CONTROLLER TUNING

CONTROLLER
CONTROL LOOP OPTIMIZATION

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 OUTLINE

• Control loop dynamics

• B i principle
Basic i i l off controller
t ll tuning
t i

• Controller parameter optimization

• The problems with control loop performance

• Tips for controller optimal tuning

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 LEARNING OBJECTIVES

When you have completed this unit

unit, you should:

• Have developed insight into the fundamental concept

of tuning feedback controllers;
• Be able to calculate the tuning parameters for a
feedback
f db k controller
t ll using
i ZiZiegler
Ziegler-
l -Nichols
Ni h l ultimate
lti t
period;;
period
• B
Be able
bl tto calculate
l l t ththe ttuning
i parameters
t ffor a
feedback controller using the process reaction curve
method.
• Understand the importance of controller tuning
tuning..

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 WHAT MEANS GOOD TUNED CONTROLLER?
CONTROLLER?

• A feedback control system

y is of little value if it is improperly
p p y tuned;;

• It is important for chemical engineers to understand how the

controller in a feedback control system
y should be tuned.

• To tune a controller you need to determine the optimum values of

the controller gain (KC), integral time (τi) i and derivative time (τd);

• The first problem you encounter when tuning controllers is

determining what is good control – it does differ from one process
to the other process;

• Good control usually means fast reaction and fast response on

disturbances and set point (SP) changes, as small as possible
overshoot, stability and robustness (flexibility on changing in
overshoot
process condition).

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 COMMON PROBLEMS IN CONTROL LOOPS

Continuous PV oscillation around SP

PV deviation from the SP

Sluggish PV response after a SP change

Overshoot and long-

long-term PV oscillation after a SP change

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 SPEED VERSUS STABILITY
• The fundamental trade-off in control loop tuning is between a
control loop’s
loop s speed and stability
stability;
• If you increase increase speed of response to a SP change or
disturbance, the loop becomes less stable.
• If you increased
i d th
the speedd more, ththe control
t l lloop will
ill eventually
t ll
become completely unstable;
• If a control loop is tuned for stability its speed of response suffers.

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 ROBUSTNESS

• Process characteristics can be influenced by production rate,

valve characteristic, eqipment in service, feedstock
composition, operating conditions, etc.
• Control loop robustness is defined as the amount of change in
process characteristics the loop can tolerate before it become
unstable;
• Instabilty can be caused with increased process gain,
increased dead time and decreased time constant;
• If control
t l loop
l is
i ttuned
d ffor stability,
stability
t bilit its
it robustness
b t will
ill b
be high,
hi h
and it can tolerate substantial changes in process
characteristics;
• If a control loop is tuned for speed
speed, it will have low robustness
and can easily become unstable.

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 CONTROL LOOP SPEED OF THE RESPONSE
• Due to the dynamic characteristics of the process and
control loop actual deviation from SP can not be removed
instantly;
• The control loop can always be adjusted for a slower
response but can not always be adjusted for faster response
response,
because it will become unstable;

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 CONTROLLER TUNING PROCEDURE
• In controller tuning the term PROCESS includes all elements between
the output of the controller (OP) and the input to the controller (PV);

• To optimize
T ti i the
th controller
t ll it iis necessary tto d
determine
t i ththe
following process parameters:
process gain
p g ((K),
),
dead time (Θ),
time constant (τ) ili integrating rate (ri)
© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
CONTROLLER TUNING – PARAMETER OPTIMIZATION
DV
CV

PROCESS
PROCES S
MV
Time constant - τ
Process (steady state) gain-
gain- k
Dead time - θ

Final control Transducer

element

Controller gain-
gain- kc
PV
Integral time - τi
D i ti time
Derivative ti - τd
OP e
ALGORITHM

CONTROLLER SP
© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 CONTROL LOOP TUNING METHODS

1 Trial
1. Trial--and
and--error tuning

- an iterative tuning method based on making small changes to

controller
t ll setting
tti and d monitoring
it i th their
i effect
ff t on th
the lloop
2. Standard (traditional) tuning methods

- Ziegler
Ziegler--Nichols method appeared in 1942;
- Cohen
Cohen--Coon method published roughly decade after;
- Most other methods are modifications of the above methods;
3. Automatic tuning (self
self--tuning
tuning))

- Simple methods embedded in controllers (autotuning

(autotuning function);
function);
- Dedicated software package for loop analysis and optimization.

