Vol.

8, 2023-05

PID Controller Stability and Tuning using a Time-Domain Approach

Hugo Hernandez① and Silvia Ochoa②

① ForsChem Research, 050030 Medellin, Colombia

[email protected]

② Universidad de Antioquia, 050010 Medellín, Colombia

[email protected]

doi: 10.13140/RG.2.2.26330.62400

Abstract

Automatic control is an important tool in process systems engineering for achieving high
productivity and quality with reduced variability and improved safety during operation.
Nowadays, the most common type of automatic controller used in industry is the PID
controller. The PID controller is a general type of linear controller incorporating three effects:
Proportional to feedback error, to the derivative of error and to the integral of error over time.
While the PID is highly versatile and can provide a good balance between performance and
implementation costs in most processes (including nonlinear processes), the main
disadvantage of this controller is that an adequate set of tuning parameters is required for
achieving such performance. PID tuning is best done if an approximate linear model of the
process is known. Reaction curve methods can be used for identifying the parameters of the
linear model. Almost all PID tuning methods have been obtained considering the model in the
Laplace-transform domain. The purpose of this report is illustrating the fact that PID
parameters ensuring stability can be obtained from a linear model solely expressed in the time
domain. A set of tuning equations for PID and PI controllers for different types of systems (self-
regulating, runaway and capacitive) is presented. The performance of this tuning method is
illustrated considering different examples of control in chemical processes.

Keywords

Capacitive Systems, Dynamic Systems, Eigenvalues, Feedback Control, Linear Control, PID
Controller, Process Control, Runaway Systems, Self-regulating Systems, Stability, Time Domain,
Tuning

Cite as: Hernandez, H. & Ochoa, S. (2023). PID Controller Stability and Tuning using a Time-Domain
Approach. ForsChem Research Reports, 8, 2023-05, 1 - 32. Publication Date: 13/03/2023.
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

1. Introduction

1.1. General Description of Linear Dynamic Systems

Let us consider a linear dynamic system described by the following SISO (single input/single
output) continuous time-invariant state-space representation without a direct feedthrough or
feedforward component:

( )
( ( ) ) ( ( ) )
(1.1)
( ) ( )
(1.2)

where represents the state variable of the system to be controlled, is a manipulated

variable (input variable influencing the state variable), is a measured output variable directly
reflecting the behavior of the state variable, is the reference or initial value of the
manipulated variable, and , , , and are characteristic constants of the system. A
negative value of indicates that the system is self-regulating; a positive value of represents
an open-loop unstable or runaway system; and a zero value of corresponds to a capacitive or
integrating system [1].

In many situations, a nonlinear system can be approximated by the linear system presented in
Eq. (1.1) and (1.2) for a relatively narrow range of operation. However, this approximation is
usually good enough for control purposes.

( )
The linear system is said to be at equilibrium at time if . This occurs when:

( ) ( ( ) )
(1.3)

For ( ) , the equilibrium state is simply ( ) , which can also be expressed

as follows:

( )
(1.4)

However, and , as well as all other system parameters, might be unknown and must be
identified.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (2 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

1.2. Identification of Linear Systems

The identification of the most relevant characteristic constants of the linear systems can be
done by means of a step change in the manipulated variable. First of all, the system must be at
the equilibrium state (measured as ) when the manipulated is maintained at . Of course,
this requires that no external disturbance alter such equilibrium. At time , with the system
still at equilibrium, the value of the manipulated variable is arbitrarily changed to a different
value . This change must proceed as quickly as possible in order to obtain a true step change.
The magnitude of the step should be large enough in order to obtain significant changes in the
state of the system, but it should be small enough in order to guarantee a true step change,
and to avoid critical situations that may affect the safety of the operators or the integrity of the
system (for example if the process is a runaway system).

After the step change, we must now choose two different arbitrary times ( and ). These
times must be selected in such a way that the corresponding measurements and are
significantly different from and from each other.

The behavior of the system state as a result of the step change can be expressed analytically by
solving (integrating) Eq. (1.1). First, let us rearrange Eq. (1.1) as follows:

( )
( ( ) ) ( )
(1.5)
and use the following substitution:

( ) ( ( ) ) ( )
(1.6)
such that
( ) ( )
(1.7)
Using Eq. (1.6) and (1.7) in Eq. (1.5) we obtain:

( )
( )
(1.8)
The solution to Eq. (1.8), integrating from , to an arbitrary time yields:

( )
( )
( )
(1.9)
The previous expression can be rearranged to obtain:

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (3 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

( ) ( ) ( )

(1.10)
Using Eq. (1.6) again, and solving for ( ) (considering that ( ) ):

( ) ( )( ( )
)
(1.11)
In terms of the measured variable , we then find (using Eq. 1.2):

( ) ( )( ( ) ( )( ( )
) )
(1.12)

Now, let us consider the change in the output variable obtained measured at the characteristic
time with respect to its initial equilibrium value:

( ) ( )( ( )
)
(1.13)
and therefore:
( )
( )( ( ) )
(1.14)

Thus, we have found an expression for determining the value of the term . This, however,
requires knowing the value of .

