Proportional, Integral and Derivatives (PID)

Controllers Tuning
MUADZ BIN CHE ABDULLAH (2019451942), MUHAMMAD ZA’IM BIN MD FARIDUL ATHAS
(2019872684)

Abstract—This experiment presents typical techniques permanent state error. The error is distributed over a
used for PID tuning. The techniques are developed for PID time until the error is zero. It holds the value to the final
tuning and optimization techniques applied for tuning control unit with error zero. Comprehensive control
purposes. Based on the process linearity graph, shows reduces the output if negative error occurs. It decreases
slightly nonlinear relation of process response and reaction speed and affects system stability. Reaction
manipulated variable. the higher the value of the
rates are improved by decreased integrated Ki gain.The
proportional gain, Kc, the lower the value of the process
output response to an error produces a marginally higher
dead time. An increases in PB and I values, make the MV
initial deviation from that which can be achieved
action slower. A slower action of MV results in slower
individually by means of a proportional plus integral
process response because PB and I are inversely
proportional to MV. Thus, the new process response curve regulation and used in the output signal (J.Hall, 2017). It
shift to the right. Although the values of Proportional Band is because of a change to the integral control signal.
(PB) and Integral (I) obtained in this rule is not Therefore, the oscillation decreases and no offset can be
generalized to work with all processes, the starting values obtained. This control mode is used commonly because
will work on many processes. the proportional component is suitable in a cycle where
moderate changes are happening whereas the integral
Keywords—PID tuning; process linearity; proportional
gain; disturbances; controller settings. component makes large changes to the load and
eliminates the offset. A derivative (D) reaction
I. INTRODUCTION anticipates a process load shift and sends a correction
The PID is the most common control algorithm used in
signal to reduce the error. Depending on the adjustment
industry and is universally accepted in industrial control.
of the variance from the setpoint, this action corrects.
Part of the success of PID controllers is due to their
Related action results in proportional movement of the
reliable performance in a wide variety of operating
valve more rapidly and more than would usually occur
conditions and partly to their flexibility, which makes it
with just proportional operation. Typically, temperature
fast and convenient to run
control systems require differential operation. Such
P is a short term for proportional but PB is a common methods are subject to lengthy processes and calculation
concept in most industrial controllers with a unit lags. Even derivative action is used for different
percentage. The proportional factor is easier to applications, including anti-surge systems where a quick
understand. The product of the gain and measurement valve reaction is imperative (A.Poe, 2012). Derivative
error is the output of the proportional factor. Therefore, control has the property that the controller output
greater proportional gain or error produces a higher changes are commensurate with the intensity. Change
proportional factor output. Setting the proportional gain from shorter to longer times in order to change the
too high causes a monitor to override the set-point again derivatives manually. The longer the derivative time is
and again, leading to oscillation. When the error is too set, the greater and the oscillatory the corrective action.
small, the loop output becomes negligible. The
downside of a proportional only loop. Therefore, there is
always error even though the proportional loop is
constant (Avery, 2009). As the p-controller is limited
b.

when an offset occurs between the method variable and

the fixed point, the I-controller is required to avoid the
 Process linearities can be determined by:
II. OBJECTIVES
1. To determine the process linearity of the process response.
2. To determine the process control characteristic using tangent
method.
3. To determine the process control characteristics using
tangent method, reformulated tangent method.
4. To calculate the time constant (Tc) based on tangent
method, tangent-and-point method and 2-point method.
5. To calculate the dead time (Td) and response rate (RR)
value using tangent method, reformulated tangent method and
numerical method.
6. To determine the optimum controller setting based on
Ziegler-Nichols / Cohen Coon method using process control
characteristic from previous lab.

III. THEORY

PROCESS LINEARITY
Process linearity is a relationship between elements of a
process and the output. Linearity represents a sensor’s ability In determining the linearity of the process,
to respond to changes in a measured variable over the entire
range in the same way. It is to determine whether the 0% change is linear
relationship between dependent variable and independent 50 ≥ % change > 0 is slightly linear
variable is linear or nonlinear. Figure 1 compares the process 50 < % change is highly nonlinear
behaviour between linear and nonlinear processes.