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 CONTROL LOOP TUNING PROBLEMS

• Some loops cannot be tuned by any approach;;

• Common problems that render a loop untunable are as follows::

1 Deficiencies in the process and instrumentation (P&I) diagram

1.

2. Process nonlinearites

3. Problems with the final control element (especially valves).

4 Measurement noise
4.

5. Large dead times

6. Control loop interaction

7 Improper
7. I nesting
ti off cascade
d loops
l

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 1 – TRIAL AND ERROR METHOD

• An iterative tuning method based on making small changing

to the controller setings and monitoring effect on the loop;
• It is the most p
popular,
p , but unfortunately
y it is the least effective
of all tuning method;
• It is suitable for fast and simple standard control loops with
the following g characteristics:
dead time + time constant < 5 to10 min

• It is not suitable for slow (sluggish) and the complex control

schemes:
dead time + time constant > 30 min

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 1 - TRIAL AND ERROR METHOD

Control loop FC PC LC TC AC
Controller type PI PI P, PI PI, PID PID
P [%/%] 0,1-3 1-20 0,2-10 0,5-10 0,1-5
I [min] 0,1-6 0,5-20 15-1000 5-100 10-200
0-2 0,5-4
D [min] - 0.5-3 0,5-4
rarely rarely
Settling time [min] 0,5-2 0,5-6 --- 15-90 60-720
Noise offtten rarely somettiimes rarely occasionalllly
Nonlinearity quadratic linear linear exists exists

• The method works on the principle of making a change to one

of the controller parameter (e.g. control gain) and seeing what
effect it has on the loop’s performance;
• If the change made the loop perform better,
better keep on chaging in
the same direction. If it made the loop perform worse, change
it in the opposite direction.
• After each
Aft h change,
h monitor
it the
th effect
ff t off the
th change
h on the
th
loop’s performance.
© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 1 – TRIAL AND ERROR METHOD

• Once you have reached an optimum value for the first

parameter
t (controller
( t ll gain) i ) begin
b i the
th same process with
ith the
th
next controller setting (integral time);

• Once you have the optimum for the second parameter,

parameter
revisit the first parameter because its optimum value might
be different now that the second setting has been adjusted;

• If you are adventorous and have enough time, you can also
change the third controller parameter (derivative time) and
interatively go back to the other two until you finally have an
optimally
ti ll tuned
t d controller;
t ll

• Table on th next slide represents guidance on trial-and-error

tuning.
tuning

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 1 – TRIAL AND ERROR METHOD

OBSERVATION
Loop seems sluggish Loop oscillates or
overshoots set point
p

Corrective action
Control g
gain Increase Decrease
Proportional band Decrease Increase
Integral time
Decrease Increase
(min or sec)
Integral gain
Increase Decrease
(repeat / min or sec)
Derivative time or Try a small decrease.
derivative gain Try a small increase If condition
If condition worsens,
worsens, decrease.
(min or sec) increase.

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 TRIAL AND ERROR – P CONTROLLER

The impact of controller gain changes on P-only controller

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 TRIAL AND ERROR METHOD – PI CONTROLLER

The impact of integral time changes The impact of controller gain

on PI controller on PI controller

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 TRIAL AND ERROR METHOD – PID CONTROLLER

Vrijeme (min)

The impact of derivative time changes on PID controller

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 2 – STANDARD (TRADITIONAL) TUNING METHODS

• Open
O
Open-
-loop
l methods
th d

Ziegler--Nichols reaction curve method

Ziegler
Cohen--Coon method
Cohen
Lambda test

• Closed
Closed--loop methods

Ziegler--Nichols continous oscillation method

Ziegler
Reley method

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 2 – CONTROLLER TUNING PROCEDURE

p 1 – Process testing
Step g
• It is necessary to cause OP step change (step test
test)).

