Dynamic models

A dynamic (or unsteady-state) model is used to characterize the

transient behavior of a process.
Benefits of process dynamic models
✓Improve understanding of the process
✓Train plant operational personnel
✓Develop a control strategy for a new process
✓Optimize process operating conditions
Classifications of dynamic models
• Theoretical models
• Empirical models and
• Semi-empirical models are developed using a combination of
theoretical and empirical models.
 Cont’d …
Example: a stirred-tank blending process
The unsteady-state model is the situation where the process variables
vary with time.

Form of the unsteady-state mass balance

rate of accumulation rate of rate of
= +
of mass in the tank mass in mass out
 Cont’d …
Mass of the liquid is the product of its volume, 𝑉 and density, ρ.
𝑑(𝑉ρ)
= w1 + w2 − w −−−−−−−−− −(𝟏)
𝑑𝑡
Similarly, the unsteady-state material balance for component A
𝑑(𝑉ρ𝑥)
= w1 x1 + w2 x2 − wx −−−−−−−−− −(𝟐)
𝑑𝑡
By setting the accumulation terms equal to zero, we obtain the
corresponding steady-state model.
Dynamic model of the blending process
Model equations can not incorporate all the features of the real process, but
are at best an approximation.
“ all models are wrong, but some are useful”
Important conservation laws that are used in theoretical models
• Conservation of mass
• Conservation of component
• Conservation of energy
 Cont’d …
Assuming constant liquid density
𝑑(𝑉)
ρ = w1 + w2 − w
𝑑𝑡
𝑑(𝑉𝑥)
ρ = w1 x1 + w2 x2 − wx
𝑑𝑡
Expanding the accumulation term using chain rule
𝑑(𝑉𝑥) 𝑑𝑥 𝑑𝑉
ρ = ρV + ρx
𝑑𝑡 𝑑𝑡 𝑑𝑡
Substituting in to the mass balance gives us the general model of the
blending process.
𝑑𝑉 1
= (w1 + w2 − w)
𝑑𝑡 ρ
𝑑𝑥 w1 w2
= x1 − x + (x2 −𝑥)
𝑑𝑡 𝑉ρ 𝑉ρ
 Cont’d …
Example; a stirred-tank blending process with a constant liquid holdup
of 2m3 is used to blend two streams whose densities are both
kg
approximately 900 3 . The density does not change during mixing.
m
a) Assume that the process has been operating for a long period of
kg kg
time with flow rates of w1 = 500 and w2 = 200 and feed
𝑚𝑖𝑛 𝑚𝑖𝑛
compositions (mass fractions) of x1 = 0.4 and x2 = 0.75. what is
the steady-state value of 𝑥?
Solution;
Denoting initial state conditions using a bar and substituting values give
the steady-state value of 𝑥 = 𝑥.ҧ
𝑤
ഥ1 𝑥ҧ1 + 𝑤
ഥ2 𝑥ҧ2 (500)(0.4) + (200)(0.75)
𝑥ҧ = = = 0.5
𝑤
ഥ 700
b) Suppose that w1 changes suddenly from 500 𝑡𝑜 400 𝑘𝑔/𝑚𝑖𝑛 and
remains at the new value. Determine an expression for 𝑥 𝑡 and
plot it.
 Cont’d …
Solution;
Rearranging the component balance using assumptions of constant 𝑉 𝑎𝑛𝑑 𝜌.
𝑑(𝑉ρ𝑥)
= w1 x1 + w2 x2 − wx
𝑑𝑡
𝑑𝑥 𝑤1 𝑥1 + 𝑤2 𝑥2 ത
𝑉𝜌
𝜏 +𝑥 = 𝑤ℎ𝑒𝑟𝑒 𝑥 0 = 𝑥ҧ = 0.5 and 𝜏 =
𝑑𝑡 𝑤 𝑤ഥ
Substituting the value of 𝜏 = 3min and the right-side equation is denoted as
constant.
𝑑𝑥 ∗
𝑤1 𝑥1 + 𝑤2 𝑥2
3 +𝑥 =𝐶 =
𝑑𝑡 𝑤
Applying standard solution methods, we obtain the solution
𝑡
−3
𝑥 𝑡 = 0.5𝑒 −𝑡/3 +𝐶∗ 1− 𝑒
The constant can be calculated using the given values
∗
𝑤1 𝑥1 + 𝑤2 𝑥2 (400 𝑘𝑔/𝑚𝑖𝑛)(0.4) + (200 𝑘𝑔/𝑚𝑖𝑛)(0.75)
𝐶 = = = 0.517
𝑤 600 𝑘𝑔/𝑚𝑖𝑛
 Cont’d …
The solution becomes
𝑡 𝑡
−3 −3
𝑥 𝑡 = 0.5𝑒 + 0.517 1 − 𝑒
c) For the case where w2 changes suddenly from 200 𝑡𝑜 100 𝑘𝑔/𝑚𝑖𝑛 and
remains there.
Solution;
∗
𝑤1 𝑥1 + 𝑤2 𝑥2 (500 𝑘𝑔/𝑚𝑖𝑛)(0.4) + (100 𝑘𝑔/𝑚𝑖𝑛)(0.75)
𝐶 = = = 0.458
𝑤 600 𝑘𝑔/𝑚𝑖𝑛
And the solution is
𝑡 𝑡
−3 −3
𝑥 𝑡 = 0.5𝑒 + 0.458 1 − 𝑒

d) For the case where x1 changes from 0.4 𝑡𝑜 0.6.