Let us now consider the change in the output variable with respect to its initial equilibrium
value, obtained for both arbitrary times:

( ) ( )( ( )
)
(1.15)
( ) ( )( ( )
)
(1.16)
Taking the quotient between both equations we obtain:

( ) ( )

( ) ( )

(1.17)

Then, the value can be obtained by solving the nonlinear equation given in Eq. (1.17). Let us
now consider some particular cases.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (4 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

 Self-regulating systems: Self-regulating systems are characterized by the fact that a

new equilibrium point is eventually obtained after the step change. This is only possible
if . The identification of self-regulating systems can be simplified by considering
that a new equilibrium point , measured as , is obtained when . Eq. (1.14)
then becomes:

(1.18)
where

(1.19)
is the gain of the process, and

| |
(1.20)
is the characteristic time of the process. Notice that for self-regulating systems (where
), we have | | .
If we now arbitrarily define as:

(1.21)
Then Eq. (1.17) becomes:
( )

(1.22)
This means that finding the time required to reach of the total change in
the measured equilibrium values, we then can determine the missing constant as
follows:

(1.23)
In some cases, the dynamic response of the system after the step change may be
delayed by different causes. Such delay is denoted as the dead time of the process
( ). In this case, Eq. (1.23) and (1.18) become, respectively:

(1.24)
( )
( )( )
(1.25)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (5 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

 Runaway systems: In the case of runaway or open-loop unstable linear systems ( ),

a new equilibrium point will not be obtained. In this case, two arbitrary observation
times are selected and the constants are determined by solving Eq. (1.14) and (1.17).
Since dead time ( ) can also be present, the corrected expressions would be:
( ( ) )
( )( ( ) )
(1.26)
( ) ( )
( ) ( )
(1.27)
where . Eq. (1.27) always requires using a suitable solution method for
nonlinear equations. Apart from this, the main difficulty of this method is the
experimental performance of the identification procedure, as the system may reach a
critical or dangerous operating condition. Care must be taken to select the times
and , so that significant changes can be observed in the system, but at the same
time still being capable of controlling the system after time . Since this experiment is
done for tuning the automatic PID controller, the control of the system after will
likely need to be manual. Alternatively, if a mechanistic, nonlinear model of the process
is available, then the linear system parameters can be simply obtained after
linearization.

 Capacitive systems: Finally, let us consider the identification of capacitive or integrative

systems ( ). The numerical solution of Eq. (1.1) becomes in this case (also
considering dead time):
( ) ( )( )
(1.28)
By measuring the response at a particular time after the step change, the value of
the term can be obtained:
( )
( )( )
(1.29)
Notice that Eq. (1.29) is similar to Eq. (1.25) and therefore it could be interpreted as the
process gain divided by the characteristic time of the process. However, since ( ) is
not an equilibrium state, is an arbitrary value, and the characteristic time for this
type of processes is undetermined, such association cannot be done.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (6 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

1.3. Feedback Error Signal

Since the state variable is not directly observed, but determined through instead, the
purpose of a control system is obtaining the desired state by minimizing the difference in value
between the measured variable and the corresponding desired set point (measured at
the desired state). Such difference is denoted as feedback error, and is defined as follows:

( ) ( )
(1.30)

The key of feedback process control is using the error signal in Eq. (1.30) for determining
adequate values of the manipulated variable such that the desired set point is obtained. While
many strategies are possible, we will consider the most common approach used in industry,
which is the PID controller.

2. PID Control Algorithm

The PID controller calculates three different effects denoted as: Proportional (P), Integral (I),
and Derivative (D), according to the following general expression:

∫ ( ) ( ) ( )
( ) ( ( ) ) ( ( ) ( ))

(2.1)

where is the gain of the controller, is the integral time, is the derivative time,
( ) ∫ ( ) is the integral of the error signal, from an arbitrary time (when the
( )
controller was turned on or it was last reset), and ( ) is the derivative of the error
signal.

