Stability of Closed-Loop Control Systems
Stability of Closed-Loop Control Systems
1. Introduction
A feedback control system must be stable as a prerequisite for satisfactory control.
Consequently, it is of considerable practical importance to be able to determine under
which conditions a control system becomes unstable. For example, what values of the
PID controller parameters Kc, I, and D keep the controller process stable?
Definition of stability
Before we proceed, we introduce the following definition for unconstrained linear
systems. Notice that the term unconstrained is used to refer to the ideal situation where
there are no physical limits on the output variable.
Definition of stability. An unconstrained linear system is said to be stable if the output
response is bounded for all bounded inputs. Otherwise, it is said to be unstable.
Characteristic equation
Consider the general block diagram, which is discussed in the previous chapter. Using
block diagram algebra that was developed in the previous chapter, we obtain
Gc Gv G p GL
y= y sp + d (1)
1 + GmGc Gv G p 1 + GmGc Gv G p
or equivalently,
C = GSP R + GLoad L (2)
The stability characteristics of the closed-loop response will be determined by the poles
of the transfer functions GSP and GLoad. These poles are common for both transfer
functions (because they have common denominator) and are given by the solution of the
equation
1 + Gc GmGvG p = 0 (3)
Equation (3) is called the characteristic equation for the generalized feedback system.
Let p1, p2,., pn be the n roots of the characteristic Equation (3):
1 + Gc GmGv G p = ( s p1 )( s p2 ) K ( s pn ) (4)
Then we can state the following criterion for the stability of a closed-loop system:
A feedback control system is stable if all the roots of its characteristic equation have
negative real parts (i.e. are to the left of the imaginary axis).
If any root of the characteristic equation is on or to the right of the imaginary axis (i.e. it
has real part zero or positive), the feedback system is unstable. Figure 1 provides
graphical interpretation of this stability criterion. The qualitative effects of these roots on
the transient response of the closed-loop system are shown in Figure 2. The left portion of
each part of this figure shows representative root locations in the complex plane. The
corresponding figure on the right shows the contributions these poles make to the closed-
loop response to a step change in the set point. Similar responses would occur for a step
change in load.
Figure 1 Stability regions in the complex plane for roots of the characteristic
equation.
The root locations also provide an indication of how rapid the transient response will
be. A real root at s = p1 corresponds to a closed-loop time constant of 1 = 1/p1. Thus, real
roots close to the imaginary axis result in slow responses. Similarly, complex roots near
the imaginary axis correspond to slow response modes. The further the complex roots are
away from the real axis, the more oscillatory the transient response will be.
Remarks
The product GOL = Gc Gm Gv G p is called open-loop transfer function because it relates
the measurement indication ym to the set point if the feedback loop is broken just before
the comparator
ym = GOL ysp (5)
Note that the same characteristic equation occurs for both load and set-point changes
since the term, 1+GOL, appears in the denominator of both terms in Equation (1). Thus, if
the closed-loop system is stable for load disturbances, it will also be stable for set-point
changes.
Example 1
Consider a process with the following transfer functions:
10
Gp = , G f = Gm = 1, Gc = K c
s 1
Determine the range of Kc values that result in a stable closed-loop system.
Solution
The corresponding characteristic equation is
10
1 + G f GmGc G p = 1 + 1 K c 1 = 0
s 1
s 1 + 10 K c
= =0
s 1
which has the root
p = 1 10 K c
The system is stable if p<0 (i.e. Kc > 1/10).
Example 2
Consider a process with the following transfer functions:
1 1
Gp = , G f = Gm = 1,
Gc = 1001 +
s + 2s + 1
2
0.1s
Determine weather the PI controller can stabilize the system.
Solution
The corresponding characteristic equation is
1 10
1 + G f GmGc G p = 1 + 1 K c 1 + 1 = 0
s + 2s + 1
2
s
The equation above yields
s 3 + 2s 2 + 102 s + 100 = 0
with roots -7.185, 2.59 + 11.5j and 2.59 11.5j. The closed-loop system is unstable
because two roots of the characteristic equation have positive real parts.
polynomials in s. Thus, the Routh Stability Criterion is not directly applicable to systems
containing time delays, since an e-s term appears in the characteristic equation.
The Routh Stability Criterion is based on a characteristic equation that has the form
an s n + an 1s n 1 + L + a1s + ao = 0 (6)
We arbitrarily assume that an > 0. If an < 0, we multiply Equation 6 by -1 to generate a
new equation that satisfies this condition.
First test. A necessary (but not sufficient) condition for stability is that all of the
coefficients ( an , an 1 , K , a1 , ao ) in the characteristic equation must be positive. If any
coefficient is negative or zero, then at least one root of the characteristic equation lies on
the right of, or on, the imaginary axis, and the system is unstable.