PROCESS CHARACTERISTICS
Process characteristics determination such as dead time,
time constant and response rate is observed using graphical
and numerical analyses. Graphical analysis contain Tangent
method and Reformulated Tangent Method to identify and
calculate the process dead times and response rates of self
regulating and non-self regulating processes. Figure 3 shows
the analysis of Dead time and Response Rate using Alternative
Mathematical Approach

Comparison between Linear and Nonlinear processes

Manipulated variable changes proportionally with linear

process and process change for nonlinear is not proportional
with MV and hard to predict. Some processes are hard to
regulate because in some typical chemical process application,
there are many linearities are not necessarily working together.
Process linearity can be observed graphically and
mathematically as shown as in below:

Figure of Dead Time and Response Rate using

Alternative Mathematical Approach

Calculate response rate using

 formulating calculation for numerical technique.

Where,

Time constant identification is to measure the time

taken by a process to reach certain quantity when Figure of Tangential Approximation using Three Point Central
a change in manipulated variable is made. Difference Technique
Figure 4 & 5 shows three distinct techniques to
calculate the time constant.

Using Tangent Method, time constant can be

obtained using formula below

The formulation of calculation numerically is based

𝑇𝑐 = 𝑇100% − 𝑇𝑑
on tabulated data.

IV. PROCEDURES

a. DCS: DELTA-V EMERSON

Open Loop Test

𝑇𝑐 = 𝑇63.2% − 𝑇𝑑
1. The control loops is selected:

i. i. FIC21-flow ctrl 100%

Figure Time Constant based on Tangent
ii. ii.FIC21 – flow ctrl 20%
Method(a) and Tangent-and-Point Method(b)
iii. iii.FIC31 - numerical
iv. iv.LIC11 – level ctrl 31.1%

2. The Faceplate is opened by double click at the

Controller.(example using PIC92 air pressure
ctrl)

Figure Time Constant based on 2-Point Method

𝑇𝑐 = 𝑇28.3% − 𝑇𝑑

In numerical technique topic, the dead time, time constant and

response rate are calculated numerically for process response that is
in discrete form. Figure 6 shows the fundamental concept in
 integral term can cause the present
value to overshoot the setpoint value.
6. The optimum PI controller setting is recorded.

Load Disturbance Test

1. The controller is set in manual mode
2. Change is done to MV of about 10% of
current MV value
3. After 3 seconds, the controller is set to
automatic mode.
4. Wait until the response become stable.

Set point test

1. From load disturbance test, set point test is
continued.
2. System must be in automatic mode
3. The trend/graph is viewed by clicking the 3. A change in set point is done about ±10% of
Process History View. current operating process value
4. The process is stabilized either in manual 4. Wait until the response become stable.
(MAN) or Automatic (AUTO) mode. 5. Both response curves is combined and the
5. The initial value of manipulated variable (MV graph is printed.
in %) is recorded.
6. The AUTO mode is switched to MAN mode
once the process is stabilized. b. DCS: FOXBORO
7. Step change is made to the manipulated Open Loop Test.
variable between 5-20%. 1. The Plant is selected -Water Flow and Level
8. The system is set to AUTO mode once, the Control Plant (WLF922)
new steady state is reached: For selfregulating 2. The selected control loops is activated.
system b) the slope can be calculated from i. FIC31
process response: For non self-regulating ii. LIC31
system 3. To view trending
9. The response curve is printed. a. File>Additional FoxView is selected
10. Response rate (RR), time delay (Td) and time b. Change Env is clicked
constant (Tc) is determined using Tangent c. The Environment is changed to
Method or Reformulated Tangent Method. Operator
11. The PI controller setting is determined using d. At new FoxView, Plant WLF922 is
Ziegler Nichol’s or Cohen Coon’s method. selected.
e. Controller is double clicked.
f. TREND is clicked.