From Eq. (1.30) and (1.2), and assuming a constant set point value, the term ( ) can be
expressed as follows:

( ) ( ) ( )
( ) ( ( )) ( ( ) )
(2.2)

Then, using Eq. (1.1) in (2.2) we obtain:

( ) ( ( ) ) ( ( ) )
(2.3)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (7 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Eq. (2.3) can be differentiated to yield:

( ) ( ) ( ) ( )
( ( ) ) ( ( ) )

(2.4)
The integral error, on the other hand, can be differentiated into:

( ) ∫ ( )
( )
(2.5)

Therefore the derivative of Eq. (2.1) becomes:

( ) ( ) ( ) ( )
( )

( ( ) ) ( ( ) ) ( ) ( ( ) )
( )
( ( ) )
(2.6)

( )
Since the term emerges in the right hand side of the equation, it can be solved out:

( ) ( ) ( )
( ( ) ) ( ( ) ) ( )
( ) ( ) ( )
(2.7)

An expanded state-space system [2] with three states instead of one can be obtained from Eq.
(1.1), (2.7) and (2.3) as follows:

( )
( ) ( )
(2.8)
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
(2.9)
( )
( ) ( )
(2.10)
where
( ) ( )
(2.11)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (8 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

( ) ( )
(2.12)
and ( ) are the states of the expanded system.

The expanded system can be expressed in matrix form as follows:

( ) ( )
[ ] [ ]
( ) ( ) ( )

[ ]
(2.13)

3. Stability

The stability of the closed loop with a PID controller can be analyzed considering the
eigenvalues of the coefficient matrix given in Eq. (2.13) [3]. The eigenvalues ( ) can be obtained
by solving:

( ) ( )
| |
( ) ( ) ( )

(3.1)
resulting in:

( ) ( ) ( )
( )( )
( ) ( ) ( ) ( )

(3.2)
which can be equivalently expressed as:

( )
( ( ) )
( ) ( )
(3.3)
Three solutions are obtained:

( ) √( ( )
)
( ) ( ) ( )

(3.4)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (9 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

The first solution is zero, indicating that one state drifts with no effect on the stability of the
system. Particularly, this eigenvalue represents the transformed manipulated variable . The
transformed state and the error signal are stable as long as the real parts of and are
negative. This is clearly influenced by the values of the parameters of the PID controller: ,
and . For this reason, an adequate tuning of the PID controller is required to guarantee a
stable operation of the closed loop.

In the following section, a PID controller tuning strategy based on the stability of the expanded
state-space system is proposed.

4. Controller Tuning

The selection of values of the controller parameters may result in a wide range of dynamic
behaviors. For example, if the eigenvalues and are complex, the system response will be
oscillatory. This occurs when the expression inside the root is negative. On the contrary, if the
expression is positive, the roots will be real. However, as the value inside the root increases,
the probability of obtaining a positive root (instability) also increases. For that reason, a critical
condition is obtained when the expression inside the root is exactly zero. Thus, the system is in
the limit of oscillatory behavior, while minimizing the risk of instability. The critical oscillation
condition can then be expressed as:

( )
( ) ( )
( ) ( )
(4.1)
From Eq. (4.1), an expression for the integral time constant is obtained:

( )
( )
( )
(4.2)
Once Eq. (4.1) is satisfied, the stability condition becomes:

( )
( )
(4.3)
This inequality can be expressed alternatively as follows:

( )
( )
(4.4)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (10 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

where is an arbitrary constant such that for and for (or in general,
).

The derivative time constant can be obtained from Eq. (4.4), resulting in:

( )
(4.5)
If we consider , then

(4.6)
and

(4.7)
or equivalently,

(4.8)
This result is obtained since

(4.9)

Such configuration corresponds to a PID controller, where the integral effect is dominant over
all other effects. Unfortunately, it is not a viable configuration as it yields infinite values of the
manipulated variable.

Another particular configuration is obtained when

(4.10)
and therefore:

(4.11)
and

( )
(4.12)

This configuration represents a PI controller, where no derivative effect is considered.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (11 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

In between these two extreme conditions we may find a suitable PID controller. Let us consider
a PID tuned exactly in the middle between these extremes, such that:

(4.13)
This implies:

(4.14)
and (from Eq. 4.2):

( )
(4.15)

This way, both the derivative and integral times are calculated as a function of the controller
gain and the system parameters.

For this case, the stability condition becomes (from Eq. 4.3):

(4.16)

If the value is less or equal than , then the system becomes unstable. If the value
is too large, the manipulated variable will change abruptly, and it will likely become saturated
resulting in a “bang-bang” controller, switching between the minimum and maximum possible
values of the manipulated variable [4]. Thus, a reasonable controller gain value must be used,
resulting in an adequate dynamic behavior of the manipulated variable. The dynamic response
of the manipulated variable should not be slower than the dynamic response of the process.
Otherwise, an inefficient control would be obtained. It is then possible to determine the value
of the controller gain by relating the dynamic response of the manipulated variable with the
dynamic response of the process.

