You are on page 1of 10

1 of 10 AIRCRAFT CONTROL Having discussed the behavior of an aircraft around a prescribed flight or reference condition, the question of the

legitimacy or applicability of such a scenario may arise. Although our general 6 DOF equations (Eqs. 87 90) are not restricted to this condition, all of the analyses regarding the various longitudinal and lateral modes came from our linearized versions of these equations! The study of the general equations leads to the concept of Open Loop control, where the aircraft motion as a function of specified control inputs is examined. However, in practice much of an aircraft's flight time is spent trying to accomplish/maintain a set task (i.e. steady level flight for cruise, or following a specific track/glidepath for an ILS approach and landing, etc.) and as such deviations from this prescribed condition (think reference condition) are seen as errors. This leads to the concept of Closed Loop control where sensors on board the aircraft are used to collect various state variables (i.e., q, , V, etc. ) and compare them to the desired values. Most modern aircraft incorporating either type of control usually employ an Automated Control System, where the relation between a pilot's input and the desired aircraft reaction is not direct. This bridge between input and output can be as simple as a gearing ratio, or as complicated as a digital computer, where in each case the significant feature is the adjustability of this bridge to produce a desired outcome.

Open Loop

Closed Loop

When the automatic control is used to provide suitable damping or natural frequencies that without which would create an aircraft which is difficult or impossible to fly, this is referred to as a Stability Augmentation System (SAS). If the system is designed so as to produce a particular response to a given control input it is called a Control Augmentation System (CAS) (e.g. achieving a maximum roll rate), while if it is simply designed to provide pilot relief it is referred to as an Autopilot (e.g. turn coordinator, altitude holding, etc.). If we go back to our general 6 DOF equations, we note that they can be expressed in the form,

2 of 10 where x is a vector containing state variables (i.e. quantities which describe the status of the aircraft) and c is a vector containing control variables (i.e. a, e, etc.).

At this point we will make use of our knowledge of Laplace transforms. These are useful in that they allow a differential equation to be expressed as a unique function of an algebraic expression which is easier to solve. But due to the uniqueness criteria, a solution to the Laplace version of the equation can be directly related to the solution to our original differential equation. It also helps that much of classical control theory is built around the manipulation of Laplace transforms. By definition the Laplace transform is expressed as,

Taking the Laplace transform of many of our governing equations allows us to define a key parameter in aircraft control known as the Transfer Function, G(s), which relates a particular input, c(s), to a particular response, x(s). Taking the Laplace transform of our system of equations yields ......

if we assume initial quiescent conditions then x(0) = 0 (and x(dot) (0) = 0 .... etc.) and one can write,

which can be re-arranged to obtain,

3 of 10 Therefore, for a given c(s) vector, a knowledge of G(s) allows one to predict the resulting behavior in the absence of non-zero initial conditions,

Starting from our expression for G(s) in Eq. 105, we can note that the inverse of the required matrix can be written as,

where the adjunct of the matrix can be obtained from the transpose of the matrix of minor determinants (or co-factors),

while we can note that the denominator of the resulting transfer function is nothing more than the characteristic polynomial for the given dynamic mode under consideration (i.e. longitudinal or lateral). The points at which the transfer function goes to infinity (i.e. the denominator goes to zero) are called the poles, and as we have just shown, are equal to the roots of the characteristic polynomial (Eq. 97b). The zeros of the transfer function are the points at which the numerator goes to zero. In general, the transfer function can be expressed as a ratio of factored polynomials ....

4 of 10 where Gij(s) is the ith, jth element of G(s), or the transfer function from the j th input to the ith output (note: i > j). The co-efficient k is known as the static loop sensitivity when all the co-efficients of s are unity. This idea of poles and zeros is integral to the classical approach to control system design, as systems are often designed around the idea of producing acceptable dynamic behaviour (i.e. obtaining specific roots) by modifying the root locus plot (where the stability and dynamic characteristics of a particular dynamic mode are shown). However, before getting into the details of a root locus plot, let us clarify our transfer function a bit. As shown in Eq. 106, this represents the open loop or forward path transfer function since it involves no feedback elements. However, for a closed loop system the transfer function becomes,

