process.

The Energy Exchanger

In many other processes, the exchange of materials alone, or of materials and energy, may be involved. In reference to
Fig. 1, energy is put into the process, passes through a series of exchanges, and emerges as energy output. The rate of
energy output equals the rate of energy input minus (1) the rate of energy lost and (2) the rate of energy storage in the
process.

Balanced Condition

When input energy equals output energy, the process is in a steady-state condition; i.e., it is in balance. Any disturbance
to either the input or the output of energy will upset this balance and consequently cause a change in the values of the
process variables. In this example, one of the process variables is output-water temperature. When the heat energy
output balances the heat energy input, the output-water temperature remains at a constant value until this heat energy
relationship is upset.

Self-Regulation

Some processes possess an inherent characteristic which aids in limiting deviation of the controlled variable. In Fig. 1,
when the steam input increases, the water temperature reaches a point of balance at some new, higher value; i.e., the
water temperature does not increase indefinitely. This inherent ability of a process to balance its energy (or material)
output and input is termed self-regulation.

Function of Automatic Control

An automatic controller can be defined as a system that measures the value of a process variable and operates to limit
the deviation of that variable from the set point. A process variable held within limits is termed a controlled variable. The
automatic controller regulates the controlled variable by making corrections in another variable of the process, which is
termed the manipulated variable.

Process Time Lags

All that appears necessary is to observe the hot water thermometer and to correct the steam valve opening accordingly
so as to hold or change the water temperature to the set point. However, processes have the char¬ acteristic of delaying
and retarding changes in the values of the process variables. This process char¬ acteristic greatly increases the difficulty
of control. These delays and retardations generally are termed process time lags.

Capacity and Resistance

The parts of a process that resist transfer of energy (or material) between capacities are termed resistance. In Fig. 1 the
walls of the steam coils and the insulating effect of the layers of steam and water on either side of them resist the
transfer of heat energy between the steam in the coils and the water on the outside of the coils. The combined effect of
supplying a capacity through a resistance produces time retardations in the transfer of energy between capacities. Such
resistance-capacity (RC) time retardations are often called capacity lags or transfer lags.

Dead Time

This time lag is not just a slowing down or retardation of a change, but a discrete time delay (or dead time) during which
no change whatever occurs at the output. The length of this dead time depends on both the velocity with which the
change is transported and the distance over which it is carried. Thus, dead time is also termed distance-velocity lag or
transportation lag.
Types of Process Disturbances

1. Supply Disturbances. This is a change in the energy (or materials) input to a process. In the heat exchanger of Fig. 1,
changes in steam quality, steam pressure, or steam valve opening are supply disturbances.

2. Demand Disturbances. This is a change in the energy (or material) output from a process. In Fig. 1, changes in cold
water temperature and rate of water flow are demand disturbances.

Effect of a Sudden Supply-Load Change

Good mixing of the water is assumed in these illus¬

trations. Each curve indicates how the temperature starts to increase immediately after the load is changed—and then
how it increases at a slower and slower rate until it finally rises to a new steady- state value. Worthy of note is how the
full response of the temperature is retarded in time; i.e., as the heat storing capacity of the process is increased, more
time is required for the temperature to reach its final value.

Effect of Both Supply- and Demand-Load Changes

The significant point to note in curve

a is that the temperature immediately begins to change as soon as the demand disturbance occurs. Curve b shows the
effect of a sudden supply-load change made at time zero and represents the increased steam supply just sufficient to
exactly correct for the demand disturbance represented by curve a. Curve c shows the effect of simultaneously applying
the same demand-load change (curve a) and the exact supply correction for it (curve b).

Multicapacity Processes

Although many processes have more than two resistance-capacity pairs of relatively equal size, the behavior of these
multicapacity processes is similar to that for the two-capacity proce
ss as shown by the reaction curves in Figs. 6 and 7. Thus
these reaction curves can be considered typical for all processes (that have no dead time) with two or more RC pairs of
relatively equal size. In general, inclusion in a process of two or more capacities of about the same size greatly increases
the difficulty of control.

Effect of Dead Time in Multicapacity Processes

The reac¬ tion curves are simply shifted along the time scale by the amount of dead time. The addition of dead time
does not change the shape or size of the reaction curves. The transportation of pro¬ cess energy or material through a
distance can cause dead time wherever it occurs, and, there¬ fore, dead time can exist in any part of a single¬ capacity or
multicapacity process.
Manual Control

When the operator feels this temperature drop, the operator must compare it with the temperature desired—then
mentally compute how much and in what direction the steam valve must be repositioned—and then manually make this
correction in the valve opening. Some time is required, of course, to make these decisions and to correct the valve
position. It is also true that some time will elapse before the effect of the valve correction on the output- water
temperature can reach the output and thus be sensed by the operator.

Problem of Control

Feedback Control

A correction for a disturbance cannot be made until the effect of the disturbance is known. But process time lags require
that some time must elapse before this can be ascertained. Also, some time is needed to measure the deviation and
make the correction. Then, in turn, process time lags require again that time pass before the effect of the correction can
be known.

Practical Meaning of Process Reaction Curves

1. Single-Capacity Process. Processes dominated by one resistance-capacity pair are easier to con¬ trol because:

a. They begin to react immediately to load changes—so deviations can be known and corrections applied without delay
(Fig. 4).
b. Corrections are immediately effective in reducing deviations (Fig. 5).