Solution; Solving for simultaneous changes in x1 𝑎𝑛𝑑 w2
∗
𝑤1 𝑥1 + 𝑤2 𝑥2 (500 𝑘𝑔/𝑚𝑖𝑛)(0.6) + (100 𝑘𝑔/𝑚𝑖𝑛)(0.75)
𝐶 = = = 0.625
𝑤 600 𝑘𝑔/𝑚𝑖𝑛
Then,
𝑡 𝑡
−3 −3
𝑥 𝑡 = 0.5𝑒 + 0.625 1 − 𝑒
 Cont’d …
e) Plot the normalized response for parts b and d.
Solution;
They have the same normalized responses.
x t − 𝑥(0) 𝑡
−3
𝑥𝑁 (𝑡) = =1−𝑒
𝑥 ∞ − 𝑥(0)
Where
𝑥 ∞ 𝑎𝑛𝑑 𝑥 0 𝑎𝑟𝑒 𝑡ℎ𝑒 𝑓𝑖𝑛𝑎𝑙 𝑎𝑛𝑑 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 𝑠𝑡𝑒𝑎𝑑𝑦 𝑠𝑡𝑎𝑡𝑒 𝑣𝑎𝑙𝑢𝑒𝑠 𝑜𝑓 𝑥 𝑡 .
 Cont’d …
Degrees of freedom (𝑵𝑭 ) expression
𝑁𝐹 = 𝑁𝑉 − 𝑁𝐸
Where; 𝑁𝑉 is the total process variables and
𝑁𝐸 is the number of independent equations
Categories of modeling problems
i. 𝑁𝐹 = 0 the process model is exactly specified, has a solution
ii. 𝑁𝐹 > 0 the process model is underspecified, infinite solution
iii. 𝑁𝐹 < 0 the process model is over specified, has no solution
Degrees of freedom analysis
• List known constants in the model (or specified parameters)
• Determine 𝑁𝑉 𝑎𝑛𝑑 𝑁𝐸 , 𝑡𝑖𝑚𝑒 (𝑡) is not considered to be a process variable.
• Calculate the number of degrees of freedom, 𝑁𝐹 .
• Specifying a process variable as disturbance variable or manipulated
variable increases 𝑁𝐸 and utilizes 𝑁𝐹 .
 Cont’d …
Example; analyze the degrees of freedom for the blending model , for the condition of
constant volume (V).
Solution;
𝑑𝑥
ρV = w1 x1 + w2 x2 − wx
𝑑𝑡
• 2 parameters or constants: 𝑉, ρ
• Process variables (inputs and outputs) 𝑁𝑣 = 4 : x, 𝑥1 , 𝑤1 𝑎𝑛𝑑 𝑤2
• Equation 𝑁𝐸 = 1
• Then degrees of freedom 𝑁𝐹 = 𝑁𝑉 − 𝑁𝐸 = 4 − 1 = 3
In order for the equation to have a solution, three input variables are specified as functions
of time.
• 2 disturbance variables: 𝑥1 , 𝑤1 and
• 1 manipulated variable: 𝑤2
Example; analyze the degrees of freedom for the blending system model, for the condition
of variable volume (V).
𝑑𝑉 1
= (w + w2 − w)
𝑑𝑡 ρ 1
𝑑𝑥 w1 w2
= x1 − x + (x2 −𝑥)
𝑑𝑡 𝑉ρ 𝑉ρ
Design and performance analysis of PID controller
Standard feedback control algorithms, are also called control laws.
Proportional integral controller (or PID) is a control algorithm and is widely used in
process industries.
Analog devices use continuous (or analog) input and output signals, which can be
either pneumatic or electronic.
Proportional, integral and derivative (PID) controller, known as a three-mode
controller became commercially available in the 1930’s.
Basic control modes
➢ Proportional control
In the proportional control, the controller output is proportional to the error signal.
𝑝 𝑡 = 𝑝ҧ + 𝐾𝑐 𝑒(𝑡)
Where; 𝑝 𝑡 = controller output, 𝐾𝑐 = 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑙𝑒𝑟 𝑔𝑎𝑖𝑛, 𝑝ҧ =
bias or steady state value.
Error signal, 𝑒 𝑡 is the difference between setpoint, 𝑦𝑠𝑝 𝑡 and measured value,
𝑦𝑚 𝑡 .
𝑒 𝑡 = 𝑦𝑠𝑝 𝑡 − 𝑦𝑚 𝑡
 Cont’d …
• The controller gain is adjustable and can be tuned after the controller is
installed.
• For general purpose controllers, 𝐾𝑐 is dimensionless.
• The controller gain is adjusted to change the controller output.
• The controller gain sign increases or decreases the controller output.
• Bias is used to adjust the controller output to the nominal value
• When the error is zero, the controller output equals to the bias value.
Ideal proportional control characteristic (𝑲𝒄 is the slope )
 Cont’d …
Proportional control transfer function
𝑝 𝑡 = 𝑝ҧ + 𝐾𝑐 𝑒(𝑡)
𝑝′ 𝑡 = 𝐾𝑐 𝑒 𝑡 𝑤ℎ𝑒𝑟𝑒 𝑝′ 𝑡 = 𝑝 𝑡 − 𝑝ҧ
Taking Laplace transform
𝑃′ 𝑠
𝐾𝑐 =
𝐸 𝑠
Actual proportional control characteristic (𝑲𝒄 is the slope )
Actual controller saturates when the output reaches a physical limit, either
𝑃𝑚𝑎𝑥 𝑜𝑟𝑃𝑚𝑖𝑛 .
 Cont’d …
Disadvantage of proportional only control
A steady-state error (or offset) occurs after a setpoint change or
sustained disturbance.
✓Offset can be eliminated by manually resetting the setpoint or bias.
✓Automatic offset reset is realized by incorporating an integral control
action.
➢ Integral control
The controller output depends on the integral of the error signal over
time.
Integral control action is also referred to as reset or floating control.
1 𝑡 ∗
𝑝 𝑡 = 𝑝ҧ + න 𝑒(𝑡 ) 𝑑𝑡 ∗
𝜏𝐼 0
Where; 𝜏𝐼 𝑖𝑠 𝑡ℎ𝑒 𝑖𝑛𝑡𝑒𝑔𝑟𝑎𝑙 𝑜𝑟 𝑟𝑒𝑠𝑒𝑡 𝑡𝑖𝑚𝑒
 Cont’d …
