Sunteți pe pagina 1din 172

1.

1 CONTROL SYSTEMS
A control system is a system which controls the output response based on the input stimuli. Such a system consists of subsystems and processes (or plants) assembled for the purpose of obtaining a desired output and performance, given a specified input. The advantage of building control systems include: 1. 2. 3. 4. Power amplification Remote control Convenience of input form Compensation for disturbances

1.2 SYSTEM CONFIGURATIONS

A generic open loop system shows that there is no feedback that can be used to compensate for disturbances and errors.

Such a system consists of:

The input is sometimes called the reference, while the output is associated with the term controlled variable.

Ed

The disadvantages of open loop systems, namely sensitivity to disturbances and inability to correct these disturbances are overcome in a closed loop system. Input transducer: converts the form of the input to the form used by the controller Output transducer: measures the output response and converts it into the form used by the controller Feedback path: provides a means of giving feedback from the output transducer Actuating signal: the input signal (which may contain feedback) that is fed into the controller.

An input transducer: which converts the form of the input to that used by the controller Controller: that drives a process or plant Summing junctions: that add input signals together

un

Li

1.3 ANALYSIS & DESIGN OBJECTIVES


Analysis is the process by which a systems performance is determined, while design is the process of creating or changing a systems performance. A control system is dynamic and responds to transient responses before reaching a steady state response that generally resembles the input. Thus, the objectives of system analysis and design are to: produce the desired transient response reducing steady state error achieving stability robust design

TRANSIENT RESPONSE
The transient response is the sum of the natural and forced responses when the natural response is large. If we design an elevator with a slow transient response, passengers will become impatient, whereas an excessively rapid response will cause discomfort.

STEADY-STATE RESPONSE

The steady state response is the sum of the natural and forced responses when the natural response is small. In essence then, we wish the transient response to decay to zero, otherwise we sacrifice the accuracy of the steady state response the quantitative error described as steady state error

STABILITY

The natural response describes the way the system dissipates or acquires energy, and the form of this response is dependent only on the system and not on the input. However, the form and nature of the forced response is dependent on the input. We note that the total response for a linear system can be expressed as:

Ed
For a control system to be useful:

The natural response must decay to zero, leaving only the forced response It must oscillate

In some systems which do not meet either of these conditions, the natural response grows without bound, leading to instability. Such instability often leads to the destruction of a physical device.

OTHER CONSIDERATIONS
Obviously, any engineering task involves the consideration of finances and the need for robust designs.

un

Li

1.4 DESIGN PROCESS


In general, there are 6 steps to designing a control system: 1. 2. Transform requirements into a physical system Draw a functional block diagram a. A functional block diagram describes the component parts of the system, such as the input transducer and controller Transform the system into a schematic Develop a mathematical model a. Using KVL, KCL, Newtons Law we can develop a simplified mathematical model. Other tools used may include the use of linear differential equations, Laplace transform, state space representation and the transfer function Reduce the block diagram a. Subsystems models are interconnected to form block diagrams of larger systems, where each block has a mathematical description. Analyse and design a. The engineer analyses the system to see if the response specifications and performance requirements are met by simple adjustments of system parameters, if not, additional hardware may be designed to achieved the desired performance. Test input signals are used both analytically and during testing, to verify the design.

3. 4.

6.

m Ed

The functional block diagram shows that the input command is an angular displacement, which in turn is converted into a proportional voltage. Similarly, the output angular displacement is converted to a voltage by the potentiometer in the feedback path. The signal and power amplifiers boost the differential signal between the input and output voltages such that the actuating signal can drive the plant.

un

1.5 ANTENNA AZIMUTH: INTRODUCTION TO POSITION CONTROL SYSTEMS


In the Antenna Azimuth problem, we wish to position the antenna using a potentiometer. The system must also be able to adjust for disturbances in the environment. The detailed layout is shown. The next step is to derive the functional block diagram: We note that the purpose of this system is to have the azimuth angle output of the antenna track the input angle of the potentiometer .

Li

5.

Note that if we increase the gain of the signal amplifier, the actuating signal may drive the motor too hard, the transient response is overdamped, and causes the motor to overshoot the final value and the system is forced to make corrections. This indicates the possibility of damped oscillations about the steady state value. In most systems the steady state error the difference between the output and the input after the transients have disappeared will still be non-zero; for these systems a simple gain adjustment to regulate the transient response is either not effective or leads to a trade off between the desired transient response and the desired steady state accuracy. To solve this problem we can implement a filter as a compensator. Finally, we seek to transform the system into a schematic.

Input Impulse

Function

m
Description

1.6 INPUT TEST SIGNALS

un
Sketch Use Transient response modeling Transient response & steady state error Steady state error

Ed
Step Ramp

Li

Parabola

Steady state error

Sinusoid

Transient response modeling, steady state error

Ed

un

Li

2.1 LAPLACE TRANSFORM


The Laplace transform is defined as:

Where = + . The inverse Laplace transform is given by:

{()} = () = ()
0

The following table summarizes some of the most common Laplace pairs:

Ed

EXAMPLE 2.1: Find the Laplace transform of () = ()


0

() ()

() ()

()

In many cases, we need not work from first principles to find the Laplace transform of a function. Instead, we can make use of the Laplace theorems to assist us.

un
() = (+) = +

()

()

() () ()

Li
() ! + + + +

1 {()} =

1 + () = ()() 2

2.2 PROPERTIES OF LAPLACE TRANSFORMS


Theorem Linearity Frequency Shift Time shift Scaling Differentiation Integration Final value
1 2

Name {1 () + 2 ()} = 1 ( ) + 2 () { ()} = ( + ) { ( )} = ( ) 1 { ()} = ( ) 1 (0 ) = . (0 ) 1 ( ) () = 01 () = lim ()

Initial value

Convert a linear differential equation into the Laplace domain Use partial fractions to reduce the equation in the Laplace domain into known Laplace transform such that we can obtain the inverse Laplace transform

Conversion into the Laplace domain yields:

m Ed
1

EXAMPLE 2.3: Find the Laplace transform, () if the initial conditions are (0) = 1, = 1 for:
2 () (1) 1 + 2( ) 2 + 3() = () ( 2 + 2 + 3)() = () + 3 2 +2 + 3 = () 2

For this theorem to work, all roots of the denominator of () must have negative real parts, and no more than one can be at the origin. 2 For this theorem to work, () must be continuous or have a step discontinuity at = 0

un
2 () + 12 () + 32 () = ()[ 2 + 12 + 32] = () = ( 2 32 32 + 12 + 32)

EXAMPLE 2.2: Find the Laplace transform () given that all initial conditions are zero for:
2 + 12 + 32 = 32 () 2

d
32

Li

We can make use of these Laplace transform theorems and known Laplace transform pairs since we can:

(0+ ) = lim ()

EXAMPLE 2.4
Find () if the initial conditions are zero for:

( ) =

( 2

32 32 = = + + + 12 + 32) ( + 4)( + 8) + 4 + 8

2 + 12 + 32 = 32 () 2 32 =1 (0 + 4)(0 + 8) 32 =1 (8)(8 + 4)

Using the Laplace transform table then:

Find the inverse Laplace transform of

m
3

EXAMPLE 2.5

Ed

2 3 3 + 1 1 3 3 1 + = 2 + 2 1 5 5 ( + 1)2 + 22 2 ( + 1)2 + 22 5 5 2

3 3 6 6 + + 3 3 5 5 5 5 = 5 2 + 2 + 5 5 ( + 1)2 + 22 3 3 2 + 1 1 = + 5 5 ( + 1)2 + 22 2 ( + 1)2 + 22

un
( 2 +2+5)

() = () 2 4 () + 8 ()

using partial fractions with complex roots.

( 2

3=

+ 3 = + 2 + 2 + 5) + 2 + 5 3 2 ( + 2 + 5) + 2 + 5

6 3 = + 2 + + + 3 5 5 3 6 3 = , = , = 5 5 5

() =

1 2 1 + + 4 + 8

Li

32 = 2 (4)(4 + 8)

2.3 TRANSFER FUNCTION


The transfer function shows the relationship between the input and output of a system. Suppose we have an n-th order linear, time invariant differential equation: () 1 () () 1 () + 1 + + 0 () = + 1 + + 0 () 1

Taking the Laplace transform and rearranging for ()/() yields the transfer function:

Note that when we took the Laplace transform, we evaluated it with zero initial conditions.

Knowing the transfer function now, allows us to draw the block diagram of a subsystem as shown.

EXAMPLE 2.6
Find the transfer function ()/ ( ) for:

2 2 3 + 3 2 + 7 + 5 = 2 + 4 + 3 3

Ed
EXAMPLE 2.7

3 () + 3 2 ( ) + 7 () + 5 () = + 2 () + 4 () + 3 ( ) () 2 + 4 + 3 = 3 () + 3 2 + 7 + 5

un
2 + 6 + 2 = 2 + 2

Find the differential equation corresponding to the transfer function () =

2 () + 6 () + 2 () = 2 () + ()

d
2 +6+2 2+1

Li

+ 1 1 + + 0 () = + 1 1 + + 0

EXAMPLE 2.8
Find the ramp response for a system whose transfer function is () = ( The Laplace transform of a ramp response is () =
2 1 +4)(+8)

() =

1 4 1 8 1 32 32 16

2.4 ELECTRICAL NETWORK TRANSFER FUNCTIONS

m
1 () 0 () = () ()

In this section we apply the transfer function to the mathematical modeling of electric circuits. We present 3 of the most basic passive linear components used in electric circuit modeling. Component Capacitor Resistor Voltage-Current Current-voltage () () () = 1 () = () 0 () = Voltage-charge () () () = 2 () () = 2 () = Impedance () () = 1
()

un

d
Admittance:
() =

Ed

Inductor

() =

MESH ANALYSIS

Mesh analysis is the electric circuit analysis technique which aims at using voltage loops (KVL). When we are given a circuit with capacitors and inductors, the resulting voltage loops will be differential equations. We can transform the circuit into the Laplace domain by: Redrawing the original network showing all the time variables as Laplace transform (), (), ( ). Replace the component values with their impedance values.

Li
1 1
()

() = () ( + 4)( + 8) 1 () = ( + 4)( + 8) 1 1 1 = 32 16( + 4) 32( + 8)

Example 2.9 Find the transfer function relating the capacitor voltage () to the input voltage (). Taking a KVL loop: () + () + () + () =0

But:

2 () + () + () = () () 1 = 2 () + + 1

In more complex circuits, mesh analysis involves multiple loops and nodes. Again we: 1. 2. 3. 4. 5. Replace passive element values with their impedances Replace all sources and time variables with their Laplace transform Assume a transform current and a current direction in each mesh Write Kirchoffs voltage law around each mesh Solve the simultaneous equations for the output

Find the transfer function 2 ()/ ()

Ed
For each mesh loop: We then rearrange the equations:

m
2 () 1 () + 2 2 () + 1 1 () + 1 () 2 () = () 2 () =0 1 ( ) + + 2 + (1 + )1 () 2 () = () 1 () = 0 2

Example 2.10

un

Li

( ) =

() () = ()

By the Cramer rule: + () 1 0 = () 2 () = 1 + = 1

Where:

+ 2 +

NODAL ANALYSIS

Example 2.11

Find the transfer function ()/()of the above circuit using nodal analysis.

Ed
Rearranging we get:

() ( ) () () () + + =0 1 2 () + () () =0 2

1 + 2 +

Again using Cramers rule, solving yields:

1 2 () = ( ) ( + ) 2 + 1 2 + + 2 1 2

un

1. 2. 3. 4. 5. 6.

Replace passive element values with their admittances Replace all sources and time variables with their Laplace transform Replace transformed voltage sources with transformed current sources Write Kirchoffs current law at each node Solve the simultaneous equations for the output Form the transfer function

2 () + (2 + ) () = 0

1 () 2 () = ()1

Like in the time domain, nodal analysis makes use of Kirchoffs current law and sums the currents at the nodes. This also applies in the Laplace domain. The general procedure is to:

Li

2 2 () = () (1 + 2 ) 2 + (1 2 + ) + 1

2.4 OPAMPS
Recall that an ideal operational amplifier has the following characteristics: Differential input Infinite input impedance Zero output impedance Infinite gain amplification

Like in normal circuit analysis in the time domain, we can use the same techniques to find the transfer function of operational amplifiers in the Laplace domain.

Find the transfer function ()/ () for the circuit shown.

The components in the feedback loop give an impedance of: 1 =

The components at the input give an impedance of:

Ed

m
2 =

1 1 1 = + 1 = + (5.6 106 ) 2 1 360000 1 + 5.6 106 360000 () () = 2 1 1 =

Nodal analysis at the inverting terminal yields:

10 106 1 + 2.016 () = + 220 103 () 360 103 10 106 + 20.16 106 + 220 103 + 443520 2 = 360 103 2 + 49.95 + 22.55 = 1.232

un

1 10 106 + 2 = + 220 103 2

d
360 103 1 + 2.016

Li

Example 2.12

The previous circuit is an inverting configuration because the transfer function is negative. In the non-inverting configuration, we find that the transfer function is positive.

Example 2.13 Find the transfer function ()/ () for the circuit given.

For the series connection the impedance is:

And for the parallel components:

Ed
Then by nodal analysis:

2 1 ( ) =1+ (2 2 + 1)(1 + 1 1 ) () 2 1 =1+ 2 2 + 2 1 1 2 2 + 1 1 + 1 2 1 2 1 2 + (2 2 + 1 2 + 1 2 ) + 1 = 2 1 2 1 2 + (2 2 + 1 1 ) + 1

un
1 () = 1 + 1 1 2 () = 1 2 + 2 2 2 = 2 2 2 + 1 () = () 1 + () () () = 2 () 1 ( ) 2 ( ) 1 ()

Li

2.5 TRANSLATIONAL MECHANICAL SYSTEM TRANSFER FUNCTIONS


So far we have looked at modeling electric circuits in the Laplace domain. We can do the same thing for mechanical systems both translational and rotational. There are many parallels between mechanical and electrical networks mechanical systems also have three passive, linear components, notably, the spring and the mass, which are energy storage elements and the viscous damper which dissipates energy. The energy storage elements are similar to the capacitor and inductor in electric networks, while the viscous damper parallels the passive resistor element. To begin with we define: = =

The following table presents a force-velocity, force-displacement and impedance relationship for each component.

d
= () = () = 2 () 2

Component

Force-Velocity

Force-Displacement

un
0

= ()

Ed

m
=

= ()

()

Many mechanical systems are similar to multiple loop and multiple node electrical networks, where more than one simultaneous differential equation is required to describe the system. In mechanical systems, the number of equations of motion required is equal to the number of linearly independent motions Linear independence implies that a point of motion in a system can still move if all other points of motion are held still. The number of linearly independent motions is commonly referred to as the degrees of freedom.