Second test. If all of the coefficients are positive, we can construct the following Routh
array:
Row1 an an 2 an 4 K
2 an 1 an 3 an 5 K
3 b1 b2 a3 K
4 c1 c2 c3 K
M
n +1 z1
where
a a a n an 3
b1 = n 1 n 2
an 1
an 1an 4 an an 5
b2 =
an 1
M
b1an 3 an 1b2
c1 =
b1
b1an 5 an 1b3
c2 =
b1
M
Routh Stability Criterion. A necessary and sufficient condition for all roots of the
characteristic equation to have negative real parts is that all of the elements in the left
column of the Routh array are positive.
Remarks
The number of sign changes in the elements of the first column is equal to the number
of roots to the right of the imaginary axis.
Example 3
Determine the stability of a system that has the characteristic equation
s 4 + 5s 3 + 3s 2 + 1 = 0
Solution
Since the s term is missing, its coefficient is zero. Thus, the system is unstable (First
test).
Example 4
Find the values of controller gain Kc that make the following feedback control system
stable.
1 1
G p = GL = , Gf =
5s + 1 2s + 1
1
Gm = , Gc = K c
s +1
Solution
The characteristic equation is
10 s 3 + 17 s 2 + 8s + 1 + K c = 0 (7)
All coefficients are positive provided that 1+Kc > 0 or Kc > -1. The Routh array is
10 8
17 1 + Kc
b1 b2
c1
where
17(8) 10(1 + K c )
b1 = = 7.41 0.588K c
17
b2 = 0
b1 (1 + K c ) 17(0)
c1 = = (1 + K c )
b1
To have a stable system, each element in the left column of the Routh array must be
positive. Element b1 will be positive if Kc > 7.41/0.588 = 12.6. Similarly, c1 will be
positive if Kc > -1. Thus, we conclude that the system will be stable if
-1 < Kc < 12.6
This example illustrates that stability limits for controller parameters can be derived
analytically using the Routh array; that is, it is not necessary to compute the roots of the
characteristic equation nor specify a numerical value for Kc before performing the
stability analysis.
Example 5
Use the direct substitution method to determine Ku for the system described in
Example 4.
Solution
Substitute s = ju and Kc = Ku into Equation (7):
10 j 3 17 2 + 8 j + 1 + K u = 0
or
(1 + K u 17 2 ) + j (8 10 3 ) = 0
Solve both equations
1 + K u 17 2 = 0
8 10 3 = (8 10 2 ) = 0
(9)
2 = 0.8 = 0.894
K u = 12.6
Thus, we conclude that Kc < 12.6 for stability. Equation (9c) indicates that at the stability
limit, Kc = Ku = 12.6, a sustained oscillation occurs that has a frequency of = 0.894
rad/min., if the time constants have units of minutes. The corresponding period = 7.03
min.
Introduction
The function of a feedback control system is to ensure that the closed-loop system has
desirable dynamic and steady state response characteristics. Ideally, we would like the
closed-loop system to satisfy the following performance criteria:
In typical control problems, it is not possible to achieve all of these goals since they
involve inherent conflicts and trade-offs. For example, PID controller settings that
minimize the effects of disturbances tend to produce large overshoots for set point
changes. On the other hand, if the controller is adjusted to provide a rapid, smooth
response to a set point change, it usually results in sluggish control for disturbances.
Thus, the trade-off is required in selecting controller settings that are satisfactory for both
load and set point changes.
where the error signal e(t) is the difference between the set point and the measurement.
Notice that for P controller, where offset occurs, the integral given by Equation does not
converge. In these cases, one can use a modified integrand, which replaces the error
by y () y (t ) , since this term does approach zero as t goes to infinity.
The ISE will penalize the response that has large errors, which usually occur at the
beginning of a response because the error is squared. The ITAE will penalize a response
which has errors that persist for a long time. The IAE will treat all errors in a uniform
manner; thus, it allows larger deviation than ISE. In general, ITAE is the preferred
integral error criterion since it results in the most conservative controller settings.
Design relations that minimize the ITAE performance index are shown in Table 1.
These relations are based on the first-order plus time-delay model, i.e. Equation (10), and
the ideal PID controller. Note that the optimal controller settings are different depending
on whether step responses to load or set point are considered. For load changes, the load
and process transfer functions are assumed identical.
Table 1. Controller design relations based on ITAE performance index and a first-
order plus time delay model.
Example 6
For the process model,
4e 3.5 s
G(s) =
7s + 1
Compare PI and PID controller settings based on ITAE tuning relations for both load and
set point changes.
Solution
0.977
3 .5
KK c = 0.859
7
K c = 0.423
0.68
3 .5
I = 0.647
7
I = 6.48
The ITAE settings are shown below:
Controller/Design method Kc I D
PI / load 0.423 6.48 -
PI / set point 0.276 7.39 -
PID / load 0.654 4.98 1.34
PID / set point 0.435 9.69 1.13
Figure 5 Comparison of controllers design using ITAE criteria for (a) set point and
(b) load changes.