The controller output 𝑝 𝑡 , automatically changes with time until it attains
zero steady state error 𝑒 𝑡 ∗ = 0.
Controller saturation occurs when the setpoint change grows so large
beyond the manipulated variable range.
Integral controller by itself takes little control action until the error signal
persists for some time.
Where as proportional control takes immediate corrective action as soon as
an error is detected.
Consequently, integral control action is used in conjunction with
proportional control as proportional-integral(PI) controller.
1 𝑡 ∗
𝑝 𝑡 = 𝑝ҧ + 𝐾𝑐 𝑒 𝑡 + න 𝑒 𝑡 𝑑𝑡 ∗
𝜏𝐼 0
The corresponding transfer function
𝑃′ 𝑠 1
= 𝐾𝑐 1 +
𝐸 𝑠 𝜏𝐼 𝑠
 Cont’d …
Response of PI controller to unit step change in 𝒆(𝒕)
Proportional action at 𝑡 = 0, changes the controller output
instantaneously.
Then integral action repeats the proportional action amount, for 0 <
𝑡 = 𝜏𝐼 by causing a ramp increase in 𝑝(𝑡).
 Cont’d …
Disadvantages of integral control
• Integral action tends to produce oscillatory response of the controlled
variable which reduces the stability of the feedback control system.
✓Too much integral action can be avoided by proper tuning or
✓Including derivative action which tends to counteract the destabilizing
effects.
• Reset windup
✓Integral term becomes quite large when a sustained error occurs and
causes the controller output eventual saturation.
✓Further buildup of the integral term while the controller is saturated is
known as reset or integral windup.
✓The large overshoot occurs because the integral term continues to increase
until 𝑒(𝑡) changes sign at 𝑡 = 𝑡1 .
✓After sufficiently small integral term, the value of 𝑝 𝑡 moves away from
the saturation limit.
 Cont’d …
Reset windup occurs when PI controller encounters a sustained error.
For example; during startup of a batch process or after a large set-point
change, 𝑒(𝑡) fails to decrease to zero.
E.g. a large sustained disturbance beyond the manipulated variable,
due to physical limitation of fully open/shut control valve.
✓Reset windup is prevented by temporarily halting the integral action
when 𝑝 𝑡 saturates.
 Cont’d …
➢ Derivative control
Derivative action anticipates the future behavior of the error signal by
considering its rate of change.
Derivative action is also known as rate action, pre-act or anticipatory control.
Example; suppose a reactor temperature increases rapidly 10𝑜 𝐶 𝑝𝑒𝑟 3𝑚𝑖𝑛.
This situation requires a quick corrective action to reduce the temperature.
✓Only derivative action could take appropriate action anticipating the rate
of change in temperature.
✓Proportional action reacts only to a deviation in temperature.
✓Integral term takes action depending on the duration of deviation.
The derivative action is proportional to the rate of change of error signal or
controlled variable.
𝑑𝑒(𝑡)
𝑝 𝑡 = 𝑝ҧ + 𝜏𝐷 𝑤ℎ𝑒𝑟𝑒, 𝜏𝐷 𝑖𝑠 𝑡ℎ𝑒 𝑑𝑒𝑟𝑖𝑣𝑎𝑡𝑖𝑣𝑒 𝑡𝑖𝑚𝑒.
𝑑𝑡
 Cont’d …
Derivative
𝑑𝑒(𝑡)
action is equal to the nominal value as long as error is constant,
= 0.
𝑑𝑡
Derivative action is always used in conjunction with proportional or PI
control.
Ideal PD controller transfer function
𝑃′ 𝑠
= 𝐾𝑐 1 + 𝜏𝐷 𝑠
𝐸 𝑠
Derivative action tends to stabilize the controlled process.
Derivative action also teds to improve the dynamic response of the
controlled variable by reducing the process settling time.
Disadvantages of derivative control
Derivative action amplifies noisy (i.e. high frequency, random fluctuations) or
unfiltered process measurement.
Practical derivative control including a derivative filter
𝑃′ 𝑠 𝜏𝐷 𝑠
= 𝐾𝑐 1 + 𝑤ℎ𝑒𝑟𝑒, 0.05 < 𝛼 < 0.2
𝐸 𝑠 𝛼𝜏𝐷 𝑠 + 1
 Cont’d …
➢ Proportional-integral-derivative control
Three most common forms of PID
i. Parallel form of PID control
Each of the three modes are combined to operate in parallel.
1 𝑡 ∗ ∗
𝑑𝑒(𝑡)
𝑝 𝑡 = 𝑝ҧ + 𝐾𝑐 𝑒 𝑡 + න 𝑒 𝑡 𝑑𝑡 + 𝜏𝐷
𝜏𝐼 0 𝑑𝑡
And the transfer function
𝑃′ 𝑠 1
= 𝐾𝑐 1 + + 𝜏𝐷 𝑠
𝐸 𝑠 𝜏𝐼 𝑠
 Cont’d …
ii. Series form of PID control
The controller is constructed so that a PI element and PD element operate in
series.
A derivative filter is applied to one of the derivative term.
𝑃′ 𝑠 1 𝜏𝐷 𝑠 + 1
= 𝐾𝑐 1 +
𝐸 𝑠 𝜏𝐼 𝑠 𝛼𝜏𝐷 𝑠 + 1