2. Multicapacity Processes. Processes having two or more resistance-capacity pairs of comparable size are more difficult
to control because:

a. They do not begin to react to load changes immediately (Fig. 6), so deviations do occur and corrections can be applied
only after some time has passed.

b. Corrections are not immediately effective in reducing deviations (Fig. 7).

3. Dead Time. Dead time (transportation lag) increases control difficulty wherever it is found.

Basic Functions of Control

the operator measures the output temperature, compares it to the desired value, computes how much to open the
steam valve, and makes that correction to the steam supply. Thus, the basic functions performed by the operator in
manual control are (1) measurement, (2) comparison, (3) computation, and (4) correction.

A Typical Control System

The foregoing theoretical discussion can be clarified by considering how the basic control functions are performed by an
actual automatic-control system. As a typical example, Fig. 12 shows schemat¬ ically an air-operated control system
applied to control of the water heater process of Fig. 1.

Measuring Means

1. Thermometer Bulb (Primary or Sensing Element). This is the part of the measuring system that is directly sensitive to
the controlled variable (temperature). The primary element converts energy from the controlled medium (hot water) to
a measurable signal (fluid pressure).

2. The Capillary Tubing (Transmitting Means). This part of the system carries the signal from the primary element to the
receiving element.

3. Bourdon Pressure Element (Receiving Element). This part of the system evaluates the signal from the primary element
and converts it to scale readings, chart records, and actuation for the error detector.
Amplifier

In order not to restrict the sensitivity or accuracy of the measuring system, an error detector must take very little power
from the measuring system. Thus, error signals usually are very weak. In order to operate most final control elements,
the error signal must be greatly amplified in power. Therefore, an automatic controller usually contains a power amplifier
which employs auxiliary power to increase the strength of the error signal.

Motor Operator

In most control systems, this requires some form of motor to operate the final control element. In the air- operated
control system of Fig. 12, the motor operator that positions the steam valve is the diaphragm air motor. The output air
pressure (amplified error signal) from the pilot valve is applied to the motor-operator diaphragm.

Control-System Time Lags

Automatic-control systems have time lags that can have a serious effect on the performance of fast¬ acting control loops.
The same kinds of lags—resistance-capacity lags and dead time—which are

found in processes are also encountered in control systems. Further, the lags found in controllers are caused by the same
properties—inertia, capacity, resistance, and transportation

Measuring-Means Lags

The resistance to fluid flow in the capillary tubing and the volumetric capacity of the tubing and bourdon spring also
form a resistance-capacity pair which has a time lag. Therefore, long lengths of thermometer capillary tubing should be
avoided in control systems. For the same reason, any pneu¬ matic or hydraulic transmitting means used between the
primary element and the receiving element in any control system should be as short as possible.

Final-Control-Element Lags

In general, every part of an automatic-control system must be designed, selected, installed, and maintained to avoid time
lags as much as possible. It must be stressed that a given time lag has just as bad an effect on control-loop stability when
it occurs in parts of the automatic-control system as when it occurs in the process itself. This can be illustrated by
examining how the time lags of the control system shown in Fig. 12 can be reduced by use of solid-state electronics

Microprocessor-Based System and Time Lags

Measuring Means

The thermometer bulb (primary or sensing element) and the capillary tubing (transmitting means) have been replaced by
a pair of thermocouple wires. This has eliminated the RC lag of the thermometer bulb and capillary tubing.
Control Actions

In manual control (Fig. 9), when the operator senses that the temperature is deviating below the desired value, he can
make steam valve corrections in several ways:

1. Instantly open the valve wide.

2. Open the valve slowly at constant speed as long as deviation continues.

3. Open the valve more when the deviation occurs rapidly.

4. Open the valve a constant amount for each unit of deviation.

The operator also can use other methods or combinations of methods of valve manipulation.

Two-Position Action

In this mode of control, the final control element is moved relatively quickly from one of two fixed positions to the other
at a single value of the controlled variable. Inasmuch as these two positions of the final control element are usually open
and closed, this mode is often termed on-off control
Two-Position Differential-Gap Control Action

This is a common variation of the two-position action. Here the final control element is moved rel¬ atively quickly from
its first position to its second when the controlled variable reaches a set value from one direction, and can return to its
first position only after the variable has passed in the oppo¬ site direction through a range of values (called the
differential gap) to a second value

Multiposition Control Action

In this action the final control element is moved to one of three or more fixed positions, each corre¬ sponding to a
definite range of values of the controlled variable. The control instrument of Fig. 17b can produce a three-position action
if it is used

with a final control element that takes a third position when the variable value is within the differential gap.

Time Proportioning Control Action

A constant speed motor rotates a cam that causes the right-hand electric contact to swing back and forth continuously.
The bourdon tube positions the left-hand contact in accordance with the value of the temperature. Suppose that contact
settings are such that the set point is 500° and the proportional band is 20° wide, extending from 490 to 510°. When the
temperature is below the proportional band (below 490°), the contacts are open all the time and the valve is open all the
time.

Floating Control Action

In this control action there is a predetermined relation between the deviation and the rate of travel of the final control
element. The final control element moves relatively slowly toward either one or the other of its two extreme positions,
depending on whether the controlled variable is above or below the set point.
Single-Speed Floating Control Action

In this control action the final control element moves at a single rate regardless of the amount of deviation. The single-
contact on-off controller of Fig. 166 can produce single-speed floating action if used with a slow-running reversible
electric-motor-driven valve. When the temperature is at or above the set point, the valve runs toward its closed position
at a single speed (see Fig. 20a).