Li
Impedance =

In our analysis of mechanical systems, we make use of linear independence for each free body diagram we superimpose the forces acting on the body when every other point of motion is held still and let that point move, and with the reverse situation whereby all other points are in motion while the point of interest is held still. We can do this because of Newtons Laws of motion.

Example 2.14 Find the transfer function 2 ()/ ().

We then hold 1 still and let all other bodies move and analyse the forces on 1 :

Ed
Superimposing, we get:

For mass 2 , we do the same thing, and this yields:

[3 + 2 ]1 () + [2 2 + (2 + 3 ) + 2 + 3 ]2 () = 0

[1 2 + (1 + 2 ) + (1 + 2 )]1 ( ) [3 + 2 ]2 () = ()

un
3 2 () 2 2 ()

() = 1 1 () + 1 1 () + 2 1 () + 3 1 () + 1 2 1 ()

We first focus on 1 s motion, holding every other motion still:

Li

By Cramers Rule: 1 2 + (1 + 2 ) + (1 + 2 ) (3 + 2 ) 2 () = 2 ( ) 3 + 2 = () = 1 2 + (1 + 2 ) + (1 + 2 ) (3 + 2 ) 3 + 2 [2 2 + (2 + 3 ) + 2 + 3 ] () 0 ()(3 + 2 )

Where:

Example 2.15

Write, but do not solve, the equations of motion for the mechanical network shown.

Ed
Then 2 : Then 3 :

There are three degrees of freedom, since each of the three masses can be moved independently while the others are held still. We focus on mass 1 first: [1 2 + (1 + 3 ) + (1 + 2 )]1 () 2 2 () 3 3 () = 0 2 1 ( ) + [2 2 + (2 + 4 ) + 2 ]2 () 4 3 () = () 3 1 () 4 2 () + (3 2 + (3 + 4 ))3 () = 0

un

Li

2.6 ROTATIONAL MECHANICAL SYSTEMS


In rotational mechanical systems, we deal with torque and angular displacement rather than forces and translational displacement. We define: =

= =

Component

Torque-angular velocity

Torque-angular displacement () = ()

Impedance
() ()

()
0

() = ()

d
() = () () = 2 () 2

un
() = ()

Ed

The concept of degrees of freedom carries over to rotational systems, except that we test a point of motion by rotating it while holding still all other points of motion. The number of points of motion that can be rotated while all others are held still equals the number of equations of motion required to describe the system.

Again, in our analysis, we make use of the superposition of torques we rotate a body while holding all other points still and place on its free body diagram all torques due to the bodys own motion. Then, holding the body still, we rotate adjacent points of motion one at a time and add the torques due to the adjacent motion to the free body diagram. Note that the torques will sum up to zero due to an equilibrium system. This process is repeated for each point of motion.

Li
2

() =

Example 2.16 Find the transfer function 2 ( )/ (), for the rotational system shown. The rod is supported by bearings at either end and is undergoing torsion. A torque is applied at the left, and the displacement is measured at the right.

For the free body diagram of 1 , we show the torques due to its own motion, due to the motion of the other bodies and then superimpose:

Then for the free body diagram of 2 : Using Cramers rule:

Where:

m
=

2 1 + 1 + 2 ( ) =

Ed

un
1 2 + 1 +

1 () + (2 2 + 2 + )2 () = 0 () 0 = ()

d
2 2 + 2 +

(1 2 + 1 + )1 () 2 () = ()

Li

There are 2 degrees of freedom, sine each inertia can be rotated while the other is held still. There are thus 2 equations of motion:

Example 2.17 Find the transfer function 2 ( )/ () for the system shown.

There are 2 degrees of freedom the cylinder and the point between the 2 springs. For the cylinder:

Ed

2 + + 1 ( ) (s + 1)T(s) ( + 1) 0 2 () = 2 = 3 s + s + 1 (s + 1) 2s + 3s 2 + 2s + 1 (s + 1) 2s + 2 1 2 () = 2 () 2 + + 1

un

For the point 2 :

( + 1)1 () + (2 + 2)2 () = 0

( 2 + + 1)1 () ( + 1)2 () = ()

Li

2.7 TRANSFER FUNCTIONS FOR SYSTEMS WITH GEARS


Gears provide mechanical advantage to rotational systems. This because gears allow you to match the drive system and the load; a tradeoff between speed and torque. In this section we ignore the effect of backlash the situation in which the drive gear rotates through a small angle before making contact with the meshed gear. The ratio of the angular displacement of the gears is inversely proportional to the ratio of the number of teeth. In an idealized case, the energy transferred between gears is conserved. Thus:

Example 2.18 Derive a relationship for the transfer function 1 ()/1 ( )

We note that 1 can be reflected to the output by multiplying by 2 /1 to give us 1 . This results in: ( 2 + + )2 () = 1 () 2 1

Ed

Now: 2 =

1 2

We generalize the result by showing that:

1 2 1 2 1 2 2 + + 1 () = 1 ( ) 2 2 2

2 1 ( 2 + + ) 1 () = 1 () 2 1

Rotational mechanical impedances can be reflected through gear trains by multiply the mechanical impedance by the ratio: 2

Where the impedance to be reflected is attached to the source shaft and is being reflected to the destination shaft.

un

Li

1 2 1 1 = = = 2 1 2 2

Example 2.19 Find 2 ()/1 ()

When we reflect the input shaft to the output shaft, we clearly see that there is only 1 degree of freedom: Reflecting the inertia and the viscous damper, we get:
= 1 1

Thus we have:

And analysis of the resulting free-body shows that:

m Ed
Thus: 2 1 = 1 2 4 =

With a gear train, we can continually reflect the angular displacement by multiplying through by the ratio:

un
= 1 = 1 2 2 + 2 1 ( 2 + + 2 )2 ( ) = 1 () 2 2 () 1 = 2 1 () + + 2 2 1 1 3 5 2 4 6 1

2 2 + 2 1

2 2 2 2 , 1 = 1 1 1

Li

Example 2.20 Find 2 ()/ (), for the rotational mechanical system with gears.

We shall reflect everything to the input shaft noting that the gears are not lossless since we need to take into account their inertia and viscous friction.

Reflecting the third shaft to the middle shaft we get: 3 2 (4 + 5 ) 4

And the middle shaft (including the load of the bottom shaft):

Thus, the total inertia in the system is:

1 2 3 2 1 2 2 + 3 + (4 + 5 ) + 2 2 4 2 1 2 1 3 2 + (4 + 5 ) 2 2 4

And total viscous friction:

Ed
The transfer function becomes:

= 1 + (2 + 3 )

un
1 2 = 1 + 2 2 ( 2 + )1 () = 1 () 1 () 1 = 2 1 () +

Li

ELECTROMECHANICAL SYSTEMS
We focus on deriving the transfer function electromechanical component that yields a displacement output for a voltage input. The basic motor schematic is shown in which an external magnetic field causes the interaction of the magnetic field generated through the armature coils carrying a current of (). The armature feels a force of () when it is perpendicular to the external magnetic field, and the resulting torque turns the rotor: () = ()
() ()

of an electric motor from first principles. A motor is an

Since the armature cuts through magnetic flux, Lenzs Law and Faradays Law suggest the existence of a voltage and its associated magnetic which opposes the change which caused it - the back emf, which is given by: () = ()

Where is a back emf constant and yields:

( )

is the angular velocity of the motor. The Laplace transform of it

The resulting loop equation thus gives us:

m Ed

( + )( 2 + ) () + () = ()

Since the armature inductance is small, then we approximate the equation to be: ( 2 + ) + = () ()

un
() = () () = ( 2 + ) ( )

( + ) () + () = () The typical mechanical loading on a motor is shown in the figure, with being equivalent inertia at the armature and includes both the armature inertia and the load inertia reflected to the armature. is the equivalent viscous damping at the armature.

() + () + () = ( )

Li

() = ()

Where:

1 2 1 2 = + , = + 2 2

To find the electrical constants in the transfer function, we perform a dynamometer test which measures the torque and speed of a motor under the condition of a constant applied voltage. From this we can measure and and hence calculate the electrical constants. Going from the previous equation:

Taking the inverse Laplace transform:

Ed

un
() + = ()

And letting = 0, then we can rearrange for , the constant torque which results when a DC voltage is applied causing a constant angular velocity .

( + ) () + () = ()

From the dynamometer test, we see a linear relationship: =

Upon rearranging we find that:

= =

d
=

Li

Example 2.21 Given the system and torque-speed curve shown, find the transfer function ( )/ () The general equation for the motors transfer function is given by: () = () We find that: + 1

Thus:

Now since we have:

Ed

5 () 0.417 12 = = () + 1 10 + 5(2) ( + 1.667) 12 () 2 2 = () = () 1 () 1

un
= = 100 =2 50

1 2 1 2 = + = 2 + 800 = 10 2 10 500 = = =5 100

0.0417 () 1 () = = () 2 () ( + 1.667)

1 2 1 2 = + = 5 + 700 = 12 2 10

Li

LINEARISATION
A linear system possess two characteristics: Superposition Homogeneity

Which can be summarized as: {1 + 2 } = {1 } + {2 } Where H is the transformation (the systems transfer function). If a system does not possess these qualities, then the system is said to be nonlinear.

An electronic amplifier is linear over a specific range, but exhibits the nonlinearity called saturation at high input voltage. A motor that does not respond at very low input voltages due to frictional forces exhibits a nonlinearity called dead zone. Gears that do not fit tightly exhibit a nonlinearity called backlash. In many situations, it is possible to make a linear approximation to a nonlinear system with small variations around the point of interest. When we linearise we: Recognise the nonlinear component and write the non linear differential equation We then linearise it for small signal inputs about the steady state solution when the small signal input is equal to zero We linearise the nonlinear differential equation and take the Laplace transform, assuming zero initial conditions. Consider a point A on the curve which we want to linearise. Suppose that the differential at point A yields , then using the point gradient formula of a straight line: ( ) (0 ) ( 0 )

Ed

un
Thus: Since: ( )

d
( ) (0 ) +

Li

Example 2.22 Linearise: 2 +2 + = 0 2


4

About =

If we let = + then:

Since:

Ed

2 2 cos + = cos sin = 2 2 4 4 4

cos + cos = cos 4 4 |= 4 = sin 4

un

( ) (0 )

2 2 2 +2 = 2 2 2

d
| =0

cos = cos ( + ) 4

+ 4 =

Li

2 + 2 4 = 2 2

3.1 STATE SPACE


The state space approach is a unified method for modelling, analysing and designing a wide range of systems. It provides a more power tool for analysis because it can also be used to represent non linear systems, systems with nonzero initial conditions, time varying systems and multiple input multiple output systems. We first enumerate the steps involved and then present an example to consolidate the process. 1. 2. 3. 4. 5. We select a particular subset of all possible system variables and call the variables in this subset state variables For an nth order system, we write n first order differential equations in terms of the state variables. These simultaneous differential equations are known as state equations If we know the initial condition of all of the state variables at 0 as well as the system input for we can solve the simultaneous differential equations for the state variables for 0 We algebraically combine the state variables with the systems input and find all of the other system variables for 0 . We call this algebraic equation the output equation The representation of the system using the system and output equations is known as the state space representation.

Where is a set of vectors and are constants. Linear independence only occurs if the only solution to the above equation is trivially = 0 Example 3.1

Find the state space representation of the system.

Ed
Rearranging yields: Since:
( )

Typically, we select the state variables to be the variables with the differentials in this case the inductor current and the capacitor voltage. A KVL loop yields: () + () + () + () = 0

Note that this first order differential equation is in terms of the state variables, inductor current and capacitor voltage and has the input of the voltage source. = ()
( )

These two differentials are the state space representation of the system.

m
=
1

un
1 + 2 + + = 0 ()

() 1 = () () + ()

The minimum number of state variables required to describe a system is equal to the order of the differential equation (or independent energy storage elements in the system). If we can define more state variables than the minimum, we must eliminate the state variables which do not form part of a linearly independent set i.e no state variable can be written as a linear combination of the other state variables:

Li

3.2 GENERAL STATE SPACE REPRESENTATION


Let us formally define: State variable: any variable that responds to an input or initial condition in a system State variables: the smallest set of linear independent system variables such that the values of the members of the set at time 0 along with known forcing functions completely determine the value of all system variables for all 0. State vector: a vector whose elements are the state variables State space: the n dimensional space whose axes are the state variables State equations: A set of n simultaneous, first order differential equations with n variables, where n variables to be solved are the state variables Output equation: the algebraic equation that expresses the output variables of a system as a linear combination of the state variables and the inputs

A state space representation of a system (state equation, output equation) can be written as:

Where:

= , = = , = , = =

Example 3.2

Ed
1.

Given the electrical network shown, find the state space representation of the output is the current through the resistor.

Select state variables check the number of independent storage elements. We select the inductor current and capacitor voltage as the state variables: = , =

= ,

un

= +

= +

Li

2.

3.

Write , as a linear combination of the state variables , and the input (). = But since R is parallel to C: = 1 = Now, the KVL loop around the voltage source-inductor and capacitor yields: () + + = 0 1 = ( () ) The output equation is: =

4.

The final state space representation is:

When we have a dependent source, the approach is much the same we just find the relationship to describe the dependent sources either in a KVL loop or a KCL node.

Example 3.3

Ed

2 Find the state and output equations for the electrical network shown if the output vector is =
2

We choose out state variables to be the inductor current and capacitor voltage so: = , =

Now let us find expressions for the right hand side of those differentials in terms of . and the input (). KVL loop 1 shows: = + 2 (1)

un

d
1 0

1 = 1

1 0 + 1 () 0

Li

But at node 2 we have: () + 4 = 2

We now wish to find in terms of the state variables at node 1: = ()

(1) =

1 ( + 2 ) 1 42

= () 1

We now have a simultaneous equation:

(2) 1

Thus:

By the Cramer Rule:

m Ed
Where: Thus: = 1 42 1 1

() =

un
= () 1 2 1 = () = = (1 42 ) 2 1

(1 42 ) 2 =

+ 2 2 ()

d
1

= ()

1 ( + 2 ) (1) 1 42 (2) 1

(1 42 ) (1 42 ) () +

2 = 1 4 2

2 2 = = 1 2

1 2 1 + 1 42 () 1 C 1 2 1 + 1 41 + 1 () 1

Li

3.3 STATE EQUATIONS FOR MECHANICAL SYSTEMS


It is convenient when working with mechanical systems to obtain the state equations directly from the equations of motion rather than from the energy storage elements. Our state variables in mechanical systems will always be the position and velocity of each point of linearly independent motion.