iii. Expanded form of PID control

This is Less used than the well-known series and parallel forms.
𝑡
∗ ∗
𝑑𝑒(𝑡)
𝑝 𝑡 = 𝑝ҧ + 𝐾𝑐 𝑒 𝑡 + 𝐾𝐼 න 𝑒 𝑡 𝑑𝑡 + 𝐾𝐷
0 𝑑𝑡
The three gains, 𝐾𝑐 , 𝐾𝐼 𝑎𝑛𝑑𝐾𝐷 are used to independently adjust each of the
terms.
 Cont’d …
Features of PID controllers
Elimination of derivative and proportional kick
Sudden change in setpoint (or error) causes the derivative term momentarily
to become very large.
This behavior provides a derivative kick to the final control element.
Derivative kick can be avoided by basing the derivative action on the
measurement signal, 𝑦𝑚 than on the error.
𝑑𝑒 𝑑𝑦𝑚
Consider the parallel form of PID, replacing 𝑏𝑦 gives
𝑑𝑡 𝑑𝑡
1 𝑡 ∗ ∗
𝑑𝑦𝑚 (𝑡)
𝑝 𝑡 = 𝑝ҧ + 𝐾𝑐 𝑒 𝑡 + න 𝑒 𝑡 𝑑𝑡 + 𝜏𝐷
𝜏𝐼 0 𝑑𝑡
 Cont’d …
Setpoint weighting makes a more flexible control algorithm, that also
eliminates the proportional kick due to the step change in setpoint.
A modified PID algorithm defined by a different error in each term.
1 𝑡 ∗ ∗
𝑑𝑒𝐷 (𝑡)
𝑝 𝑡 = 𝑝ҧ + 𝐾𝑐 𝑒𝑃 𝑡 + න 𝑒 𝑡 𝑑𝑡 + 𝜏𝐷
𝜏𝐼 𝑠 0 𝑑𝑡
Where;
𝑒𝑃 𝑡 ≜ 𝛽𝑦𝑠𝑝 𝑡 − 𝑦𝑚 𝑡
𝑒 𝑡 ≜ 𝑦𝑠𝑝 𝑡 − 𝑦𝑚 𝑡
𝑒𝐷 𝑡 ≜ 𝛾𝑦𝑠𝑝 𝑡 − 𝑦𝑚 𝑡
This control algorithm is known as the parallel PID controller with
proportional and derivative mode weighting, or as the beta-gamma
controller.
To eliminate derivative kick, 𝛾 is set to zero and
To eliminate proportional kick, 𝛽 is set to zero.
 Cont’d …
Reverse acting controller
For proportional control, when 𝐾𝑐 > 0, (Fig. a) the controller output,
𝑝 𝑡 increases as the input signal decreases, 𝑦𝑚 .
𝑝 𝑡 − 𝑝ҧ = 𝐾𝑐 𝑦𝑠𝑝 𝑡 − 𝑦𝑚 𝑡
Direct acting controller
When 𝐾𝑐 < 0, (Fig. b) the controller output, 𝑝 𝑡 increases as the input
signal increases, 𝑦𝑚 .
 Cont’d …
Automatic/Manual control modes
Manual mode of controller operation situations are considered during a
plant startup, shutdown or emergency.
A manual/automatic switch or software equivalent is used to transfer
between the modes.
Abrupt changes are prevented by using bump-less transfers.
On-off controllers
Ideal on-off controller has only two possible values
𝑝𝑚𝑎𝑥 𝑖𝑓 𝑒 ≥ 0
𝑝 𝑡 =
𝑝𝑚𝑖𝑛 𝑖𝑓 𝑒 < 0
On-off control is also known as two-position or bang-bang control.
On-off control can be considered a proportional control special case with
very high gain.
They are simple and inexpensive feedback controllers, commonly used as
thermostats in home heating and domestic refrigerators.
Disadvantages: they can produce excessive wear on control valves, but have
less effect for solenoid valves.
 Cont’d …
Typical responses of feedback control systems
The behavior of a controlled process after a step change in a disturbance variable occurs.
The nature of response depends on the choice of controller settings and the process dynamics.
Proportional control
✓Speeds and reduces offset of the process
✓Increased gain 𝐾𝑐 , makes the process response less sluggish or slow
✓Too large gain results in undesirable oscillation and unstable response
Integral control
✓Eliminates offset but makes oscillatory response
✓Increased integral time makes the control sluggish
✓Large values of 𝜏𝐼 , results in very a slow response
Derivative control
✓Reduces oscillation degree and the response time
✓Increased derivative time improves response time, reduces deviation and oscillation.
✓But too large 𝜏𝐷 , amplifies noise and results in oscillatory response.
 Cont’d …
The 𝐶𝑉 (𝑦) represents deviation from initial steady state value.
 PID controller tuning techniques
The procedure of adjusting controller settings to achieve the desired
closed-loop system performance is known as controller tuning.
Closed loop systems performance criteria
Feedback control system ensures that the closed loop system has
desirable dynamic and steady state response characteristics.
✓The closed loop system must be stable
✓Provide good disturbance rejection
✓Provide Good set point tracking
✓Eliminates offset (steady state error)
✓Avoids excessive control action
✓Robust control system (insensitive to changes in process conditions)
It is not possible to achieve all the goals simultaneously, since they
involve inherent conflicts and trade offs.
 Cont’d …
Model based design methods
❑Direct synthesis method
Controller design is based on a process model and a desired closed
loop transfer function.
Block diagram of a standard feedback control system
 Cont’d …
Closed loop transfer function for set point changes
𝑌 𝐾𝑚 𝐺𝑐 𝐺𝑣 𝐺𝑝
=
𝑌𝑠𝑝 1 + 𝐺𝑐 𝐺𝑣 𝐺𝑝 𝐺𝑚
Simplifying with, 𝐺 ≜ 𝐺𝑣 𝐺𝑝 𝐺𝑚 𝑎𝑛𝑑 𝐺𝑚 = 𝐾𝑚
𝑌 𝐺𝑐 𝐺
=
𝑌𝑠𝑝 1 + 𝐺𝑐 𝐺
Solving for the feedback controller gives
1 𝑌Τ𝑌𝑠𝑝
𝐺𝑐 =
𝐺 1 − 𝑌Τ𝑌𝑠𝑝
Derivation of the design equation
𝑌
Replacing the unknown actual process, 𝐺 𝑏𝑦 𝑎 𝑚𝑜𝑑𝑒𝑙 𝐺෨ 𝑎𝑛𝑑 𝑏𝑦 𝑎 𝑑𝑒𝑠𝑖𝑟𝑒𝑑
𝑌𝑠𝑝

