Beruflich Dokumente
Kultur Dokumente
1
xdot*sqrt(m)
-1
-2
-3
-4
-6 -4 -2 0 2 4 6
x*sqrt(c )
Version 1.0
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Properties of nonlinear systems . . . . . . . . . . . . . . . . . . . . . 6
2.1 Superposition does not hold . . . . . . . . . . . . . . . . . . . 7
2.2 Commutativity does not apply . . . . . . . . . . . . . . . . . . 7
2.3 Transfer function methods fail . . . . . . . . . . . . . . . . . . 8
2.4 Equilibria (local stability) . . . . . . . . . . . . . . . . . . . . 8
2.5 Limit cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.6 Chaotic behaviour . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7 Jump and hysteresis effects in the frequeny response . . . . . . 15
2.8 Generation of new frequencies for periodic inputs . . . . . . . 15
2.9 Inversion is not always possible . . . . . . . . . . . . . . . . . 15
2.10 Finite escape time . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Analysis methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1 Stability definition . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Stability analysis tools . . . . . . . . . . . . . . . . . . . . . . 18
3.3 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Techniques for determination of time-domain solutions . . . . 22
A.1 Examples of non-linear closed-loop systems . . . . . . . . . . . . . . . 25
A.2 System with saturation . . . . . . . . . . . . . . . . . . . . . . . . . . 25
A.3 Nonlinear closed-loop . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
A.4 Hystheresis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
A.5 Simulation of Duffing’s equation . . . . . . . . . . . . . . . . . . . . . 27
A.6 Dry friction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1 Introduction
All physical systems exhibit non-linearities and time-varying parameters to some
degree. Linear systems, described by linear differential equations and functions are
always some idealistic description of real physical systems. As many analysis and
synthesis techniques have been developed for linear systems, non-linear systems are
often analyzed and controlled withthe help of linear techniques. In many cases of
small non-linearities, satisfactory results can be obtained. In general, however, linear
methods become restrictive in their application and quite often unrealistic. There-
fore many methods have been developed for different kinds of non-linear systems.
In this document, non-linear systems are considered which can described by
ordinary differential equation and algebraic functions. Systems described by partial
differential equations are not part of this lecture.
The title “Nonlinear systems” includes the followingx words:
• System: Used as term for the plant, process or open loop system as well as
for the feedback control system. The plant or process includes the process
dynamics as well as the actuators and sensors needed to realize the control.
Actuators and sensor dynamics are sometimes also integrated into the con-
troller. In the following, the term process is used for this part of the control
system.
• Nonlinear: A system property that applies to the process, the controller and
the overall control system.
This first part gives an overview of combinations of controller and process. The
process (including actuator and sensor) is divided into two classes: linear and non-
linear ones.
Figure 1.1 shows the block diagram of a closed-loop control system.
r + u y
Controller Actuator Process
_
ym
Sensor
Based on the classes above there exist six combinations of controller and process.
In the following an example will be given for each combination togehter with the
main non-linear characteristics as well as with the possible methods for analysis and
design.
Discontinuities
up up up
u y u y u
lo lo lo
Saturation Dead Zone Rate Limiter
Dynamic Dynamic Dynamic
y (n) = f (y (n−1) , . . . , y, d, u, t)
u y ( n ) = f ( y ( n −1) ,...., y , u, d ) y
Initial conditions
parameters
• Control input u
• finite escape time: solutions of nonlinear systems may not exist for all times
for t ≥ 0 by superposition of the free response f (t) which depends on the initial con-
ditions and the forced response g(t)∗u(t) which depends on the input. Unfortunately
this approach is no longer valid for nonlinear systems because the superposition or
addition of solutions can no more be used. Therefore the impulse or step responses
cannot be used for describing a nonlinear system.
The linearization around the first two equilibrium points leads to the linear dif-
ferential equation:
△ẍ + k△ẋ + 2△x = △u
which has two poles in the left half plane, while the third equilibrium point leads to
△ẍ + k△ẋ − △x = △u
1
xdot
-1
-2
-3
-4
-3 -2 -1 0 1 2 3 4
x
Figure 1.4: Trajectories in the phase plane of the van der Pol equation with e = 2
-1
-2
-3
-4
0 20 40 60 80 100
time
Figure 1.5: Limit cycle oscillation of van der Pol’s equation as a function of time
The term on the right hand side is equal to F0 when v0 > ẋ, and −F0 when
v0 < ẋ. Then the following solutions for the phase paths in these two regions are
m v0
The variable x1 represents the prey and x2 the predator. Equilibrium points are at
(0, 0) and (c/d, a/b). Figure 1.8 shows limit cycle trajectories.
