Professional Documents
Culture Documents
net/publication/4151346
Root-locus dynamics
CITATION READS
1 1,095
2 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Ciann-Dong Yang on 12 January 2014.
Root-Locus Dynamics
Ciann-Dong Yang1, and Chia-Hung Wei
Institute of Aeronautics and Astronautics
National Cheng Kung University
Tainan 701, Taiwan
Abstract II.Generalized Root Locus (GRL)
In this paper, the discussion of root locus is taken from Consider a general transfer function
the point of view of field theory by treating root locus as s b1
E1
s b2
E2
" s bm
Em (1)
some kind of potential flows. This approach throws a new G s D1 D2 Dn
eO1s O2 s
s a1 s a2 " s an
light on root locus and suggests a physical modeling of
root locus in terms of the streamlines in flow field and where, the parameters Oi , ai , bi , D i , and E i are all assumed
electric field. Based on potential theory we derive the to be complex numbers with Re D i , Re E i R . This
governing equations of root locus for time-varying systems, class of transfer function may contain irrational terms such
in terms of which the force interaction existing within root as s 1 , s j j , and e s 1 s , etc., where j 1 . The
loci can be explained, and root sensitivity and root systems defined by Eq.(1) covers both lumped systems
robustness can be defined. Furthermore, the superposition
governed by ordinary differential equations and some
of elementary potential flows makes it possible to
reconstruct open-loop transfer function from the desired distributed systems governed by partial differential
root locus - the so-called inverse root-locus problem. equations.
Key words: Root locus, Fluid Dynamics, Field Theory Conventional root loci (CRL) of G(s) are defined by
I. Introduction the following set
Though root locus method is so popular and familiar
to control engineers, we still do not know theoretically
CRL ^ s 1 kG ( s )
0, k R ( R ) ` (2)
why some rules of experience for constructing root locus The major advantage of using root-locus method is
should be. For example, a rule of thumb in plotting root that one can examine the effect of changing open-loop gain
locus says that adding zeros to the open-loop transfer or plant parameters to aid in achieving best overall control
function G s has the effect of moving the root loci design. Instead of merely changing the open-loop gain k ,
here we will consider a more practical and general case of
toward the left-half of the s-plane and adding a pole to
changing the dynamics compensator K ( s ) . In this case the
G s in the left half of the s-plane has the effect of
roots of 1 K s G s 0 for varying dynamic compensator
pushing the original root loci toward the right-half plane. K s are to be determined. In terms of Eq.(2) we can see
Indeed there still lacks a theoretical verification of the that if the parameter k is replaced by the compensator
above phenomena using the existing knowledge about root
K ( s ) , the magnitude and the phase of parameter k are
locus.
varying simultaneously. With this consideration the
For a control engineer who has some experiences
conventional definition of root locus is modified as
in fluid visualization, he may find an interesting link
between the flow pattern of fluids and the trajectory of root following.
locus. In fluid laboratory we can see the phenomena that
½
GRL(k ) ° s 1 kG(s) 0, k k e jkT , k R °
(3)
T ® r r ¾
adding a source to a flow field will push the streamlines °¯ °¿
away from the source, while adding a sink will pull the In this definition k is no longer restricted to be
streamlines toward the sink. The effect of adding a pole in positive real or negative real; instead, k is released to be
root locus is very similar to the effect of adding a source in any complex number.
flow field, and the effect of a zero is similar to a sink. To establish this analogue, the complex potential
These analogies motivate the present study. The main function is introduced
concept we want to introduce here is that the movement of :( s ) ln G ( s ) (4)
root locus is a phenomenon of potential flow. Like other
and define
fields of potential flow, such as electric field, magnetic
: (V jZ ) ) (V , Z ) j < (V , Z ) (5)
field, fluid field, gravitational field, and temperature field,
etc., the investigation of root locus can be generalized to Let C0 be the complex plane excluding the
field theory [2]. singularities contained in : s . Since : s is analytical
within C0 , ) and < must be conjugate harmonic
functions in C0 , satisfying the Laplace equations.