From the block diagram we can note that e = d r, but the feedback signal r can be expressed as a function of the feedback transfer function,

while the output signal is related to the error signal through the forward path or open loop transfer function in Eq. 106. Combining both the open loop and feedback transfer functions we can write the desired signal as,

which can be manipulated to obtain,

From Eq. 111 one can see that if the term F(s)H(s) = -1 (this is also known as the loop transfer function since F(s)H(s) = R(s)/E(s) = ratio of feedback to error), the overall closed loop transfer function goes to infinity and hence we are at the poles of the transfer function (which are the roots of the characteristic polynomial). Since every transfer function can be expressed as a ratio of polynomials as in Eq. 108, using this form on the loop transfer function yields for the characteristic polynomial of the closed loop system,

In this form one can see that there is a direct link between the system gain, k, and the roots of the system, s = n i where for a given root (either actual or desired...) the magnitude of the gain can be found using (this is sometimes referred to as the magnitude criteria),

5 of 10 In this form one can see that the magnitude of k must be zero at the poles. Conversely, as the magnitude of k is increased to infinity, the loop transfer function goes to 0 and hence we are at the zeros. Therefore, this implies that the roots of the system start at the m poles for k = 0 and tend towards the n zeros as k approaches infinity. If we want to show this a little more clearly, perhaps a graphical approach would help....... Root Locus Plots The importance of the roots of a system has already been introduced when we examined the period of the Phugoid mode as a function of centre of gravity position. However, instead of plotting T vs. h to show that as the static margin is approached, the Phugoid period increases and eventually becomes unstable, this same trend can be shown more generally on a Root Locus Plot. This type of plot is simply a graph of the various roots of the system in the complex (real vs. imaginary) plane, as some parameter is changed (in this case the static margin). From what we know about complex roots, several key dynamic characteristics can be immediately inferred from this type of plot:

Any roots (poles) that lie on the left hand side of the imaginary axis represent a stable motion (i.e. the response decays with time, where the farther to the left of the Im axis, the faster the motion decays [i.e., the larger the value of n in Eq. 98]) Any roots that lie on the right hand side of the imaginary axis are unstable Any roots that lie on the real axis are nonoscillatory (since = 0) All roots lying on the same vertical line have the same time to half (or double) amplitude All roots lying on the same horizontal line have the same damped frequency (and hence period) (Eq. 104) All roots lying on the same radial line extending straight from the origin have the same damping ratio (Eq. 99) All roots lying on the same circular arc around the origin have the same undamped natural frequency (Eq. 100) From the root locus plot for the Phugoid mode of our Piper Cherokee we can see that the curves are symmetric about the real axis as expected (n i), where starting at A and traveling towards B the static margin is continually reduced. The number of closely spaced points near A is indicative of our conclusion drawn earlier that the Phugoid mode remains fairly constant until near the neutral point, as this shows both n and are nearly uniform near A. Closer to B, the static margin approaches zero and the frequency changes rapidly (although the motion

6 of 10 remains stable and at nearly the same time to half amplitude). As the neutral point is passed (K n < 0) the pair of complex roots become two real roots, one positive and one negative, where the positive root creates a dynamic instability. If the centre of gravity is moved even further back, another stable, oscillatory root appears but the unstable root merges with the Short Period mode (which is also unstable at this point). It is interesting to note that by simply shifting the centre of gravity from A to B we move from having a lightly damped system to a critically damped system (i.e., no oscillation but still stable), a change it would be nice to have direct control over . Since the behavior of the loop transfer function (r/e) depends on the roots of the polynomial F(s)H(s), the reason our initial block diagram included a controller in the forward path now becomes a bit clearer. Since F(s) is simply the product of the controller transfer function ( J(s) = /e ) and the aircraft transfer function ( A(s) = x/ ), depending on the choice of controller, the resulting poles and zeros can be manipulated to produce a desired response. The choice of controller is also influenced by the data we have available and thus the type of feedback ( H(s) ) can also be used to modify the resulting transfer function. In order to determine what type of controller or feedback to implement, let us examine how the addition of a pole or zero can affect the resulting root locus plot. Going back to Eq. 108 and starting with a simple transfer function:

F(s)H(s) =

Stable for 0 < k <

F(s)H(s) =

The addition of a simple pole (both p 2 > p1 and p2 < p1) will bend the root locus to the right, thereby limiting the stable range of k (past the value of k where the locus crosses the Imaginary axis, the behavior becomes unstable)

F(s)H(s) =

The addition of a zero causes the locus to bend towards the left side of the Imaginary axis thereby stabilizing the system.