Single-Speed Floating Control Action with a Neutral Zone

In actual practice, single-speed floating control is usually employed with a neutral zone. The two- contact controller (Fig.
17a) when used with a relatively slow reversible-motor valve produces such neutral-zone action (see Fig. 206). When the
value of the controlled variable is in the neutral zone between the contact setting, no contact is made and the valve
remains motionless—its rate of travel is zero.

Proportional-Speed Floating Control Action

In this control action the position of the final control element is changed at a rate that is proportional to deviation. In
other words, the greater the deviation becomes, the faster the valve moves, as shown in Fig. 20d. This is also called
integral control action

Integral Control Action

The factor of primary importance in connection with integral control action is that as long as deviation continues, the
correction to the valve position con¬ tinues. Integral control responds to both the amount and time duration of the
deviation. Thus, this mode of control continues to operate until it produces an exact correction for any load change. This
is a unique advantage of integral control action over any other action.

Proportional Control Action

In proportional control action (throttling or modulating control action), there is a continuous linear relation between the
value of the controlled variable and the position of the final control element (within the proportional band). In other
words, the valve moves the same amount for each unit of deviation of the controlled variable from the set point.
Gain (Sensitivity

Another concept for expressing proportionality is gain or sensitivity. These terms describe the ratio between the input
and output to and from a control device. Mathematically, gain and sensitivity are reciprocal to proportional band.

Offset (Droop)

Proportional control action can produce an exact correction for only one load condition; at all other loads, there must
always be some deviation (error) left. This error is called offset (droop) and is an inescapable characteristic of
proportional control action.

Proportional-plus-Derivative Control Action

In this action (sometimes referred to as rate action), there is a continuous linear relation between the rate of change
(i.e., first derivative) of the controlled variable and the position of the final control element. In other words, the valve
motion is proportional to the speed at which the temperature is changing; the faster the temperature changes, the more
the valve is moved.
Derivative Time

The derivative adjustment is expressed in a der¬ ivation time, which is the time in minutes by which derivative action
advances the effect of propor¬ tional action on the final control element. Figure 30 shows this graphically. Proportional
action alone would produce the valve position shown by the solid curve.

Proportional-plus-Integral-plus-Derivative Control Actio

1. The proportional action component corrects the valve position by an amount proportional to the deviation and thus
produces a temporary increased energy input represented by area A.

2. The integral action (reset) component corrects the valve position at a rate proportional to deviation and thus produces
a permanent increased energy input represented by area D.

3. The derivative action rate component corrects the valve position by an amount proportional to the rate of change of
the controlled variable. Derivative correction first added the energy repre¬ sented by area B and then subtracted the
energy represented by area C.

Selection of Control Action

Probably the most troublesome decision in designing a control system is selection of an adequate yet economical control
action. The solution is usually a compromise between the quality of control

obtained and the cost of the control system. The control system must be sufficient to meet the toler¬ ance of the
process, but it should not include refinements beyond those required, or its cost will be excessive.

Stability of Control Loops

Feedback control loops can be either stable or unstable. A feedback control loop is said to be stable when it is in such a
condition that after a disturbance its controlled variable eventually reaches a steady, noncyclic value. It is the purpose of
the following discussion to explain in a simple, practical way why and when control loops will be stable or unstable and
to give some practical criteria for expressing their stability.
Types of Oscillation

1. Increasing Amplitude. If at the frequency that gives a 360° phase shift the gain around the loop is over one, the
amplitude of oscillation will increase. This condition is shown in Fig. 34a.

2. Constant Amplitude. If at the frequency that gives a 360° phase shift the gain around the loop is just one, the
oscillation will continuously sustain itself at constant amplitude (see Fig. 34b.).

3. Decreasing Amplitude. If at the frequency that gives a 360° phase shift the gain around the loop is less than one, the
oscillations will decrease in amplitude and eventually stop (Fig. 34c). This is the stable condition sought in automatic
control

Effect of Control Actions on Stability

Proportional Action

The problem in feedback control is to improve control performance through the use of more feedback, but this tends to
produce instability and self-excited oscillation. Therefore, proportionality (or gain) in most automatic controllers is made
adjustable.

Integral Action

The addition of integral action (reset) to proportional action causes a negative phase shift in the response of the
controller, as shown in Fig. 36. Note that the proportional component of output lags

just 180° behind the input variable because of the built-in negative feedback (i.e., point B lags point A by 180°). Also, the
integral action component lags behind the proportional component by 90° (point C lags point B by 90°).

Derivative Action

It has been stated that one remedy for oscillatory tendencies in feedback loops is to use means that correct for phase lag
by creating phase lead. Such a means is drivative or rate action. Note that, while the integral component lagged the
proportional by 90° (Fig. 36), the derivative component leads the proportional by 90° in Fig. 37 (point C leads point B by
90°).

Effect of Load Changes on Stability

Anything that changes the balance between energy or (material) flow of the input and output of a control loop can be
termed a load change. The size, rate, frequency, and duration of load changes greatly affect the quality of control
obtained.