Example 3.4 Find the state space representation of the translational mechanical system

1.

2.

Take the inverse Laplace transform:

3.

Assign our state variables to be 1 , 1 , 2 2 and rewrite the above equations: 1 = 1 (1 1 + 2 ) 1 1 (() + 1 2 ) 2 1 1 0 0 0 1 0 2 0

Ed
4. Write the matrix:

0 1 1 1 = 2 0 2 2

un
1 = 2

Find the Laplace transformed equations of motion for the two linearly independent motions: (1 2 + + )1 2 = 0 1 + (2 2 + )2 = 2 1 1 + + 1 2 = 0 2 2 2 1 + 2 + 2 = () 2

0 0 0 1 1 + 0 () 1 2 1 2 0 2

Li

3.4 CONVERTING A TRANSFER FUNCTION TO STATE SPACE


One advantage of the state space representation over the use of transfer functions is that physical systems can be simulated on the computer. To perform a conversion from a transfer function to a state space, we select phase variables, where each subsequent state variable is defined to be the derivate of the previous state variable. Consider the differential equation: 1 + 1 1 + + 1 + 0 = 0

We choose the phase variables to be:

2 = =

On differentiating both sides, we get:

m
0 1 0 2 0 3 = 0 1 0 0 2 1 0 0

In Vector matrix form:

Ed
Yielding an output equation of:

un
2 = 3 0 3 0 1 0 0 4 0 0 1 0 5 0 0 0 0 0 0 = [1 0 0 0 0 1 2 3 0] 1

= 2 1

= 0 1 1 2 1 + 0 1 0 0 2 3 + 0 1 1 0 1 0

1 1

Li

1 =

Example 3.5 Find the state space representation and draw the equivalent block diagram.

We first transform the transfer function to a differential equation:

Let

In Matrix-vector form:

1 = 2 ,

1 0 = 0 2 3 24

Ed

To draw the functional block diagram, we note that it is an order 3 system, so we need 3 integrators in series which is the output from the summer as shown. with the input being 3

un
2 = 3 , 1 = [1 0 0] 2 3 1 0 26

Thus:

3 =

3 = 241 262 93 + 24 0 1 0 1 2 + 0 9 3 24

2 =

1 =

Li

2 3 +9 + 26 + 24 = 24

In the above example, the numerator was a constant term. If the transfer function had been a polynomial in s, we would handle the numerator and denominator separately by splitting the system up into two subsystems. Consider the system shown and how it is split up into two systems. The first subsystem is the same as the previous example. The second transfer function would yield: () = 2 2 1 1 + 1 + 0 1 2

After taking the inverse Laplace transform with zero initial conditions. In the second subsystem, the output is an output equation, so in conformance to what is shown above, the result would be writing the terms in reverse order: () = 0 1 + 1 2 + 2 3 Example 3.6

Find the state space representation for the system shown.

1.

We split the system up as shown:

Ed
2. 3.

With subsystem 1 we have: 3 1 2 1 1 + 9 2 + 26 + 241 = 3 Let 1 = 1 , 2 = 1 . 3 = 1 , then: = 241 262 93 + 3 The resulting matrix-vector representation is: 1 0 1 0 1 0 = 0 2 0 1 2 + 0 3 24 26 9 3 1 For subsystem 2: 2 1 1 = +7 + 21 2 Let 1 = 1 , 2 = 1 , 3 = 1

un

Li

Which yields:

4.

1 = [2 7 1] 2 3 The block diagram for subsystem 1 is similar to that of the previous example. The second subsystem collects the derivates from subsystem 1 and sums them as shown.

() = 3 + 72 + 21

3.5 CONVERTING FROM STATE SPACE TO A TRANSFER FUNCTION


Given the state and output equations:

We take the Laplace transforms assuming zero initial conditions:

Ed
Now in (1): And in (2): Which yields the transfer function:

un
= + = + () = () + () (1) () = () + () (2) () = ( )1 ()

() = ( )1 () + ()

() = ( )1 + ()

Li

Example 3.7 Find the transfer function given: 0 = 0 1 1 0 2 0 10 1 + 0 3 0

= [1 0 0]

This means: 0 1 = 0 0 1 2 0 10 1 = 0 = [1 0 0] = 0 3 0 1 2 0 1 + 3

Now:

m Ed
Now:
1

( 2 + 3 2) 1 ( + 3) ( + 3) (2 + 1) 1 2 = 3 2 + 3 + 2 + 1 + 3 ( 2 + 3 + 2) 1 ( + 3) (2 + 1) = 3 + 3 2 + 2 + 1 1 2

1 0 1 0 2 + 3 1 + 3 1 2 1 0 0 1 1 + 3 1 2 2 + 3 1 0 0 1 1 0 1 0 = 3 + 3 2 + 2 + 1

un
( )1 = ( )1 det( ) () = ( )1 +

So:

This is an adjoint matrix.

= 0 1

Li

State space representation can also be used to represent systems with nonlinearities. For small perturbations about an equilibrium point, state space representations can be linearised.

Example 3.8 Represent the system in stage space. Assume the mass is evenly distributed with the centre of mass at L/2. Then linearise the state equations about the pendulums equilibrium point the vertical position with zero angular velocity.

Ed
The sum of torques is given by:

m
Recall that the torque experience due to gravity is given by: = = We select the state variables 1 , 2 as phase variables and let 1 = . 2 = 1 = 2 , 2 = 1 + 2
2

un
2 + = 2 2

In order to be able to convert these state equations to transfer functions, we must linearise them. We let 0 = 0, 1 = 0, 2 = 0 be perturbed:

Li

LINEARISATION

10 [( 2 + 3 + 2) + 3 1] 0 0 = 3 + 3 2 + 2 + 1 10( 2 + 3 + 2) = 3 + 3 2 + 2 + 1

( 2 + 3 + 2) + 3 ( 1 + 3) (2 + 1) = [1 0 0] 3 2 + 3 + 2 + 1

1 2 10 0 0

Now we wish to linearise 1 :

1 = 1 , 2 = 2 (1 ) (0 ) = 1 0 = ( )|=1 1
1 =0

1 = 1 1 = 2

1 |

1 1

Ed

un

Li

2 =

1 + 2

4.1 POLES, ZEROS AND SYSTEM RESPONSE


Recall that the output response of a system is the sum of two responses:

So far we have neglected to focus on the poles and zeros of a transfer functions even though they serve of great importance in analysing a systems response. What we find is that: The poles cause the transfer function to become infinite, and forms part of the natural response The poles of the input function generates the form of the forced response The zeros and the poles generate the amplitudes for both the forced and natural responses.

Ed
transient response. When :

4.2 FIRST ORDER SYSTEMS

A first order system without zeros can be described as:

If we find the output and take the inverse transform we get:

The most significant (and only) parameter in this system is a, the exponential frequency, which determines the

un

Li

Consider a unit step response into the system as shown:

TIME CONSTANT
The time constant is defined as:

The time it takes for the step response to rise to 63% of the final value.

RISE TIME
The rise time is defined as the time for the wave to go from 0.1 to 0.9 of its final value.

For a first order system this would be:

SETTLING TIME

The time it takes for the response to reach and stay within 2% of its final value:

Ed

un

Li

Example 4.1 The curve shown shows the response of a system subjected to a step input. Find the transfer function given that it is a first order system:

Note that the output equation is given by:

The final value is 0.72, which means:

When

, the output has fallen 63%:

Ed

From the curve, it takes a=0.13s to reach 0.45, thus:

un

Li

4.3 SECOND ORDER SYSTEMS


Whereas varying a first order systems parameter simply changes the speed of the response, changes in the parameters of a second order system can change the form of the response. The forms of the response vary widely, and can be summarised as either overdamped, underdamped, undamped or critically damped. System Pole-zero plot Criteria Two real poles Response

Two complex (conjugate ) poles

m Ed
Thus: 1. 2. 3. 4. Overdamped response: Two real poles at Underdamped response: Two complex poles Undamped response: Two imaginary poles Critically Damped: Two real poles

un

d
Two completely imaginary poles Double real roots

Li

4.4 GENERAL SECOND ORDER SYSTEM


The natural frequency, is defined to be: is the frequency of oscillation of the system without damping. The damping ratio

We can represent a second order transfer function as:

Solving for the poles of the transfer function yields:

Ed

un

Li

Example 4.2 Find the value of and the type of response expected a)

b)

c)

Ed

4.5 UNDERDAMPED SECOND ORDER SYSTEM

In underdamped systems, it is possible to relate transient specifications with the pole locations. Recall that a general second order system with step input can be written as:

un

Li

The inverse Laplace transform is:

Where

Note that the lower the value of , the more oscillatory the response. From our definition of and define other parameters: Rise time Peak time the time required for the waveform t go from 0.1 to 0.9 of the final value : the time to reach the first maximum or peak

, we can

Percent overshoot: %OS: the amount that the waveform overshoots the steady state value at the peak time, expressed as a percentage of the steady-state value Settling time, : the time required for the transients damped oscillations to reach and stay within of the steady state value.

To evaluate the peak time, we differentiate and find the stationary point:

Completing the squares yields:

Ed
Which gives us: Thus, the stationary point is at:

un

EVALUATION OF

Li

EVALUATING %OS
We evaluate and :

For a unit step input. But:

Notice that the percent overshoot is a function only of the damping ratio .

In order to evaluate the settling time,

Which gives us:

Ed
RISE TIME
The rise time cannot be calculated analytically.

And the cosine term will tend towards 1 when the transient response dies down, then we are left with:

un

EVALUATING

must stay within 2% of the steady state value. Thus since

Li

4.6 RELATIONSHIP BETWEEN POLES AND THE SECOND ORDER


UNDERDAMPED SYSTEM PARAMETERS
Consider the pole plot shown for an underdamped second order system. What we find is that: The angle subtended by the poles location and the x -axis gives us The distance between the pole and the origin is the natural frequency The imaginary component, the damped frequency oscillation is given by , the damping ratio.

The real part, the exponential damping frequency is given by

We thus find:

Example 4.3

Determine what parameters are changing and which are constant, for each of the pole-zero plots.

Ed
a)

Since is inversely proportional to the imaginary part of the poles, lines of constant imaginary value (horizontal lines) imply that each pair of poles produce a constant peak time.

un

Li

b) Lines of constant real value (vertical lines) show that the settling time is constant since

Ed
In summary:

The line also indicates that the real part of is greater, indicating that the settling time is shorter.

Since the imaginary part is greater for , this indicates that the peak time is shorter.

To decrease the settling time, we move the pole to the left. To decrease the peak time, we move the pole higher. To decrease the percent overshoot, we make the angle shallower

d) We find that the line % yields a smaller angle than , this means that the damping factor is greater for since which gets its maximum when .

un

Li

c)

Since , radial lines are lines of constant . And since percent overshoot is only a function of , radial lines are thus lines of constant percent overshoot.

Ed m un d Li

5.1 BLOCK DIAGRAMS


In control system block diagrams, the main elements are signals, summing junctions, pickoff points and systems. Using these elements we can represent a system with different topologies. As we shall see, there are several common topologies.

5.1.1 CASCADE FORM

In cascade form, each subsystem is lined up one after the other, and the resulting output of each stage, is the product of the input and the transfer function e.g . For the system shown then, the equivalent transfer function of the system is . This assumption holds as long as the interconnected subsystems do not load adjacent subsystems.

Ed
5.1.2 PARALLEL FORM

Parallel subsystems have a common input and output formed by the algebraic sum of the outputs from all the subsystems. The equivalent transfer function is thus:

un

Li

5.1.3 FEEDBACK FORM


The feedback system forms the basis for control systems engineering given that it forms a close loop system. We can easily derive the transfer function for negative and positive feedback. Now:

But:

5.2 MOVING BLOCKS TO CREATE FAMILIAR FORMS

Ed

In some situations, we may need to move blocks to create familiar forms. This may involve moving the blocks to the left or right of summing junctions and pickoff points. The figures show the equivalent block diagrams after the movement of the blocks.

un

Li
.

Note that represents the negative feedback system while feedback system. The open loop gain of this system then is given by

represents the positive

Example 5.1 Reduce the system shown to a single transfer function.

We begin by noting the feedback loop between

and

. Thus the transfer function block will be:

We also shift the pickup point in the last summer to

The feedforward path consists of

m
and and

un
. This simplifies to . We also shift with to the right of and

Ed
with . Note that parallel blocks,

the second summer so that we can remove one of the summers. We thus replace

. Thus, we replace this block with one block labelled .

d
.

We also simplify the cascaded block with one block labelled

Li

We finally note the feedback loop which has a transfer function of:

This then yields a cascaded system, which simplifies down to:

Ed

un

Li

5.3 ANALYSIS AND DESIGN OF FEEDBACK SYSTEMS


In some situations, systems can reduce down to second order systems. Consider a situation in which we have a feedback control system with a open loop gain of The closed loop transfer function is thus: .

Where models the amplifier gain, the ratio of the output voltage to the input voltage. As K varies, the poles move through the three ranges of operations of second order systems overdamped, critically damped and underdamped. Note that when the poles are given by:

The poles are real, and the system is overdamped. When a critically damped system. Whereas if increases beyond

, then the system a double pole, which yields then the poles are complex:

while the settling time, which is given by

Example 5.2

Find the peak time, percent overshoot and settling time for the system shown.

Ed

The closed loop transfer function is:

un
remains constants.

And the system is underdamped. We find that the peak time decreases and the percent overshoot increases,

Li

5.4 SIGNAL FLOW GRAPHS


Signal flow graphs consist only of branches, which represent systems and nodes, which represent signals. This offers an alternative to block diagrams. The conversion from a block diagram to a signal flow diagram is simple: 1. 2. 3. Draw a node for each signal Then draw branches to represent each transfer function that is associated with the particular signal Simplify

Example 5.3 Convert the block diagram into a signal flow graph.

1.

We first identify all the signals and draw nodes for them:

And the transfer functions:

Ed
2. Start at and move through the signals. This is a unity transfer function multiplied by the input, . Thus we connect a line from , labelled 1. also subtracts off , so there is a line labelled -1. Then is produced by and label them appropriately. We continue for all the signals. . We thus connect the signals and

m
to

un

Li

3.

To simplify, we combine branches with unity gain with other transfer functions. For example we combine and to be . This reduced the signal flow graph to:

Ed
1. 2. 3.

5.4 SIGNAL FLOW GRAPHS OF STATE EQUATIONS


In this section we draw signal flow graphs from state equations. Identify the nodes to be the variables and their derivatives Identify the input and output nodes.

Interconnect the state variables and their derivatives with the defining of integration, .

un

Li

Example 5.4 Drawn the signal diagrams based on the following state and output equations:

1.