This system exhibits some pecularities:
• There are infinitely many limit cycles and there is no attraction region for any
of them. So there are also no transient trajectories possible.
1
xdot*sqrt(m)
-1
-2
-3
-4
-6 -4 -2 0 2 4 6
x*sqrt(c )
20
15
predator
10
0
0 10 20 30 40 50 60
prey
• The limit cycles are neither stable nor unstable. They are indifferent and right
at the border line between stable and instable cycles. Any small deviation from
the limit cycle results in a neighbourhood cycle.
• Linearization around the nonzero equilibrium point will reveal two poles on the
imaginary axis. So a small deviation from the equilibrium shows an undamped
oscillatory behaviour.
Example 7 Relay-control
A relay controlled system is an example of nonlinear system that shows a limit cycle
behaviour. Figure 1.10 shows the block diagram of a relay control system, Figure 1.9
3 Scope
Step s+1
Add Relay Transport
Transfer Fcn
Delay
Figure 1.9: Relay control system, first order system with delay
2.5
set point, actual value, relay output
1.5
0.5
set point
relay
actual value
0
0 5 10 15
Time in sec
Figure 1.10: Relay control (hysteresis of -0.5, 0,5) of first order system with delay
of 1 sec
This example cannot be displayed in the phase plane due to the delay element.
The effect of the hysteresis is that the larger the hysteresis is the larger is the am-
plitude and the period of the oscillation.
For a continuous system to exhibit chaotic behaviour, the following three neces-
sary but not sufficient conditions must be fulfilled:
• The system order must be of order 3 or higher. The third dimension can be
the time in the case of a time varying parameter or if the system has a time
dependent input. In three dimensions much more complicated solution curves
can exist than for two dimensions.
• The system must be nonlinear and has stable and unstable equilibrium points.
This ensures that solution curves are repelled and attracted depending on the
type of equilibrium point the solution curves move around.
• The nonlinearity must be such that there exist a high sensitivity of the solution
with respect to initial conditions. This ensures that solutions starting very
close to an existing solution do not remain neighbouring solutions as time
goes on but can move apart to the existing solution.
Example 8 Duffings’equation
Duffings’equation with periodic input ẍ + k ẋ − x + x3 = A sin(ωt).
This version of the Duffing’s equation has negative linear damping and positive
cubic restoring terms. The corresponding autonomous √ system (A=0) has equilibrium
points at x = ±1, (stable spirals when 0 < k < 2 2), and at x = 0 (saddle point).
With increasing amplitude A of the input, the system shows for a given frequency ω
an increasingly complex type of limit cycles until, at a certain amplitude (A = 0.5),
it becomes chaotic. If the amplitude is further increased, the system behaviour settles
again to a limit cycle. Figure 1.11 shows the Duffing’s equation states as a function
of time.
1.5
0.5
0
v
−0.5
−1
−1.5
1
0.5 600
0 400
−0.5 200
x −1 0
time [s]
Other examples of chaotic systems are the Lorenz model for the weather dynamics
or Chua’s double scroll family for a nonlinear electronic circuit.[2]
Example 9 Duffing’s equation with damping (with positive linear spring force and
cubic restoring force, and periodic forcing term)ẍ + k ẋ + x + βx3 = A cos(ωt). [1].
ẋ − ex = 0
14
12
10
8
x
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
t [s]
3 Analysis methods
A state trajectory is stable if, for a given maximal deviation from the origina trajec-
tory, initial conditions close to the original initial conditions such that the trajectory
deviation remains within the specified “vicinity” of the original trajectory.
A trajectory which is not stable is unstable. This means that even with an
arbitrary small perturbation the trajectory leaves the specified “vicinity” of the
original trajectory.
A trajectory is asymptotically stable if after a perturbation of the initial con-
ditions the trajectory eventually converges to the original trajectory. The largest
perturbations under which the solution returns to the old equilibrium is the domain
of attraction of this equilibrium.
Of course the definition above can be applied to equilibrium points, as equilib-
rium points are “static” trajectories.