Kinds of Load Changes

Load changes are of four general kinds:

1. Supply-load changes

2. Demand-load changes

3. Set-point changes

4. Ambient variable changes


Nonlinearity of Control-Loop Components

Thus far, control loops have been considered as though they were linear, i.e., as though the loop phase shift and gain
were constant for all conditions of load and set point. This is an oversimplified concept, because many control-loop
components do change their phase shift or gain under different conditions.

Nonlinear Phase Shift

Many control-loop components produce different amounts of phase shift (lag) under different con¬ ditions. The
following are some practical examples: (1) In liquid-level control, the capacity lag of a tank with nonparallel sides changes
as the set point is moved up and down; (2) in flow-rate control

processes, the transfer lag changes with the set point, viscosity, and density; (3) in many processes, the dead time (phase
lag) changes greatly with load

Nonlinear Gain

Control-loop components can produce different gain under different conditions. As an example, many controllers have
nonlinear (such as square root) scales, so that their proportional band (or

gain) is different in every part of their range.

Practical Results of Nonlinearity

Load Changes

These facts explain the common observation that controller adjust¬ ments which may produce good stability at one load
may produce severe cycling at an increased or

decreased load
Set-Point Changes

This explains the common observation that controller adjustments which produce good stability at one set point may
produce severe cycling at higher or lower set points.

Stability Changes with Time

Or friction and lost motion can develop in parts of the controller, diaphragm operator, and valve because of wear,
corrosion, and dirt. Particularly common is the development of static friction in control-valve stems,

bearings, and packings. Excessive wear of the control-vlave ports or leaks in the output-air system (of a pneumatic
controller) can change the proportionality (gain).

Problem of Noise

Similar noise problems can occur in connection with most signals, e.g., (1) random pulsations in pressure signals, (2)
waves in liquid-level signals, (3) turbulence in differential-measured flow signals, and (4) induced alternating or direct
currents in thermocouple circuits.

Solutions to Noise Problem

1. Reduction or elimination of the noise at its source.

2. Filtering or averaging the noise out of the signal. For example, in Fig. 12 the source of the thermal noise can be
eliminated by better mixing of the hot and cold water in the tank or by using an averaging-type thermometer bulb that
measures temperature over a considerable length instead of at one point.

Adjustment of Automatic Controllers

The quality of regulation obtained from automatic-control systems depends greatly on the adjust¬ ments made in their
various control actions. To obtain the best control, a systematic adjustment method must be used; haphazard
adjustments rarely find the best settings.

1. Systematic trial method - This method of controller adjustment requires that the controller and the process be
completely installed and operating in their normal manner.

2. Ultimate-sensitivity calculated method* - A more exact way of using this self-excited closed-loop analysis is the
ultimate-senstivity method developed by Ziegler and Nichols (1942). This method permits calculating all three controller
adjustments from data obtained in a simple quick test of control-loop character¬ istics.

3. Reaction-curve calculated method- The general procedure in the reaction-curve method of controller adjustment is to
open the control loop just before the final control element and to create a small, sudden step change in process input.

Multielement Control (Multivariable, Multiple-Loop)

Multielement control occurs when two or more input signals jointly affect the action of the control system. Examples of
multielement control are:

1. Cascade control

2. Ratio control

3. Auctioneering control (override control, limiting control)


Cascade Control

In cascade control, the output of one controller is the set point for another controller. Each controller has its own
measured variable with only the primary controller having an independent set point and only the secondary controller
providing an output to the process.

Ratio Control

In ratio control, a predetermined ratio is maintained between two or more variables. Each controller has its own
measured variable and output to a separate final control element. However, all set points are from a master primary
signal that is modified by individual ratio settings.

Auctioneering Control (Override Control, Limiting Control)

1. Suction and Discharge Pressure Compressor Control. The discharge control valve is nor¬ mally regulated from the
discharge pressure (Fig. 49).
2. Flow and Pressure Control of Gas Distribution Systems. The distribution control valve is normally regulated from
the discharge pressure (Fig. 50).
3. Temperature or Vacuum and Differential Pressure Control for a Packed Distillation Column (Fig. 51). The column
is controlled either from a vacuum set point, with the steam flow on manual for constant heat input, and a
differential-pressure override to prevent flooding
Feedforward Control

Feedback control, by definition, requires that a deviation exist between the measured variable and the set point before
control action can take place. The feedback controller changes its output until the deviation (error signal) between the
set point and the measured variable is negligible, which is essentially a trial-and-error process characterized by the
oscillatory nature that feedback control

often exhibits.

Multiple-Variable Interactive Control

A method for evaluating the degree of interaction between control loops involves use of the rel¬ ative gain concept or
the Bristol array.* A two-output-variable interactive system is shown in Fig. 54. The Bristol array for this process is given
in Fig. 55. The concept is similar for a system of any size. Figure 55 is filled out from either a system test or an analytical
analysis.
GENERAL CONTROL-SYSTEM DESIGN PROCEDURE

To help ensure that all necessary considerations are included in control-system design, the fol¬ lowing procedure is
suggested. This procedure is built around the generation of three key documents:

1. Description of equipment operation

2. Control system specification

3. Control system test specification

PROCESS CONTOL TECHNIQUES


The presentation is in five parts covering the following topics:

1. System characteristics and representation

2. Basic control concepts

3. Continuous and sampled-data control systems

4. Stability

5. System design and control strategies

SYSTEM DEFINITION

The first step in the design of a control system is to develop a thorough understanding of process operation and control
objectives. The next step is the development of a model for the system. System boundaries are defined and inputs and
outputs are specified. An example is shown in Fig. 1; the formal representation is shown in Fig. 2. The relationships are
often expressible in terms of algebraic equations and Laplace transforms, or z transforms in the case of sampled-data
systems.
Linearity

The defining characteristics of a system in an engineering sense are dependent on the system inputs and the intended
use. In many cases, if the system is to be subjected only to small disturbances, the assumption of linearity may be made.
In a linear system, the output signal does not vary in form or in relative magnitude with varying size, level, or direction of
the input.