We select

and their derivatives to be the nodes. The input is r and the output is y.

2. 3.

Join

to

with a line labelled . Do the same

Ed

In the next section, the signal flow model will help us visualise the process of determining alternative representations in state space of the same system. We will see that even though a system can be the same with respect to its input and output terminals, the state space representations can be many and varied.

un

Starting from which is represented by We then proceed with and .

in the signal flow diagram, we interconnect the signals.

Li

5.5 ALTERNATIVE REPRESENTATIONS IN STATE SPACE


So far, we have represented a system in state space with the phase variable form. Here, we see that there can be many representations that yields the same output for a given input. Such variations in representations allow us to select a model which allows us to determine a solution easily.

5.5.1 CASCADE FORM


We can represent a system:

For each block, in the form of:

That yields a inverse Laplace transform:

Ed

We can now easily determine the state equations from the signal flow graph by looking at each node that represents a differential.

The output equation is written by inspection:

We can represent the block as shown with an integrator and a closed loop. We thus cascade the system as shown.

un

Li

As a cascaded system of first order blocks as shown. From this, we can determine the signal flow graph, and consequently the state equation in cascade form.

The state space representation in vector-matrix form is thus:

5.5.2 PARALLEL FORM


By considering the same transfer function, we can perform a partial fraction expansion:

This shows the sum of three terms, with each term being the first order subsystem with as the input. Each term is thus parallel with the other 2 terms, as shown. The state equations are then:

And in Vector Matrix form:

Ed
Example 5.5

This representation always yields a diagonal system matrix, A. The result is that each equation is a first order differential equation in only one variable such that the equations are decoupled. Note that with repeated real roots, we will not obtain a diagonal matrix but a Jordan canonical form.

Determine the parallel representation of the state space equations for the transfer function:

Since there are 3 terms, there are 3 parallel branches. The

un
contains 2 feedback loops as shown.

Li

The state and output equations are thus:

And in vector-matrix form this is:

This form is obtained from the phase variable form simply by ordering the phase variables in the reverse order i.e becomes .

Example 5.6

Determine the controller canonical form for the transfer function:

Which has a phase variable form given by:

Ed

Then rearranging in ascending order:

un

5.5.3 CONTROLLER CANONICAL FORM

Li

5.5.4 OBSERVER CANONICAL FORM


The observer canonical form yields left companion system matrix. The process of obtaining the state equations in observer canonical form is done over a few processes: 1. 2. 3. Divide by the highest power of Cross multiplying, obtain terms of life power of integration Identify the state variables as the outputs of the integrators

The result is a form:

Where

Example 5.7 Determine the observer canonical form:

Ed
Thus in vector matrix form:

Now, we let the innermost bracket represent

m
the input translate to r and for each translate to

un

Li

We can now draw the signal flow diagram either using the vector matrix form, or the factorised expression for . For the factorised expression for , we start with the outermost bracket and find that the output is connected to an integral with the integral being fed the input and a feedback of . Then we do the same for the other state variables.

Ed

un

Li

6.1 STABILITY
An unstable system cannot be designed for a specific transient response or steady state error requirement. Recall that the total response of the system is given by:

Using this, we define a: Linear, time invariant stable system: A system which has a decaying natural response Linear, time invariant unstable system: A system which has a natural response that grows without bound Linear, time invariant marginally stable system : A system which has a natural response that does not decay nor grow but remains constant or oscillates.

An alternative definition of stability is the BIBO stability:

A system is stable if and only if every bounded input yields a bounded output.

Ed

We also find that stable systems have closed loop transfer functions with poles only in the left half plane, since this yields an exponential decay or damped sinusoidal natural response as time tends to infinity. Poles in the right half plane yield either pure exponentially increasing or exponentially increasing sinusoidal natural responses, leading to an unstable system. Moreover, poles of multiplicity greater than 1 on the imaginary axis lead to the sum of responses of the form where , which also approaches infinity.

A marginally stable system is obtained when the closed loop transfer functions have only poles on the imaginary axis of multiplicity 1 and all other poles in the left half plane.

un

Li

6.2 ROUTH-HURWITZ CRITERION


Using this criteria, we can determine how many closed loop system poles are in the left half plane and in the right half plane. The process required : 1. 2. Generate a Routh Table Determine how many times the numbers change sign to determine how many right half plane poles

The power of the Routh-Hurwitz criterion allows one to determine what values a parameter can be to yield a stable design.

THE BASIC ROUTH TABLE


Consider the transfer function:

We then fill the remaining entries by noting that each entry is a negative determinant of entries in the previous two rows divided by the entry in the first column directly about the calculated row. The left hand column of the determinant is always the first column of the previous two rows, and the right hand column is the elements of the column above and to the right.

Ed

un

To create the Routh Table, we label the rows with powers of s starting from the highest power. We then start with the coefficient of the highest power of s in the denominator and list, horizontally in the first row, every second coefficient. In the second nd row, list horizontally, starting with the 2 highest power of s, and every other coefficient that was skipped in the first row.

Li

Example 6.1 Make a Routh Table for the system shown.

The closed loop transfer function is given by:

1 10

31 1030 0

1030

d
0

INTERPRETING THE ROUTE TABLE

The number of roots of the polynomial that are in the right half plane is equal to the number of sign changes in the first column

m Ed

un

Thus, if a closed loop transfer function has all poles in the left half of the s-plane, the system is stable. For the previous example, there were 2 sign changes, which indicates 2 poles in the right half plane and 1 pole in the left half plane.

Li
0 0 0 0

Example 6.2 For the system shown, find the range of values of gain K (which is to be positive), for which the system is stable.

1 18

77 K

d
0

Ed

un

Li
0 0 0 0

6.3 STABILITY IN STATE SPACE


We can determine the stability of a system represented in state space by finding the eigenvalues of the system matrix, A and determining their locations on the s-plane. We stated that the solution of:

Give us the poles of the transfer function:

Example 6.3 Given the system

Find out how many poles are in the left half plane and the right half plane.

We must first find

m Ed

Since there is one sign change in the first column, the system has one right half plane pole and two left half plane poles. Thus, it is unstable.

un
1 -7 0 0 -6 -1 -26 -26

Li

6.4 STEADY STATE ERROR


Recall that the steady state error is the difference between the input and the output for a prescribed test input as . Test waveforms such as the step, ramp and parabola allows us to determine the state of the system under different conditions: Step input: represent constant position and are useful in determining the ability of the control system to position itself with respect to a stionary target Ramp input: represent constantvelocity inputs to a position control system by their linearly increasing amplitude. Parabolas: represent constant acceleration inputs to position control systems and can be used to represent accelerating targets.

Ed

Thus, the larger the value of K, the smaller the value of of .

On the other hand, if an integrator is placed in the forward path as shown, there will be zero error in steady state for a step input. This is because as increases, the error decreases until there is zero error. Note that there will still be a value for since an integrator can have a constant output without any input.

un

The steady state errors that we look at here arise from the configuration of the system itself and the type of applied input. Consider the system shown which has an error of In steady state, if equals , will be zero. But with a pure gain, K, the error, cannot be zero if is to be finite and nonzero. Thus, by virtue of the configuration of the system, an error must exist:

would have to be to yield a similar value

Li

6.5 STEADY STATE ERROR FOR UNITY FEEDBACK SYSTEMS


Consider the figure shown.

But

To find the steady state error, we use the final value theorem:

Let us consider now a feedback control system as shown. As we have shown previously:

If the system is stable then:

From this, we determine the steady state error for the 3 most common test signals.

A step input has a Laplace transform of 1/s so:

Ed

is considered the DC gain of the forward transfer function, since s, the frequency variable, is

approaching zero. In order to have zero steady state error then

This occurs if there is at least one pure integration in the forward path resulting in the forward transfer function having the form:

6.5.1 STEP INPUT

un

Li

6.5.2 RAMP INPUT


A ramp input is given by: . We thus obtain:

To have zero steady state error then:

In other words, there must be at least two integrations in the forward path. If only one integration exists in the forward path then will be finite and given by

. This leads to a constant steady state error for a

would be infinite.

For a parabolic input:

. Hence:

In order to have zero steady state error for a parabolic input, we must have:

Ed
Which can only occur if:

If only 2 integrations exist in the forward path, we obtain an finite and constant steady state error, while a lower number of n will yield an infinite steady state error.

un

6.5.2 PARABOLA INPUT

ramp input. If there are no integrations in the forward path result in

Li
, and the steady state error

Which indicates that

must be in the form:

Example 6.4 Find the steady state errors for the inputs of to the system shown.

We verify that the closed loop system is indeed stable by using the Routh-Hurwitz criterion. We then use the final value theorem on the step input:

Where For a ramp input though:

For a parabolic input:

Ed
But

un

Li

6.6 STATIC ERROR CONSTANTS & SYSTEM TYPES


The steady state error performance specifications are called static error constants. Recall that:

The three limit terms are known as static error constants with the position constant

And the acceleration constant

given by:

m Ed

un
Since the steady state errors are dependent upon the number of integrations in the forward path, we give a name to this system attribute. We define the system type to be the value of n in the denominator, or, equivalently, the number of pure integrations in the forward path. Thus indicates a type 0 system.

The velocity

given by:

Li
defined as:

6.7 STEADY STATE ERROR SPECIFICATIONS


Static error constants can be used to specify the steady state error characteristics of control systems. For example, if a control system has the specification , we can draw from this: That the system is stable The system is a Type 1, since only type 1 systems have a finite A ramp input is the test signal. Since is specified as a finite constant, and the steady state error for the ramp input is inversely proportional to , we know that the test input is a ramp The steady state error between the input ramp and the output ramp is per unit of input slope.

Given the control system in the figure, find the value of K so that there is a 10% error in the steady state.

Applying the Routh-Hurwitz criterion, we see that the system is stable at this gain.

Ed

un

This is a type 1 system since there is 1 pure integration in the forward path gain. The input signal must also be a ramp, in order to yield a finite error in a type 1 system. Thus:

Li

Example 6.5

6.8 STEADY STATE ERROR FOR DISTURBANCES


The advantage of using feedback is that regardless of disturbances, the system can be designed to follow the input with small or zero error. Consider the following system subject to disturbances between the controller and the plant. The transform of the output is given by:

Using

To find the steady state value of the error, we apply the final value theorem to obtain:

Where

is the steady state error due to

disturbance. If we assume a step disturbance

Can be decreased by increasing the DC gain of (which lowers the value of that is fed back) or decreasing the DC gain of , which yields a smaller value of as predicted by the feedback formula.

Ed

Example 6.6

Find the steady state error component due to a step disturbance with

This shows that the steady state error produced by the step disturbance is inversely proportional to the DC gain of .

un
, and , then:

d
is the steady state error due to the

Li

6.9 STEADY STATE ERROR FOR NONUNITY FEEDBACK SYSTEMS

The steady state error is given by:

If we now consider step input and disturbances then:

For zero error:

Ed
1. 2. 3. 4.

It is possible that the steady state error is zero if: The system is stable is a type 1 system is a type 0 system H(s) is a type 0 system with a DC gain of unity

un

Li

Example 6.7 Determine the system type, error constant, and the steady state error for a unit step.

We first create a system with unity feedback as shown.

Ed

This is a type 0 system since there is no pure integration. The appropriate static error constant is

The negative value for steady state error implies that the output step is larger than the input step.

un
. Thus:

We then simplify and note that

This then creates a feedback loop. Let

Li

6.10 SENSITIVITY
The degree to which changes in system parameters affect system transfer functions and hence performance, is called sensitivity. A system with zero sensitivity is ideal. Mathematically, sensitivity is the ratio of the fractional change in the function to the fractional change in the parameter as the fractional change of the parameter approaches zero:

Example 6.8

Find the sensitivity of the steady-state error to changes in parameter K and parameter a for the system shown.

This is a type zero system so the steady state error is:

For parameter a:

Ed
For parameter k:

un

Li

6.11 STEADY STATE ERROR FOR SYSTEMS IN STATE SPACE


There are two methods for calculating the steady state error: 1. 2. Analysis via final value theorem Analysis via input substitution

6.11.1

FINAL VALUE THEOREM APPROACH

Consider a closed loop system represented in state space by:

The Laplace transform of the error is:

But:

, thus:

Where

is the closed loop transfer function given by

6.11.2

INPUT SUBSTITUTION APPROACH

Ed
Where is constant. Also, We thus get:

STEP INPUTS

Given the state equations as shown initially, and an input unit step where for is:

This method avoids taking the inverse of systems.

un

and can be expanded to multiple input and multiple output

. This was proved previously.

Li
, a steady state solution,

But the steady state error is the difference between the steady state input and the steady state output. The final result for the steady state error for a unit step input into a system represented in state space is:

RAMP INPUTS
For unit ramp inputs, , a steady state solution for is:

Where

and

are constants. Hence,

The state equations are then:

From (1):

Substituting into (2) yields:

Ed
This yields a steady state error:

un

Li

7.1 VECTOR REPRESENTATION OF COMPLEX NUMBERS


Any complex number, , can be represented by a vector with a magnitude M and angle : complex number if substituted into a complex function , another complex number will result. For example, we can represent 1. 2. in two ways: . If the

The traditional radius vector approach The vector drawn from the zero of the function to the point s.

Using the segment approach, we find that the magnitude M of

at a point s is given by:

Where

is the magnitude of the vector drawn from the zero of F(s) at

Ed

un

length being the magnitude of the vector from the pole of F(s) at at the point s is then given by:

Li
to the point s, and the pole to the point s. The angle of

7.2 ROOT LOCUS DEFINITION


The root locus technique can be used to analyse and design the effect of loop gain upon the systems transient response and stability. Consider the system shown and the table which represents the nature of the change in the gain with respect to changes in the poles. The representation of the paths of the closed loop poles as the gain is varied is called the root locus. It provides solutions for systems of order higher than two Describes qualitatively the performance of a system as various parameters are changes Gives a graphical representation of a systems stability

Ed

7.3 PROPERTIES OF THE ROOT LOCUS

Consider the general control system which has a transfer function:

From this, we see that a pole exists (and are part of the root locus) when the denominator becomes zero. This condition can also be expressed as:

un

From the given root locus, we can see that the system is overdamped for gains less than 25 (since the poles are completely real). When the gain is 25, the system is critically damped, and beyond this, the system is underdamped. When the system is underdamped, the settling time is constant since the real part of the poles do not change. Also, as we increase the gain, the damping ratio diminishes, and the percent overshoot increases.

Li

The gain K, at a point for which the angles add up to an odd multiple of 180 degrees is found by dividing the product of the pole lengths by the product of the zero lengths.