Closed Loop Test 4. To enlarge the trending

1. The controller is set in AUTO mode a. Change Env is clicked.
2. The controller setting is set by clicking Detail b. The Environment is changed to
icon at Faceplate. Process_Eng
3. The PI controller setting (Kc and I) value is c. Then, the trend is right clicked and
inserted at Gain and Reset Section. moveable is selected.
4. Fine tuning is done to achieve stable condition d. The trend is maximized.
once the response oscillates.
5. If the response oscillates: 5. To perform open loop test, the process is
a. Kc is reduced - an excessively large stabilized either in manual (MAN) or Automatic
proportional gain will lead to process (AUTO) mode.
instability and oscillation. Or b) I is 6. The initial value of manipulated variable (MV in
increased – an excessively small %) is recorded.
7. Once the process stabilized, AUTO mode is current operating process value
switched to MAN mode. 10. Wait until the response become stable.
(Remember: open loop test must be carried out in 11. Both response curves are combined.
MAN mode) 12. HyperSnap-DX program is used to print the
8. A step change is made between 5 – 20% to the graph:
manipulated variable.
9. The system is set to AUTO mode after, a. START > Program > HyperSnap-
a. For self-regulating system: the new steady DX>HyperSnap- DX
state is reached. b. In HyperSnap-DX Pro, Capture is
a. For non self-regulating system: slope can be selected > Active window and the
calculated from process response. trending is clicked.
c. Once the trending appear in
10. Data is collected from AIM Historian Data HyperSnap –DX program, go to Image
Display at Desktop >Invert Black and White
a. The desired data is tagged d. The inverted graph is printed
b. The desired Date/Time is set
c. View Data is clicked. V. RESULTS AND DISCUSSIONS
1. PROCESS LINEARITY
11. The response rate (RR), time delay (Td) and
time constant (Tc) is determined using Numerical The experiment is based on a flow controller FIC 21 using the
Analysis.(Remember: Value of PV and MV MUST be manual mode of Emerson system and has a constant setpoint of
in %) 2 (m3 /h).
12. The PI controller setting is determined using
Ziegler Nichol’s or Cohen Coon’s method.

Closed Loop Test

7. The controller is set in AUTO mode.
8. The PI controller setting (PB and I) value is set
at PBAND and INT Section.
9. If the response oscillates, fine tuning is done to
achieve stable condition.
10. If the response oscillates:
a. The Kc is reduced- an excessively large
proportional gain will lead to process
instability and oscillation. Or
b. The I is increased– an excessively small
integral term can cause the present value to Process Steady State Values for each Manipulated Variable
overshoot the setpoint value.

11. The optimum PI controller setting is recorded.

Load Disturbance Test

6. The controller is set in manual mode
7. MV is changed of about 10% of current MV
value
8. Wait for 3 s.
9. After 3 s, the controller is set to automatic
mode.
10. Wait until the response become stable.

Set point test

7. From load disturbance test, set point test is
continued.
% Change = 41% -> Slightly Non-Linear
8. System must be in automatic mode
9. A change is made in set point about ±10% of From the graph we can observe that the process over the
manipulated variable is slightly nonlinear as the process response Data taken for objective 2 is from the open loop process
gradually moves itself from trendline. This can also be response conducted on a controller named FIC21 Emerson
determined by calculating the % change which indicates 41% functioned as liquid flow controller implementing the PI mode
which is slightly non-linear.The flow control loop undergoes controller. A change in manipulated variable was made from
several stage tests by adjusting the manipulated variable from 10% to 20%.The process dead time (Td) and the response rate
10% to 100% continuously. Based on the process linearity (RR) were determined from the graph using tangent method. The
method graph shows slightly shifts away from the linear thread process span is 6 m3 /h. The process dead time calculated using
of the response to the process, showing that the response to the data from the graph is 1.3953 seconds which is observed as a
manipulated variable is slightly nonlinear. This shows that, in low value where the response rate is determined to be 0.05288 s-
practice, the process will not achieve 100 % efficiency, resulting 1 . This may be due to a high controller gain value that is
in a slight nonlinear outcome. Although this is the case, the inversely proportional to process dead time. Therefore, the
relationship between the process response, PV and the higher the value of the proportional gain, Kc, the value of
manipulated variable can still be considered to be directly process dead time will be decreased. Due to the fact that tangent
proportional. method is the least accurate method compared to reformulated
tangent and numerical method, several tuning is needed to
achieve more accurate result. (Ishak 2012).
2. TANGENT METHOD
This experiment is based on FIC21
3. REFORMULATED TANGENT METHOD