Let us simply consider that (from Eq. 4.4, 4.13, and 4.14):

( )

(4.17)
Replacing this value in Eq. (4.14) and (4.15) results in:

( )
(4.18)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (12 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

( )
( )
(4.19)

Using these values, the characteristic time of the manipulated variable ( ) will be (from the
corresponding coefficient in Eq. 2.9):

| | | |
( ) | |
(4.20)
where is the characteristic time of the process.

Now, in the absence of process dead time, a reasonable assumption is , indicating that
the manipulated variable is times faster than the process. Larger values may result in
unnecessary drastic changes in the manipulated variable. However, in the presence of dead
time ( ), the dynamics of the manipulated variable should not be faster than the dead time or
otherwise, the system might become unstable. In this case we may say that . So, in
general, let us consider the following expression (satisfying the condition ):

( ) ( )

(4.21)

Using this expression, the final tuning of the PID becomes (using Eq. 1.19 and 1.20 in the case of
self-regulating systems):

(( ) ( ) ) (( ) ( ) )

(4.22)

(( ) ( ) ) (( ) ( ) )

(( ) ( ) ) (( ) ( ) )

(4.23)

(( ) ( ) ) ( ) ( )

(4.24)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (13 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Notice that the tuned value obtained for the derivative time has an opposite sign to that of the
integral time. Since certain PID controllers do not support negative derivative or integral times,
a PI controller with no derivative time can be used as an alternative. In this case, the tuned PI
parameters become (following a similar procedure as before):

| | ( ) ( )
( )

(4.25)

( )| | ( ) ( )

(( )| | ) (( ) ( ) )

(4.26)

In the particular case when the system is capacitive or integrating (that is, when ),
proceeding similarly we obtain the following controller parameters for a PID:

(4.27)

(4.28)

(4.29)

where in this case is a positive number related to the characteristic time of the manipulated
variable (the desired dynamic behavior of the manipulated variable)§:

(4.30)
where .

And for a PI controller (when ):

(4.31)

§
This result is obtained by replacing Eq. (4.27) and (4.29) in Eq. (4.20), for .

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (14 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

(4.32)

As a final notice, let us remark that the presence of dead time negatively influences the
performance of the controllers tuned with this method, particularly for PID. In fact, for large
dead times, the PI controller can be found to perform better than the PID, using the proposed
tuning equations. For self-regulating systems PI works better if ( ). In the case of
runaway systems, even the PI controller fails to control the system when ( ). For
capacitive systems, the PI controller also performs better. This behavior can be explained by
the fact that dead times were not considered in the time-domain model used in the derivation
of controller parameters.

Table 1 summarizes the tuning equations for the controller parameters (considering PID and PI
control) obtained in this work, for the different possible types of linear systems: Self-
regulating, runaway and capacitive. Notice that process gain and characteristic process time
are only considered for self-regulating systems. For all other systems, the model terms and
are used. The identification of these parameters was considered in Section 1.

Table 1. Proposed Equations for determining the Tuning Settings for PID and PI Controllers
Case
Self-regulating ( ) ( ( ))( ( )) ( ( ))
Process (PID) ( ( )) ( ( )) ( )
Self-regulating ( ( ))
Process (PI) ( ( )) ( ( ))

Runaway Process ( )( )
(PID) ( ) ( )

Runaway Process ( )
(PI) ( )
Capacitive Process
(PID)
Capacitive Process
(PI)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (15 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

5. Application Examples of PID Tuning

5.1. Heating Tank with Evaporation

As a first example of a self-regulating process with small dead time, let us consider the system
depicted in Figure 1. A stirred tank filled with water is heated using an electric heater with
variable power. The temperature of water is measured using a thermocouple and this signal is
sent to a PID controller.

Figure 1. Water heating tank with temperature control

The behavior of the temperature of water is described by the following system of differential
equations, considering water evaporation [5] and heat losses to the environment, and a time
delay in the action of the heater of seconds:

( ( ))
(5.1)

( ) ( ) ( )
(5.2)

where is the mass of water in the tank in , is temperature of water in , is time in ,

( ) is the percentage of heating power used (manipulated variable), and the initial
conditions are ( ) , ( ) .

The response of the system to a step change in the heating power from to is shown in
Figure 2. From this response, the following process parameters are obtained: Process gain

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (16 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

, process characteristic time , process dead time . The

temperature set point considered is .

Different PID controllers are tuned following common methods found in the literature [5,6]
(summarized in Table 2) as well as the equations proposed in this work for the time-domain
stability and tuning (Table 1).