Example 7.1 Given a unity feedback system with a forward transfer function:

Calculate the angle of at the point by finding the algebraic sum of the angles of the vectors drawn from the zeros and poles to the given point. b) Determine if the point specified is on the root locus c) If the point specified in part (a) is on the root locus, find the gain, K, using the lengths of the vectors.

a)

From the plot, we can easily find the sum of the angles:

b)

The point can be on the root locus since the sum of angles is an odd number of 180 degrees. The only other criteria is that:

Ed
c) We use:

un

a)

We first find the poles and zeros:

Li

7.4 SKETCHING THE ROOT LOCUS


1. 2. 3. Number of branches: is equal to the number of closed loop poles. The root locus is symmetrical about the real axis On the real axis, for K>0, the root locus exists to the left of an odd number of real axis , finite open loop poles and/or finite open loop zeros. The root locus begins at the finite and infinite poles of zeros of . and ends at the finite and infinite

4.

5.

Every function of s has an equal number of poles and zeros if we include infinite poles and zeros as well as the finite poles and zeros. The root locus approaches straight lines asymptotes as the locus approaches infinity. The equation of these asymptotes is given by the real axis intercept and the angle :

Example 7.2

Sketch the root locus for the system shown.

Ed

We can easily plot the open loop poles and zeros and determine the root locus on the real axis, as shown.

un

Li

We then wish to find the asymptotes which are given by:

Thus, the asymptotes cut the real axis at -4/3. The angle of intersection is given by:

Ed

un

Since there are more open loop finite poles than zeros, then there must be zeros at infinity. The asymptotes tell us how we get to these zeros at infinity.

Note that the number of asymptotes obtained is equal to the difference between the number of finite poles and number of finite zeros.

Li

7.5 REFINING THE ROOT LOCUS


The following rules help us to find specific points on the root locus such that we can better understand what the root locus looks like.

REAL AXIS BREAKAWAY AND BREAK-IN POINTS


The point where the locus leaves the real axis, called the break away point occurs where the gain is a maximum, and breaks into the real axis at the break in point where the gain is a minimum, We can find the break away and break in point by using differential calculus and noting that the points are on the real axis such that . But on the real axis we get:

The breakaway and break-in points can then be found by solving:

The breakaway and break-in points also satisfy the relationship:

Where

and

are the negative of the zero and pole values of

un

d
We then solve for

Example 7.3

Ed
And

Find the breakaway and break-in points for the root locus shown using differential calculus, if we are given that the open loop system has a root locus of:

To find the breakaway and breakin points on the real axis we note that they lie on the root locus so:

which yield the breakaway and break-in points without differentiating.

Li

Thus:

, which are the breakaway and break-in points respectively. We do not need to check that they are indeed minimums or maximums since we know that there is a maximum between -1 and -2 and a minimum between 3 and 4. We verify this using the transition method:

AXIS CROSSINGS

The axis crossing is a point on the root locus that separates the stable operation of the system from the unstable operation. The value of at the axis crossing yields the frequency of oscillation.

R OUTH -H URWITZ C RITERION

Ed
Example 7.4

As already seen, forcing a row of zeros in the Routh table will yield the gain. Going back one row to the even polynomial equation and solving for the roots yields the frequency at the imaginary axis crossing.

Find the frequency and gain, K, for which the root locus crosses the imaginary axis for the closed loop system shown with a closed loop transfer function:

un

Li

We start off by drawing the Routh table: 1 7 90-K 14 8+K 21K 3K

21K

Somce K>0, then the only line that can be zero is the

line.

Then in the

line we substitute to get the polynomial equation:

Solving for this, we find that . Thus, the root locus crosses the We conclude that the system is stable for .

S UM OF A NGLES
At the

axis crossing, the sum of angles from the finite open loop poles and zeros must add to

un

By searching on the

axis for when this situation occurs, we can determine the

Ed

C HARACTERISTIC E QUATION

By letting , in the characteristic equation, and equating both the real and imaginary part to zero, we can solve for and K.

Li
axis at at a gain of 9.65. . axis crossing.

ANGLES OF DEPARTURE AND ARRIVAL


We can calculate the root locus departure angle from the complex poles and the arrival angle to the complex zeros. Recall that on the root locus:

Ed

To determine the angle of arrival at a complex pole, consider a point on the root locus close to the complex zero. Again, we sum the angles drawn from all finite poles and zeros and equate this to an odd multiple of 180 There is only one unknown angle in the sum, the angle drawn from the zero that is close, that corresponds to the angle of arrival:

un

Li

If we assume a point on the root locus close to a complex pole, the sum of angles drawn from all the finite poles and zeros to this point is an odd multiple of . Except for the pole that is close to the point, we assume all angles drawn from all other poles and zeros are drawn directly to the pole that is near the point. Thus, the only unknown angle in the sum is the angle drawn from the pole that is close, which corresponds to the angle of departure.

Example 7.5 Given the unity feedback system, find the angle of departure from the complex poles and sketch the root locus.

Now:

PLOTTING AND CALIBRATING THE ROOT LOCUS

If a few test points along the

Ed
This is shown in the figure.

And the corresponding gain by:

Let us assume we want to find the exact point at which the locus crosses the 0.45 dmping ratio line and the gain at that point. We plot the root locus plot. line are selected, we can evaluate which points yield:

un

Li

We note that the open loop poles are found by solving: , giving us , and there is one open loop zeros at s=-2. Since the complex poles are complex conjugates, we only need to work with one of the poles and use symmetry to find the angle of departure.

Example 7.6 Given a unity feedback system that has the forward transfer function of:

b) Find the imaginary axis crossing We can use the Routh Table since the closed loop transfer function is:

We force a row of zeros for row since it is the highest row which can be all zero. Solving this yields . Then using the row above we find:

c)

Find the gain K, at the imaginary axis crossing As shown in the last section, the gain, K=1.

Ed
d) Find the break-in point Since

must lie between 2 and 4, then

Thus, the

axis crossing is at

un
.

Li

a) Sketch the root locus Recall that the number of branches is equal to the number of closed loop poles. This happens to be 2. Solving yields . These two finite poles will be part the start of the root locus, and the two zeros will be the end of the root locus, as shown.

e)

Find the point where the locus crosses the 0.5 damping ratio line. Searching along for the 180 point, we find

f)

Find the gain at the point where the locus crosses the 0.5 damping line.

This yields:

g)

Find the range of gain, K, for which the system is stable.

Since the root locus starts on the left side and ends on the right, the maximum gain for which the system is still stable. Thus:

7.6 TRANSIENT RESPONSE DESIGN VIA GAIN ADJUSTMENT

1. 2.

3.

Ed
1. 2. 3. 4.

Higher order poles are much farther to the left of the s plane that dominant second order pair of poles. Closed loop zeros near the closed loop second order pole pair are nearly cancelled by the close proximity of higher order closed loop poles Closed loop zeros not cancelled by the close proximity of higher order closed loop poles are far removed from the closed loop second order pole pair.

The design of a high order system can be summarized as: Sketching the root locus Assuming the system is a second order system without any zeros and then find the gain to meet the transient response specification Justify the second order assumption by finding the location of all higher order poles and evaluating the fact that they are much farther from the axis than the dominant second order pair a factor of 5 times approximately. If closed loop zeros are not cancelled by higher order closed loop poles, then they should be far removed from the dominant second order pole pair. Otherwise simulate the results.

un

Recall that our analysis of transient response was limited to second order systems. However, if we can justify that a second order approximation is possible, then we may use the same techniques:

Li
crossing gain represents the

7.7 GENERALISED ROOT LOCUS


In many cases, we wish to know how the closed loop poles change as a function of another parameter. Consider the open loop forward path transfer function to be:

We now find the closed loop function, but also in the process want :

to appear as the forward path gain

Then divide by is in the form

so that the denominator :

This conceptually implies that:

We can thus sketch the root locus now. The zero is at s=-2 and the poles are at .

Ed

un

Li

8.8 MATLAB
For the unity feedback system with a forward path of:

a) b) c) d)

Find the operating point at which the damping ratio is 0.45. The axis The breakaway point The range of K within which the system is stable

Plot the root locus >> >> >> >> >> >> sgrid(z,wn) clf clear all numg=[1 -4 20]; deng=poly([-2 -4]); GH=tf(numg,deng)
5 4 3 2
Imaginary Axis

1 0 -1 -2 -3 -4

>> rlocus(GH)

un
-5 -5 -4

d
-3 -2 -1 Real Axis 0 1 2 3

Transfer function: s^2 - 4 s + 20 -------------s^2 + 6 s + 8

Zoom in

Imaginary Axis

>> axis([-3 1 -4 4]) %axis([xmin xmax ymin ymax])

Ed

-1

-2

-3

Sgrid show damping line and natural frequencies

-4 -3

-2.5

-2

>>z=0:0.05:0.5;

4 0.5 3 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05

%damp ratio from 0-0.5 in steps of 0.5 >> wn=0:1:10; %natural frequency in steps of 1 from 1-10 >> sgrid(z,wn) %plot on graph
Imaginary Axis

-1

-2

-3 0.5 -4 -3 -2.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 -2 -1.5 -1 Real Axis -0.5

Li
Root Locus -1.5 -1 Real Axis -0.5 0 0.5 1
Root Locus 4 3 2 1 1 2 3 4 0 0.5 1

Root Locus

Select points (3 points on graph) >> for k=1:3 [K,p]=rlocfind(GH) end

Root Locus 4 0.45 3

1
Imaginary Axis

-2

-3

0.45 -4 -3 -2.5 -2

un
-1.5 -1 Real Axis -0.5 0 0.5 1

-1

Ed
Smoothing the root locus >> K=0.005 Rlocus(G,K)

a) b) c) d) Using the plot of the 3 points, we find that the maximum gain occurs on the imaginary axis, with an associated gain of 1.5. Thus the system is stable for 0<k<1.5

%specify range of gain to smooth

Li

Input command Allows custom keyboard entry of value >> pos=input(type %OS)
Step Response 1.4

1.2

Find closed loop transfer function After selecting the operating point using the [K,p]=rlocfind(G) command:
Amplitude

0.8

>> T=feedback(K*G,1)

0.6

0.4

Simulate design step input


0.2

>> step(T)

d
0 0 1 2 3

Ed

un

Li
4 5 6 7 8 Time (sec)

8.1 IMPROVING TRANSIENT RESPONSE & STEADY STATE ERROR


TRANSIENT RESPONSE
Our goal here is to get the desired transient response even though the operating point is not on the original root locus. Rather than change the existing system, we augment, or compensate, the system with additional poles and zeros, so that the compensated system has a root locus that goes through the desired pole location for some value of gain. The compensated system can be realized using a passive or active network (ideal compensators) and presents little need to interfere with the power requirements of the system or the load for that matter. A disadvantage is that the compensated system is of a higher order making it necessary to simulate the system after the design is complete. We have seen that the transient response will be improved if we insert a differentiator in the forward path in parallel with the gain since large changes yield a large derivative signal that drives the plant, while small changes yield a small derivative signal, and the output from the differentiator becomes negligible compared to the output from the gain.

STEADY STATE ERROR

We have seen that the steady state error can be improved by adding an open loop pole at the origin in the forward path, thus increasing the system type and driving the associated steady state error to zero. This additional pole at the origin is realized with an integrator.

In summary:

Improvements in transient response require differentiators Improvements in the steady state error require integrators in the forward path

Ed
There are two techniques:

8.2 IMPROVING STEADY STATE ERROR VIA CASCADE COMPENSATION


Compensated systems can be arranged using either cascade compensation or a feedback compensation. Here, we will only be looking at the cascade compensation, whose system diagram is as shown.

Ideal integral compensation: uses a pure integrator to place an open loop, forward path pole at the origin, thus increasing the system type and reducing the error to zero. This type of system is called the proportional-plus-integral (PI) controller, since the implementation consists of feeding the error and the integral of the error forward to the plant. Lag compensation: does not use a pure integration and places the pole near the origin, but is unable to reduce the error to zero.

un

If we use dynamic compensators, compensating networks can be designed that will allow us to meet transient and steady state error specifications simultaneously.

Li

IDEAL INTEGRAL COMPENSATOR


To show how steady state error is improved without affect the transient response, consider the root locus shown. In this case, the root locus goes through point A. Suppose that this is the desired operating point. To reduce the steady state error, we then add a pole at the origin; the root locus no longer passes through A. To solve this problem, we add a zero closed to the pole at the origin as shown. The total angular contribution of the compensator zero and pole cancel out, and point A is now on the root locus.

A compensator with a pole at the origin and a zero close to the pole is called an ideal integral compensator.

LAG COMPENSATION TO IMPROVE STEADY STATE ERROR

Ed

With lag compensation, the pole and zero are moved to the left of the origin. This does not increase the system type and improves the static error on an uncompensated system. Similarly, we find that the transient response is not changed much since the total angular contribution of the compensator pole and zero is approximately zero degrees. Similarly, the gain remains relatively unchanged since the length of the vectors drawn from the lag compensator are approximately equal.

un

Li

Without loss of generality consider a type 1 static error constant:

The new static error constant after lag compensation is:

To minimize the error, the static error constant must be larger, and this is achieved by placing both the pole and zero close to the origin, with the zero left of the pole.

Lag compensation offers an improvement of the static error constant by a factor of

Example 8.1

Compensate the system shown, whose root locus is shown, so that the steady state error is improved by a factor of 10 if the system is operating with a damping ratio of 0.174.

Ed

As seen in the root locus, the desired operating point is at yields:

m
, with a gain of K=164.6. This The new steady state error must be:

un

Li
.

The improvement in

is the required ratio of the compensator zero and pole:

We choose

We then must check:

Where the dominant second order poles are and the associated gain: Where the third and fourth closed loop poles are: -11.55, -0.101 Whether the fourth pole of the compensated system cancels its zero.

Ed

un

Li

8.3 IMPROVING TRANSIENT RESPONSE VIA CASCADE COMPENSATION


There are two ways to improve the transient response of a system via compensation: 1 2 Ideal derivative compensation: uses a pure differentiator and is referred to as a proportional-plusderivative (PD) controller. This technique is prone to high frequency noise. Lead compensation: This does not use a pure differentiation but rather a passive network which has a zero and pole closed to the origin, with the pole being more distant.

PD CONTROLLER
By adding an additional zero to the forward path, we speed up the original system. The PD controller does just that and has a transfer function of:

The basic process of designing a PD controller is to: Evaluate the sum of angles from the open loop poles and zeros to a design point that is the closed loop pole that yields the desired transient response. The difference between 180 and the calculated angle must be the angular contribution of the compensator zero. Trigonometry is then used to locate the position of the zero to yield the required difference in angle

Example 8.2

Given the system, design an ideal derivative compensator with 16% overshoot with a threefold reduction in settling time. The operating point with the associated overshoot is . Assume a second order approximation.