Nonlinearities

When significant nonlinearities are present, the analysis of stability and performance becomes more complicated. There
are two notable cases where nonlinearities cannot be ignored. First, in chemical reactions the reaction rate is a strong
function of temperature as indicated by the Arrhenius law

SYSTEM REPRESENTATION

Two types of diagrammatic representations are common in the literature on control systems. Both types permit complex
systems to be represented in compact, meaningful form. By following set rules, the diagrams permit system
characteristics to be studied and investigated. Continuous-Data System Diagrams and Sampled-Data System Diagrams.

CONTROL CONCEPTS

Characteristics of Feedback Control Systems

The symbols in the block diagrams represent the transfer functions of the respective blocks. In gen¬ eral, such functions
have a dynamic part and a steady-state part; when they are combined, they specify the output response to changes in
the input.

Key Ratios

GH, the Loop Ratio. A signal passing once around the loop from any point will be multiplied

by minus this operator.

1 /(I + GH), the Actuating Ratio. The ratio of the output of a summing point to the external input signal entering the
summing point. A third and fourth algebraic operator are important where the reference input (or set point) varies.

G/(l + GH), the Control Ratio. The ratio of the controlled variable C to the reference input

R.

GH/(1 + GH), the Primary Feedback Ratio. The ratio of the response of the primary feed¬ back B to reference input R
Feedforward Control

Closed-loop feedback control tends to effectively reduce disturbances and to force the variable to the new command
level. This is true, however, only when a high “gain” can be attained without insta¬ bility. Processes that contain
significant dead time require both low proportional gain and slow inte¬ gral gain for stability. The result is a slow
response to set-point changes and low attenuation of disturbances with significant high-frequency components.

SMITH LINEAR PREDICTOR

CONTINUOUS SYSTEM ANALYSIS


A thorough understanding of the dynamic response of linear systems is essential if one is to design control systems. The
topics below cover linear system dynamics and summarize basic techniques for the analysis of continuous linear systems.
Linear discrete-data systems are piecewise linear systems and are discussed in the next subsection

Dynamic Response

Following a disturbance a system exhibits two types of response: transient and steady state. The transient is the part of
the response of a system to a disturbance which tends toward zero (or infinity) as time increases without limit. The
steady state is the part which remains when the transient response is removed

Step Response

Transient response, to either a step or pulse input, is frequently used as a quick check on the response of a system or
component. Table 7 gives the step-response equations for several simple systems which often occur as components in
complete control systems. It can be seen from the partial fraction expansion that two factors influence the effect of any
root sk on the transient response.

Frequency Response

A steady-state method of determining the transfer function of a system is to observe its response to a sinusoidally
varying input. If the system under test is linear, the output response will also be sinusoidal with time, and the frequency
of the output sinusoid must be identical to the frequency of the disturbing sinusoidal signal

Polar Plots of Frequency Response

Graphical techniques exist for the design of feedback control systems using polar frequency-response plots. However,
polar plots are chiefly useful in stability analysis (see Nyquist Criterion). The log magnitude and phase plots described
below are generally used for system design.

Log Magnitude and Phase Plots

If separate plots are made of the logarithm of the magnitude ratio versus the logarithm of frequency and of the (linear)
phase versus the logarithm of frequency, several advantages are gained. Since the magnitude ratios are in logarithmic
form, the overall log magnitude ratio of a series of noninteracting components at each frequency is obtained by addition
of the values.
STATE VARIABLE THEORY

The methods presented previously are convenient for dealing with single-input-single-output (SISO) control systems and
other simple configurations. For more complex multiple-input-multiple-output (MIMO) systems, methods employing
linear algebra and matrix techniques are very useful. System description is based on the concept of state: “

Observability and Controllability

Observability and controllability are mathematical limit tests that define conditions for which control is not possible.

Observability. A system is said to be observable if measurements of the output c contain suf¬ ficient information
to enable us to completely identify the state x.

Controllability. A system is controllable if it is possible to bring a state variable to any arbitrary value in a finite
period of time using the manipulated variables that are available.

Observability and controllability are both necessary conditions for control, but they are not suf¬ ficient to define the
degree of control possible.

SAMPLED-DATA SYSTEMS

Sampled-data techniques apply when a control system involves a digital computer, a microprocessor, or an analyzer with
a sampled output. A brief summary of techniques is given below. More detailed discussions and derivations are given in
Ragazzini and Franklin (Ref. 36), in Kuo (Ref. 37), and Tou (Ref. 38). For a treatment oriented to digital process control
systems, see Ash (Ref. 4).

Algorithms for Direct Digital Control (DDC

There are three possible approaches to the design of discrete-data controllers for conventional appli¬ cations such as a
single-loop control and cascade control: (1) Convert conventional control equations to difference equations. (2) Use a
formula such as the Ragazzini and Franklin formula (see Ref. 36). (3) Use a special purpose design employing basic
discrete-data techniques (Refs. 21 and 37).

