1

Digital Control
Summary # 8: Proportional, Integral and Derivative: Tuning methods

pre–act control and anticipatory control. The PD continuous

time transfer function is
Kp (1 + τD s) (4)

Proportional, integral and derivative.

PI and PID control have been dominant control algorithms
for process control for decades (about 95% of process
controllers utilize some form of PID). The continuous time
transfer function for PID controller is given by
1
Kp (1 + + τD s) (5)
τi s
There exist several variants of the PID control.

Position and velocity algorithms for digital controllers

Two alternatives for digital PID control exist, namely
position form and velocity form

Fig. 1. Ideal and actual proportional control

Position algorithm
The discrete time equation for the digital PID controller
output is given by
Proportional control  
T ∑k
τ
u(k) = u0 +Kp e(k) + (e(k) − e(k − 1))
D
The error between the desired set point and the actual e(j) +
τI j=1 T
output is defined as
(6)
e(t) = r(t) − y(t) (1) where T is the sampling time. This form is called position
where r(t) is the desired set point. The set point can be algorithm
constant or time varying. The controller output is given by
Velocity algorithm
u(t) = u0 + Kp e(t) (2)
In the position PID control, we calculated the actual value
where u0 is the a bias and Kp is the controller gain. For dig- of the PID controller output. In the velocity form we calculate
ital controllers, u is usually expressed as a percentage. The the change in the controller output. The velocity form can
actual behavior of the proportional controller is illustrated in be derived from the position form as follows:
figure 1. The saturation shown in the figure is an important
limitation that needs to be taken into account.  
T ∑
k−1
u(k − 1) = u0 + Kp e(k − 1) + e(j) (7)
τI j=1
PI control
[τ ]
D
The proportional integral control is widely used because +Kp (e(k − 1) − e(k − 2)) (8)
T
of its important practical advantage: it eliminates the steady
The velocity form is based on the difference given by
state error. The continuous time transfer function of a PI
controller is ∆u = u(k) − u(k − 1) (9)
1
Kp (1 + ) (3)
τi s and therefore

[ ]
PD control T
∆u = Kp e(k) − e(k − 1) + e(k) (10)
τI
PD control has two terms, a proportional term and a [τ ]
D
derivative term. This controller is also called rate action, + (e(k) − 2e(k − 1) + e(k − 2)) (11)
T
Digital Controls, spring 2015 Proportional, Integral and Derivative: Tuning methods

• Robust, the closed loop system is insensitive to

Step Response
changes in the plant conditions and molding inaccu-
racies.
1
• Good disturbance rejection
Amplitude

− −
It is not possible to achieve all these goals simultaneously
+
0.5 + because of the conflicts and trade offs that may exist.

0 PID control design methods

0 5 10 15 20 25
Time (seconds) Several techniques are used
Step Response • Model based design methods
1 • Controller tuning relationships
• Frequency response techniques
Amplitude

• Computer simulations
0.5 + • Online tuning
Methods 1-4 are based on an approximation of the plant
model and method 5 (online tuning) is purely experimental.
0
0 1 2 3 4 5 6 7 8 Methods based on the plant model allow for initial setting
Time (seconds)
of the controller. Computer simulation (Matlab and Labview)
allow for comparison between the different alternatives.
Fig. 2. Illustration of integral wind up. Top: sign of the error alternates,
positive and negative values cancel out. Bottom: error sign is always
positive, saturation is more likely to happen
Model based design methods
Two methods are discussed here: the direct synthesis
method and the internal model control.
It is possible to obtain an explicit formula for the controller
as follows Direct synthesis method
[ ] The controller is designed based on the desired closed
T
u(k) = u(k − 1) + Kp e(k) − e(k − 1) + e(k) (12) loop transfer function. The method can be used to design
τI
[τ ] but not limited to PID controllers. The closed loop transfer
D
+Kp (e(k) − 2e(k − 1) + e(k − 2)) (13) function is given by
T
Y (s) Gc (s)G(s)
= (14)
R(s) 1 + Gc (s)G(s)
Problems with PID controllers: Derivative kick and integral
windup where
• Gc (s) is the controller transfer function
A sudden change in the error that usually results from • G(s) is the open loop transfer function.
the change in the set point will cause the derivative part to
Rearranging the terms and solving for the controller transfer
become very large. This spike is undesirable. One way to
function yields the following equation
solve this issue is by using −dy de
dt instead of dt . Integral wind
up (or reset wind up) is another problem that happens when Y (s)
1 R(s)
the integral output becomes large and the controller output Gc (s) = (15)
becomes saturated. The build up of the integral terms is G(s) 1 − Y (s)
R(s)
called integral windup. Consider the time response of figure In order to solve for the controller’s transfer function we need
2. In figure 2–top, the integral term initially increases but to know:
begins decreasing again when the error changes sign. The
• the desired closed loop transfer function
positive and negative terms cancel out and the controller
• the open loop transfer function.
moves away from the saturation point. In figure 2–bottom,
the integral term keeps increasing and this may result in in We can define a desired closed loop transfer function based
integral wind up. The PID velocity form is called anti-integral on the desired characteristics of the system. We call this
wind up because the summation is eliminated. function ( )
Y (s)
(16)
R(s) d
Performance criteria for closed loop systems Assuming we can determine an approximation G̃ of the open
The performance criteria are as follows loop transfer function, the controller equation becomes
( )
• The closed loop system is stable Y (s)
1 R(s)
• Steady state error is eliminated Gc (s) = ( d)
(17)
• Good transient response G̃(s) 1 − Y (s)
R(s)
d