Ed
The settling time is: This is the real part of the root locus (which must be in the left plane for stability) that intersects the damping line . The imaginary part is thus given by:

un

Li

Now, the additional zero must have an angular contribution equal to that of the sum of the other open loop zeros and poles:

With lead compensation:

Ed

A pole and zero pair must be introduced, with the pole farther from the imaginary axis than the zero, resulting in the angular contribution of the compensator to be still positive, and thus approximating an equivalent single zero. Implementation is done using passive components There is less sensitivity to noise We arbitrarily select a lead compensator pole or zero during design There are an infinite number of lead compensators that can meet the transient response requirements, but the design usually is constrained by the static error constant, gain and second order approximation

LEAD COMPENSATION

un

Li

Example 8.3 Design a lead compensator that will reduce the settling time by a factor of 2 while maintaining 20% overshoot. Assume a second order approximation.

The new operating point is thus

Using trigonometry, we can determine the pole location, as shown in the figure: The pole contribution angle must be

Ed

un

Let us choose . We then find the angle contribution of the current zeros and poles to be:

Li
.

8.4 IMPROVING STEADY STATE ERROR & TRANSIENT RESPONSE


The process of improving both the steady state error and transient response is a two step process: We first improve the transient response by using a ideal derivative or lead compensator Then we improve the steady state error by using an ideal integrator or lag compensator

A problem with this approach is that improvement in the transient response in some cases yields deterioration in the improvement of the steady state error, but this is not always the case. Such a design process yields A PD controller followed by a PI controller, or alternatively a proportional-plus-integral-plusderivative (PID) controller. If we first design a passive lead compensator and then design a passive lag compensator, the resulting compensator is called a lag-lead compensator.

PID CONTROLLER DESIGN


The general transfer function is:

The process of designing such a PID controller involves: 1. 2. Evaluate the performance of the uncompensated system to determine how much improvement in transient response is required. Design the PD controller (lead compensator) to meet the transient response specifications zero and loop gain or for a lead compensator, also the pole location. Verify the system has met the requirement. Redesign if not. Design the PI controller (lag compensator) to yield the steady state error Determine the gains Simulate the system to be sure all requirements have been met.

Ed
3. 4. 5. 6.

Which has two zeros plus a pole at the origin one zero, pole pair used as an ideal integral compensator and the other zero to be used as an ideal derivative compensator.

un

Li

Example 8.4 Given the system shown, design a PID controller so that the system can operate with a peak time, two thirds of that of the uncompensated system at 20% overshoot and with zero steady state error.

Step 1

We thus find the dominant poles to be the intersection of the root locus and the damping line of 0.456:

Step 2

We determine the original peak time and find the new peak time:

Ed
The real part must be: Step 3

The new peak time is to be 2/3 of 0.297=0.198. The imaginary part of the new dominant poles is thus:

We then find the total angular contribution of the poles and zeros:

un

Li

The zero must then contribute:

Then, using trigonometry:

The PD controller is described by:

Step 4 We now design the ideal integral compensator for zero steady state error. Any ideal integral compensator will work as long as the zero is placed close to the origin:

We then check the root locus characteristics:

Ed
Step 5

We now determine the 3 values for the gain:

un

Li

We then simulate the result to check the reduction in the peak time.

Step 6 We verify with simulations that we have indeed improved both the transient and steady state error. The complete PID controller further improved the steady state error with appreciably changing the transient response designed with the PD controller it is slightly longer than the PD controller.

If we want a faster response, we must move the zero farther from the origin for the PI controller.

Example 8.5

Design a lag lead compensator with 20% overshoot and a twofold reduction in settling time and a 10 fold improvement in steady state error for a ramp input.

Ed
Step 1: Evaluate performance: 192.1. Step 2: Evaluate settling time:

corresponds to a 20% overshoot. We find the dominant poles at

m
, with a gain of The new real part of the new dominant pole in the negative direction is:

un

Li

The imaginary real part of the dominant pole is:

Step 3: Design lead compensator A lead compensator has both a zero and pole. We select the compensator zero to be -6. Now, the sum of the angles to the design point is:

The angular contribution of the pole must be

By using Matlab, we find the gain at the design point is 1977. We also verify results using simulation.

Ed

un

Li

Step 4: Steady state error using lag compensator The steady state error of the uncompensated system is:

The new steady state error must be:

Note that the current lag compensated system has a steady state error of

Step 5: Select values for lag compensator We select a pole to be close to the origin: 0.01. Thus:

Step 6: Combine lag-lead system

Ed
The new dominant, closed loop poles are at with a gain of 1971. Notice that the lag-lead compensation has indeed increased the speed of the system, as witnessed by the settling time or the peak time. The steady state error for a ramp input has also decreased by about 10 times.

The final transfer function of the open loop is:

un

Li

8.5 PHYSICAL REALISATION


With active circuit realizations, the transfer function can be formed by the use of an inverting operational amplifier. By judicious choice of , this circuit can be used as a building block to implement compensators and controllers. By using the basic building blocks, we can cascade compensators to build larger compensators such as the lag lead compensator.

Ed

un

Li

We may also choose to do passive circuit realization. Note that cascading these networks require buffers since one network can load the other thus changing the overall transfer function.

8.6 SUMMARY
Function Improve steady state error Improve steady state error Improve transient response Improve transient response

Compensator PI

un
Transfer Function

m
Lag PD Lead PDI Lead-lag

Ed
Improve steady state error and transient response Improve steady state error and transient response

d
Characteristics
Increases system type Error is zero Zero at is small and negative

Li
Active circuit implementation
Error improved Pole is small and negative Zero is left of pole Passive implementation Zero selected to put design point on root locus Can cause noise and saturation Active circuit implementation Zero and pole selected to put design point on root locus Pole is to the left of zero Passive Implementation Lag zero left of pole at origin improves steady state error response

Lead zero improves transient Active circuit implementation Can cause noise and saturation
Lag zero left of pole at origin improves steady state error

Lead zero improves transient


response

Passive Implementation

9.1 CONCEPT OF FREQUENCY RESPONSE


In this chapter, we present the design of feedback control systems through gain adjustment and compensation networks from the perspective of frequency response. This method has distinct advantages: Transfer functions can be modeled from physical data We can design compensators that meet both steady state error and transient response requirements We can find stability in nonlinear systems Settles ambiguities when sketching a root locus Before doing so we define the concept of frequency response of a system consisting of a relationship that describes the changes in the output magnitude and phase component of a signal when given a particular input. We consider a mechanical system with a block diagram as shown. We find that that the system can be describes as:

We thus show that the magnitude response is:

And the phase response is:

More generally, we can express it as:

Ed

The analytical expression for the frequency response given a transfer function

m
is thus:

un

Li

Example 9.1 Find the analytical expressions for the magnitude and phase responses of

Rationalising:

The resulting bode plot is shown. We shall consider in the next section how we can get approximate sketches of the bode plot by hand.

Ed

un

Then:

Li

9.2 BODE PLOTS


Bode plots are widely used in the engineering discipline because they characterize both the magnitude and phase response of a system. In order to obtain bode plots, we must first look at the transfer function, and if we are doing so by hand, we make generalizations about the transfer function. As we have just seen, any transfer function can be written in the form:

Where

denotes the DC gain, i.e

. The frequencies

represent the zeroes and the

BODES RULES
The task of plotting the magnitude and phase response becomes tedious, and for this reason, we make the following approximations for first order systems: As passes each pole frequency, the slope of the magnitude plot decreases by 20 dB/dec a tenfold change in H for a tenfold increase in frequency. As passes each zero frequency, the slope of increases by 20 dB/dec. We determine the phase response at low frequencies, the break frequencies and at high frequencies.

Ed

The table shows a summary of bode straight-line magnitude and phase plots.

un

Li

poles of the transfer function respectively.

BODE PLOTS FOR


We first normalize the expression:

We then let

. Note that as

. This equates to a magnitude response of:

This magnitude is approximate in the range of When

to

We note that this is a 6 dB increase for the doubling of the frequency. Finally for high frequencies:

Which is equivalent to:

If we plot against , we get a straight line with a slope of 20, that is a slope of 20 dB per decade or a slope of 6dB/octave, where an octave is a doubling of the frequency.

Ed
To get the phase response, we note that at low frequencies, the phase is completely real, that is high frequencies we have for the zero. , which is imaginary so , and at . The slope is 45 per decade upwards

un

Li

BODE PLOTS FOR

When

Which in dB is a constant:

In dB is a 20dB decade slope downwards:

We can normalize the bode plot by removing the 1/a term

un
and plotting

response.

From the above analysis, we also find that at low frequencies, the response is real, while at high frequencies the phase is . Again we normalize the plot so that the corner frequency is at .

Ed

BODE PLOTS FOR G(S)=S AND G(S)=1/S

Li
. This gives the following magnitude

When

Example 9.2 Construct the bode plot for the transfer function We first put the transfer function in standard form:

We can double check by converting the transfer function into decibels:

Ed
approximately zero, while at

The dotted lines show where each term is. We then add the lines together to get our magnitude plot. To get our phase plot

We then consider the cases in which:

Now, we draw the dotted lines for the terms of inverse tan. We note that at , the phase is approximately

un
, the phase is . We then add the lines together.

Li

This indicates that there are 2 corner/break frequencies at and a zero at 0. Hence, the plot should initially have a 20dB/dec increase, then countered by a -20dB/dec decrease to bring it linear, and then finally another 20dB/dec decrease.

Example 9.3 Sketch the bode plots for:

We first rearrange so:

The break frequencies are at 1,2 and 3. Since s=0 is a pole, then the magnitude plot begins a decade below the lowest break frequencies and extend a decade above the highest frequencies. The magnitude K determines whether the curve should be moved up or down the magnitude plot, but has not effect on the phase plot. We first take k=1 and denormalise later. We make a table as shown: Frequency (rad/s) 1 (start at -1) 2 (start at -2) -20 -20 -20 -20 0 -20 0 0 -40 -60

Thus using the table, we draw the required magnitude plot:

Ed

un

Pole at 0 Pole at -1 Pole at -2 Zero at -3 Total Slope (dB/dec)

0.1 (start at 0) -20 0 0 0 -20

Li
3 (start at -3) -20 -20 -20 20 -40

We take a similar approach in the phase response but the breaks exist at a decade above and below the break frequency. Thus, our table will consist of 0.1, 0.2, 0.3, 0, 20, 30. The zero at 0 is shifted a decade below to approximately 0.1 but is not shifted to 10 since 10 is significantly far away from 0. Frequency (rad/s) 0.3 0 -45 0 -45 -45 45 45 -45 0

Pole at -1 Pole at -2 Pole at -3 Total slope (deg/dec)

0.1 -45

0.2 -45 -45 -90

20 0 45 45

30

-45

0 0

We show that the contributions of the breaks in the decade below the break frequency and we show that the contribution is negligible at a decade above the breakfrequency.

Ed

un

Li

9.3 SECOND ORDER BODE PLOTS


The second order polynomial is of the form:

Without correction, the asymptotic approximation and the actual frequency response can be great for some values of . We will thus introduce a correction technique. For a general second order system, at low frequencies:

At high frequencies though,

So, the magnitude response becomes:

This indicates that the slope is at 40 dB/decade or 12 dB/octave. We also note that the break frequency is and at that point:

un

d
, and that the phase plot increases by 90 degrees per decade .

And this indicates that the phase is

Ed
NORMALISATION

For convenience, we normalize and scale our findings by dividing the magnitude by , and the scale frequency by . As such we plot:

This results in having a low frequency asymptote at 0 dB and a break frequency of 1 rad/s.

Li

As such, the magnitude response is:

CORRECTION TO SECOND ORDER BODE PLOTS


At the natural frequency, the correction on a normalized log magnitude:

On an unscaled magnitude plot, the correction at the natural frequency gives a magnitude of is:

SECOND ORDER PLOTS


In this case, the plots are reverse in polarity:

Ed

The magnitude curve breaks at the natural frequency and decreases at a rate of -40 dB/decade. The phase is 0 at low frequencies and decreases at at until , where it levels off at . The correction is on the normalized log magnitude plot or a magnitude of on an unscaled plot.

un

Li

9.4 SUMMARY OF BODES RULES


In summary we state that: As passes each pole frequency, the slope of the magnitude plot decreases by 20 dB/dec As passes each zero frequency, the slope of increases by 20 dB/dec. We determine the phase response at low frequencies, the break frequencies and at high frequencies. For second order systems, the correction to the normalised magnitude plot is or for the unscaled magnitude plot. For second order system, the phase plot breaks at a decade below the break frequencies and has negligible impact on the phase at a decade above the break frequency, as shown in the diagram.

Example 9.4 Draw the bode plots for

We draw a magnitude plot after normalizing the transfer function:

Ed
Pole at -2 Zero at -3 Total slope (dB/dec) Start plot 0 0 0 0

Note the 3/50 DC gain, we thus start the plot at Start at pole -2 -20 0 0 -20 Start at zero -3 -20 20 0 0 Start at -20 20 -40 -40

The corrected magnitude at the natural frequency is thus the magnitude is , as shown.

un
or the correction to

Li

We now turn to the phase plots: 0.2 (start pole at -2) -45 0.3 (start zero at -3) -45 45 0 0.5 (start at -5) -45 45 -90 -90

Pole at -2 Zero at -3 Total slope (dB/dec)

-45

20 (end pole at -2) 0 45 -90 -45

Li
30 (end zero at -3) 0 -90 -90 50 (end 0 0 as shown.

Since the second order system is dominant, the result is that the plot ends at

9.5 NYQUIST CRITERION

Ed
By Letting: We show that: From this:

The Nyquist criterion relates the stability of a closed loop system to the open loop frequency response and open pole location. We can thus make statements about the stability of the closed loop system knowing the open loop system, just like we did with the root locus.

1. The poles of 2. The zeros of

un
are the same as the poles of are the same as the poles of

, the open loop system , the closed loop system.

We also make a note that contours are a set of points drawn out in a mapping where we take points drawn out by contour A and substitute it into a function:

Assuming a clockwise direction for mapping the points on contour A to contour B:

Ed
From the last point about rotation cancelling we extend this idea to show that: Where N is the number of counterclockwise rotations of contour B about the origin P is the number of poles of inside contour A, or the number of poles of Z is the number of zeros of inside contour A, or the poles of the closed loop system

un

Li

1. The contour B maps in a clockwise direction if contour A encircles only zeros or only does not encircle poles 2. The contour B maps in a anti-clockwise direction if contour A encircles a pole 3. The contour B encircles the origin if contour A encircles either a pole or zero. 4. The contour B does not encircle the origin if contour A encircles a pole and zero due to rotation cancelling.