Figure 2. Dynamic response of heating tank to step change in heating power.

Table 2. Common Equations for determining the Tuning Settings for PID Controllers (self-
regulating processes) [6,7]
Method

Ziegler-Nichols

Shinskey-Lipták

( )
Cohen-Coon

3C ( ) ( ) ( )

Ciancone-Marlin [7]
(Graphical method)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (17 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Using these values, the PID tuning parameters using different methods are given in Table 3,
and their corresponding performance (time-average integral of square error 〈 〉) during
hours is summarized in Figure 3.

Figure 3. Dynamic performance of the PID/PI controller for the heating tank example,
considering different tuning methods.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (18 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Table 3. PID parameters and control performance obtained using different tuning methods for
the heating tank problem
Method (h) (h) 〈 〉
Time-domain
(PID, self-regulating)
Time-domain
(PI, self-regulating)

Ziegler-Nichols

Shinskey-Lipták

Cohen-Coon

3C

Ciancone-Marlin
(PI set point response)
Ciancone-Marlin
(PI disturbance
response)

Particularly for the Ciancone-Marlin method, only the PI configuration was used because the
PID configuration resulted in zero derivative times. Due to the small dead time, most common
tuning methods (with the exception of Ciancone-Marlin) yield very large controller gain values
and very low integral times, resulting in bang-bang behavior for about hours. Ciancone-Marlin
provides two sets of tuning parameters (suggesting PI controllers) with a non-oscillating
approach to the set point value, and without saturating the manipulated variable. Particularly,
the disturbance response parameters presented a small overshoot, not observed with the set
point response parameters. The PID and PI controllers tuned using the time-domain approach
(the expressions proposed in this work), were also non-oscillating and reached the set point
value faster than Ciancone-Marlin, but the manipulated variable became saturated at the
beginning. This is not necessarily an operational issue, as long as sudden changes in the
manipulated variable are not observed (as in the case of bang-bang control). The best
performance in terms of sum of square error was provided by the time-domain PID, closely
followed by the time-domain PI. In addition, the PI controller tuned with the time-domain
approach presented a small overshoot (similar to Ciancone-Marlin’s disturbance response), not
observed in the corresponding time-domain PID controller (similar to Ciancone-Marlin’s set
point response).

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (19 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

5.2. Wastewater Treatment

As a second example, let us consider the neutralization of an acid aqueous effluent generated
as a by-product in a chemical process, as depicted in Figure 4. The acid waste enters the stirred
tank at the bottom, and exits the process at the top by overflow. The acid concentration is
determined at the outlet by an online analyzer with a long time delay. The total dead time of
the process is . The signal received from this analyzer is used to manipulate the flow
of sodium hydroxide solution (the manipulated variable is the opening percentage of the
valve).

Figure 4. Wastewater treatment with composition (acid concentration) control.

The process can be described by the following system of differential equations:

( )
(5.3)
( )
(5.4)

where is the concentration of acid in , is the concentration of base (sodium

hydroxide) in , is the inlet flow of waste acid solution in , is the
concentration of acid in the waste solution in , is the percentage of opening of the
linear valve in the sodium hydroxide solution line, and is time in minutes. Both and
are unmeasured disturbances to the process. During process identification, they will be kept
constant at and For the purpose of control performance
simulation, they will be considered normal random Markov chains changing every , with
standard deviation of for and for .

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (20 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

The system identification is performed using a step change in the valve opening from to
, starting from the initial conditions ( ) , and ( ) . The response
of the system is presented in Figure 5. From these results, the system can be approximated by
a linear model with the following parameters: , , and .

Figure 5. Dynamic response of the wastewater neutralization process to a step change in valve
opening.

The current tuning strategy along with the common tuning methods presented in Table 2, are
used to obtain different PID/PI parameters, which are summarized in Table 4. These sets of
PID/PI tuning parameters are then used to regulate the acid concentration at the outlet at a set
point value of . The system is initially at the following conditions: ( ) ,
and ( ) , and then it is subject to Markov chain disturbances in and as
illustrated in Figure 6. While the disturbances were randomly generated, the same disturbance
profiles are used for all tuning parameters in order to provide a fair comparison. The
performance of the controller during minutes using the different tuning parameters is
summarized in Figure 7.