Step 2 – Process dynamics analysis

a. Standard methods are based on response graphical analysis for
determination of controller parameters;;
b. Automatic tuning (self
self--tuning
tuning) using test which introduced smaller
disturbance in control loop and process;
c
c. Computional(numeric
Computional numeric) methods in dedicated software for controller
parameter determination based on control loop history database.

Step 3 – Controller parameter calculation

• For particular controller (P, PI, PID, etc.) empirical relationship is
applied to determine optimal controller setting;
• By using specialized software the controller parameters are
optimized applying numerical optimization methods.
© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 ZIEGLER--NICHOLS CLOSED LOOP METHOD
ZIEGLER

• J.G. Ziegler and N.B. Nichols published in 1942 two empirical

methods for PID controller tuning;

• One method is called process reaction-

reaction-curve method
method, and
another method off critical oscilation is called ultimate
ultimate--cycling
method;
method

• Th put the
They h controll lloop iin sustained
i d cycling;
li

• Based on the period and amplitude of oscillation they

determined controller parameters;

• For modern heat-integrated processes this method is not

practial.
practial

For many
y loops,
p , the severity
y of such a test is unacceptable.
p

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 Z–N CLOSED LOOP METHOD TUNING PROCEDURE

TUNING PROCEDURE:

1. Put the controller in automatic mode. Remove integral and

derivative modes;

2. Increase the controller gain until the loop cycles continuously

( PV and OP oscillations are visible);

3. Note the period of cycle (time) PU (ultimative period)

and ultimate gain KCU which causes this oscillation;

4. Substitute the values of the ultimate gain and the ultimate

period into the tuning equations (table on the following slide!)

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 Z–N METHOD: Continous oscillation test

Temperature
(°C)
Vallve opening
(%)

Time(min)

P controller PI controller PID serial PID parallel

Kc = 0.5*Kcu Kc = 0.45*Kcu Kc = 0.6*Kcu Kc = 0.75*Kcu
τi = Pu / 1.2
12 τi = 0.
0 55*P
Pu τi = 0.625
0 625*P
Pu
τd = Pu / 8 τd = Pu / 10
© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 OPEN LOOP METHOD

• Execute the process test with the controller on manual mode

(open loop);

• Analyse PV-a response on the OP step change;

• OP change must be as small as possible so as not to

disturb the process too much, but larger than measurement
noise and valve hysteresis;

• The problem can be appereance of disturbances or process

load change during the open loop test;

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 ZIEGLER–
ZIEGLER–NICHOLS OPEN LOOP METHOD

TUNING PROCEDURE:
PROCEDURE:

1. Put the controller in manual mode and wait for PV to

stabilize;;
stabilize

2. Make previously defined OP step change;

change;

3. Graphically determine the slope of the process reaction

curve at the inflection point (maximum slope), and related
dead time;
time;

4
4. Based on empirical relationship ( table on the following slide!)
slide!)
calculate controller parameters.
parameters.

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 ZIEGLER–
ZIEGLER–NICHOLS OPEN LOOP METHOD

Tangent
perature
((°C)
Temp

Steady-state
ve opening
(%)
Valv

Time (min)

© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb
 Z–N OPEN LOOP METHOD

Reaction curve characteristics:

LR – Reaction lag [min]

– Time elapsed from step change to the point where tangent

intersect the baseline

RR – Reaction rate (reaction rate

rate),
), [% / min]
– slope of the tangent

• The values are then substituted into the equations in the

following table:

P controller PI controller PID serial PID parallel

KC = ∆OP / (LR*R
RR) Kc = 0.9
0 9*∆OP
∆OP / (LR*R
RR) Kc = 1.2
1 2*∆OP
∆OP / (LR*R
RR) KC = 11.5
5*∆OP
∆OP / (LR*R
RR)
τi = 3.33*LR τi = 2.0*LR τi = 2.5*LR
τd = 0.5*LR τd = 0.4*LR
© Laboratory for Automatics and Measurement, Faculty of Chemical Engineering and Technology University of Zagreb