Appndx D Mae175
Appndx D Mae175
Appndx D Mae175
The main body of this book has introduced the idea that the pilot rating of an aircraft can be
predicted using a mathematical model of that aircraft's behavior. The early appendices have
shown how the aircraft itself can be modeled in the time domain, but it is evident from much of the
material in the book that we need to present the aircraft dynamics, including control system
components, in the manner of classical control theory. This appendix gives a minimal
Most of us are familiar with the time domain representation and solution of differential equations.
complementary
is : y = + particular integral
function (transient
(steady state)
response)
homogeneous equation (i.e. the left side) and solving the resulting characteristic equation for
R(t ) . Arbitrary constants are found using the method of undetermined coefficients. The same
note that this equation is stated sufficiently generally to allow the example equation D.1 to be cast
in this form.
Conventional differential equations are messy. Fortunately, the method of Laplace transforms
gives the engineer a method that is not only compact, but very insightful. The only difficulty is
that the time domain equations are mapped into another world (the 'frequency domain') that at
first is unfamiliar. After our manipulations are performed there, an inverse Laplace transformation
can return us to the time domain. However, many engineers 'think' in the frequency domain. The
advantages of the Laplace approach include algebraic determination of the transient and steady
state solutions together, even for cascaded physical elements, as well as direct correspondence
Definition D.1: Laplace transformation: L f (t ) F ( s) f (t )e st
dt e
st
f (t )dt .
0 0
For our purposes, this applies to linear differential equations with constant coefficients. Notice the
use of lower and upper case for the time-domain Laplace-domain quantities respectively.
Definition D.2: Laplace of differentiation: L ( f ( t )) sF (s) f ( 0 ) .
-where the dot indicates differentiation with respect to time. For zero initial conditions we remark
that the Laplace transform of differentiation n times, consists simply of replacing the number of
So for integration we divide by the number of Laplace variables s corresponding to the number of
integrations.
Although the integral expressions in these definitions appear somewhat daunting, for zero initial
conditions they allow us to use the Laplace variable as a simple operator for differentiation (by
Pairs simply describe the translation between the time and frequency domains. For example, we
have already defined the pairs for differentiation and integration (with zero initial conditions) as
follows:
f (t ) F ( s)
d
differentiation, ( x) sX
dt
integration, xdt X
s
The most important remaining pairs are shown in the following Table. Although at one time, a
voluminous book of Laplace transform pairs was required for practice in this area, software is
available today that fulfils this purpose much better. One example is Program CC, and another is
MATLAB.
TABLE A.1
f (t ) F ( s)
a
step: size = a
s
unit impulse 1
The IVT and FVT tell us that for a step input to our function, its initial value and steady state
values can be found by substituting s and s 0 respectively. (This is true for a step
because the 1 / s , the transform of the step, handily cancels the s in the left-hand-side of both
definitions.)
Transfer functions
These are a simple extension of the Laplace transform idea. Consider a mass-spring system as
Summing the forces dues to inertia, the spring and the viscous damper leads to
m x c x kx f (t ) where the dots denote differentiation with respect to time and the f (t ) is
ms2 X ( s) csX ( s) kX ( s) F ( s)
or, X ( s)(ms2 cs k ) F ( s)
We can write an algebraic expression for the output deflection X(s) for an input F(s) as follows:
X ( s) 1
F ( s) ms cs k
2
Output (s)
or, Transfer Function .
Input (s)
This is the transfer function for the deflection x as output for the force f as input.
So, if we know F(s) or X(s) then we can use the transfer function to calculate the unknown, i.e.
Each of the factors represents a first order lag, which appears on the s-plane as in Figure D.2.
Knowing that the Laplace transform of a step is 1/s, we can calculate the step response as the
F(s) f(t)
1 1
(1 e at )
s( s a ) a
1 1
(1 e bt )
s( s b ) b
1 1 1
1 (be at ae bt )
s( s a )( s b) ab ( a b)
The higher the value of a or b, the faster the step time response due to that element reaches
steady state, as shown in Figure D.3 for a generalized first order lag.
1.0
.95
Output
.63
1
F ( s)
s(s 1)
0
0
2 3
Time, sec
Figure D.3 General form of time response of a first order lag to a step
Returning, however, to our mass-spring system, we can use it to illustrate some useful points.