The best performance (in terms of 〈 〉) is obtained using the 3C tuning method, closely
followed by the Cohen-Coon, the time-domain PI and the Ciancone-Marlin methods,
respectively. The time-domain PID had the worst 〈 〉 value, due to its slow response.
However, only the time-domain PI and PID, as well as the 3C method were able to satisfactorily
regulate the outlet concentration. All other methods had difficulties reaching or maintaining
the set point value. No saturation was observed for any tuning method. The 3C method and the
time-domain PI also resulted in a smooth change in the valve opening. On the other hand, the
concentration obtained with the 3C method was actually below the set point value. While this
is good from the perspective of successful acid neutralization, it implies a larger consumption
of neutralizing solution, and thus, higher operational costs.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (21 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Table 4. PID parameters and control performance obtained using different tuning methods for
the wastewater neutralization problem
Method (min) (min) 〈 〉
Time-domain
(PID, self-regulating)
Time-domain
(PI, self-regulating)

Ziegler-Nichols

Shinskey-Lipták

Cohen-Coon

3C

Ciancone-Marlin
(PID disturbance
response)
Ciancone-Marlin
(PI disturbance
response)

Figure 6. Dynamic profile of the disturbances (inlet flow and inlet acid concentration) for the
wastewater neutralization process.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (22 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Figure 7. Dynamic performance of the PID/PI controller for the wastewater neutralization
example, considering different tuning methods.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (23 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

5.3. Exothermic Reaction

The following example corresponds to a runaway system, consisting on a continuous stirred

tank reactor carrying out a highly exothermic isomerization reaction. The reaction temperature
is controlled by the flow of fresh reactant added to the reactor. However, this stream not only
cools the mixture but also increases the concentration of the reactant, favoring the exothermic
reaction. This hypothetical and not recommended, unstable configuration is illustrated in
Figure 8. The reactor level is kept constant by an overflow stream.

Figure 8. Continuous exothermic reactor with temperature control.

Linear runaway models are seldom found in practice due to two main reasons: First, no process
will reach a state value ( ) as a result of an open-loop disturbance; and second,
because real processes are usually nonlinear and therefore, the linear runaway behavior is only
observed for a limited range of operating conditions. The continuous exothermic reaction
considered in this example is not an exemption. At low temperatures, the reactor is self-
regulated because the rate of reaction is negligible. At intermediate temperatures, as the rate
of reaction increases, the runaway behavior is observed. At higher temperatures, self-
regulation emerges due to larger heat losses to the environment, and most importantly,
because the reactant is depleted, extinguishing the reaction. A hypothetical example of such
behavior is illustrated in Figure 9. Initially the system is in a runaway condition until the reactant
is depleted, reaching extremely high temperatures that the reactor itself would not be able to
sustain. Then, the system becomes self-regulating until it accumulates enough reactant,
reaching a runaway condition again.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (24 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Figure 9. Illustrative dynamic behavior of a highly nonlinear exothermic reactor.

For illustrative purposes, let us assume that the process is operating in a narrow region where
the behavior can be approximately described by the following linear runaway system:

( ) ( ) ( )
(5.5)

where is the temperature measured in the reactor in , is the percentage of opening of the
linear valve in the stream of fresh reactant, is the atmospheric temperature (a potential
disturbance), and is time in minutes. The dead time is considered negligible. As it can be seen
in Eq. (5.5), the equilibrium condition is for and . However, any
disturbance in the process will destabilize the system. Figure 10 shows the dynamic response of
the system when a step change in the manipulated variable (from to ) is done
at time , when the system is already at equilibrium. As a result, the reaction temperature
decreases due to the larger amount of cold reactant entering the process and the decrease in
residence time in the reactor.

Figure 10. Dynamic response of the exothermic reactor to a step change in valve opening of
fresh (cold) reactant.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (25 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Since a new equilibrium point is obtained, a linear model of the process can be obtained by
measuring temperatures at two different times after the step change (as described in Section
1.2). Considering as reference times and , then we obtain
( ) and ( ) . Solving Eq. (1.24) and (1.25) results in
and , consistent with Eq. (5.5). In highly nonlinear systems,
the particular selection of times may have a large influence on the parameter values.

Let us now consider the effect of the environmental temperature on the reactor temperature.
Figure 11 shows a particular random profile for the environmental temperature and the
corresponding dynamic of the reactor temperature while keeping a constant flow of reactant.
Even when the environmental temperature fluctuates only within in hours, the system is
ultimately destabilized by the disturbances.

Figure 11. Dynamic response of the exothermic reactor to random disturbances in

environmental temperature. Left plot: Environmental temperature profile. Right plot: Reactor
temperature and %opening of fresh feed valve.

The exact same disturbance profile is used to test the proposed tuning of PI and PID controllers
with the time-domain approach using . The process parameters found earlier are
used for tuning the PI and PID controllers, as summarized in Table 5, along with the
corresponding control performance during hours. The dynamic performance can also be
observed in Figure 12. Common tuning methods are designed for self-regulating systems. Of
course, there are many methods available for non-self-regulating systems [8], but they are not
considered here since most of them do not work well for negligible dead time processes.