2
Digital Controls, spring 2015 Proportional, Integral and Derivative: Tuning methods

Several possibilities exist for the desired closed loop system,

Step Response
Ideally we want the closed loop response to be equal to the
input, that is 2
( ) open loop
Y (s)
=1 (18)
R(s) d
1.5
This is too perfect and is not possible in practice. A more
realistic approach is to use a first order system.
τc = 1

Amplitude
First order system
1
In this case the desired closed loop system is a simple τc = 10
first order system
( ) 0.5
Y (s) 1
= (19)
R(s) d τc s + 1 τc = 3
where τc is the closed loop time constant. The controller
equation is reduced to the following equation 0
0 10 20 30 40 50 60 70
Time (seconds)
1 1
Gc = (20)
G̃(s) τc s Fig. 3. Closed loop response for different values of the time constant
If the process has a delay, we can use the closed loop
transfer function with delay
we get
( ) −θs
Y (s) e τ
= (21) Kp = (27)
R(s) d τc s + 1 K(τc + θ)
where θ is a time delay. The equation for the controller is τI = τ (28)

1 e−θs
Gc = (22) C. Second order approximation
G̃(s) τc s + 1 + e−θs
The transfer function is approximated by a second order
Assuming a small delay, the delay can be approximated as
transfer function as follows
e−θs = 1 − θs (23)
Ke−θs
and the controller as
G̃(s) = (29)
(τ1 s + 1)(τ2 s + 1)
1 e−θs By substituting in the controller equation we get
Gc = (24)
G̃(s) (τc + θ)s τ1 + τ2
Kp = (30)
Clearly, equations (20) and (24) depend on the approxima- K(τc + θ)
tion of the open loop system. This discussed below. τI = τ1 + τ2 (31)
τ1 τ2
τD = (32)
A. Open loop transfer function approximation τ1 + τ2
Several methods exist to approximate the open loop where
1
transfer function such as the step response and the fre- Gc = Kp (1 + + τD s) (33)
τI s
quency response. Here we consider first and second order
approximations with delay.
Example
Consider the following approximation of the open loop
B. First order approximation
transfer function
The open loop system is approximated by a first order 2e−1s
transfer function as follows G̃(s) = (34)
(10s + 1)(5s + 1)
Ke−θs Design a PID controller when τc = 1s, 3s, 10s and simulate
G̃(s) = (25)
(τ s + 1) the system behavior.
where τ is the open loop time constant. Using equation (24) • τc = 1s ⇒ Kp = 3.7500, τI = 15; τD = 3.3333
and knowing that the controller’s equation is • τc = 3s ⇒ Kp = 1.8750, τI = 15; τD = 3.3333
• τc = 10s ⇒ Kp = 0.68, τI = 15; τD = 3.3333
1
Gc = Kp (1 + ) (26) A simulation is shown in shown in figure 3
τI s

3
Digital Controls, spring 2015 Proportional, Integral and Derivative: Tuning methods

Fig. 4. Standard and IMC models Fig. 5. Illustration of the continuous cycling method

