Control Design by Pole Placement 0
Control Design by Pole Placement 0
1. Introduction to control
•If a system does not have desired response ( peak time too large,
time response large, much oscillation, large overshoot),
then it can be modified with control.
• 𝒕𝒅 (Delay Time): is the time needed for the response to reach half of its final value the very
first time.
• 𝒕𝒓 (Rise Time): is the time required for the response to rise from 10% to 90%.
• 𝒕𝒑 (Peak Time): is the time required for the response to reach the first peak of the
overshoot.
• 𝑴𝒑 (Maximum percent Overshoot):is the maximum peak value of the response curve.
• 𝒕𝒔 (Settling Time): is the time required for the response curve to reach and stay within 2%
of the final value. Dr Nassim Ammour CEN455 King Saud University 2
1.1 Control architecture (feed-forward control)
• Open-loop control (feed-forward control).
If we ignore D(s)
Y(s) = P(s) C(s) R(s) .
-This is very simple and cheap (no feedback so no sensor needed, no software to interact with sensor, no signal processing
to use with the sensor).
-Not robust. (model error -> bad controller C(s)= 1/P(s) is not robust, disturbances change the plant , U(s) is not the only input)
-More robust:
by measuring the output we can see the
fact of no convergence of the output to the input (due
to bad modeling or disturbances), and the controller
changes the control signal to improve the system behavior.
• Approach to Control design: Translate engineering specifications into control requirements, then design a controller to
meet those specifications.
1. Determine what the system should do and how to do it (design specification),
2. Determine the controller or compensator configuration relative to how it is connected to the controlled process.
3. Determine the parameter values of the controller to achieve the design goals.
• Example: example in cars, increases the comfort and safety of driving a car or reduces the fuel consumption and exhaust gas
emissions.
• Examples transient specifications (system step response): Time constant𝜏 (speed of response of a first order
system), over shoot 𝑀𝑝 , peak time 𝑡𝑝 , settle time 𝑡𝑠 and rise time 𝑡𝑟 . (feasible analytically only for second-order systems, or
systems that can be approximated by a second-order system).
• Example steady-state specifications: Typically it is required that steady-state error be less than some amount, for
example ess <0.02. We can use the final value theorem for different types of reference inputs ess = 𝑙𝑖𝑚𝑠→0 𝑠𝐸(𝑠)
• This is a nonlinear controller which is very simple and it does not need any design. The On-Off Controller is defined as:
𝑈𝑚𝑎𝑥 𝑖𝑓 𝑒 𝑡 > 0
𝑢 𝑡 =
𝑈𝑚𝑖𝑛 𝑖𝑓 𝑒 𝑡 < 0
Where : 𝑒 𝑡 = 𝑟 𝑡 − 𝑦 𝑡 is the tracking error and u(t) is the applied control system.
• Control signal u(t) can have only two possible values. (fully-on, High u(t) =Umax or fully-off, low u(t) =Umin) depending if error
is positive or negative. The main idea in this way of control, which only two control levels achieve desired value of the
controlled variable in shortest time possible.
• The control signal will oscillate between two levels (high frequency, damage the actuators) and good tracking accuracy is
achieved.
• The control signal never reach the zero value which means the process consumes energy (cost is high).
𝑈𝑚𝑎𝑥 𝑖𝑓 𝑒 𝑡 > +𝜀
𝑢 𝑡 =
𝑈𝑚𝑖𝑛 𝑖𝑓 𝑒 𝑡 < −𝜀
Dr Nassim Ammour CEN455 King Saud University 6
2.1 Advantages and Drawbacks of ON-OFF controller
Control the water level system by two-position controller. The transfer function 1
of the water level system is: ℎ = 𝑢(𝑡)where h(t) is the water level system. 0.8
SOLUTION 0.6
• Simulation using: SIMULINK : With eps=0.1 and Umax=5 and Umin=-5, 0.4
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (sec)
-2
-4
-6
Or 𝑢 = −5𝑣 when 𝑒 𝑡 < 1; 𝑦𝑟𝑒𝑓 − 𝑦 𝑡 < 1 𝑜𝑟 𝑦 𝑡 > 𝑦𝑟𝑒𝑓 − 1 = 7 Y(t) > 𝑦𝑟𝑒𝑓 − 𝜀
The problem is when 7 < 𝑦 𝑡 < 9, the control signal keeps the previous value.