First, if the damping constant is reduced to zero, Table D.1 tells us that the mass and spring
oscillate sinusoidally indefinitely. The value in the Table becomes n , the undamped natural
frequency of our mass-spring system. If the damping is re-introduced, the system exhibits a
X ( s) 1/ m
2
F ( s) s 2 n s 2n
where is the damping ratio. We can plot the roots, a and , ot the transfer function
undamped
natural
damped natural frequency, n
d
frequency, d
=damping ratio
resonant 1
frequency r sin
Real
The effect of damping change at a constant frequency can be seen in the step time responses of
Figure D.5.
The main ideas of the s-plane plot can be summarized in the sketch of Figure D.6.
Examination of the Laplace transform pair table shows that if the real part of a denominator root is
positive, the function will grow to as t because of the growing positive exponent in the
eat term in the time response. For our purposes, we will refer to systems of this kind as being
unstable.
If the real part is identically zero, then a second order response with 0 will appear as in
Figure D.5 and will exhibit a neutrally stable response consisting of a sustained oscillation with
Inverse Laplace
As already mentioned, the time response of any known transfer function can be obtained from a
The first and second order lag contributions are thus added together. The formal rules for
achieving expansion of the Laplace transformed response are the Heaviside expansion
theorems.
body. He was castigated for not demonstrating the ties between his work, which used
because he did not understand the process of digestion, that did not mean that he
Notice that so far we have covered linear systems. By this we mean that the coefficients are
constant with respect to time, and that there are no products or other functions of variables
involved.
1. Doubling the size of the input to a stable linear system will double the output.
2. A stable linear system with a sinusoidal forcing function at a constant frequency will eventually
frequency in radians per second, yields much insight about the steady state response of the
1
1. Establish L
-1
e at
sa
3. Laplace transform 2. using 1. gives
s 2
2
4. Cascade the sine wave with the system transfer function to get the output to
s 2
2
a sine wave
5. Take the partial fractions of 4. The stable transient modes die out.
Definition D.4 - The gain or amplitude ratio of a frequency response is defined by Figure A.8.
It is simply the ratio of the amplitude of the output sine wave to that of the input.
Definition D.5- The phase of a system frequency response is defined by Figure D.7.
The notation of these figures refers to the magnitude and argument of the complex number which
information about the fequency response of a system is to plot values of the gain and phase
which apply to a number of frequencies. There are a number of ways of doing this. We will
D.2.1 Bode Plots- The gain of the response is first converted to decibels using:
Output Amplitude
Amplitude Ratio (or gain) = 20 log10
Input Amplitude
Bode plots are then plots of gain (dB) vs. frequency in radians/second, and phase (degrees) vs.
frequency, again in radians/secon. The main features are sketched in Figure D.8.
D.2.2 Nichols Chart- This a means of converting open-loop gain and phase to closed-loop. First it
is necessary to specify strictly in the context of the Nichols chart, what is meant by open and
closed loop. Any system involving feedback or feedforward loops can be written in the form of
Figure D.9
If the open loop transfer function is A( s) , then the closed loop transfer function is easily shown
A( s )
to be .
1 A( s)
The Nichols chart plots, at any desired frequency, the gain and phase of A( s) and reads the
A( s )
gain and phase of as desired, as shown in Figure D.10
1 A( s)
.
D.3 Root Locussk
This is a powerful yet simple tool in linear systems analysis and synthesis. Probably the best way
Suppose we wish to find the roots of a third order polynomial, that is, we want those values of x
x 3 ax 2 bx c 0
where a, b, c are given fixed numerical values.
Suppose we modify this problem slightly to say that we want to find the roots of the polynomial
for all values of, say, the parameter a from zero to infinity.
x 3 bx c 0
i.e., when a is zero, the roots of our original polynomial are the roots of the polynomial in the
denominator in the basic root-locus equation. Putting it another way, when a is zero, the poles of
the polynomial ratio, given by x 3 bx c 0 , are the roots of the original polynomial.
or,
ax 2
1 .
x 3 bx c
We would like to substitute our values of a in this equation and determine what the roots are.
For example, if a 0 , then we go back to our original equation and see that
x 3 bx c 0 .
So the roots are the roots of the polynomial in the denominator of our basic equation. These can
and x 2 0 when a .