𝑐𝑙𝑜𝑠𝑒𝑑 − 𝑙𝑜𝑜𝑝 𝑡𝑟𝑎𝑛𝑠𝑓𝑒𝑟 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛, 𝑌Τ𝑌𝑠𝑝 we get

𝑑
1 𝑌Τ𝑌𝑠𝑝 𝑑
𝐺𝑐 = −−−−−−−−− − 𝟑
𝐺෨ 1 − 𝑌Τ𝑌𝑠𝑝
𝑑
 Cont’d …
Desired closed-loop transfer function
DS controller performance strongly depends on the desired transfer function,
𝑌Τ𝑌𝑠𝑝 𝑑 specification.
For ideal situation, set-point changes with out any error and is called perfect
control.
𝑌Τ𝑌𝑠𝑝 𝑑 = 1
• The first order model, for a process with out time delay
𝑌 1
=
𝑌𝑠𝑝 𝑑 𝜏𝑐 𝑠 + 1
𝑤ℎ𝑒𝑟𝑒 𝜏𝑐 = 𝑑𝑒𝑠𝑖𝑟𝑒𝑑 𝑐𝑙𝑜𝑠𝑒𝑑 𝑙𝑜𝑜𝑝 𝑡𝑖𝑚𝑒 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡
The model has unity steady-state gain and no offset for set point changes.
Solving for 𝐺𝑐 using the controller design equation
1 𝑌Τ𝑌𝑠𝑝
𝑑
𝐺𝑐 =
𝐺෨ 1 − 𝑌Τ𝑌𝑠𝑝
𝑑
1 1
𝐺𝑐 = 𝑤ℎ𝑒𝑟𝑒 𝜏𝑐 = 𝑡𝑢𝑛𝑖𝑛𝑔 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟 of the integral control
෨
𝐺 𝜏𝑐 𝑠
 Cont’d …
• Process transfer function with known time delay, 𝜃
𝑌 𝑒 −𝜃𝑠
=
𝑌𝑠𝑝 𝑑 𝜏𝑐 𝑠 + 1
Physically, the response to setpoint change at 𝑡 = 0 𝑖𝑠 𝑖𝑚𝑝𝑜𝑠𝑠𝑖𝑏𝑙𝑒
𝑏𝑒𝑓𝑜𝑟𝑒 𝑡 = 𝜃.
Solving the controller equation
1 𝑒 −𝜃𝑠
𝐺𝑐 = −−−−−−−−− − 𝟒
𝐺෨ 𝜏𝑐 𝑠 + 1 − 𝑒 −𝜃𝑠

The Direct Synthesis method is also known as the lambda tuning

method, where 𝜏𝑐 𝑖𝑠 𝑟𝑒𝑝𝑙𝑎𝑐𝑒𝑑 𝑏𝑦 𝜆.
 Cont’d …
Derivation of PID controllers for simple process models
The denominator time delay term is replaced with Taylor series
expansion, and the controller equation 𝟒 becomes
1 𝑒 −𝜃𝑠
𝐺𝑐 = 𝑤ℎ𝑒𝑟𝑒, 𝑒 −𝜃𝑠 ≈ 1 − 𝜃𝑠
𝐺෨ 𝜏𝑐 + 𝜃 𝑠
First-Order-Plus-Time-Delay (FOPTD) Model
Assuming perfect model, 𝐺෨ = 𝐺.
Consider the standard FOPTD model
𝐾𝑒 −𝜃𝑠
𝐺෨ 𝑠 =
𝜏𝑠 + 1
 Cont’d …
Substituting, 𝐺෨ in to the controller equation gives a PI controller
1 𝑒 −𝜃𝑠 1 𝑒 −𝜃𝑠
𝐺𝑐 = = −𝜃𝑠
𝐺෨ 𝜏𝑐 + 𝜃 𝑠 𝑒 𝜏𝑐 + 𝜃 𝑠
𝜏𝑠 + 1
1
𝐺𝑐 = 𝐾𝑐 1 +
𝜏𝐼 𝑠
Controller settings
1 𝜏
𝐾𝑐 = 𝑎𝑛𝑑 𝜏𝐼 = 𝜏
𝐾 𝜃 + 𝜏𝑐
If the product 𝐾𝑐 𝐾 𝑟𝑒𝑚𝑎𝑖𝑛𝑠 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 the characteristic equation and
stability characteristics of the closed-loop system doesn’t change.
Second-Order-Plus-Time-Delay (SOPTD) Model
𝐾𝑒 −𝜃𝑠
𝐺෨ 𝑠 =
𝜏1 𝑠 + 1 𝜏2 𝑠 + 1
 Cont’d …
Substituting, 𝐺෨ in to the controller equation gives a PID controller
1 𝑒 −𝜃𝑠 1 𝑒 −𝜃𝑠
𝐺𝑐 = =
෨
𝐺 𝜏𝑐 + 𝜃 𝑠 𝐾𝑒 −𝜃𝑠 𝜏𝑐 + 𝜃 𝑠
𝜏1 𝑠 + 1 𝜏2 𝑠 + 1
1
𝐺𝑐 = 𝐾𝑐 1 + + 𝜏𝐷 𝑠
𝜏𝐼 𝑠
Controller settings
1 𝜏1 + 𝜏2 𝜏1 𝜏2
𝐾𝑐 = , 𝜏𝐼 = 𝜏1 + 𝜏2 𝑎𝑛𝑑 𝜏𝐷 =
𝐾 𝜃 + 𝜏𝑐 𝜏1 + 𝜏2
Example; use the DS design method to calculate PID controller settings for the
process:
2𝑒 −𝑠
𝐺=
10𝑠 + 1 5𝑠 + 1
Consider the desired, 𝜏𝑐 = 1,3, 𝑎𝑛𝑑 10.
a) Evaluate the controllers for a unit step change in the setpoint. Assume the
process model is perfect 𝐺෨ = 𝐺.
 Cont’d …
Solution;
Controller settings
1 𝜏1 + 𝜏2 𝜏1 𝜏2
𝐾𝑐 = , 𝜏𝐼 = 𝜏1 + 𝜏2 𝑎𝑛𝑑 𝜏𝐷 =
𝐾 𝜃 + 𝜏𝑐 𝜏1 + 𝜏2