STABILITY

The basic idea of stability is this: “If a system has a suitable response for a given set of inputs or initial conditions, and if
small changes in the inputs or initial conditions occur, then the new response should be close to the original one”
(Brockett, Ref. 43). For linear systems the definition of stability is relatively simple. It depends on pole and zero locations
and dead time only, and is not a function of initial conditions or of the disturbances.

Linear System Stability


A stable linear system or element is one in which the system response is always bounded (does not increase without
limit) for any bounded system input. An example of a stable system is G(s) = (1 + TV)-1; the system G(s) = (1 — Ts)~] is an
example of an unstable system. A step input is a bounded input.

Systems Equations and Eigenvalues

Linear system stability depends on the eigenvalues or poles of the closed-loop system equations. When the state variable
equations are used, computer solutions may provide the eigenvalues. For simpler configurations where the transfer
function approach is employed, it is necessary to evaluate stability based on the characteristic equation

Nyquist Criterion

Like the Routh-Hurwitz criterion, the Nyquist criterion permits a determination of absolute stability to be made without
actually solving for the roots of the closed-loop characteristic equation. The Nyquist criterion, being a graphical method,
often gives a certain physical interpretation which assists in stabilizing the system.

Describing-Function Techniques

If a nonlinear system or component is subjected to a sinusoidal input, the output will not necessarily

be a sine wave. However, in many practical cases, the output response will contain the input fre¬ quency as the
fundamental Fourier-series components. In this case, the magnitude ratio and phase angle of the Fourier fundamental
versus input amplitude, the frequency, and other factors form the describing Junction. The describing function may be
determined either analytically or experimentally.

Stability Definitions for Nonlinear Systems

1. Local Stability or Stability in the Small. Applies to a small region about a singular point.

2. Global Stability or Stability in the Large. Applies to the entire state space of the system.

3. Finite Stability. Applies to a region larger than that of definition 1 but smaller than that of definition 2.

Ordinarily definitions 1 to 3 do not exclude the possibility of limit cycles.

4. Asymptotic Stability. For any initial conditions within the region, the state point approaches arbitrarily close to the
singularity as time approaches infinity. Dynamic equilibrium such as a limit cycle is excluded.
5. Monotonic Stability. Similar to asymptotic stability for fixed-parameter systems. The above generally apply only to
autonomous (undriven) systems. Total stability for a nonautonomous (driven) system indicates that for every input
within a reasonably small bound, the output is bounded.

Liapunov’s Second Method

This provides several theorems which when applied to the so-called Liapunov V functions will indicate stability or
instability. The result is obtained without solving the system equations, and the method is sometimes called the indirect
method. It applies to systems that may be autonomous or forced, linear or nonlinear, stationary or time-varying, or
deterministic or stochastic (Ref. 31, p. 498).

SYSTEM DESIGN TECHNIQUES

In addition to meeting the requirement of stability, a control-system design must satisfy certain design criteria which
specify performance and design margins. A number of criteria are in common use. These are somewhat related to the
design techniques employed and to the problem at hand.

Continuous-Control System Design

Complete specification of system performance generally includes a definition of steady-state errors as well as a
description of dynamic performance. Systems with nonzero steady-state errors include certain systems with proportional
control and feedforward control systems without feedback. Given the performance requirements, the basic design
problem is to meet these requirements by applying (1) compensating control functions (lead and lag functions), and (2)
feedback paths.

Frequency-Response Techniques

For a system in which the set-point response is of concern, a dominant second-order system is often assumed for the
closed-loop response. The performance specification can then be made in terms of f and (*>„, as discussed previously.
Frequency- response plots for the second-order system are shown in Fig. 43.
Sampled-Data System Design

Sampled-data control systems can be designed using approximations based on continuous control techniques—however,
there are significant advantages in designs based on purely sampled-data techniques.

1. A wider range of control functions can be realized. (In particular, realistic simulation of dead time.)

2. Input and disturbance time functions can be specified.

3. Computations and logic are more easily accommodated, and special control algorithms are not difficult to develop,
when necessary.

4. Digital computer design aids and simulation programs make it feasible to design in the time domain. R

CONTROL STRATEGIES

The selection of a control strategy requires that two types of information be available: (1) the char¬ acterization of the
process, including a dynamic model with a definition of disturbances; and (2) a clear understanding of control objectives
and criteria for evaluating performance. The strategies for various types of process model characteristics are discussed
below.

Process Characteristics

The selection of a control strategy depends on the type of process model, as shown in Fig. 49. The assumption has been
made here that normal operation involves significant changes in process param¬ eters. For this reason, a preference for
feedback control is indicated, even where feedforward control is used. With this approach, the feedback control will
usually compensate for the effect of process parameter variations.

Single-Variable Control without Dead Time

If the dead time in the process model is small compared to the lags involved, a simple feedback control loop will often be
adequate. If the techniques discussed in System Design indicate that performance will not be adequate, cascade control
may be employed, as shown in Fig. 50.
Cascade Control

The method of analysis is to design the secondary loop first. Franks and Worley (Ref. 68) present the results of an analog
computer study indicating improvements which are possible using cascade control versus single-loop control. The results
compare the performance of the two config¬ urations using the ratio of ITAE for single-loop to ITAE for cascade versus
the ratio of dominant lags in the primary and in the secondary.