10
Output
Control Signal
-5
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (s)
𝜔𝑛 2
2nd order system transfer function: 𝐺(𝑠) =
𝑠 2 +2𝜉 𝜔𝑛 +𝜔𝑛 2 𝜉 is Damping Ratio
𝜔𝑛 is the Natural frequency
The poles of the system: 𝑃1,2 = 𝜎 ± 𝑗𝜔𝑑 = −𝜉𝜔𝑛 ± 𝑗𝜔𝑛 1 − 𝜉 2
𝜋−𝛽 4 1+0.7 𝜉
Rise Time: 𝑇𝑟 = Settling Time: 𝑇𝑠 = Delay Time: 𝑇𝑑 =
𝜔𝑑 𝜉 𝜔𝑛 𝜔𝑛
𝜋𝜉
𝜋 𝜋 −
Peak Time: 𝑇𝑝 = = 1−𝜉2
𝜔𝑑 𝜔𝑛 1−𝜉 2 Maximum Overshoot: 𝑀𝑝 = 𝑒 × 100%
Controller Design
• Choose controller gains (for ex: 𝐾𝑝 , 𝐾𝐼 , and 𝐾𝐷 ) so that poles of the close-loop system meet given requirements.
• If the system is simple ( canonical first order system or canonical second order system) can employ algebra (precise
relationships between poles and the shape of the step response).
• If the system is non-canonical (higher order system) the relationships is not exact. One idea is to place one or two dominant
poles to meet our requirements and place the rest of poles to be significantly faster, so that we can one sense neglect the effect of
the dynamics for the other transient response. Dr Nassim Ammour CEN455 King Saud University 10
3.1 Example
• Find the close-loop transfer function for a cruise control system with PI controller (How to design a PI controller, in other words
how to choose the gains proportional gain 𝐾𝑝 and integral gain 𝐾𝐼 for a first order system).
Forward
Controller Plant
Loop
1 1
𝑉(𝑠) 𝑓𝑜𝑟𝑤𝑎𝑟𝑑 (𝐾𝑝 + 𝐾𝐼 𝑠 𝑚𝑠 + 𝑏 )
)( 𝐾𝑝 𝑠 + 𝐾𝐼 𝐾𝑝 𝑠 + 𝐾𝐼
= = = =
𝑉𝑑𝑒𝑠 (𝑠) 1 + 𝑙𝑜𝑜𝑝 1 + (𝐾 + 𝐾 1)( 1 ) 𝑠 𝑚𝑠 + 𝑏 + 𝐾𝑝 𝑠 + 𝐾𝐼 𝑚 𝑠 2 + 𝑏 + 𝐾𝑝 𝑠 + 𝐾𝐼
𝑝 𝐼 𝑠 𝑚𝑠 + 𝑏
• we can see that the two gains affect the coefficients of the denominator.
• with two parameters (𝐾𝑝 and 𝐾𝐼 ) can place poles of this second order system anywhere we like (2 d.o.f. control)
• To see the effect of changing the control gains we match this denominator to the canonical second order system
denominator :
2 2 2
𝑏 + 𝐾𝑝 𝐾𝐼
𝑠 + 2 𝜉 𝜔𝑛 𝑠 + 𝜔𝑛 = 𝑠 + 𝑠+
𝑚 𝑚 𝜔𝑛 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑠
𝜎 constant Im
𝑏 + 𝐾𝑝 𝐾𝑝 will directly affect 𝜎 which is the real
𝜉 𝜔𝑛 = 𝜎 = part of our poles. 𝜎 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑠 𝜔𝑑
2𝑚 𝜔𝑛 constant 𝜔𝑛
2
𝐾𝐼 𝐾𝐼 𝐾 directly affects 𝜔 which is the 𝛽
𝜔𝑛 = 𝜔𝑛 = 𝐼 𝑛
𝜎
𝑚 𝑚 naturel frequency. Re
4 𝜋
𝑡𝑠 = 𝜎 , 𝑡𝑝 = 𝜔 , 𝑀𝑝 = 𝑒 −𝜉𝜋/ 1−𝜉 2
𝑑
• Increasing 𝐾𝑝 (holding 𝐾𝐼 constant ) ->𝜎 increases (𝑡𝑠 smaller), 𝜔𝑛 = constant (the poles move on a circle of radius 𝜔𝑛 ),
𝜔𝑑 decreases (𝑡𝑝 increases), 𝛽 decreases → 𝜉 larger less oscillations (towards real poles, 𝑀𝑝 decreases).
