Root-locus plots

Its time we start designing systems ourselves. But how can we make sure our system gets the properties we want? In this chapter well investigate one method for that: the root-locus method.

General drawing rules

First we will examine the basic rules of the root-locus method. In the next part we will examine how they work.

1.1

The basic ideas

Suppose we have a (rather familiar) closed-loop system, as is shown in gure 1. We know that the transfer function is F (s) = G(s)/(1 + G(s)H (s)). We also know that the properties of the system strongly depend on the poles of F (s). We would like to have some inuence on these poles.

Figure 1: Again, a basic closed loop system. Well, if we want to nd the poles of F (s), we should set its denominator to zero. So that gives us the equation (s z1 )(s z2 ) . . . (s zm ) G(s)H (s) = K = 1. (1.1) (s p1 )(s p2 ) . . . (s pn ) This equation is called the characteristic equation. Every s satisfying this equation is a pole of F (s). By the way, the variables z1 , . . . , zm and p1 , . . . , pn in the above equation are the zeroes and poles, respectively, of G(s)H (s). (Not to be confused with the zeroes and poles of F (s).) We assume that all these zeroes and poles of G(s)H (s) are already dened. There is, however, one variable left for us to dene, being K . Luckily it turns out that the poles of F (s) (which we want to dene) strongly depend on K . We therefore would like to do plot these poles as a function of K . In other words, we want to plot all s satisfying the above equation. This is, however, more dicult than it seems. Especially because the above equation is a complex equation. So how do we know which s satisfy it? Well, we know that if two complex numbers z1 and z2 are equal, then both their arguments (angles) are equal (arg(z1 ) = arg(z2 )) and their lengths are equal (|z1 | = |z2 |). Lets apply this to equation (1.1). If we examine lengths, we nd the length-equation |G(s)H (s)| = |K | |s z1 | . . . |s zm | = | 1| = 1. |s p1 | . . . |s pn | (1.2)

This isnt a very useful equation yet. So we also examine the arguments. This time we get the argumentequation arg(K ) + arg(s z1 ) + . . . + arg(s zm ) arg(s p1 ) . . . arg(s pn ) = arg(1) = (2k + 1) 180 , (1.3) 1

where k Z can be any constant. (Note that multiplying complex numbers means adding their angle, while dividing complex numbers means subtracting their angle.) We can even simplify the above equation a bit. We know K must be a real number. And usually its also positive. This means that arg(K ) = 0. So the above equation doesnt even depend on our variable K ! And theres even more. If we can nd an s satisfying this equation, then we can simply adjust our K , such that it also satises the length-equation. So all we need to do is nd an s that satises the argument-equation. Most of the rules in the upcoming paragraph use this fact.

1.2

Root-locus drawing rules

There is no set of rules which will always enable you to draw a perfect root-locus diagram. However, there are rules which can give some info about its shape. So in the following plan of approach, we will mention most of these rules. 1. First we should draw a complex plane. In it, we should draw all the poles and zeroes of G(s)H (s). We denote the n poles by a cross (x) and the m (nite) zeroes by a circle (o). (We will examine the innite zeroes at point 4.) 2. Next we should examine which points s on the real line are part of the root-locus plot. Just pick a point on this axis. If the total number of poles and zeroes to the right of this point is odd, then this point is part of the root-locus plot. Otherwise, it is not. (By the way, this follows from the argument-equation.) In this way we can nd all segments of the real axis which are part of the root-locus plot. 3. We will now nd the breakaway points. These are the points where the root-locus lines break away from the real axis. To nd them, we rst have to solve equation (1.1) for K . We will nd K= (s p1 )(s p2 ) . . . (s pn ) . (s z1 )(s z2 ) . . . (s zm ) (1.4)

We then use the condition dK/ds = 0. We will nd several solutions si for this. However, not all of these solutions represent breakaway points. In fact, to nd the actual breakaway points, we need to substitute the values of s back into the above equation. Only if a certain value si gives a positive K , then si is a valid breakaway point. 4. We can sometimes also nd the asymptotes of the graph. (However, this only works if H (s)G(s) has enough innite zeroes. With enough, I mean at least 2. The number of innite zeroes of G(s)H (s) is equal to n m.) To nd the asymptotes, we rst have to nd the asymptote center. It is located at the point s= z1 + z2 + . . . + zm p 1 p 2 . . . p n . nm (1.5)

From this point, we now draw asymptote lines in n m direction. These directions have angles asymptote angles = (2k + 1) 180 . nm (1.6)