Noise

Noise is a factor in the design and the tuning of some control systems. Most random noise found in industrial control
applications can be characterized as exponentially correlated noise. Fertig (Ref. 14) describes a method for characterizing
noise amplitude and noise bandwidth in terms of easily measured parameters.

Practical Considerations

Three examples are cited. Shunta (Ref. 74) describes an application requiring sampled-data control of level with a dead
time. The storage tank receives product from a distillation train and maintains the level within tank limits by control¬ ling
feed to the train.

MODELING AND SIMULATION

Process models are necessary for control-system design and are fundamental to all practical types of optimal or adaptive
control. The appropriate type of model may be quantitative or qualitative. For most applications a quantitative model is
necessary. (For an example of a qualitative model see Hill

Climbing.) It is convenient to classify the method of derivation of quantitative models as fundamental or empirical
models. The resulting model may be composed of algebraic or differential equations
and may be either linear or nonlinear.

Model Formulation

The first step in the formulation of any model is to establish the boundaries of the portion of the process to be modeled.
The choice of the boundaries should be made so as to:

1. Exclude as many variables and equations as possible

2. Eliminate as many unmeasured disturbances as possible

3. Hold measured (uncontrolled) disturbances to a minimum

4. Allow explicit relationships between measured variables and factors that disturb the process

5. Include all significant variables

Model Strategies

In many types of systems the difficulty of dealing with a large model having many nonlinearities can be circumvented by
splitting the model into two or more parts. One approach is to build a simple linearized model including the dynamics
that represent a portion of the system at a particular oper¬ ating point. The other portion of the model would include
little or no dynamics and could be very complex and detailed.

Statistical Models

A statistical model deals chiefly with future probabilistic behavior, irrespective of the past history of the process.
Statistical models are in general most useful in describing the environment around the process being controlled. When
they are applicable in a process model it is frequently for some type of discrete operation or filtering as in a Kalman filter.

Simulation

Simulation permits testing and validation of control schemes and devices under a variety of condi¬ tions, some of which
would be difficult, costly, or dangerous to duplicate on the actual process. Sim¬ ulation also permits a comparison of
alternative strategies under identical conditions. Time can often be compressed in the simulation, thus decreasing test
time. Simulation is often used for (1) testing control strategies prior to installation, with or without the control hardware;
(2) validating a model for optimization or for adaptive control under a wide range of conditions and parameter mismatch;
and (3) operator training, especially in the handling of contingency situations. This is useful both for preinstallation and
for ongoing training.

Digital Simulation

The two types of digital simulation, continuous and discrete, are both done by discrete methods. There are a number of
simulation programs of the continuous variety which hopefully simplify this operation: Madas, Pactolus, Mimic, DSL/90,
and CSSI, to mention a few.

OPTIMIZATION

Optimal control is a control system that maximizes (or minimizes) some performance index. As a practical engineering
consideration, a less than theoretically perfect example (suboptimal control) is often used. In some instances, especially
those involving large processes, simple controllers, and sim¬
ple interconnections of control devices (when considered as a single loop) do not provide the most economic operation.
In these cases, advanced control strategies should be investigated. These spe¬

cialized techniques are classified as (1) optimization, trying to obtain the best results from a known process; and (2)
adaptive control, in which part of the process is not known or is highly variable and the identification problem is
combined with the optimization problem.

Incentives and Constraints

Constraints are the limitations of the system. These include such items as safe operating condi¬ tions, miminum and
maximum power available, valve stroke limits, storage capacity, maximum (or minimum) allowable pressures,
temperatures, and speed, composition, and in general any hard lim¬ its. Imposed constraints include the practicality of
measuring certain state variables and certain con¬ trol actions.

Linear Programming

The simplex method can be used with digital computers or analog com¬ puters to solve the linear programming problem.
The solution is always in a “corner,” that is to say,

up against some of the constraints. Linear programming is widely used for blending and other systems which meet the
linearity requirements and the necessity for steady-state operation (no dynamics involved).

Nonlinear Programming

When the system of algebraic equations is not too nonlinear, the solution can sometimes be obtained by nonlinear
programming, particularly if the solution is in a corner. An analog computer solution is often applicable. The problems
involved in the analog solution are in the implementation and in the occurrence of an occasionally local maximum.

Hill Climbing

If the problem of finding the largest value of a function of two variables is given, the geometric interpretation of finding
the highest point on a surface, the name hill climbing is apropos. Other names are gradient method and method of
steepest ascent (descent).
Dynamic Programming

Continuous systems are sampled and treated as data sequences; hence the opti¬ mum control is treated as a multistage
decision process—the principle of optimality. A policy that is optimal over the interval 0 —»• N — 1 is necessarily
optimal over any subinterval n N — 1, where 0 < n < N — 1.

Kalman Filter

The Kalman filter is a method for providing an optimal estimate of variables in the presence of noise. Because of the
similarity of a filter design to a controller design, Kalman also provides a design method for an optimal controller for
some types of systems.
ADAPTIVE CONTROL

Adaptive control is on-line identification of process gains and time parameters and the subsequent use of this
information to improve performance of a control loop or strategy. In the simplest form, this is analogous to control loop
tuning; in more complex cases, it is analogous to on-line design of

a strategy.