The contour mapping is the same for and , except that the previous is translated one unit to the left and this calls for us to count the rotations about -1 instead of about the origin. Finally, we state that the Nyquist stability criterion is: If a contour A, that encircles the entire right half plane is mapped through , then the number of closed loop poles, Z, in the right half plane equals the number of open loop poles, P, that are in the right half plane minus the number of counterclockwise revolutions, N, around -1:

Example 9.5 Determine the number of right half plane poles in the following mapping.

Since contour A does not encircle any poles or zeros, the contour B rotates clockwise. Also since contour A does not encircle any poles or zeros, the mapping does not encircle -1. As such:

Which means that there are no right half plane poles and the system is stable.

Ed

Example 9.6

Determine the number of right half plane poles in the following mapping.

Since there are two unknown zeros of , or alternatively the poles of the closed loop system, there are two clockwise encirclements about -1 in the Nyquist plot. P=0, N=-2 so, Z=-2. The system is then unstable.

un

Li

9.6 SKETCHING NYQUIST PLOTS


A sketch can be obtained by looking at the vectors of and their motion along the contour. When an open loop pole is situated along the contour enclosing the right half plane, we detour around the poles on the contour. If we detour to the right around the pole, the pole vectors rotate through . If we detour to the left of open loop poles, the pole vectors rotate through an angle of as we detour around it.

Example 9.7

Ed
Since:

Conceptually, we substitute points in the right half plane into:

The resultant vector R is the product of the zero vectors divided by the product of the pole vectors. As we move in a clockwise direction around the contour from A to point C, the resultant angle goes from to , since the poles gain in a counterclockwise direction, which explains the decrease in angle of the function . The resultant goes from a finite value at zero frequency to zero magnitude at infinite frequency since the length of the 3 infinite poles lengths is infinite.

un

Given the following system, sketching the Nyquist diagram for the system.

Li

Then:

The Nyquist plot starts at 50/3 with an angle of imaginary part remains negative until crosses the negative real axis:

. Now as

increases the real part remains positive and the

. When the imaginary part goes to zero, the Nyquist diagram

The real part of this is:

Which is approximately zero at

m Ed

un

d
Around the infinite semicircle from point C to point D, the vectors rotate clockwise each by , and thus the resultant undergoes counterclockwise rotation of , starting at point C and ending at point D. At point C, the angles are all 90 , and hence the resultant is given by: Which is, , and for point D, the angles are all , so the resultant is . By selecting intermediate points, we can verify the spiral. The mapping of the imaginary axis is a mirror image of the mapping of the positive imaginary axis. Thus the mapping of the section of the contour from points D to A is the mirror of the mappings from points A to C.

At infinite frequency,

Li

Example 9.8 Sketch the Nyquist plot of

Which yield:

Moving along BCD, the magnitude stays at 0, but the vectors change; at point C, the total angle contribution is , but at point D, the total angle contribution is . Thus the total change in the vector from to is .

Ed
And finally at A again:

The mapping from D to E is the mirror of the mapping of A to B. Finally over the detour, the resultant magnitude approaches infinity, since substituting the pole would yield an infinite magnitude. At point E:

Since the zero has negligible contributions, but each of the poles contribute

un

At point B, the zero makes an additional contribution of , and the total contribution is magnitude is also zero. Alternatively consider the low and high frequencies:

At point A, the two open loop poles contribute and the zero contributes nothing. This gives a total angle at point A of , since the angle contribution is the sum of zero angles minus the sum of poles angles. The zero is negligible. Close to the origin, the fuction is infinite, thus point A maps into point A, located at infinity at an angle of -180 .

Drawing a test radius shows one counterclockwise revolution and one clockwise revolution, yielding zero encirclements.

Li
at B. The . And at F:

9.7 STABILITY VIA THE NYQUIST DIAGRAM


We previously used the Routh-Hurwitz criterion and the root locus to find system stability. If we manipulate the gain, we manipulate the resultant by a constant factor in the Nyquist diagram. 1. 2. 3. Set K=1 and sketch the Nyquist diagram Consider the critical point to be at -1/K rather than at -1 Adjust the value of K to yield stability

If the Nyquist diagram intersects the real axis at -1, then the system is marginally stable.

We use for stability, the number of right half plane poles, Z must be zero:

Sketch the Nyquist diagram and determine what values of K yield stability for:

We first set K=1, and then sketch the Nyquist plot. The right half contour encircles 2 open loop poles, and we thus require the number of counterclockwise encirclements to be N=2 about -1/K. Starting at the origin we have , then moving to B, the contribution of the angles is zero, and the

magnitude is small. To find the negative axis crossing:

Ed
We take the imaginary part and solve it for the roots: Thus the negative axis is cut at -1.33. Finally, we note that since poles are encircled, the number of counterclockwise revolutions is 2.

un

Example 9.9

Li

So long as the critical point at can be decreased until:

, which yields -1.33 in

encircles

For a stable system, and if

STABILITY VIA MAPPING ONLY THE POSITIVE

Ed
Consider the diagram shown, which represents a system that is stable at low values of gain and unstable at high values of gain. Since the contour does not encircle open loop poles, the Nyquist criterion tells us that we must have no encirclements of -1 for the system to be stable. Thus, if the gain is small, the mapping will pass to the right of -1 for a stable system. For such a system then:

The loop gain K, must be so that the open loop magnitude is less than unity at that frequency where the phase angle is i.e the negative real axis.

un
, the system is marginally stable.

AXIS

Li
, the system is stable. Thus, K

For a system with stability achieved at high gains, the contour looks like that shown. In this case, we require two counterclockwise encirclements of the critical point for stability. Thus:

The system is stable if the open loop magnitude is greater than unity at that frequency where the phase angle is i.e the negative real axis.

Find the range of gain for stability and the gain for marginal stability for the unity feedback system with:

Let K=1. The point A maps to , and the curve moves clockwise since there are no encirclements required. Now at B, the total angle contribution yields , but the magnitude is zero so the Nyquist diagram is as shown.

Ed

Since the open loop poles are all in the left half plane, the Nyquist criterion tells us we want no encirclements about -1 for stability. Thus a gain of less than unity is required on the negative real axis.

un

Example 9.10

Li

To find the negative axis crossing:

We set the imaginary part to zero and get:

We substitute this back into the open loop transfer function and this yields -1/20. Thus, the gain can be increased to:

The system is marginally stable for K=20 (frequency of oscillation is

) and unstable for K>20.

Ed

un

Li

9.8 GAIN MARGIN AND PHASE MARGIN VIA THE NYQUIST DIAGRAM
We define:

Gain margin : The gain margin is the change in open loop gain, expressed in dB, required at 180 of phase shift to make the closed loop system unstable.

Phase Margin : The phase margin is the change in open loop phase shift required at unity gain to make the closed loop system unstable.

Ed

For the given diagram, since no encirclements are required, the gain margin expressed in dB is the log of the reciprocal of the real axis crossing:

The angle drawn out by the negative real axis and a line drawn from the origin to the point Q, the intersection of the Nyquist plot and the unit circle is the phase margin. Unfortunately, this is hard to find by hand, but can be done using Bode plots very easily or by computational tools

un

Li

Example 9.11 Determine the gain margin for the unity feedback system with the forward transfer function:

We thus find the negative axis crossing:

Thus,

Which yields a real part of:

is:

9.9 STABILITY, GAIN MARGIN, PHASE MARGIN USING BODE PLOTS


Stability can be determined using bode plots.

Ed
This is the case for

If we require no encirclements, then the gain should be less than unity, when the bode phase plot is

If we take K=40, the bode plots look like that shown. At , the magnitude plot is -20 dB. As such a gain of +20 dB is possible before the system is unstable. Since the plot is scaled for a gain of 40, the gain for instability is . Thus, stability is achieved when 0<k<200.

un

Thus, the gain margin can be increased by

before the real part becomes -1. Hence the gain margin

Li

Gain margin : Is how far below unity the magnitude plot is at a frequency phase plot is at Phase Margin : The phase margin is how far above , where the gain is unity.

where the

the phase plot is at the frequency

If:

Ed

Find the gain margin and the phase margin.

We use the bode plots for K=40. Thus, now that K=200, the magnitude plot would start at Since the plots have taken 20log40 as 0dB, the plot required should be 13.98 dB higher. Finding the gain margin is done by look at when the phase reaches the magnitude plot shows a gain of . The correction needed is then . This occurs at . Thus:

m
. . And must be made, so we are now at . Thus the

Example 9.12

The phase margin is found when the gain is 0dB. However, a correction of seeking the phase margin at the point -13.98 dB on the plot. The phase is gain margin is .

un

Li

9.10 RELATION BETWEEN CLOSED LOOP TRANSIENT AND CLOSED LOOP


FREQUENCY RESPONSES
Consider the second order feedback control system shown. The closed loop log magnitude plot is shown.

By finding the magnitude of the system and finding the maximum through differentiation, we can find the peak value of the closed loop magnitude response by just knowing the damping ratio:

At a frequency,

Ed

m
of: This also indicates that is also related to the percent overshoot. We also note that is not the natural frequency, but for low values of the damping ratio, we can assume the peak occurs at the natural frequency. Note that no peak will occur at frequencies above zero if .

un

Li

The bandwidth of the closed loop frequency response is defined as the frequency, magnitude response curve is 3 dB lower than its initial value at DC:

, at which the

Since we know the relationships:

and

, then we find the bandwidth is related to the

settling and peak time of a second order system.

Example 9.13

Find the closed loop bandwidth required for 20% overshoot and a 2 second settling time.

Ed

un

Li

9.11 RELATION BETWEEN CLOSED LOOP TRANSIENT AND OPEN LOOP


FREQUENCY RESPONSE
In this section, we consider situations in which we are given the open loop frequency response and wish to find information about: The system bandwidth The peak time The settling time The rise time

We simply state without proof that the phase margin can be determined as:

The closed loop bandwidth can be approximated as the frequency at which the open loop magnitude response is between -6 and -7.5 dB if the open loop phase response is between and -223

Ed

un

Li

To do so, we must find the phase margin in terms of the damping ratio and also a relationship between the closed loop bandwidth and open loop frequency response.

Example 10.14 Given the system shown and the bode diagrams (open loop transfer function always), find the settling time and peak time.

We attempt to find by first selecting the region between -6 and -7.5 dB and 135 and 225 . This is approximately 3.7 rad/s. The phase margin is about . We then use the following graph or to determine

Ed

Then:

un

Li

9.12 STEADY STATE ERROR CHARACTERISTICS FROM FREQUENCY RESPONSE


We show how to find values of the static error constants for equivalent unity feedback systems from unnormalised and unscaled bode plots.

POSITIVE ERROR CONSTANT


A type 0 system has the form:

Which will always yield a constant term at DC when we take the logarithm. Given that:

We find that this value can be obtained from the low frequency axis on an unscaled magnitude plot yielding:

VELOCITY ERROR CONSTANT

Ed
Which is the same as the velocity error constant.

A Type 1 system always starts with a -20 dB/dec drop. The extension of a -20 dB/dec slope to the frequency axis yields the velocity error constant,

un

The position error constant is found at the intersection of the magnitude 0 system always starts with a horizontal magnitude plot.

Since a type 1 system has an open loop transfer function of the form:

Which always yields a 20 dB decrease in the magnitude from 0 frequency. If we extend this, the intersection of the -20 dB/dec drop with the frequency axis yields:

Li
at DC. A Type

ACCELERATION CONSTANT
For a type 2 system:

There is an initial -40 dB/dec drop due to the term part of the function:

And the intersection with the frequency axis yields:

A Type 1 system always starts with a -40 dB/dec drop. The extension of a -40 dB/dec slope to the frequency axis yields the square root of the acceleration error constant,

For each unnormalised and unscaled bode plot, find the system type and the appropriate static error constant.

Ed

This is a type 0 system due to the initial horizontal magnitude. Thus:

This is a type 1 system due to the -20 dB initial drop. The zero dB crossing is approximately at 0.55 rad/s, and as such .

un

Example 10.15

Li

This is a type 2 system with initial 40 dB drop. The zero crossing is at 3 rad/s and this corresponds to .

so,

MATLAB COMMANDS

There are two ways to create a transfer function: zpk() and tf(). The first is used when we are given the transfer function is factorised form, while the second method is used when we know the coefficients of the transfer function numerator and denominator.

To get the bode plot/Nyquist for G=zpk([-20], [-1 -7 -50], 1) Bode(G); grid on Nyquist(G)

To plot

Numg[ 1 3];

Ed
G=tf(numg, deng) Bode(G)

Deng=conv([1 2], [1 2 25]);

[mag, phase, w]=bode(G); Points=[20*log10(mag(:,:), phase(:,:), w]

To find information about margins: [Gm, Pm, Wcg, Wcp]=margin(G);

un

Li

10.1 TRANSIENT RESPONSE VIA GAIN ADJUSTMENT


As we have already seen, we can change the transient response by adjusting the gain of a system. This was previously done by mathematical analysis and by the root locus method. We now use the frequency response tools to design systems.

The procedure involves: 1. 2. Drawing the bode plots for a convenient value of gain We determine the required phase margin from the percent overshoot:

4.

segment CD. We must then change the gain by an amount AB to force the magnitude curve to go through 0 dB at .

Example 10.1

Ed
1. 2. 3.

For the position control system shown, find the value of preamplifier gain, K, to yield a 9.5% overshoot in the transient response for a step input.

For convenience, we normalize the bode plot with K=3.6 such that 0 dB is at An overshoot of 9.5% corresponds to . We locate the point on the phase plot which is above i.e rad/s.

un

3.

We find the frequency

on the phase diagram that yields the desired phase margin given by the

Li
. This occurs at 14.8

4.

The gain adjusted open loop transfer function is then:

Ed

un
will yield the required phase margin.

At a frequency of 14.8 rad/s, the gain is -44.2 dB. This magnitude has to be raised to 0 dB to yield the required phase margin. Since the log magnitude plot was drawn for K=3.6, a 44.2 dB increase or

Li

10.2 LAG COMPENSATION


The function of the lag compensator is to: 1. Improve the static error constant by increasing only the low frequency gain without any resulting instability Increase the phase margin of the system to yield the desired transient response

2.

The general procedure is: 1. 2.

Set the gain, K, to the value that satisfies the steady state error specification and plot the bode plots Find the frequency where the phase margin is to greater than the phase margin that yields the desired transient response. This step compensates for the fact that the phase of the lag compensator may still contributed to of phase at the phase margin frequency. Select a lag compensator whose magnitude response yields a composite bode magnitude diagram that goes through 0 dB at the frequency found at step 2. a. Draw the compensators high frequency asymptote to yield 0 dB for the compensated system at the frequency found in step 2. b. Select the upper break frequency to be 1 decade below the frequency found in step 2 c. Select the low frequency asymptote to be at 0 dB d. Connect the compensators high and low frequency asymptotes with a -20 dB/decade line to locate the lower break frequency Reset the system gain, K, to compensate for any attenuation in the lag network in order to keep the static error constant the same as that found in step 1.