This experiment is based on LIC11 level controller

An open-loop response when a manipulated variable change was

made from 30% to 40%. From figure below, a tangent method is
used for determining the process dead time (Td), response rate LIC11 (Level Control)
(RR) and time constant (tc). The time frame is 6 m3/h.
Data Determined from Graph
= 16%
= 42%
= 28%
= 48%
ΔMV = 12%
ΔMV = 6%
Scaling factor (a)
a= (100mm/1000mm)(100%/16mm) = 0.625
a = 10% / 16 mm = 0.625 %/mm
b= 60s/43mm = 1.3953s/mm
Scaling factor (b)
RR = tan10/6%)(0.625/1.3953) = 0.013163/s
b = 60s / 43 mm = 1.3953 s/mm

Dy = 17 mm 𝑇 = 12mm * 1.3953 = 16.7436 s

Dx = 2 mm Objective 3 used data from LIC11, an level controller. It is an

open loop process response when a change in manipulated
Dead Time, td variable was made from 20% to 10%. The process dead time
(Td) and response rate (RR) determined from the graph using
𝑇 (length) = 1.0 mm reformulated tangent method are 16.7436 s and 0.013163/s
respectively. This controller implemented PI controller in which
𝑇 (Time) = Td (length) x b = 1.0 mm x (1.3953 s/mm) = the tuning rules will be decided in objective 6. In choosing the
1.3953 s right tuning rule, the controller performances are examined in
terms of time response specifications or set point change
Response Rate, RR disturbances (Ishak, 2012).

RR= 0.05288/s
4. TIME CONSTANT LIC11 (Level Control) – EMERSON

FIC21 (Liquid Flow Control) - EMERSON

b=60s/42mm = 1.4285s/mm

Tangent Method
b=60s/43mm = 1.3953s/mm
𝑇 = 70mm

Tangent Method 𝑇 = 11.5 mm

𝑇 = 11.5mm 𝑇 = (70mm-11.5mm)*(1.4285) =83.56725s

𝑇 = 9 mm
2-Point Method
𝑇 = (11.5-9)*(1.3953) =3.48825s
dy = 10mm

= 2.83mm , T28.3 = 31mm

2-Point Method = 6.32mm , T63.2 = 46mm

dy = 16mm 𝑇 = 1.5(11mm-10mm)(1.4285) = 32.141255s

= 4.528mm , T28.3 = 10mm

Tangent & point
= 10.112mm , T63.2 = 11mm
𝑇 = 46mm

𝑇 = 1.5(11mm-10mm)(1.3053) = 2.09295s 𝑇 = (46mm - 11.5mm)(1.4285) = 49.2825 s

The process time constant (Tc) is determined for how fast the PV
is responding. The distinctive methods that are used for this
process are tangent method, 2-point-method and tangent-and-
Tangent & point point method. The values obtained varies from one another.
Time constant value as calculated based on tangent method,
𝑇 = 11mm tangent and point method and 2-point method for FIC21 are
3.48825s, 2.09295s and 2.7906s respectively. For LIC11 tangent
method, tangent and point method and 2-point method The
𝑇 = (11mm - 9mm)(1.3953) = 2.7906s
values vary because of the different sets of equation provided.
5. NUMERICAL METHOD In Objective 5, time constant describes how fast the PV responds to
a change in CO.in this experiment values Tc is 2.7872 s. However,
the variation of values is expected due to the methods having
different sets of equations (Ishak, 2018). Furthermore, the values of
maximum response rate (RR) and dead time (Td) was obtained at
0.288675 s-1 and 33.8928 s via the numerical method. Dead time can
be defined as the delay from when a controller output (CO) is being
issued until when the process variable (PV) begins to respond. From
the dead time obtained above, it is observed that the value is quite
high but does not exceed the values of time constant obtained which
could be an issue. Since the dead time is the denominator of
controller gain, KC, the higher the dead time gets, the smaller the KC
which implies a less active controller which makes it more cautious
and conservative approach.This might be due to the condition
around the sensor such as high flowrate which could add
troublesome delay for the detection of flow (Control Guru, 2015)

6. OPTIMUM CONTROLLER SETTINGS

this data experiment is based on FIC21 and FIC31

Tables of tuning rules The tuning rules by Ziegler-Nichols is as

stated;

Based on the data given, FIC31 flow cotroller response rate, dead
time and time constant can be determined.