So the roots are the roots of the polynomial in the numerator of our basic equation. These can be
plotted on the s-plane as zeros. See figure D.11 for example poles and zeros on the s-plane.
the parameter a . For intermediate values, the roots follow a locus which can be plotted simply
by incrementing a from zero to large values, or by using two criteria, known as the angle criterion
and the amplitude criterion. Which method is used is immaterial if a computer program is used to
develop the locus. There are also a number of rules (about a dozen) for the locus which aid
We want
ax 2
1 D.19
x 3 bx c
or,
ax 2
1 D.20
( x p)( x q rj )
Using these criteria, plus the additional rules, or simply the digital computer, the complete locus
in Figure D.13.
1. The locus appears on the real axis to the left of an odd number of singularities
2. If there are n more poles than zeros, then the locus asymptotically approaches the
Here we used the notation of Langill (1965) but there are many texts (mostly of that vintage) that
Notice that we have used root locus merely as a means of finding roots. How then do we apply it
In general, this method is insightful for exploring the changes in transfer function numerators and
denominators as a single variable changes. For example, the numerator of the sideslip to control
transfer function contains the dimensional derivative N r . The roots of this numerator (which for
the dutch roll approximation is second order) can be quickly explored for all values of N r , and
many engineers can sketch variations like these rapidly for quick insight.
Finally (and leaving this aspect until last has been quite deliberate) we can of course explore the
roots of a transfer function denominator as one parameter varies. This is the most common use
of the method, because it directly illustrates the sensitivity of the stability of the system to any
parameter. In fact, the root locus is considered by many to be a only a method for investigating
stability. As we have seen, by viewing it as a root finder, we can use its insight in other ways
also.
In the equation x 3 Ax 2 bx c 0 A is a positive number. How would you use root
locus to show the effects on the roots of the polynomial, of incrementing or decrementing A
by some amount ?
We shall use this example to demonstrate a useful technique which we will call the
x 3 ( A ) x 2 bx c 0 , so that
x 3 Ax 2 bx c x 2 0 . Now divide by
x 3 Ax 2 bx c 0 to get
x 2
1 0 and we are ready with this to do a root locus analysis,
x 3 Ax 2 bx c 0
except that the poles show our ‘baseline’ dynamics. We show the effects of incrementing
using the method as we have described it already, and the effects of decrementing by
plotting a ‘zero degree’ instead of a ‘180 degree’ locus since we can represent the negative
value of by changing only the sign of unity in the right-hand-side of the basic root locus
equation. Thus our locus now passes through the baseline roots.
The commonest use of root locus is in examining the closed-loop stability of a dynamic system
like an aircraft with feedback control. In our earlier discussion we wrote the closed loop transfer
function as
A( s )
1 A( s)
where A( s) is a polynomial, or ratio of polynomials, in the Laplace parameter s . The roots of
the denominator of this transfer function determine the modes of motion- the frequency, damping,
etc. -of the closed loop system. We might use a table of transform pairs or the Heaviside
theorems to determine these. Suppose A( s) includes a varible parameter k . Then the closed
and we use root locus to locate the roots (i.e., values of s similar to our example values of x ) of
the denominator for various values of k , and therefore see directly the effect of k variation on
and for root locus contruction the zeros are the roots of the open loop transfer fucntion numerator,
and the poles are the roots of the open loop transfer function denominator.
D.B.Groll has proposed a very simple algorithm that facilitates root locus analysis of very large
systems. These systems are often analyzed by computer programs that accept values of
polynomial coefficents (for linear dynamic flying qualities analysis, the polynomial coefficients are
aerodymamic derivatives and control system gains, time constants, natural frequencies,
dampings etc). The program then assembles high order polynomials and uses a technique
similar to Cramer’s Rule to produce transfer functions for the needed input/output pairs. The
result can be of quite high order- up to a hundred is not unknown. Sorting out the poles and
Suppose a root locus is required for variation of one parameter b . We want the zeroes N ( s)
and the poles D( s) in D( s) bN ( s) 0 .
Groll’s method simply determines the system polynomial first for b 0 and then for b 1 . The
first calculation, of course, gives us the poles D( s) . The second gives us the poles plus the
zeroes, D( s) N ( s) . Evidently if we subtract the result of the first calculation from the latter,
The simplest and most graphic way of describing a system consisting of a number of physical
components is a block diagram. For a linear system, the blocks contain gains or transfer
functions. Block diagrams are identically equivalent to systems of equations, however for readers
who ordinarily have little contact with block diagram conventions, Figure D13 should be helpful.
Note throughout that A, B etc. are polynomials or ratios of polynomials in s . and may also be