7 of 10

F(s)H(s) =

With an idea of how a root locus plot will behave under a given set of changes, it would be helpful to review how to obtain these plots. Noting that a root locus plot is always symmetrical about the Real axis, we will draw the plot for the following loop transfer function,

1. Every pole is the starting point for a branch (k = 0) which either tends towards an open loop zero or infinity (at k = ). in our case, 2. The number of branches that tend towards infinity is equal to n m (n = number of poles, m = number of zeros) in this case, 3. Segments of the Real axis which have an odd number of poles and zeros to the right form part of the root locus

4. Branches tending towards infinity do so along straight line asymptotes which intersect the real axis at the centre of gravity of all finite poles and zeros, , where

= here, = 4.33 5. The angle the asymptote makes with the Real axis is, where q = 0, 1, .... (n m - 1) here, q = 0,1,2 and thus a = 60o, 180o, 300o

8 of 10 6. If a portion of the root locus on the Real axis lies between two poles, the branch must break away from the Real axis so that it can tend towards a zero (or infinity) as k . This break away point is found by solving the relation F(s)H(s) + 1 = 0 for k and setting the derivative dk/ds = 0. Only points that lie on the locus are of interest.

7. The angle of departure from a pole is p = 180o (2q +1) + where is the sum of all the angles for lines drawn from every pole and zero to the pole of interest 8. The angle of arrival at a zero is z = 180o (2q +1) + where is the sum of all the angles for lines drawn from every pole and zero to the zero of interest

9 of 10 Controllers Since the loop transfer function is composed of both the controller transfer function and that of the aircraft, clearly it becomes possible to alter the poles and zeros of the system by modifying the form of J(s) (there are other ways as well, such as adding lead or lag compensation, but we will focus here on feedback controllers). If the control deflection is a function of the difference between a given variable's sensed value and it's desired value (i.e., an error), this constitutes a form of feedback control. The simplest form of feedback control is Proportional control, where as the name implies, the control deflection is directly proportional to the error,

J(t) = kp e(t)

J(s) = kp E(s)

The disadvantage of this is that there may be a steady state error that persists, one which can be eliminated by adding an Integral control, where the control deflection is proportional to the accumulated error in the system,

J(t) = ki e(t) dt

J(s) = ki/s E(s)

The disadvantage of integral control is that the system becomes less stable through the addition of a pole at the origin (1/s), where we have already established that the addition of a pole bends the root locus to the right thereby reducing the range of stable values for k. However, accepting this limitation, with the steady state error eliminated the system can be further improved by reducing the oscillations about the desired state. This can be done using a Derivative control where,

J(t) = kd de/dt

J(s) = kd s E(s)

As with the integral control, this control algorithm is not perfect either. The major disadvantage is that without a change in the error, no corrective control deflection will be produced! Also, since the control deflection depends on the rate of change in the error, this type of controller will react significantly to noise in the error signal, and hence the transfer function is often modified slightly to attenuate high frequencies. The combination of all three types of control constitute a PID controller, where

10 of 10 The use of a PID controller is aimed at improving the response of a system, and it implicitly assumes that the system is stable to begin with. For example, two researchers by the name of Ziegler and Nichols studied PID controllers and found that the integral of the absolute error was at a minimum when a system responded as shown, which can be obtained for either a simple P, a PI, or a PID controller by setting the gains to : P: PI: PID: kp = 0.5 kp,u kp = 0.45 kp,u ki = kp/(0.83 Tu) kp = 0.6 kp,u ki = kp/(0.5 Tu) kd = kp (0.125 Tu)

where kp,u is the ultimate gain, or the value of k in Eq. 108 at which the root locus (with k i and kd set to zero) crosses the Imaginary axis. This can be found from,

G ( s )=

k ps + z 1...s + z m k p , u (s + z 1...s + z m) = s + p1...s + pn s + p 1...s + p n

where one requires the value of s at which the locus crosses the Imaginary axis. With the ultimate gain determined, one can calculate the period of oscillation at this gain, T u as 2/ where is simply the value of the purely imaginary roots at this point, s = i. The only restrictions on this procedure are that:

The root locus plot must be marginally stable (with ki = kd = 0), i.e. as kp is increased, the locus crosses the imaginary axis All other roots of the system must lie to the left of the Imaginary axis, i.e., they must all have a negative real component

You might also like