Process Span = 6 m3/h

The tuning rules by Coheen-Coon is as stated;
dMV=10%

PV(%) and RR will be calculated between time 9:01:21 until

9:01:31 because flowrate difference happen in that time interval.

Flow rate
time (s) (m3/hr) PV(%) RR

09:01:21 0.515098 8.585 0

09:01:22 0.515098 8.585 0

09:01:23 0.515098 8.585 0.015475

09:01:24 0.533672 8.8945 0.202145

(max)09:01:25 0.757676 12.6279 0.288675 I. FIC21

09:01:26 0.880078 14.668 0.153615
RR= 0.0134
09:01:27 0.942012 15.7002 0.0765625 𝑇 = 4.400157 (from lab 4)
𝑇 =15.3488 s
09:01:28 0.971953 16.1992 0.03774417

09:01:29 0.987302 16.4551 0.0215825

Ziegler-Nichols (Flow controller) use PI

09:01:30 0.997852 16.6309 -0.413771 PB(%)= 111.1(0.0134)(15.3488)

= 22.8504%
09:01:31 0.997852 16.6309 0
I (time) = 3.33(15.3488)
=51.1115s
PV(%) and RR calculated using formula

𝑇 = 36-2(1) (14.668-8.585)/(14.668-8.8945) = 33.8928 s Cohen-coon (Flow controller) use PI