When |s| the root-locus plots will converge to these asymptotes. 5. Sometimes we can also nd the points where the root-locus plot crosses the imaginary axis. (In fact, this usually only works if n m 2.) On these points, we can write s = i. So we can insert this condition into the characteristic equation (1.1). We then nd a complex equation. If we equate both the real and complex parts of this equation, we can nd . 2

6. Finally, we can determine the angle of departure/angle of arrival of the complex poles/zeroes. (This, of course, only works if G(s)H (s) has complex poles or zeroes.) We can use the argumentequation for this. Lets suppose we want to nd the angle of departure/arrival of the pole p1 . We examine some point sa , innitely close to p1 , but at an angle = arg(sa p1 ). If we insert this into the argument-equation, we nd that arg(sa z1 ) + . . . + arg(sa zm ) arg(s p2 ) . . . arg(sa pn ) = (2k + 1) 180 . (1.7) Since sa is innitely close to p1 , we can simply substitute sa = p1 in the above equation. We can then solve for the angle of departure at this pole. Once we have gone through all of these rules, we should be able to draw a more or less accurate graph. In fact, we only have to draw one half of the graph, since root-locus plots are always symmetric about the real axis. When drawing the root-locus plot, you do have to keep in mind that every line should go from a pole to a (possibly innite) zero. If you have a zero or a pole without a line, then you probably havent completed your root-locus plot yet.

Example problems

The root-locus drawing rules might seem rather vague. So lets clarify them with some examples.

2.1

An example problem without innite zeroes

Lets consider the basic feedback system with G(s) and H (s) given by G(s) = K (s + 2)(s + 3) s(s + 1) and H (s) = 1. (2.1)

We want to plot the root-locus graph for this system. In other words, we want the possible range of poles of F (s), if we can vary K . To nd this, lets simply apply the rules. 1. The function G(s)H (s) has n = 2 poles, being s = 0 and s = 1. It also has m = 2 zeroes, being s = 2 and s = 3. We draw a complex plane, and insert these four points. (For the nal solution, in which these four points are also drawn, see gure 2 at the end of this section.) 2. Lets examine a point s on the real axis and count the number of poles/zeroes to the right of it. This number is only odd if 1 < s < 0 or 3 < s < 2. So only these parts of the real axis are part of our root-locus plot. We mark these sections in our graph. (Again, see gure 2.) 3. Lets nd the breakaway points. We know that K= s(s + 1) . (s + 2)(s + 3) (2.2)

We can take the derivative with respect to s. This gives us the condition dK (s + 2)(s + 3)(2s + 1) s(s + 1)(2s + 5) = = 0. (2.3) ds (s + 2)2 (s + 3)2 1 (3 3). So we have two solutions for s. We do have to check Solving this for s gives s = 2 1 whether they are actual breakaway points. The solution s = 2 (3 3) implies that K = 0.0718, while s = 1 3) gives K = 13.93. Both values of K are positive, so both values of s are 2 (3 + breakaway points. We can thus draw breakaway points at these positions. 3

4. Since n = m, there are no innite zeroes. So there are no asymptotes either. This rule therefore isnt of much use. 5. Since there are no asymptotes, there arent any points where the root-locus plot crosses the imaginary axis. 6. There are no complex poles/zeroes, so we cant nd any angles of departure either. We have got some nice info on the shape of our root-locus diagram already. Experienced root-locusdrawers can already predict its shape. If you are, however, new to root-locus drawing, you might want to nd some extra points. We could use the argument-equation for that. We could also simply try to nd some point in a brute force kind of fashion. Well use that second method now, since its a bit more exact. We know that root-locus lines always run from a pole to a zero. So there must be some point on the root-locus plot with real part Re(s) = 3/2. What point would that be? To nd that out, we can insert s = 3/2 + i into the characteristic equation. If we do this, we can solve for . We will nd that = 3/2. So we just found another point of our root-locus plot! If you want, you can continue to do this for other points. You will nd some more points of the root-locus diagram. When you feel youre sure what the shape of the plot will be, you can draw it. And then you should hopefully have found the plot of gure 2.

Figure 2: The solution of our rst example problem.