Controller
𝜏𝑐 = 𝟏 𝜏𝑐 = 𝟑 𝜏𝑐 = 𝟏𝟎
settings

𝐾𝑐 3.75 1.88 0.682

𝜏𝐼 15 15 15
𝜏𝐷 3.33 3.33 3.33
PID Controller TF
1 1
𝐺𝑐 = 𝐾𝑐 1+ + 𝜏𝐷 𝑠 = 3.75 1 + + 3.33𝑠 for 𝜏𝑐 = 𝟏
𝜏𝐼 𝑠 15𝑠
 Cont’d …
The values of 𝐾𝑐 decrease as 𝜏𝑐 increases and the response become
sluggish.
See the figure, 𝜏𝑐 = 10 𝑎𝑛𝑑 𝐾𝑐 = 0.682.
 Cont’d …
❑ Internal model control (IMC)
IMC is model-based like the DS method where the controller settings are
calculated from the assumed process model.
Advantages of IMC approach
Allows model uncertainty and
Allows tradeoffs between performance and robustness
IMC block diagram and conventional feedback control
 Cont’d …
Notations;
o 𝐺෨ = process model
o P = controller output
o 𝑌෨ = model response
o 𝑌 = actual response
o 𝐺𝑐∗ = IMC controller
Relation between 𝑮𝒄 and 𝑮∗𝒄
Gc∗
Gc =
1 − Gc∗ 𝐺෨
Closed-loop relation of the IMC block diagram
Gc∗ 𝐺 1 − Gc∗ 𝐺෨
Y= Ysp + 𝐷
1+ Gc∗ 𝐺 − 𝐺෨ 1+ Gc∗ 𝐺 − 𝐺෨
෩ and the relation becomes
For a perfect model, G = G
෨ −−−−−−−−− − 𝟓
Y = Gc∗ 𝐺Ysp + 1 − Gc∗ 𝐺𝐷
 Cont’d …
Design of the IMC controller
i. The process model is factored as
𝐺෨ = 𝐺෨+ 𝐺෨−
Where 𝐺෨+ contains,
o Any time delays and right half plane zeros
o Steady state gain equal to one
ii. The IMC controller is specified as
1
Gc∗ = 𝑓
෨
𝐺−
Where 𝑓 is a typical low pass filter with steady state gain of one
1
𝑓= 𝑟
𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑖𝑛𝑡𝑒𝑔𝑒𝑟, 𝑟 = 1
(𝜏𝑐 𝑠 + 1)
 Cont’d …
෩.
Substituting the IMC controller in to equation 𝟓 , consider the ideal situation, G = G
෨
Y = Gc∗ 𝐺Ysp + 1 − Gc∗ 𝐺𝐷
Y = 𝐺෨+ 𝑓Ysp + (1 − 𝑓𝐺෨+ )𝐷
Then the closed loop transfer function for set-point changes is
𝑌
= 𝐺෨+ 𝑓
Ysp
Example; use the IMC design method to design controllers for the FOPTD model.
Assume that 𝑓 is specified with 𝑟 = 1 and consider the approximations for the time
delay.
a. Pade approximation:
𝜃
1− 𝑠
𝑒 −𝜃𝑠 ≅ 2
𝜃
1+ 𝑠
2
 Cont’d …
Solution; Substituting the approximation in to FOPTD model gives

𝜃
𝐾𝑒 −𝜃𝑠 𝐾 1− 𝑠
෨ 2
𝐺 𝑠 = =
𝜏𝑠 + 1 𝜃
1 + 𝑠 (𝜏𝑠 + 1)
2
Factors of the model, 𝐺෨ = 𝐺෨+ 𝐺෨−
𝜃 𝐾
𝐺෨+ = 1 − 𝑠 𝑎𝑛𝑑 𝐺෨− =
2 𝜃
1 + 𝑠 (𝜏𝑠 + 1)
2
Then the IMC controller can be obtained from
1 1
Gc∗ = 𝑓 𝑎𝑛𝑑 𝑓 =
෨
𝐺− (𝜏𝑐 𝑠 + 1)𝑟
 Cont’d …
𝜃
1 + 𝑠 (𝜏𝑠 + 1)
∗ 2
Gc =
𝐾 𝜏𝑐 𝑠 + 1
Then the equivalent controller is obtained from

𝜃
Gc∗ 1 + 𝑠 (𝜏𝑠 + 1)
2
Gc = ∗෨
=
1 − Gc 𝐺 𝜃
𝐾 𝜏𝑐 + 𝑠
2
This can be rearranged into the PID controller in the parallel form
1
𝐺𝑐 = 𝐾𝑐 1+ + 𝜏𝐷 𝑠
𝜏𝐼 𝑠
 Cont’d …
The controller settings will become
𝜏
1 2 𝜃 +1 𝜃 𝜏
𝐾𝑐 = 𝜏 , 𝜏𝐼 = + 𝜏 𝑎𝑛𝑑 𝜏𝐷 =
𝐾2 𝑐 +1 2 𝜏
𝜃 2 +1
𝜃
b. First order Taylor series approximation of the time delay term
𝑒 −𝜃𝑠 ≅ 1 − 𝜃𝑠
Solution; FOPTD model of the internal model
𝐾𝑒 −𝜃𝑠 𝐾(1 − 𝜃𝑠)
𝐺෨ 𝑠 = =
𝜏𝑠 + 1 (𝜏𝑠 + 1)
Factors of the model, 𝐺෨ = 𝐺෨+ 𝐺෨−
𝐾
𝐺෨+ = 1 − 𝜃𝑠 𝑎𝑛𝑑 𝐺෨− =
𝜏𝑠 + 1
 Cont’d …
The IMC controller can be calculated as
1 1
Gc∗ =
𝐺෨− (𝜏𝑐 𝑠 + 1)𝑟
𝜏𝑠 + 1
Gc∗ =
𝐾 𝜏𝑐 𝑠 + 1
Then the equivalent process controller is