root locus is also characterized by Laplace equation. where q > qV qZ @ and n is the outward unit vector
III. Phaser and Phase flux perpendicular to ds . The flux across the arc 's is
(A) “Phaser” : a moving particle d < qV d Z qZ d V (9)
In this section we attempt to give a physical On the other hand, the phase function < < V , Z in
interpretation of GRL. Root locus GRL kT can be
general can be expressed by
considered as a path line traced out by a moving medium
w< w<
which is called “phaser” here. Phaser conveys the phase d< dV dZ (10)
kT of k from one point to another along root locus wV wZ
GRL kT . One kT will determine one path line Hence, the phaser velocity components qV , qZ parallel
GRL kT as phase kT is varied, we then form a flow to the axes are given by
pattern consisting of many such path lines along which w< w< (11)
qV , qZ
phasers carry different values of phase. wZ wV
(B) Phase flux Using Cauchy-Riemann condition, flux rate in Eq.(11) can
Next we strive to introduce the concept of flux of be expressed in terms of magnitude function ) as
phaser. Let A(V 0 , Z0 ) be a fixed point in the V Z w) w) (12)
qV , qZ ,
plane which has unit thickness and P (V , Z ) an arbitrary wV wZ
point in the same plane. (See Fig.1) Now, two arbitrary or in a vector form
paths ABP and ACP are drawn between the points P and A. q ) (13)
Assume that there is no singularity within the region Along lines < V , Z constant, d < 0 and the
bounded by these curves, i.e., there is no phase created or combination of Eq.(9) and Eq.(11) gives
destroyed within the region, then the rate of phase flow § dZ · q
Z
entering the region across the curve ABP is equal to the ¨ ¸ (14a)
© dV ¹< cons tan t q
V
flow rate across the curve ACP. The term flux for the rate
of phase flow will be used hereafter. The flux across the This result shows that the velocity of phaser q is
curve ABP is equal to the flux across any curve joining A everywhere tangent to curves in the V Z plane along
to P. Since the point A is fixed, the flux is a function of the which < V , Z constant. This result is consistent with the
position of P. If the root locus passing through P is previous assumption that < V , Z constant (GRL) is the
GRL < P , then the phase flux across AP is denoted by trajectory of a phaser. On the other hand, along
< P . It must be borne in mind that the existence of phase
) V ,Z constant
function is a consequence only of the conservation of w) w)
d) dV dZ 0
phase, so a phase function is valid for nonlinear and wV wZ
time-varying systems. Therefore
Now consider two points P1 and P2 and two curves § dZ · w) wV q
V
¨ ¸ (14b)
drawn from them to the fixed point A. Let GRL <1 , © d V ¹) cons tan t w) wZ q
Z
GRL < 2 represent the root loci passing through points
Since these slopes in Eq.(14a) and Eq.(14b) are negative
P and P , respectively (see Fig.2). Then, the flux across
1 2
reciprocals, the lines are perpendicular to one another.
the curve AP is equal to the flux across the curve AP
2 1
Thus, the constant magnitude lines and GRL form an
plus that across the curve PP . Hence, the flux across the
1 2
orthogonal set of lines which completely describe the
curve PP is < < . It can be easily seen that if the
1 2 1 2 phase flow in a two dimensional field.
reference point A is replaced by another point A , the value
of the stream function < < changes by a constant,
1
IV. Superposition of Elementary Root Loci
1 2
The Laplace equation or Poissin equation, governing
namely the flux across A A . Since definition the stream
1 2
phase function for two-dimensional potential flows is in
function is constant along a root locus, when the point P1
such a simple form that some elementary solutions to
and P2 are points of the same root locus (not necessarily these equations can easily be found. Each of these
coincident), the flux across P1 P2 is '< <1 < 2 0 . It
64
solutions represents a physically possible elementary GRL. assumed to be part of the open-loop transfer function
The method of superposition is also used in solving Eq.(1). We are interested in the case where other poles and
problems in electromagnetism, heat conduction and fluid zeros of G s are far removed from the pole-zero pair,
dynamics whose governing equations are in the form of in which case the distance s0 of Fig.6 is much smaller
either the Laplace or the Poisson equation. Some of the than the distance between the pole-zero pair and any other
GRL represented by the linear combination of solutions pole or zero of G s . Such an isolated pole-zero pair is
may simulate the root loci originated from linear or called a dipole. When dipoles occur on or very near the
nonlinear systems such as in Eq.(1). There are four stability margin, the hidden modes may cause stability
elementary GRL that are commonly encountered in problems if not properly anticipated [4]. It is then worth
control problem : The time delay, the pole (or sink), the taking a close look on this kind of locus. An special GRL
dipole, and the vortex. Although the CRL for pole, sink
results when the distance s between pole and zero of
and time delay is well known, the concept of GRL 0
Hence, phase function and potential function are obtained, where O2 N e jD .Obviously, this transfer function is
respectively, as originated from a nonlinear element.