• Increasing 𝐾𝐼 (holding 𝐾𝑝 constant ) -> 𝜎 constant (𝑡𝑠 𝑖𝑛𝑐ℎ𝑎𝑛𝑔𝑒𝑑) ->𝜔𝑑 increases (𝑡𝑝 decreases) ->𝜉 smaller (𝑀𝑝 increases)
Dr Nassim Ammour CEN455 King Saud University 12
Example (Continued)
• Let m = 5 and b = 1. Choose 𝐾𝑃 and 𝐾𝐼 to achieve setting time of 2 seconds and a peak time of 1 second.
• the close-loop transfer function is:
𝑉(𝑠) 1 𝐾𝑝 𝑠 + 𝐾𝐼 𝜔𝑛2 1 𝐾𝑃 𝑠 + 𝐾𝐼
= ⟹ 𝐶𝐿𝑇𝐹 = =
𝑉𝑑𝑒𝑠 (𝑠) 𝑚𝑏 + 𝐾𝑝 𝐾𝐼
𝑠 2 + 2𝜉𝜔𝑛 𝑠 + 𝜔𝑛2 5 𝑠 2 + 1 + 𝐾𝑃 𝑠 + 𝐾𝐼
2
𝑠 + 𝑠+𝑚
𝑚 5 5
4 4 1 + 𝐾𝑃
𝑡𝑠 = ⟹ 𝜎 = = 2 Matching the coefficients 𝜎 = 𝜉𝜔 𝑛 = = 2 ⟹ 𝐾𝑃 = 2 10 − 1 = 19
𝜎 2 2 (5)
𝜋 Im
𝑡𝑝 = = 1 ⟹ 𝜔𝑑 = 𝜋 𝐾𝐼
Matching the coefficients 𝜔𝑛 = 𝜔𝑑 = 𝜋
𝜔𝑑 5 𝜔𝑛
Re
Recall that: 𝜔𝑛 = 22 + 𝜋 2 ≈ 3.72 𝜎 =2
𝜔𝑑 = 𝜔𝑛 1 − 𝜉 2 Or we can use geometry
𝐾𝐼 2
⟹ 𝜔𝑛 = ≈ 3.72 ⟹ 𝐾𝐼 ≈ 3.72 5 ≈ 69.3
5
• The close-loop system with this controller will not have settling time of 2 seconds and peak time of 1 second, because we don’t
have a canonical system (presence of a zero in the numerator). The calculated gains will be a good starting point.
• Plot region in complex plane where poles of a canonical 2nd order system must be located in order to achieve a settling time less
than 2 seconds and a peak time less than 1 second.