𝜏𝑠 + 1
Gc∗ 𝐾 𝜏𝑐 𝑠 + 1 𝜏𝑠 + 1
Gc = ∗ ሚ = =
1 − Gc 𝐺 𝜏𝑠 + 1 𝐾(1 − 𝜃𝑠) 𝐾 𝜏𝑐 + 𝜃 𝑠
1− ×
𝐾 𝜏𝑐 𝑠 + 1 (𝜏𝑠 + 1)
Rearranging into the standard PI controller form
𝐺𝑐 = 𝐾𝑐 1 + 1Τ𝜏𝐼 𝑠
 Cont’d …
Then the resulting controller settings are
1 𝜏
𝐾𝑐 = 𝑎𝑛𝑑 𝜏𝐼 = 𝜏
𝐾 𝜃 + 𝜏𝑐
IMC tuning relations
Controller tuning relations are analytical expressions used to derive PID
controller settings for various transfer function models.
Different tuning relations can be derived based on the type of low pass filter
and time delay approximation that are selected.
IMC filter for models with integrating elements
2𝜏𝑐 − 𝐶 𝑠 + 1 𝑑𝐺෨+
𝑓= 2
𝑤ℎ𝑒𝑟𝑒 𝐶 = อ
(𝜏𝑐 𝑠 + 1) 𝑑𝑥
𝑠=0
Example; a process model for a liquid storage system
𝐾𝑒 −7.4𝑠
𝐺෨ 𝑠 =
𝑠
 Cont’d …
Calculate PI and PID controller settings for 𝐾 = 0.2 𝑎𝑛𝑑 𝜏𝑐 = 8 𝑎𝑛𝑑 15.
a. Assume that 𝐺𝑑 = 𝐺෨ for unit step changes in the setpoint and disturbance.
Solution;
This is an integrating process model
𝐾𝑒 −7.4𝑠 0.2(1 − 7.4𝑠)
𝐺෨ 𝑠 = =
𝑠 𝑠
Taylor series is used for the time-delay approximation.
Factors of the model, 𝐺෨ = 𝐺෨+ 𝐺෨−
0.2
𝐺෨+ = 1 − 7.4𝑠 𝑎𝑛𝑑 𝐺෨− =
𝑠
The IMC controller
∗
1 0.2 2𝜏𝑐 + 7.4 𝑠 + 1
Gc = 𝑓= 𝑤ℎ𝑒𝑟𝑒 𝐶 = −7.4
𝐺෨− 𝑠 (𝜏𝑐 𝑠 + 1)2
 Cont’d …
Then the equivalent process controller is

0.2 2𝜏𝑐 + 7.4 𝑠 + 1

Gc∗ 𝑠 (𝜏𝑐 𝑠 + 1)2
Gc = ∗ ǁ=
1 − Gc 𝐺 0.2 2𝜏𝑐 + 7.4 𝑠 + 1 0.2(1 − 7.4𝑠)
1−
𝑠 (𝜏𝑐 𝑠 + 1)2 𝑠

Rearranging into a standard PI controller form

2𝜏𝑐 + 7.4 1 1
Gc = 2
1+
𝜏𝑐 + 7.4 2𝜏𝑐 + 7.4 𝑠
The resulting controller settings are
2𝜏𝑐 + 7.4
𝐾𝑐 = 2
𝑎𝑛𝑑 𝜏𝐼 = 2𝜏𝑐 + 7.4
𝜏𝑐 + 7.4
• For the process model with PID controller settings Pad𝑒ư time-delay
approximation is used.
 Cont’d …
PI controller settings
1 2𝜏𝑐 + 7.4
𝐾𝑐 = 2
𝑎𝑛𝑑 𝜏𝐼 = 2𝜏𝑐 + 7.4
0.2 𝜏𝑐 + 7.4
PID Controller settings
1 2𝜏𝑐 + 7.4 7.4𝜏𝑐 + 13.7
𝐾𝑐 = 2
, 𝜏𝐼 = 2𝜏𝑐 + 7.4 𝑎𝑛𝑑 𝜏𝐷 =
0.2 𝜏𝑐 + 3.7 2𝜏𝑐 + 7.4
Controller
settings 𝐾𝑐 𝜏𝐼 𝜏𝐷
(𝜏𝑐 = 𝟖)
PI 0.493 23.4 -
PID 0.857 23.4 3.12
 Cont’d …
Closed loop control response
The response becomes more sluggish and less oscillatory when,
𝜏𝑐 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑠.
Maximum deviation is larger after the disturbance.
PID controller provides better disturbance response and shorter settling
time.
 Cont’d …
On-line controller tuning
During control system design preliminary controller settings are
specified for the process.
After a controller is installed, the preliminary settings are adjusted to
achieve satisfactory control.
This on-site adjustment is known as on-line tuning, or field tuning or
controller tuning.
General observations
✓Controller tuning involves a tradeoff between performance and
robustness.
✓Controller settings don’t have to be precisely determined.
✓For most plants, it is not feasible to manually tune each controller.
✓Diagnostic techniques are used for monitoring the control system
performance.
 Cont’d …
❑ Continuous cycling method
The continuous cycling method for controller tuning is developed by Ziegler-Nichols, 1942.
It is based on the trial-and-error procedure
i. After the process steady state (or approximately)
Set the derivative, 𝜏𝐷 = 0 and integral control action,
𝜏𝐼 = 𝑙𝑎𝑟𝑔𝑒𝑠𝑡 𝑝𝑜𝑠𝑠𝑖𝑏𝑙𝑒 𝑣𝑎𝑙𝑢𝑒.
ii. Then set 𝐾𝑐 = 𝑠𝑚𝑎𝑙𝑙 𝑣𝑎𝑙𝑢𝑒
iii. Gradually, increase 𝐾𝑐 𝑢𝑛𝑡𝑖𝑙 𝑐𝑜𝑛𝑡𝑖𝑛𝑢𝑜𝑢𝑠 𝑐𝑦𝑐𝑙𝑖𝑛𝑔 𝑜𝑟
𝑠𝑢𝑠𝑡𝑎𝑖𝑛𝑒𝑑 𝑜𝑠𝑐𝑖𝑙𝑙𝑎𝑡𝑖𝑜𝑛 𝑜𝑐𝑐𝑢𝑟𝑠.
The gain value at the condition is called ultimate gain, 𝐾𝑐 = 𝐾𝑐𝑢 and
The period is known as ultimate period, 𝑃𝑢 .
iv. Calculate the PID controller settings using the Ziegler-Nichols (Z-N) tuning
relations.
v. Evaluate the Z-N controller settings by observing the closed loop response
after a small set-point change.
 Cont’d …
Experimental determination of the ultimate gain