2.5
1.5
0.5
x2
0
−0.5
−1
−1.5
−2
−2.5
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
x1
Figure 1.13: Example of phase portait. The origin is stable, for large value of |x2 |
the value of x2 diverges.
NONLINEAR
GAIN
z = yψ(y, t)
Let G(s) be the transfer function of the linear element. The Circle Criterion says
that the system in Figure 1.14 with α ≤ ψ(y, t) ≤ β is globally asymptotically stable
if the Nyquist locus G(j ω) (−∞ ≤ ω ≤ ∞) does not penetrate the disc intersecting
the real axes in − α1 and − β1 and encircles it as many times anticlockwise as G(s)
has unstable poles. Figure 1.16 shows the Circle criterion predicting closed-loop
stability for the case that the linear system G(s) in Figure 1.14 is stable.
yψ(y, t)
Im
− α1 − β1 Re
G(jω)
Figure 1.16: Circle criterion proving closed-loop stability for an open-loop stable
system
If it can be shown that the time derivative V̇ (x(t)) of this function is decreasing, i.e.
dV (t)
<0
dt
in a certain region around x = 0 then x = 0 is an asymptotically stable equilibrium
point. If the weaker condition holds that V̇ (x(t)) ≤ 0, the equilibrium is stable.
The stability or asymptotic stability of a system can be determined without
solving the nonlinear differential equation. Unfortunately there is no general method
for the construction of a Lyapunov function for a given system. One can prove that
the system is stable only if a Lyapunov function is found (by chance or by ability).
The difficulty of solving the differential equation is traded for the problem “How
to find a Lyapunov function?”. The concept of a Lyapunov function can be also
applied to time-varying systems and systems with inputs.
Lyapunov functions are in particular commnly used in proving the stability of a
controller for a given system (see more later).
Theorem 1 Assume two systems with MIMO transfer functions S1 and S2 are
connected in a feedback loop; then the closed-loop system is input-output stable if
kS1 k · kS2 k < 1.
The norm k.k can be the infinity norm, that is the size of the largest singular value
of the transfer function over all frequencies. Other induced norms can be used [3].
The supremum is taken over all T ≥ 0 and all admissible pairs (u(), y()) with the
fixed initial state x(0) = x0 . A system is considered passive if E is finite for all
initial states x(0). Otherwise, the system is considered active. Roughly speaking,
the inner product is the instantaneous power, and E is the upper bound on the
integral of the instantaneous power (i.e., energy). This upper bound (taken over all
T ≥ 0) is the available energy in the system for the particular initial condition x(0).
Passivity, in most cases, can be used to demonstrate that passive system will
be stable under specific criteria. Passive systems will not necessarily be stable
under all stability criteria. For instance, a resonant series LC circuit will have
unbounded voltage output for a bounded voltage input, but will be stable in the sense
of Lyapunov, and given bounded energy input will have bounded energy output.
Passivity is frequently used in control systems to design stable control systems
or to show stability in control systems.
3.3 Robustness
Robustness is treated as in linear systems with respect to the following points:
[4] W. S. Levine, editor. The control handbook. CRC Press, 1996. Very broad for
every aspects of control, covers many methods, differs from author to author in
the theoretical background needed to understand the individual contribution.
[6] O. Foellinger, editor. Nichtlineare Regelung I und II. Oldenburg, 1969, 1980.
Classic books in german for the analysis and design of nonlinear system via
describing function method, phase plane methods, Lyapunov function. Very
readable.
[8] K.H. Seitz, editor. Regelung mit Zwei- und Dreipunktregeln. Oldenburg, 1986.
Covers many type of discontinuous controller and linear system combinations.
Practically oriented.
[10] Anders Robertson. Nonlinear control theory. Technical report, Lund University,
2006. Lecture notes, very good summary of important results for nonlinear
systems and control.
Exercises
1
s
Sine Wave Integrator2 Saturati on Scope2
1
w + x
-1
1
_
-1
y 1
1+s
1 1
w + x
1
_ 1+s
1
y
1 -1
-1
1 1
1
w + x
-1
1
_
-1
y 1
s
2
w1 =step at t=0 of height 1
1 1
w + x
1
_ s
1
y
1 -1
-1
1 1
A.4 Hystheresis
Why is a hysteresis always a nonlinear element?
ẍ + k ẋ + x + βx3 = A sin(ωt)