𝑇 = 2(1) (16.6309-8.585)/(14.668-8.8945) = 2.7872 s

 PB%= ( )( ) variable changed from 10% to 100%. Thus, 100% efficiency would
( ) not be achieved by the process, which resulted to a slight nonlinear
= 12.23331% outcome. However, the PV and the MV can still be considered to be
directly proportional. In Objective 2, the process dead time
calculated using data from the graph is 1.3953 s which is perceived
I (time) = 3.33 ( )( 𝑇 )
to be a low value. On the other hand, the RR value was obtained at
=4.7618 s 0.05288/s.The reason for the low value is to a high controller gain
value that is inversely proportional to process dead time. Thus, the
higher the value of the proportional gain, Kc, the lower the value of
the process dead time. Due to being the least accurate method
II. FIC31 compared to reformulated tangent and numerical method, the values
obtained from the tangent method requires tuning in order to
RR= 0.288675 achieve a more accurate result than that (Ishak, 2018).The process
dead time (Td) and response rate (RR) values in Objective 3 are
= 2.7872
determined from the graph via reformulated tangent method are
=33.8928 16.7436 s and 0.013163/s respectively when a change in MV is
µ= 12.1601 performed from 30% to 40%. In order to get an optimal values of
controller settings, a suitable tuning rule in Objective 6 is applied
where the controller performances are examined in terms of time
Ziegler-Nichols (Flow controller) use PI response specifications or set point change disturbances. Other than
that, time constant values in Objective 6 as calculated based on
PB(%)= 111.1(0.288675)( 33.8928) tangent method, tangent and point method and 2-point method are
= 1087.0028% 3.48825 s, 2.7906 s and 2.9295 s respectively. The values vary
because of the different sets of equation provided (Ishak, 2018).
I (time) = 3.33(33.8928) From Objective 5, it is calculated via numerical method that the
= 112.863024 s value of time constant is 2.7872 and MAX RR is at 𝑇 . That being
said, the time constant values that was obtained varies from one
another while maximum response rate (RR) and dead time (Td) of
0.288675 s-1 and 33.8928 s. Even though the value of dead time does
Cohen-coon (Flow controller) use PI not exceed the values of time constant, the higher it gets, the smaller
the KC which implies a less active controller which makes it more
PB(%) = ( )( ) cautious and conservative (Control Guru, 2015). Last but not least,
( ) all of the process in Objective 6 uses mode controller which is PI
= 464.6959% mode controller where Ziegler-Nichols and Cohen-coon tuning rule
is utilized. The values of Proportional Band (PB) and Integral (I) for
(FIC21) obtained are 22.8504 % and 51.1115 s using Z-N ,then by
using C-C the PB and I obtained are 12.23331% and 4.7618 s. in
I (time) = 3.33 ( )( 𝑇 ) Objective 6 for FIC31 by using Z-N the values of Proportional Band
= 8.5625 s (PB) and Integral (I) obtained are 272.1782 % and 1.5895 s in
Objective 3. On the other hand, the values of Proportional Band
(PB) and Integral (I) obtained are 1087.0028 % and 112.863024 s.
Then by using C-C tuning rule, values of Proportional Band (PB)
For Objective 6, the tuning rule that was applied is Ziegler-Nichols
and Integral (I) obtained is 464.6959% and 8.5625 s in Objective 6.
and Cohen-Coon tuning rule. The controller involved is flow
However, although the optimal values of controller settings are
(FIC21) where PI mode controller would be implemented (Ishak,
obtained, this rule is not generalized to work with all processes
2018). That being said, the values of Proportional Band (PB) and although the starting values will work on many processes (Gerry,
Integral (I) obtained for FIC21 using Ziegler-Nichols is 22.8504% 2003).
and 51.1115s . then by using Cohen Coon tuning rule PB and I
obtained is 12.23331% and 4.7618 s . Not only that, the another
controller involved is (FIC31) is flow controller where PI mode
controller would be implemented (Ishak, 2018) in which the
values of Proportional Band (PB) and Integral (I) obtained using VII.REcOMMENDATIONS
Ziegler-Nichols are 1087.0028%and 112.863024 s. Other than In order to achieve more accurate and better results, few
recommendations can be made throughout the experiment.
that, by using cohen-con tuning rule, where the values of
Specifically for tangent and reformulated tangent method, the
Proportional Band (PB) and Integral (I) obtained are tangent line should be drawn straight fitting well with the curve line
464.6959%and 8.5625 s. However, despite the optimal values to prevent error in calculation later. Next, the decimal places taken
obtained, this rule is not generalized to work with all processes from the calculation is important to the numerical method and it
although the starting values will work on many processes (Gerry, must be constant throughout the experiment. Besides that, the value
2003). Thus, it is necessary to test the system offline and tool it for of RR need to be calculate for all data to determine the most
special tests of response. This is due to the fact that the rule does not accurate RR maximum value. Other than that, to determine the
give a great command tracking performance which could lead to process linearity, the step size for the manipulated variable chosen
inconsistent design setpoint and overly aggressive gains. (Microstar must be in range from 5 to 20%. As for the optimum controller
Laboratories, n.d.) setting, the value of PB obtained in this experiment was quite high.
It might be necessary to change the tuning rules to require lower PB
VI.CONCLUSION value.

Based on the process linearity method graph in Objective 1, it

indicates that the graph slightly shifts away from the linear thread of
the response to the process, showing that the response to the
manipulated variable is slightly nonlinear when the manipulated REFERENCES
[1] Ishak, A. A. (2018). PID Tuning: Self Study Tutorials & Exercises. Shah
Alam, Selangor: UiTM Print Faculty of Art Design.
[2] Control Guru. (2 April, 2015). Dead Time Is The "How Much Delay"
Variable. Retrieved from Control Guru:Practical Process Control:
https://controlguru.com/dead-timeis-the-howmuch-
delayvariable/#:~:text=Dead%20time%20is%20the%20delay,steering%2
0wheel%20and%20the%20tires.
[3] Gerry, J. (7 February , 2003). ZieglerNichols Tuning Rules And
Limitations. Retrieved from Automation.com:
https://www.automation.com/enus/articles/2003-1/ziegler-nicholstuning-
rules-and-limitations
[4] Microstar Laboratories. (n.d.). Microstar Laboratories Inc. Retrieved from
Ziegler-Nichols Tuning Rules for PID:
[5] ] A.Poe, A. l. (2012). Process Control Fundamentals. In A. l.
A.Poe,Handbook of Natural Gas Transmission and Processing (Second
Edition).
[6] Avery, P. (2009, January 01). Introduction to PID control. Retrieved from
Machine Design
[7] J.Hall, A. l. (2017). Instrumentation and control..
[8] PID Theory Explained. (2020, March
APPENDIX