Predictive Control

Given a model of the process and some measure of the state of the process, it is possible to predict what will happen to
the process for a given set of control actions. Because of noise and uncertainties, this prediction is normally limited to a
relatively short time. Predictive control can be described as utilizing this information to anticipate and hence avoid
undesired deviations.

Restrictions

The adaptive and optimizing techniques may be thought of as on-line automatic redesign of the control system. It
therefore follows that the limitations imposed on design must also be imposed on redesign, using these techniques
(Refs. 83 and 141).

Stability

In addition to stability of the individual loops, it is necessary also to have stability of the overall system and of the
computational procedures.

Control Algorithms
Control algorithms are computer programs designed to control and regulate the behavior of a system or a
process. These algorithms typically operate by monitoring the current state of the system, comparing it to a
desired state, and then calculating and executing the necessary actions to bring the system into alignment with
the desired state.

The implementation of a control function on a process requires either an analog control system, natural to the
process but with its own problems, or a digital computation. The process of sampling an analog value,
converting the resulting value to a (quantized) number, computing a control action, and outputting that value
involves issues remote from the abstractions of design control theory. This section develops one view of these
issues (Ref. 1). Other attitudes toward practical algorithm design may be of interest (Refs. 2 to 4).

NUMBER SYSTEMS AND THE EFFECTS OF THE BASIC ARITHMETIC OPERATIONS

There are two basic formats for representing control data as numbers, fixed point and floating point. The fixed-
point format represents the data as an integer, with a fixed range of possible values which must be scaled
(painfully) in much the same way analog control systems and simulations were scaled. The scale factors, as well
as multiple-precision computations and conversions, are the explicit respon¬ sibility of the programmer.
Fixed-Point Format

A fixed-point format represents an integer as a binary number stored in a word containing a fixed number of
bits, usually in what is called two’s complement format. This binary format causes the range of positive
numbers to be 0 to 2" — 1 and signed numbers to be — 2n_1 to 2n_1 — 1, n being the number of bits in the
word. In two’s complement arithmetic, negative numbers are represented in binary notation as very large
numbers, and addition is “around the clock.” Thus, in Fig. 1 (with n = 3), negative 3 ( — 3) is represented by a
binary number whose decimal equivalent is 5.

Error in Fixed-Point Arithmetic

The objective of design decisions relating to quantization and multiple precision is to avoid poor control arising
from inaccuracies not inherent in the real process. All the algorithms likely to be used are made up of
combinations of additions, subtractions, multiplications, and divisions; they define rational functions of their
data. It would be unimportant how calculations were ordered if all cal¬ culations were carried out with as much
precision as needed.

Fixed-Point Multiplication and Division

The extensive expansion of range of a multiplication is the most problematic aspect of multiplication. Among
other consequences it generates large numbers whose differences may cause large errors. It makes smaller
numbers smaller still, compared to the errors caused by the large numbers. The important issue is the ratio of
the large numbers, which make the errors, to the small numbers, which in the end constitute the result.

Floating-Point Format

With the floating-point format, all remainders, carries, and multiple-precision results of single-pre¬ cision
inputs disappear. The new floating-point standard (Ref. 8) deals with this problem by using an awkward
combination of different levels of precision and so-called guard bits. For floating-point data the errors
introduced through differences of large numbers become more severe, and at the same time less obvious
because their development is automated.

BASIC CONTROL ALGORITHMS

This section considers three basic algorithms as examples: the lag calculation, the lead-lag calcula¬ tion, and
the proportional-plus-integral-plus-derivative (PID) controller.

The Lag Calculation


Quantization and Saturation Effects

All the calculations can now be carried out. The discussion has not addressed the effects of truncation on
derivative but, as developed here, they are not more serious than for the proportional control. In both cases,
truncation will cause a very small limit cycle, on the order of the minimum quantization of the D/A converter,
when the loop operates about the steady state

OPERATIONAL EFFECTS

One of the classic controller effects is integration wind-up. It is an indefinite continuation of internal integrating
action after the output has reached a limiting value. Each of the above digital algorithms can be made to limit
the integrating action if the value hits a bound. If the limit is known, then one strategy for limiting further
integration is to saturate the output and the integration directly

IDENTIFICATION AND MATRIX-ORIENTED OPERATIONS

Current academic control theory emphasizes different matrix-oriented formulations. These are becoming more
common as more engineers are trained in them. From the author’s point of view the traditional control
elements and attitudes (including the Smith predictor and decouplers) are more effective for normal control
application efforts, but there are aspects of advanced control algorithm design where these newer methods are
appropriate.
SUMMARY

Digital control algorithms can be designed for experimental or single applications using the easiest tools:
FORTRAN and floating-point implementation. In this case, there is a reasonable hope that normal
commissioning debugging will weed out all problems. But, if a sense of workmanship moti¬ vates one, or many
people intend to use the algorithm, then attention to refinement and foolproofing

are important. Among the issues addressed in this article are:

1. The causes of numerical problems in fixed-point and floating-point algorithms

2. The documentation, control, and testing of detailed scaling, precision, and saturation within the

algorithm

3. Design for natural tuning and qualitative behavior predictable from an analog perspective

4. Nasty surprises due to quantization

5. Windup and other accommodations of the limits to control

6. Bumpless transfer and operational considerations

With the current state of programming languages, the proper design of a control algorithm is no more covered
by the usual FORTRAN textbook equation than is an automobile design covered by the styling artist’s sales
renditions.

You might also like