Figure 5 compares the ITAE controllers. Design for load changes results in large
overshoots for set-point changes, while set-point design produces sluggish responses to
load disturbances. If set-point changes and load disturbances are both likely to occur,
then a compromise in the controller settings should be employed.
This example has demonstrated that, in general, integral error criteria for set point
changes results in more conservative controller settings than for load changes.
Controller tuning
After controller installation, the controller settings must usually be adjusted until
control performance is considered satisfactory. This activity is referred to as controller
tuning or field tuning of the controller.
In order to save time and effort, it is desirable to have preliminary estimates of
satisfactory controller settings. A good first guess may be available from experience with
similar control loops. Alternatively, if a process model is available, integral error
methods can be employed to calculate controller settings. Field tuning, may still required
to fine tune the controller, especially if the available information is incomplete or not
very accurate.
1. After the process reaches steady state, remove the integral and derivative modes of
the controller, leaving only proportional control. On some PID controllers, this
requires that the integral time (I) be set to its maximum value and the derivative time
(D) to its minimum value.
2. Select a value of proportional gain (Kc), disturb the system, and observe the transient
response. If the response decays, select a higher value of Kc and again observe the
response of the system. Continue increasing the gain in small steps until the response
first exhibits a sustained oscillation, Figure 6. The value of gain and the period of
oscillation that correspond to the sustained oscillation are the ultimate gain (Kcu) and
the ultimate period (Pu). In performing the experimental test, it is important that the
controller output does not saturate. If saturation does occur, then a sustained
oscillation can result even thought Kc > Kcu. Typical results are shown in Figure 7.
Definition 1. The ultimate gain Kcu is the largest value of the controller gain Kc that
results in closed-loop stability when proportional only controller is used.
Definition 2. The ultimate period Pu is defined as the period of sustained cycling that
would occur if a proportional controller with gain Kcu were used.
3. From the values of Kcu and Pu found in the previous step, use the Ziegler-Nichols
rules given in Table 19.1 to determine controller settings (Kc, I, D). These tuning
relations were empirically developed to provide a decay ratio, Figure 8.
Figure 6 Response of
the loop with the
controller gain set
equal to the ultimate
gain Kcu. Tu is the
ultimate period.
Figure 8 Quarter decay ratio response to disturbance input and to change in set
point.
Remarks
1. Kcu and Pu can be determined by the direct substitution method if the transfer
functions of all of the components of the loop are known qualitatively.
2. The quarter decay ratio response is very desirable for disturbance inputs because it
prevents a large initial deviation from the set point without being too oscillatory.
However, it is not as desirable for step changes in set point, because it causes a 50 %
overshoot. This is because the maximum deviation from the new set point in each
direction is one-half the preceding maximum deviation in the opposite direction,
Figure 8. This difficulty can easily be corrected by reducing the proportional gain
from the value predicted by the formulas of Table 2. In fact, the decay ratio is a direct
function of the controller gain, and can be adjusted at any time by simply changing
the gain. In other words, if for a given process the quarter decay ratio response is too
oscillatory, a reduction of the gain will smooth out the response.
3. Shortcomings:
3.1. It may be objectionable because the process is pushed to stability limit.
Consequently, if external disturbances or a change in the process occurs during
tuning, unstable operation or hazardous situation could occur.
3.2. This tuning procedure is not applicable to processes that are open-loop unstable
because such processes are unstable at both high and low values of Kc, but stable
for intermediate range of values.
3.3. Some processes do not have an ultimate gain; for example, first-order and
second-order processes without time delay.
3.4. The set of tuning parameters necessary to obtain the quarter decay ratio response
is not unique, except for the case of P controller.
Figure 9 Block diagram of a control loop for measurement of the process reaction
curve.
1. After the process reaches steady state at the normal level of operation, switch the
controller to manual.
2. With the controller in manual, introduce a small step change in the controller output
that goes to the valve and record the transient, which is the process reaction curve
(Figure 10).
3. Draw a straight line tangent to the curve at the point of inflection, as shown in Figure
10. The intersection of the tangent line with the time axis is the apparent transport lag
(); the apparent first-order time constant () is obtained from:
= Bu S (14)
where Bu is the steady state ultimate value of B and S is the slope of the tangent
line. The steady state gain that relates B to M in Figure 9 is given by:
Figure 10 Typical process reaction curves: (a) non-self regulating process, (b) self-
regulating process.
K p = Bu M (15)
Note that if the process reaction curve has the typical sigmoidal shape shown in Case b
of Figure 10, the following model usually provides a satisfactory fit:
B( s) K p e s
= Gv G p Gm = (16)
M ( s) ps +1
4. Using the values of Kp, p and from step 3, the controller settings are found from the
relations given in Table 3.
Notice that the settings given in this table were developed to provide closed-loop
responses with a decay ratio of .
Remarks
The advantages of PRC method:
Example 7