Disadvantages of the continuous cycling method

• It’s a time-consuming procedure, for several trials and slow process dynamics
• The testing could result in unstable operation
• It is not applicable to integrating processes, which are unstable at both high and
low values of 𝐾𝑐 .
• Ultimate gain doesn’t exist for first-order and second-order models with out time
delays.
 Cont’d …
PID settings in the parallel form reduce the oscillatory response and large
overshoots, that occur during set-point changes.
❑ Tyreus-Luyben tuning method
Tyreus-Luyben tuning method procedure is quite similar to the Ziegler-Nichols
method but gives more conservative controller settings.
That is Tyreus-Luyben tuning method provides a higher closed-loop damping
coefficient and large time constant.
It is more suitable for applications that require robustness, e.g. chemical
process control.
Controller settings for T-L method
 Cont’d …
Controller settings for continuous cycling method

Example; For the process model compare PID controllers with the
following settings.
2𝑒 −𝑠
𝐺=
10𝑠 + 1 5𝑠 + 1
a. Ziegler-Nichols’ settings
b. Tyreus-Luyben settings and
c. DS tuning method, for 𝜏𝑐 = 3.
 Cont’d …
Solution;
a. Ziegler-Nichols’ settings
Ultimate gain and period are determined using trial-and-error methods.
• 𝐾𝑐 from the process closed loop characteristic equation
1 + 𝐾𝑐𝑢 𝐺 = 0
10𝑠 + 1 5𝑠 + 1 + 2𝐾𝑐𝑢 1 − 𝑠 = 0 𝑠𝑒𝑡𝑡𝑖𝑛𝑔 𝑠 = 𝑗𝜔
𝜔2 + 0.3 − 0.04𝐾𝑐𝑢 𝑗𝜔 + 0.02 + 0.04𝐾𝑐𝑢 = 0
2𝜋
• From the imaginary part, 𝐾𝑐𝑢 = 7.88 𝑎𝑛𝑑 𝑡ℎ𝑒𝑛 𝑃𝑢 = = 11.66
𝜔
𝑤ℎ𝑒𝑟𝑒, 𝜔 = 0.566.
• Therefore, based on the Z-N tuning relations PID controller settings are
𝐾𝑐 = 0.6𝐾𝑐𝑢 = 4.5
𝑃𝑢 𝑃𝑢
𝜏𝐼 = = 5.6 𝑎𝑛𝑑 𝜏𝐷 = = 1.4
2 8
 Cont’d …
b. Tyreus-Luyben settings
Solution;
Substituting the values in to the T-L tuning relations, gives us the PID
controller settings.
𝐾𝑐 = 0.45𝐾𝑐𝑢 = 3.55
𝑃𝑢
𝜏𝐼 = 2.2𝑃𝑢 = 25.8 𝑎𝑛𝑑 𝜏𝐷 = = 1.84
6.3
c. The DS method PID controller settings for SOPTD model
𝐾𝑐 = 1.88
𝜏𝐼 = 15 𝑎𝑛𝑑 𝜏𝐷 = 3.33
The Z-N controller provides the best control for disturbance and the
worst for the set-point changes.
The T-L and DS controllers result in satisfactory set-point responses but
sluggish disturbance responses.
 Cont’d …

❑ Step test method

A second on-line tuning technique developed by Ziegler-Nichols based on a
single step test.
The controller settings are based on the process reaction curve, that is the
open-loop step response.
This tuning technique is known as the step test method or the process
reaction curve method.
 Cont’d …
Disadvantages
The experimental test is performed under open-loop conditions with out
corrective action for disturbances.
For a nonlinear process, the test results can be sensitive to the magnitude
and direction of the step change, due to process nonlinearities, noise and
disturbances.
The method is not applicable to open loop unstable processes.
For analog controllers the method tends to be sensitive to controller
calibration errors but the continuous cycling method is less sensitive.
Self-regulating (b) and non-self-regulating (a) reaction curves
 Cont’d …
Example; consider the stirred tank blending process step test.
Controller output changes from 30% 𝑡𝑜 43% (a)
The resulting process reaction curve (b)

Solution; The FOPTD model can be developed from the process reaction
curve.
Slope of the tangent line
 Cont’d …
Normalized slope of the tangent line

The model parameters can be calculated as

Where, 𝜏 = 5.93 𝑖𝑠 𝑡ℎ𝑒 𝑎𝑝𝑝𝑎𝑟𝑒𝑛𝑡 𝑡𝑖𝑚𝑒 𝑑𝑒𝑙𝑎𝑦

The resulting empirical process model and the PI controller settings can be
calculated using other tuning relations.