Table 5. PID parameters and control performance obtained using different tuning methods for
the exothermic reactor problem
Method (min) (min) 〈 〉
Time-domain
(PID, runaway)
Time-domain
(PI, runaway)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (26 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Figure 12. Dynamic performance of the PID/PI controller for the exothermic reactor example,
considering only the time-domain tuning method.

Even though both controllers satisfactorily stabilized the process, the PI controller provided a
better performance (lower 〈 〉) with a less noisy action on the valve. Also notice that large
changes in the manipulated variable were required to stabilize the system in the presence of
relatively small disturbances. Notice that the performance values are smaller in this example
than in previous examples, because in this case, the system was already at the set point value
at the initial time considered.

5.4. Liquid Level in a Tank Filled with a Pump

As a final example, let us consider an integrating or capacitive system. In this case we have an
intermediate buffer tank providing process water to different units in the plant. The level of the
tank is regulated using a variable-speed pump replenishing the water consumed by the plant.
This system is depicted in Figure 13.

Figure 13. Buffer tank with level control, storing process water for the plant.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (27 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

The system can be described using the following differential equation:

(5.6)

where is the level of water in the buffer tank in , is the percentage of pump speed,
is the consumption of water by the plant (a disturbance) in , and is time in hours. The
dead time of this process is negligible. The maximum liquid level is , and the liquid level
set point is .

The response of the system to a step change in the manipulated variable from to
when no water is consumed by the plant is shown in Figure 14. The response obtained can be
used to determine the term (Eq. 1.29), characterizing the capacitive process. From the data
obtained we find that ( )( )
, corresponding to the coefficient of in Eq.
(5.6).

Figure 14. Dynamic response of the level in a buffer tank to a step change in pump speed
feeding fresh water.

The water consumed by the plant changes drastically when one of the units served begins
or finishes a batch process. There are also random changes in water consumption during each
batch. The disturbance profile used as reference for control performance is presented in Figure
15. It also shows the open-loop response of the tank level to such disturbances.

In order to tune the PI and PID controllers for this integrative process, it is necessary to
determine the desired characteristic time of the manipulated variable. If the characteristic time
of the manipulated variable is too short, an unnecessary stress will be exerted on the pump
drive. Thus, a value of is arbitrarily chosen. The controller parameters and their
corresponding performance during hours obtained are summarized in Table 6. The dynamic
performance of the controllers is illustrated in Figure 16. PID tuning rules for this type of
systems considering non-negligible dead times can be found in the literature [8].

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (28 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Figure 15. Dynamic response of the level in a buffer tank to random disturbances in water
consumption by the plant. Left plot: Water consumption profile. Right plot: Water level and
%pump speed.

Table 6. Parameters and control performance for the buffer tank problem
Method (h) (h) 〈 〉
Time-domain
(PID, capacitive)
Time-domain
(PI, capacitive)

Figure 16. Dynamic performance of the PID/PI controller for the exothermic reactor example,
considering only the time-domain tuning method.

The PID controller maintains the liquid level within around the set point value, whereas
the PI controller keeps the level within only. The PI controller not only reduces the
fluctuation in liquid level, providing a better control performance, but also provides a less noisy
action. For the PI controller the profile of the manipulated variable obtained resembles more
closely the actual behavior of the disturbance, than the PID controller.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (29 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

6. Final Remarks

In this report, novel tuning rules for PID and PI controllers have been obtained using various
criteria including stability, non-oscillatory behavior, and reasonable dynamic behavior of the
control action (avoiding bang-bang control). All these criteria were applied considering a linear
model approximation of the system in the time domain, with no dead time:

( )
( ( ) ) ( ( ) )
(1.1)
( ) ( )
(1.2)

This time-domain approach does not require transforming the model into the Laplace-
transform domain. Different equations were obtained for each type of processes considered:
Self-regulating, runaway or unstable, and capacitive or integrating. The resulting PID/PI tuning
equations were summarized in Table 1:

Table 1. Proposed Equations for determining the Tuning Settings for PID and PI Controllers
Case
Self-regulating ( ) ( ( ))( ( )) ( ( ))
Process (PID) ( ( )) ( ( )) ( )
Self-regulating ( ( ))
Process (PI) ( ( )) ( ( ))

Runaway Process ( )( )
(PID) ( ) ( )

Runaway Process ( )
(PI) ( )
Capacitive Process
(PID)
Capacitive Process
(PI)

where represents the actual dead time of the process, and

(1.18)

| |
(1.20)

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (30 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

| | | |
( ) | |
(4.20)

where is inversely related to the response speed of the manipulated variable, with a
suggested value of .