3.

Ed
4.

un

Where

Li

Consider the bode plot show with an unstable system since the gain is higher than unity at . The gain compensator, while not changing the low frequency gain, will reduce the high frequency gain. Thus the low frequency gain can be made high to yield a large without creating instability. The process of designing such a lag compensator is based on knowning that the transfer function of the lag compensator is:

Example 10.2 Given the system shown, use bode diagrams to design a lag compensator to yield a ten fold improvement in steady state error over the gain compensated system while keeping the percent overshoot at 9.5%.

1.

Find value of k which satisfies steady state

Recall that we found that

and the phase margin was

Thus the open loop system is:

Which is a type 1 system, with a velocity error constant of:

Ed

A ten fold increase in error means that the new velocity error constant must be:

Thus, the new value of K which satisfies the steady state is:

And the new open loop transfer function is:

un

We first find a value of K which satisfies the transient percent over shoot.

. This required K to be adjusted to:

Li

2.

Now fix the transient response by finding desired phase margin . We add . for compensation so

Since for 9.5% overshoot, the required phase margin is the phase margin is , and the phase now must be

This occurs at 9.8 rad/s. At this frequency, the magnitude plot must go through 0 dB, so we must introduce a 24 dB lag compensator at that particular frequency.

3.

We select the high frequency asymptote as -24 dB starting from a decade below the phase margin frequency, . Then starting at 0.98 rad/s, draw a 20 dB/decade line until 0dB is reached. The lower break frequency is the intersection of this line with the frequency axis, and is found to be 0.062 rad/s. To retain the value of , the compensator must have a DC gain of unity, so:

Ed
4. Thus:

un

Li

11.3 LEAD COMPENSATION


The lead compensator: 1. 2. Increases the bandwidth by increasing the gain crossover frequency Increases the phase margin and phase margin frequency to reduce the percent overshoot (larger phase margin) with smaller peak times (higher phase margin frequencies) The uncompensated system has a small phase margin (B) and a low phase margin frequency (A). Using a phase lead compensator, the phase angle plot is raised for higher frequency. At the same time, the gain crossover frequency in the magnitude plot is increased from A rad/s to C rad/s. These effects yield a larger phase margin (D), a higher phase margin frequency (C), and a larger bandwidth.

Ed

The maximum phase shift of the compensator,

m
is given by: With the compensators magnitude at being:

un

d
, with setting the compensator to unity. Some useful analytical expressions include finding the frequency , at which the maximum phase angle occurs:

Li

DESIGN PROCEDURE
1. Find the closed loop bandwidth required to meet the settling time, peak time, or rise time requirements using:

3.

Plot the bode magnitude and phase diagrams for this value of gain and determine the uncompensated systems phase margin

6. 7.

Determine the compensators magnitude at the peak of the phase curve Determine the new phase margin frequency by finding where the uncompensated systems magnitude curve is the negative of the lead compensators magnitude at the peak of the compensators phase curve. Design the lead compensators break frequencies to find T and the break frequencies Reset the system gain to compensate for the lead compensators gain

8. 9.

Ed

10. Check the bandwidth requirement is met

un

5.

Determine the value of

by using the additional phase contribution of the compensator.

4.

Find the phase margin to meet the damping ratio or percent overshoot requirement. Then evaluate the additional phase contribution required from the compensator

Li

2.

Since the lead compensator has negligible effect at low frequencies, set the gain, K, of the uncompensated system to the value that satisfies the steady state error requirement.

Example 10.3 Given a unity feedback system with a forward transfer function of compensator to yield a 20% overshoot, with a peak time of 0.1 seconds. , design a lead

1.

We first find the closed loop bandwidth requirement given

and

2.

Since

And

then:

3.

We make the bode plots.

Ed
4. Next the required phase margin is:

un

Li

Since the lead compensator also increases the phase margin frequency, we add a correction factor such that:

But currently, the systems phase margin is contribution of the lead compensator is:

and

. Thus, the total phase margin

This will start to yield a system with a phase margin of 5. 6. 7. This yields Then the compensators magnitude at is

with a bandwidth of 46.6rad/s.

Next, we find where the uncompensated system has a magnitude of -3.76 dB. This occurs at 39 rad/s, so our new phase margin frequency is

We now require to find the components 1/T,

Hence:

Ed
9.

The final compensated system is thus:

10. We now check the lead compensated open loop magnitude response. Recall that the connection between the open loop bode plots to the closed loop bandwidth is that the open loop response must be between -6 and -7.5 dB and 135 and 225 degrees. As such we estimate that , which exceeds our bandwidth requirement.

un

d
to derive the equation of the compensator

8.

Then we find T:

Li

10.4 LAG-LEAD COMPENSATION


Our approach to such design problems is to first perform the preliminary work to find the desired bandwidth, gain K to yield the static error constant, the required phase margin, selection of a new phase margin frequency around and to work out the phase contribution of a lead compensator given that the phase contribution of a lag compensator is about .

Such that:

Where

is the higher break frequency determined to be 1 decade below the new phase margin it determined to yield 0 dB (unity gain) when s=0. to be the selected new phase

frequency and

Ed
Where And: Is the lower break frequency.

Our final step is to design the lead compensator given that we know margin frequency, and we also know . Thus:

is the higher break frequency found by using:

un

Li

The next step is to find the lag compensator using the new phase margin frequency. Then using the above graph, we find a value of , to help us find:

Example 10.4 Given a unity feedback system where , design a passive lag-lead compensator using

bode diagrams to yield a 13.25% overshoot, a peak time of 2 seconds and

Preliminary Work 1. We first find the required bandwidth given

2.

Given

3.

We make the bode plots

Ed
4. The required phase margin is:

un

Li

5. 6.

We select a new phase margin frequency around

, lets say

The uncompensated phase is . The lag compensator will add about contribution so this gives a total of . We require a phase given by and as such, require the lead compensator to provide .

The lag compensator is not critical and only provides stabilization, not phase margin design. 7. Since we chose shown: , our higher break frequency will be at 0.18 rad/s. We find

We can now find our lower break frequency since:

Must be unity at DC so,

Lead Compensator 8. At

, the lag compensated system has a phase angle of , then:

un
. Sinev we know ,

Ed

Li
using the graph

Lag Compensator

11.1 DESIGN VIA STATE SPACE


In chapter 8, we used the root locus method to design a system that met both steady state and transient response. In the previous chapter, we used the frequency method to design our system. To finish off, we use the state space method to design systems which are: Non linear MIMO systems Specified for all higher poles

The disadvantage though is that we cannot specify the closed loop zero locations which frequency domain methods do allow through placement of the lead compensator zero.

CONTROLLER DESIGN

Consider an nth order feedback control system with an nth order closed loop characteristic equation:

If we introduce feedback of each state variable to the control, u:

Ed

To the state variable feedback representation:

The resulting signal flow diagram changes from the phase variable representation:

un

Let us consider a plant:

Li

We shall see that the phase variable form with its typical lower companion system matrix, or the controller canonical form, with its typical upper companion system matrix, yields the simplest evaluation of the feedback gains. The main design principles in this chapter consists of: Using the phase variable of controller canonical form Equating the characteristic equation of a closed loop system to a desired characteristic equating and then finding the values of the feedback gains

11.2 POLE PLACEMENT IN PHASE VARIABLE FORM (MATCHING COEFFICIENTS )


The procedure is to: 1. 2. 3. 4. 5. Represent the plant in phase variable form Feed back each phase variable to the input of the plant through gain Find the characteristic equation for the closed loop system represented Decide upon all closed loop pole locations and determine an equivalent characteristic equation Equate like coefficients of the characteristic equations and solve for

Ed
Where This gives:

The characteristic equation of the plant is thus:

With the closed loop state variable feedback being:

un

Recall that the phase variable representation of the plant is given by:

Li

Which gives a characteristic equation of the closed loop system of:

Given that we have a desired characteristic equation for proper pole placement, , we can compare the coefficients to find the feedback gains since

un

d
. Example 11.1

Ed
1. Thus:

Design a phase variable feedback gains to yield 9.5% overshoot and a settling time of 0.74 second for the given plant:

We first use the specifications to find the poles:

Li

Now the original system is a third order, so we must select another closed loop pole. Given a zero at -5, it would be wise to choose -5 as a pole so that it will cancel. We however choose -5.1 to show the effect of such a selection. The desired characteristic equation is thus:

2.

We first draw a signal diagram of the system and deduce the phase variable form

From the transfer function:

Using the phase variable form of the matrix (or the linear equations of them), we can draw our signal diagram with the feedback of the gains.

Ed

un

Li

The resulting closed loop systems state equation will be the same except for the system matrix which is now :

The characteristic equation is thus:

Comparing coefficients:

We thus obtain:

Since the zero is unchanged, the transfer function numerator is unchanged, only our denominator has changed which is given by

Ed

When we take a simulation of the closed loop system, there is a large steady state error given that the steady state response is 0.24 and not unity. A better system would be achieved if we had actually chosen -5 as our pole.

un

Li

The following example shows the same approach when the system is not in phase variable form. Example 11.2 Given a plant with a transfer function:

Design a state feedback for the plant represented in cascade form to yield a 15% overshoot with a settling time of 0.5 seconds.

We then break this down in the signal diagram:

We can derive the matrix representation of the cascade system by noting:

Ed

When adding feedback, our system matrix will then become:

un

Li

In cascade form:

And our characteristic equation is:

Our desired characteristic equation given:

Thus:

Matching coefficients:

11.3 CONTROLLABILITY

If any one of the state variables cannot be controlled by the control u, then we cannot place the poles of the system where we desire.

Conversely, if an input to a system can be found that takes every state variable from a desired initial state to a desired final state, the system is said to be controllable.

Ed

Consider the parallel system shown. If could not be controlled by u, and also exhibited an unstable response due to a nonzero initial condition, there would be no way to effect a state feedback design to stabilize . In such a case, a state feedback design is not possible.

un

Li

The state equation for the controllable system is:

While the state equation for the uncontrollable system shows that

is decoupled from u.

A system with distinct eigenvalues and a diagonal system matrix is controllable if the input coupling matrix B does not have any rows that are zero.

Recall that the rank of a matrix is determined by the number of linear independent rows or columns, or put simply in row echelon form, the number of non-zero rows. Using this, and the controllability matrix of an nth order plant whose state equation is

The controllability matrix must be of rank n. Another method of determining whether the rank of the determinant of is non-zero.

un

d
is n is if Given the system shown, determine its controllability

Ed

Example 11.3

Li

Thus, the state equation is:

Thus:

11.4 OBSERVER DESIGN

And an observer with the state equation

We take the difference:

Ed

Note that the speed of convergence between the actual and estimated state is the same as the transient response of the plant since the characteristic equation is the same. From the figure shown, we can increase the speed of convergence between the actual and estimated states using feedback. In designing an observer, the observer canonical form yields the easy solution for the observer gains. Thus, the state equations for the plant and observer should be in observer canonical form.

Similar to the design of the controller vector, K, the design of the observer consists of evaluation the constant vector, L, so that the transient response of the observer is faster than the response of the controlled loop in order to yield a rapidly updated estimate of the state vector.

un

Assuming a plant with state equation

Controller design relies upon access to the state variables for feedback through adjustable gains but this may not be available or viable. An observer sometimes called an estimator is used to calculate state variables that are no accessible from the plant. Here the observer is a model of the plant.

Li

The determinant of

is not zero, so the system is controllable.

With the feedback in place in the closed loop observer we find that:

Then:

The design then consists of solving for the value of

to yield a desired characteristic equation:

For an nth order plant, the observer canonical form is:

The characteristic equation for A-LC is then:

Ed

We then compare this characteristic equation with the desired characteristic equation to find the elements of .

un

We then select the eigenvalues of the observer to yield stability and a desired transient response that is 10 times faster than the controlled closed loop response.

Li

Example 11.4 Design an observer for the plant represented in observer canonical form. The observer will respond 10 times faster than the closed loop control system with poles at and -10.

1.

Draw the estimated plant in observer canonical form:

In controller form we reverse the order of the phase variables and rearrange:

Using the relationship of

Ed
2.

Take the difference between the plants actual output y, and the observers estimated output and add the feedback paths from this difference to the derivate of each state variable. Note the only difference is in the system matrix which is now A-LC

un

we derive the observer canonical form:

Li

In phase variable

3.

From step 2, we take the system matrix and find the observer error:

The observer characteristic polynomial is given by:

4.

Our desired characteristic polynomial needs to be 10 times faster. Thus the poles are 10 times greater than the original poles at and -10.

Ed
5.

Comparing the coefficients:

Again, for plants not in observer canonical form, we can do exactly the same thing although the calculations may be more difficult.

un

Li

Example 11.5 Design an observer for the phase variables with a transient response described by the plant is described by: and given

In phase variable form:

Then:

Thus:

From the problem, we find that the desired characteristic polynomial is:

Matching the coefficients then:

Our final solution is then:

Ed

un

Li

11.5 OBSERVABILITY
If any state variable has no effect upon the output, then we cannot evaluate this state variable by observing the output.

We define the observability matrix as:

Ed
Example 11.6

If the system is observable then, the rank of the observability matrix must be of rank n.

Determine whether the system show is observable

un

Given a plant in the form (note can be in phase variable, observer canonical etc):

Li

Then:

Thus:

Since the determinant of the observability matrix is 0, the system is not observable.

11.6 STEADY STATE DESIGN VIA INTEGRAL CONTROL

Ed
The error is given by: Writing the state equations we have:

un

We now extend our design of the PI controller by introducing a feedback path from the output to form the error, e, which is fed forward to the controlled plant via an integrator. The integrator increases the system type and reduces the previous finite error to zero.

Li

We then produce an augmented matrix:

But, from the diagram:

We thus obtain:

equation given by the determinant of

, we can design K and

Example 11.7

Design a controller with integral control to yield a 10% overshoot and settling time of 0.5 second given a plant:

Then:

Ed
The characteristic equation of the observer is then: The desired characteristic polynomial has poles at :

un

d
.

As the system type has been increased, we now have an additional pole to place. By using the characteristic

Li

Since this is a third order system, we must introduce a third pole. We select the third pole to be -100 i.e at least 5 times greater than the real parts of the second order poles. Thus:

By matching the coefficients:

We can find the integral controller transfer function to be:

And steady state error for unit step input to be:

Ed

un

Li

Then,

S-ar putea să vă placă și