Im
𝜔𝑑 = 𝜋
4
𝑡𝑠 = < 2 ⟹ 𝜎 >2
𝜎
Region of choosing the poles Re
( Complex conjugate )
𝜋 - 𝜎 = −2
𝑡𝑝 = <1 ⟹ 𝜔𝑑 > 𝜋
𝜔𝑑
𝜔𝑑 = −𝜋
• One way to have the error using the transfer function of the input
𝑉𝑑𝑒𝑠 𝑠 and the output the error 𝐸(𝑠) :
𝐸(𝑠) 𝑓𝑜𝑟𝑤𝑎𝑟𝑑 1 𝑠(5𝑠 + 1)
𝐺 𝑠 = = = =
𝑉𝑑𝑒𝑠 𝑠 1 + 𝑙𝑜𝑜𝑝 19𝑠 + 69.3 𝑠 5𝑠 + 1 + 19𝑠 + 69.3
1+
𝑠(5𝑠 + 1)
• We can then find the steady-state error by applying the final value theorem
𝑠 5𝑠 + 1 1 1 1 1
𝑒𝑠𝑠 = lim 𝑠𝐸 𝑠 = lim 𝑠 = = ≈ 0.0144 (𝑖𝑛𝑝𝑢𝑡 𝑟𝑎𝑚𝑝𝑒: 𝑉𝑑𝑒𝑠 𝑠 = 2)
𝑠→0 𝑠→0 𝑠 5𝑠 + 1 + 19𝑠 + 69.3 𝑠 2 69.3 𝐾𝐼 𝑠
• By increasing the integral gain 𝐾𝐼 we reduce the steady-state error 𝑒𝑠𝑠 .
• The proportional gain 𝐾𝑃 does not appear to affect the steady state error.
• If we have had a condition on the steady-state error that conflicted with the condition on the peak time or the settling time we
may have to iterate in the design of our controller.
Dr Nassim Ammour CEN455 King Saud University 15
PID Controller (three-term Controller )
• The transfer function of the PID controller is given by: 𝑈(𝑠) 𝐾𝐼
𝐶 𝑠 = = 𝑘𝑝 + + 𝐾𝐷 𝑠
𝐸(𝑠) 𝑠
• PID controller is used to increase or to achieve the desired
performances in transient and steady state response. Integral of derivative
Proportional the error Of the error
To the error
• The controller input is the error (difference between the reference and
the actual output ) and the controller output is the control signal input to Where Kp is the proportional term, KI is the
the plant).
integral term, and KD is the derivative term.
• PID controllers are ubiquitous (very widely used in the industry ).
• Provide sophistication in that the control is based not only on the current error, but also the history of the error (integral
of error) and the anticipated future error (derivative of the error).
• Matching with the parameters of a 2nd order canonical system (standard form).
Increasing Kp makes 𝜔𝑛 larger (make the system
stiffer (rigid))
• The poles are all in the left side of the s-plane because all the coefficients of the characteristic equation have the same sign
• The final value of the system does not go to one (with P control we will have
Some amount of error for unit step reference).
𝐾𝑃
10 𝐾𝑃
We have the close-loop transfer function: 𝐶𝐿𝑇𝐹 = 𝑠2
+ 0.05 𝑠 = 10
𝐾𝑃 2 + 0.05 𝑠 + 𝐾𝑃
10 𝑠 10
1+ 2
𝑠 + 0.05 𝑠
𝐾𝑃 0.05
𝐶𝐿𝐶𝐸 = 𝑠 2 + 0.05 𝑠 + 10 Matching 0.05 = 2𝜉𝜔𝑛 = 2 0.7 𝜔𝑛 𝜔𝑛 = = 0.0357
2(0.7)
𝐶𝐿𝐶𝐸 𝑑𝑒𝑠𝑖𝑟𝑒𝑑 = 𝑠 2 + 2𝜉𝜔𝑛 𝑠 + 𝜔𝑛 2 𝐾𝑃 2
For 𝜉 = 0.7 10 = 𝜔𝑛
𝐾𝑃 = 10 (0.0357)2 = 0.0127
Dr Nassim Ammour CEN455 King Saud University 20
Proportional-Derivative Controller (PD)
• We will see the effect of adding a derivative action to our P controller
• With derivative control, do not have to wait for error to get large before control action becomes large, control anticipates the
error (by using the slope of the variation of the error and even error is relatively small the control signal increase before the error
gets large, so large control signal, and then improve damping).
• Never use D control by itself, amplifies noise (for D control we use PD or PID).
• The PD controller has a constant proportional to the error plus the proportional (constant KD) to
the differentiation of the error (the term s).