Internal Model Control We have

K
G̃− (s) = (39)
The IMC method uses a simplified model of the control (0.5s + 1)(3s + 1)
loop where another controller called IMC controller is intro- for r = 1, we have
duced. The standard control loop and the IMC model are
(0.5s + 1)(3s + 1)
shown in figure 4. The IMC controller is denoted by G∗c . The G∗c (s) = (40)
first step is to establish the relationship between the two K(τc s + 1)
controllers. Assuming the block diagrams are equivalent, it After rearranging the terms, we get
is possible to write 13
Kp = (41)
G∗c K(4τc + 1)
Gc = (35)
e
1 − G∗c G τI = 3.5 (42)
There is one to one relationship between Gc and G∗c .
This τD = 3/13 (43)
implies that for each IMC controller thre is a corresponding
It is important to note that the choice of τc plays an important
standard controller. The IMC method first finds an expres-
role. High values of τc result in a more conservative con-
sion for G∗c and then derives an equation for Gc . This is
troller and low values result in more aggressive controller.
done in two steps:
• Step 1: The open loop model is written as Tuning relationship
G̃ = (G̃+ )(G̃− ) (36) These are tables that give the relationship between the
+ model parameters and the PID parameters (handouts given
where G̃ contains any time delays and right half plane
in class).
zeros.
• Step 2: The IMC controller is derived using the following
equation Experimental online tuning
1
G̃∗c = (37) Experimental online tuning is also called field tuning. It
−
(G̃ )(τc s + 1)r is based on testing, trial and error. Having a good initial
where τc is the closed loop time constant and r is a model can be very helpful in experimental tuning because
positive integer that characterizes the system order it makes the tuning process more straightforward and less
time consuming. Here we focus our discussions on the
continuous cycling and the step reactive curve methods.
Example
Continuous cycling
Use IMC to design a PID controller for the following
system Introduced by Ziegler and Nichols in 1942, it is based
−1s on trial and error. After the system has reached its steady
Ke (−1 + s)
G̃(s) = (38) state, the integral and proportional actions are eliminated.
(0.5s + 1)(3s + 1)

4
Digital Controls, spring 2015 Proportional, Integral and Derivative: Tuning methods

Step Response

System: T3 System: T3
2.5 Time (seconds): 2.07 Time (seconds): 5.75
Amplitude: 2.72 Amplitude: 2.71

2
Amplitude

1.5

0.5

0
0 2 4 6 8 10
Time (seconds)

Fig. 6. Sustained oscilations for teh continuous cycling method Fig. 7. Top: Astrom and Hagglund tuning method and bottom: step test
method

The gain Kp is set to a small value and then increased

slowly until sustained oscillations with constant amplitude
occur. The numerical value of the gain for which sustained
oscillations with constant amplitude occur is called the 1
ultimate gain (Kpu ), its corresponding period is called the 0.5
ultimate period (Pu ). Once Kpu and Pu are determined,
the table below is used to calculate the PID parameters. 0 2d
Illustration of the continuous cycling method is shown in −0.5
figure 5.
−1
  0 2 4 6 8 10 12
Ziegler − −N ichols Kp τI τD
 P 0.5Kpu −−− −−−  1
 
 PI 0.45Kpu 0.83Pu −−−  a
0.5
P ID 0.6Kpu 0.5Pu 0.125Pu
(44) 0

−0.5
Example
−1
Design a PID control for the system whose response 0 2 4 6 8 10 12
is shown in figure 6. The ultimate gain for the sustained
oscillations is Kp = 6.
It is clear from the oscillations that Pu = 3.68s. Therefore Fig. 8. Top: on–off controller output signal and bottom: sustained oscilla-
tions response

Kp = 0.6Kpu = 3.6 (45)

τI = Pu /2 = 1.8 (46) the closed loop system exhibits sustained oscillations. The
τD = Pu /8 = 0.45 (47) principle of the method is illustrated in figure 7. Astrom and
Hagglund derived an approximation of the ultimate gain as
The main drawback of this method is that it pushes the
follows
system to the limit of stability, which can result in hazardous 4d
situations. Kpu = (48)
πa
where a, d are shown in figure 8. Once Kpu , Pu are deter-
Astrom and Hagglund tuning method mined, the Ziegler–Nichols table (44) can be used to cal-
This method is an effective alternative to the continuous culate the controller parameters. One important advantage
cycling method. It uses an on-off controller with dead– of the Astrom and Hagglund method is that it can be easily
zone. After the on-off controller is connected to the plant, automated.

5
Digital Controls, spring 2015 Proportional, Integral and Derivative: Tuning methods

Step Response
Fig. 9. Experiment setting and step response reaction curve
4
Step test method
This is another method proposed by Ziegler and Nichols in 3
1942. This method is also called the plant reaction curve or
Ziegler and Nichols open loop method. After the system has
2
Amplitude

reached its steady state, a small step change is introduced,

the plant reaction curve is obtained as shown in figure 9-
bottom. The controller parameters are derived based on
the reaction curve parameters using table 51. The open
1
loop transfer function whose response is shown in figure
9-bottom is
Ke−θs 0
G̃ = (49)
τzn s + 1
where K, τzn , θ can be obtained form the reaction curve. −1
Let us define
τzn
0 5 10 15 20
Time (seconds)
K0 = (50)
Kθ
The PID parameters can be determined from the table below Fig. 10. Step test reaction curve
 
Ziegler − −N ichols Kp τI τD
 P K0 −−− − − −
  (51)
 PI 0.9K0 3.3θ − − −
P ID 1.2K0 2θ 0.5θ

Example
Consider the reaction curve shown in figure 10. The goal
is derive the PID control parameters. From the graph we
have: K = 3, θ = 2, τzn = 1, from which we get
Kp = 0.2 (52)
τI = 4 (53)
τD = 1 (54)