2.2

An example problem with multiple innite zeroes

This time, lets consider a system with G(s) = K s(s + 1)(s2 + 4s + 13) and H (s) = 1. (2.4)

Lets use the drawing rules to draw a root-locus plot. 1. We nd that G(s)H (s) has no (nite) zeroes. (So m = 0.) It does have four poles. (So n = 4.) They are s = 0, s = 1 and s = 2 3i. We draw those poles in the complex plane. 2. We examine all points on the real axis which have an odd number of poles/zeroes to the right of it. We nd that this condition is only valid for 1 < s < 0. We mark this part in our plot. 4

3. Lets nd the breakaway points. We have K = s(s + 1)(s2 + 4s + 13). Taking the derivative will give dK = 4s3 + 15s2 + 34s + 13 = 0. (2.5) ds This is a cubic equation. So we let a computer program solve it. We get as solutions s = 0.467 and s = 1.642 2.067i. Inserting the rst solution in the characteristic equation gives K = 2.825, which is positive. So s = 0.467 is a breakaway point. We mark this one in our plot. However, inserting the other solutions results in a complex K . They are therefore not breakaway points. 4. We have n m = 4. So there are four innite zeroes, and thus also four asymptotes. The asymptote center is at 0 1 (2 + 3) (2 3) 5 z 1 + z 2 + . . . + zm p 1 p 2 . . . p n = = . (2.6) s= nm 4 4 The angle of the asymptotes can be found using (2k + 1) 180 . (2.7) 4 If we let k vary, we will nd angles 135 , 45 , 45 and 135 . We lightly mark these asymptotic lines in our plot. asymptote angles = 5. There are innite zeroes. So the root-locus might cross the imaginary axis as well. To nd out where it crosses it, we substitute s = i into the characteristic equation. We then nd that K = 4 + 5 3 i + 17 2 13i. If we equate both the real and complex part, we will nd that K = 4 + 17 2 and 0 = 5 3 13. (2.9) (2.8)

Solutions for are = 0 and = 13/5. The corresponding solutions for K are K = 0 and K = (13/5)2 + 17 13/5 = 37.44, respectively. The solution = 0 isnt altogether very surprising. However, we nd that the root-locus plot also crosses the imaginary axis at s = 13/5i = 1.612. We thus mark those points in our plot. 6. Finally we would like to know the angles of departure of the two complex poles. For that, we use the argument-equation. Lets nd the angle of departure for the pole p1 = 2 + 3i. First we nd the arguments we need to put into the argument-equation. They are arg((2 + 3i) (2 3i)) = arg(6i) = 90 , arg((2 + 3i) 0) = arg(2 + 3i) = 123.69 , arg((2 + 3i) (1)) = arg(1 + 3i) = 108.43 . It now follows that the angle of departure of p1 is = (2k + 1) 180 90 123.69 108.43 = 142.13 . (2.13) (2.10) (2.11) (2.12)

From symmetry we can nd that the angle of departure of p2 = 2 3i is = 142.13 . We can mark these angles in our root-locus plot. Now we should have sucient data to sketch our root-locus plot. The result can be seen in gure 3.

Finding the appropriate points in the plot

Its nice that we know what poles we can give to our system. But how do we choose the right value of K ? Thats what well look at now. 5

Figure 3: The solution of our second example problem.

3.1

Setting demands

Lets suppose we have drawn a root-locus plot. How can we use this plot to inuence our system? For that, we will examine a pole. But rst we have to make sure we examine the right pole of F (s). But which pole is the right pole? As you know, poles with a bigger real part are more inuential. So actually, the pole closest to the origin is the only pole we should focus on. (The other poles are less signicant.) Then, using certain system demands, we can nd this pole. But what kind of properties can we demand from our system? Lets make a list of that. We can set the dampening factor . (Often a value between 0.4 and 0.7 is preferred.) Lets suppose we use this criterion to nd a pole pi . We dene the argument of this pole to be = arg(pi ). It can then be shown that = cos(180 ). So how do we nd pi ? Well, we draw a line from the origin with an angle = 180 arccos . The point where it intersects our root-locus plot is the position of our pole pi . We can also set the damped natural frequency n . It can now be shown that the undamped natural frequency due to some pole pi is |pi | = n . So how do we nd pi ? We simply draw a circle with radius n around the origin. The point where it intersects the root-locus plot is where our pole pi is. Finally we can set the undamped natural frequency d . We can nd that the damped natural frequency due to a pole pi is equal to the real part of pi . So, Re(pi ) = d . So how do we nd pi ? We simply draw a horizontal line at a height d . The point where it intersects the root-locus plot is where our pole pi is.

3.2

Finding the gain K

We now know how to nd the right pole. But how do we use this to nd K ? Well, we can simply use the characteristic equation for this. Insert the pole pi we just found into the characteristic equation. From this, you can solve for K quite easily. And once you know K , all that is left to do is to nd the other poles of F (s). 6