𝜔𝑛 = 4
For Poles at −8 ± 3.9 𝑖 (using 𝑠 2 + 2𝜉 𝜔𝑛 s + 𝜔𝑛 2 = 0) 𝑠 + 0.8 + 3.9𝑖 𝑠 + 0.8 − 3.9𝑖 = 0
𝜉 = 0.2
• The open-loop system poles are 0 and -1, it did not meet the desired 3.9
specifications, we need to add a controller to have a new system.
𝐷𝑒𝑠𝑖𝑟𝑒𝑑 𝑝𝑜𝑙𝑒𝑠 −0.8
PD controller Plant
𝑋𝑑 + 1 𝑋 -1
𝐾𝑃 + 𝑠𝐾𝐷 𝑝𝑙𝑎𝑛𝑡 𝑝𝑜𝑙𝑒𝑠
𝑠2 + 𝑠
−
1
𝐾𝑃 + 𝑠𝐾𝐷 ( 2 ) 𝐾𝑃 + 𝑠𝐾𝐷
𝐶𝐿𝑇𝐹 = 𝑠 + 𝑠 = 2 𝐶𝐿𝐶𝐸 = 𝑠 2 + 𝑠 + 𝐾𝑃 + 𝑠𝐾𝐷 = 0
1
1 + 𝐾𝑃 + 𝑠𝐾𝐷 ( 2 ) 𝑠 + 𝑠 + 𝐾𝑃 + 𝑠𝐾𝐷
𝑠 + 𝑠
𝑠 2 + 2𝜉 𝜔𝑛 s + 𝜔𝑛 2 = 0
desired 𝑠 2 + 2(0.2) 4 s + 42 = 0 𝑠 2 + 1.6s + 16 = 0
Matching
𝐶𝐿𝐶𝐸 = 𝑠2 + (1 + 𝐾𝐷 )𝑠 + 𝐾𝑃 = 0 coefficients 16 = 𝐾𝑃 𝐾𝑃 = 16
𝐷𝑒𝑠𝑖𝑟𝑒𝑑 𝐶𝐿𝐶𝐸 = 𝑠 2 + 1.6s + 16 = 0 1.6 = 1 + 𝐾𝐷 𝐾𝐷 = 0.6
• Cannot use canonical relations because third order with a zero, but tends to make system slower (sum of error)and more
oscillatory.
𝐾𝑃 𝑠 + 𝐾𝐼 𝑠+𝑏 Where
•The transfer function of the controller can be written as: 𝐶 𝑠 = = 𝐾𝑃
𝑠 𝑠
For a reference
one (unit step) the
output is one
Transfer
Unit stp input
function
• Therefore, the steady-state error is zero for a step reference, even for small KI (just takes longer to reach steady-state).
•The benefit of using the integral term is to eliminate the steady-state error.
𝐾 10
𝐾𝑃 + 𝑠𝐼 (𝑠 + 3) 10 𝐾𝑃 𝑠 + 𝐾𝐼
𝐶𝐿𝑇𝐹 = = 2 𝐶𝐿𝐶𝐸 = 𝑠 2 + (10𝐾𝑃 + 3)𝑠 + 10𝐾𝐼 = 0
𝐾 10
1 + 𝐾𝑃 + 𝑠𝐼 (𝑠 + 3) 𝑠 + 3𝑠 + 10 𝐾𝑃 𝑠 + 𝐾𝐼
• Increasing Kp : Same amount of error generates a proportionally larger amount of control, makes system faster, but
overshoot more (less stable).
•Increasing KD : Allows controller to anticipate an increase in error, adds damping to the system (reduce overshoot), can
amplify noise.
• Increasing KI : Control effort builds as error is integrated over time, helps reduce steady-state error, but can be slow to
respond.
2. For systems that are not canonical first or second order, need to
use trial and error. Can look for reduced-order approximation ( try
to identify if there are any poles that dominate the response, try to
identify if there are any of zeros cancel with any of the poles, and if
we can take a higher order system and approximates it as first or
second order system.