< V ,Z m0 tan
1
Z Zp
m0T (19a) 4 Vortex
V V p This type of GRL stems from poles oz zeros with
2 2
1 2 imaginary power in Eq.(1), i.e., from the existence of
) V ,Z m0 ln V V p Z Zp m0 ln r (19b) Im D i and Im E i . The complex potential function for
We can verify by direction substitution that both < and vortex can be written as
) satisfy Laplace equation and Poisson equation. The *
: s j ln s s0 ) j< (25)
curves < c are sketched in Fig.5. From the observation 2S
of this figure, a pole at point V p , Z p can be conceived where * is a real number and s0 V 0 jZ0 is the
of as a point from which phaser emanates in equal amounts center of the vortex. Therefore, < and ) for vortex
along radial paths. Hence the GRL in this case be straight becomes
radial lines, i.e., * * 1 Z Z0 *T
qT 0, qr qr r (20) < V ,Z ln r , ) V , Z tan (26)
2S 2S V V0 2S
where qr r can be found from Eq.(13) using polar It can be seen that GRL are circles, and equi-magnitude
coordinate expression. lines are radii (Fig.8). Note if we interchange < and ) ,
1 w< m0 we obtain the GRL for pole. The phaser velocities are
qr (21)
r wT r w) 1 § w< · 1 § w) · w< *
qr ¨ ¸ 0, qT ¨ ¸ (27)
3 Dipole wr r © wT ¹ r © wT ¹ wr 2S r
Consider a pole-zero pair (see Fig.6) which is It turns out that phasers in this type of GRL moves in
circular paths with the velocities being inversely
65
proportional to the radii of the circles. numerical approach using discretized N V would be
The GRL for any pole-zero configurations with delay more desirable. We divide the dipole region into n
or dipole can be obtained by linear combinations of this segments of equal width 'K , as shown in Fig.11. We
four elementary GRL. Hence, Eq.(3) can be rewritten in designate by N j 'K , the total dipole strength within the j
an alternate form for GRL kT as segment, whose center is at a distance K j from the origin;
^ V , Z l1 < delay l2 < pole l3 < dipole l4 < vortex kT , li R` (28) N j is taken as constant, equal to the average of the exact
Fig.9 shows part of the GRL for the delay system with a distribution within the segment. N j will of course vary
dipole at origin. The CRL is composed of the V axis and from one segment to another. The dipoles within the j
the big circle centering at origin. It is seen that the loci of segment will contribute phase '< j to the phase function
dipole are isolated from those of time delay by this big at a given point P in the field. This contribution may be
circle and thus form a phase flow pattern very similar to written as
the fluid flow pattern caused by a uniform flow via a N j +KZ p
'< j (31)
cylinder with infinite length [3, 6]. 2 2
V p K j Zp
IV. Inverse Root-Locus problem
Let G s C s P s , where P s denotes the plant Hence, the approximate formula corresponding to the
model and C s denotes the compensator. The exact form of Eq.(15) is
root-locus problem is that given the compensator C s , S
n
N j +K
plot the roots of 1 kG s 0 as k is varied, while the
inverse root-locus problem is that given one segment of Zp
¦
j 1
V p K j
2
Zp
2
(32)
66
to open-loop transfer function has the effect of moving the wqV w w wpV
root loci toward the left half of the s-plane and that adding QV q x QZ q x (40)
wt wV wZ wV
poles in the left-plane s-plane has the effect of pushing the Expanding the left-hand terms and using Eq.(36) and
original loci toward the right-half plane. These phenomena Eq.(37), yields
can be well explained by the concept of phase flow. From
wqV wqV wpV 1 wpV
the observation of GRL for poles we have known that a qV qZ (41a)
pole can be conceived of as a point from which phasers wt wV wZ U wV
emanate along all radial directions. When added to the Similarly, applying the momentum theorem in the Z
original loci, these phasers emanating from the pole create direction, we have
a pressure on the loci and cause it to move away from the wqZ wqZ wpZ 1 wpZ
pole; while from the observation of GRL for zeros, we qV qZ (41b)
wt wV wZ U wZ
have known that a zero is a point into which phasers flows
Eqs.(41) can be put into a compact vector form as
radially from all directions. Thus, the phasers on the
wq p
original loci in the neighborhood of the zero will be sucked q q 0 (42)
into the zero and cause the loci to move toward the zero. In wt U
the following paragraph, we will give a quantitative This can be simplified further by noting from vector
description of above phenomena. analysis that
(B) The Continuity Equations 1 2
Consider the two-dimensional phase flow in the q q q q u u q (43)
2
complex plane, and cut out a control volume of Substituting Eq.(43) into Eq.(42), we obtain for
infinitesimal dimensions (see Fig.12). Consider the total incompressible phaser fluid
phase rate across the control surface, and the rate of
wq § p q2 ·
change of phase storage within, for a unit depth normal to qu uq (44)
¨
¨ U 2 ¸¸
the complex plane. In terms of density U and phase flux wt © ¹
rate (velocity of phaser) q of phaser fluid, the mass flux VI. Force Action of Root Loci
rate Q can be defined accordingly as Let L1 GRL ( <1 ) is a segment of the root locus with
Q > QV QZ @ U q > U qV U qZ @ (36) phase <1 . We now want to find the force exerted by the
Application of the principle of conservation of phase, remaining loci of GRL on L1 . The phase pressure force
yields acting on the surface element ds (which has a unit
wQV wQZ wU thickness perpendicular to the complex plane) is pds and is
0 (37)
wV wZ wt perpendicular to ds. The two components of the phase
or, in vector form pressure force dFV and dFZ in the positive V and
Q
0
w ʳ ʳ ʳ ʳ dFV p sin T ds pd Z , dFZ p cos T ds pdV
This continuity equation must be satisfied for any control Where T is the angle that ds makes with the positive
systems including nonlinear and time-varying systems. If V axis. These two differential force components may be
we assume that phaser fluid is “incompressible”, i.e.,
written in complex form as
U =constant, we have
q
0
67
§ d: s · 1 d: s Company, 1986.
dFV jdFZ jHds j
¨ ¸ds (49) U
2 ds © ds ¹
Noting that along L , d < 0 and d: d : d) ,
1
68