The performance of the PID/PI controllers tuned using this method was evaluated in four
different examples of chemical processes. Despite the fact that the derivative time of PID
controllers was found to be negative, which is counter-intuitive according to the conventional
use of the concept, the control performance was satisfactory, at least for small dead times
relative to the characteristic time of the process. An increase in the dead time fraction affects
the performance of the PID controller, particularly for . For significant dead times,
the PI controller performed better than PID, and was less affected by the dead time. The PI
controller also presented better performance for unstable and capacitive systems, with a less
noisy control action.

Compared to conventional tuning methods (only for self-regulating systems), it was shown
that the PI and PID tuning rules proposed in this work provided better controller performance
for small dead times, while the PI tuning rule was highly competitive for large dead times.

Finally, and only for comparative purposes, Figure 17 shows the behavior of the PID parameters
obtained by the Ciancone-Marlin correlations and by the proposed time-domain method. This
plots show that the values of the parameters tuned with the proposed method are closer to
Ciancone-Marlin’s values for PI controllers than for PID controllers.

Figure 17. Graphical representation of PI/PID tuning equations in the time-domain for self-
regulating systems (solid lines), compared to the Ciancone-Marlin correlations (dashed lines).

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (31 / 32)
 PID Controller Stability and Tuning
using a Time-Domain Approach
Hugo Hernandez and Silvia Ochoa
[email protected]
[email protected]

Acknowledgment and Disclaimer

This report provides data, information and conclusions obtained by the author(s) as a result of original
scientific research, based on the best scientific knowledge available to the author(s). The main purpose
of this publication is the open sharing of scientific knowledge. Any mistake, omission, error or inaccuracy
published, if any, is completely unintentional.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-
for-profit sectors.

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC
4.0). Anyone is free to share (copy and redistribute the material in any medium or format) or adapt
(remix, transform, and build upon the material) this work under the following terms:
 Attribution: Appropriate credit must be given, providing a link to the license, and indicating if
changes were made. This can be done in any reasonable manner, but not in any way that
suggests endorsement by the licensor.
 NonCommercial: This material may not be used for commercial purposes.

References

[1] Kuphaldt, T. R. (2019). Lessons in Industrial Instrumentation (Version 2.32). Chapter 33. Process
Dynamics and PID Controller Tuning. Self-regulating, Integrating, and Runaway Process
Characteristics. https://control.com/textbook/process-dynamics-and-pid-controller-tuning/process-
characteristics/.
[2] Hernandez, H., & Ochoa, S. (2020). Expanded State-Space Model for a Linear SISO Controller:
Stability and Tuning. ForsChem Research Reports, 5, 2020-05, 1-27. doi: 10.13140/RG.2.2.23053.90086.
[3] Zadeh, L. A., & Desoer, C. A. (1963). Linear System Theory: The State Space Approach. Reprinted in
1979 by Robert E. Krieger Publishing Co., New York. Section 5.4. Modes in linear time-invariant
systems (distinct eigenvalues). pp. 311-324.
[4] Bellman, R., Glicksberg, I., & Gross, O. (1956). On the “bang-bang” control problem. Quarterly of
Applied Mathematics, 14(1), 11-18. https://www.jstor.org/stable/43634288.
[5] Hernandez, H. (2022). Molecular Modeling of Macroscopic Phase Changes 1: Liquid Evaporation.
ForsChem Research Reports, 7, 2022-10, 1 - 20. doi: 10.13140/RG.2.2.29272.98566.
[6] Murrill, P. W., Schnelle, P. D., Lipták, B. G., Gerry, J., Ruel, M. and Shinskey, F. G. (2006). Tuning PID
Controllers. In: Lipták, B. G. Instruments Engineers’ Handbook. 4th Edition. Volume 2. Process
Control and Optimization. CRC Press, Boca Raton FL. Section 2.35. doi:
https://doi.org/10.1201/9781315219028.
[7] Marlin, T. E. (1995). Process Control: Designing Processes and Control Systems for Dynamic
Performance. McGraw-Hill International Editions, New York. Chapter 9: PID Controller Tuning for
Dynamic Performance. http://pc-textbook.mcmaster.ca/Marlin-Ch09.pdf.
[8] O’Dwyer, A. (2009). Handbook of PI and PID Controller Tuning Rules. Imperial College Press,
London. doi: 10.1142/p575.

13/03/2023 ForsChem Research Reports Vol. 8, 2023-05

10.13140/RG.2.2.26330.62400 www.forschem.org (32 / 32)