Describing Function Analysis

University of Cagliari Dept of Electrical and Electronic Eng.
Describing Function analysis

of
nonlinear systems
Prof. Elio USAI eusai@diee.unica.it
University of Leicester
Department of Engineering
System definition and problem statement

d(t) r(t)
+ _
Plant P
process
e(t)
C
controller
u(t)
A
actuator
m(t) +
+
y(t)
z(t)
F
filter
ym(t)
+
T
trasmitter
n(t)
Controller
r: reference variable e: error signal u: control variable m: manipulated input y: controlled variable ym: measure of the output z: feedback signal d: disturbance n: measurement noise
Describing Function analysis of nonlinear systems Prof Elio USAI March 2008

In many cases the system presents a nonlinear phenomenon which is fully characterised by its static characteristics, i.e., its dynamics can be neglected
m
Saturated actuators Relay control Gears backlash Hysteresis in magnetic materials
+M
-M
Dead zone in electromechanical systems

e
Saturated actuators Relay control Gears backlash Hysteresis in magnetic materials Dead zone in electromechanical systems
-E
+U
e +E
-U

m (driven gear)
-U +U u (driving gear)

m
-M0 +M0 u

m
-U +U u

Nonlinearities are not always a drawback, they can also a have a stabilising effect
r(t)
+ _
e(t)
u(t)
m(t)
y(t)
z(t)
P(s ) =
1 s 1 + 1 .4 s + s 2
)
A(s ) = N 1 O sat ( 2 )
0.5 H (s ) = (1 + 0.1s )
C (s ) = 10

Step response
20 NOT saturated actuator Saturated actuator 15
10
System output
-5
The saturated system oscillates but does not diverge
-10
-15
-20
10
12
14
16
18
20
time [s]

Bode Diagram of the difference transfer function
50 System: sys Frequency (rad/sec): 1.68 Magnitude (dB): -0.0708
Magnitude (dB)
System: sys Frequency (rad/sec): 0.94 Magnitude (dB): 12
-50
-100
-150 -90 System: sys Frequency (rad/sec): 0.941 Phase (deg): -180
The NOT saturated system is not stable since its stability margin are negatives mg = -12 dB m = - 47 deg
Phase (deg)
-180
-270
System: sys Frequency (rad/sec): 1.68 Phase (deg): -227
-360 10
-1
10
10
10
Frequency (rad/sec)

Many systems can be reduced to a simplified form in which the all linear dynamics is concentrated in a unique block and the static non linear characteristics is represented by a separate block
r(t) + _ e(t) C(j) u(t) f(u) m(t) P(j) y(t)
z(t)
H(j)
r(t)
C(j)
r(t) + _
u(t)
f(u)
m(t)
P(j)H (j)C (j)
w(t)

If constant (or very slowly varying) reference signals are considered, under some conditions it is possible to separate the low-frequency, almost static, behaviour defined by a working nominal condition and a highfrequency behaviour due to small variations around such a working point
r(t)=R C(j) r(t) + _ u(t) f(u) m(t) P(j)H (j)C (j) w(t)
u (t ) = u0 + u (t )
r(t)=kCR + _ u(t) f(u) m(t) G (j) w(t)
m(t ) = m0 + m(t ) w(t ) = w0 + w(t ) u0 = kC R kG m0

The non linear characteristics is translated so that the origin of the new reference Cartesian system is the point (u0, m0) in the original one
m
m
kC R kG
m0
kC R
u
u0
1 kG
m0 = f (u0 ) u0 = kC R kG m0
m0 = f (u ) f (u ) = f (u 0 + u ) m0
r(t)=0
+ _
u(t)
f(u)
m(t)
G (j)
w(t)

The aim of the system analysis is to define which is the steady-state behaviour of the system defined by the variations around the nominal working point
r(t)=0 + _ u(t) f(u) m(t) G (j) w(t)
If f(u) is a passive sector function absolute stability tools allow for sufficient conditions for global asymptotic stability of the variation system, i.e, the steady state is characterised by constant values of the system variables
If the origin of the variation system is not stable does the variables diverge to infinity or some periodic motion can appear?
Limit cycles
Limit cycle: a periodic oscillation around a constant working point
r=51() + _ u(t) m(t) w(t)
f(u)
G (j)
G (s ) =
10 (s + 1) s 2 + 1.4s + 1
6 f (u ) = sat (1, 11 , 16 ) 11
2
m
1.5
m=
1 1 u+ 10 2
0.5 m0
0 u0
5 11 5 m0 = 11 u0 =
-0.5
-1 -6
-4
-2
u
Limit cycles
Limit cycle: a periodic oscillation around a constant working point
r(t)=0 + _ u(t) f(u) m(t) G (j) w(t)
output w(t)
w
4
5 11 5 m0 = 11 u0 =
w0 =
0 5 10 15 20 25 30
50 = 4.545 11
time [s]
Limit cycles
Limit cycle can be a drawback in control systems: Instability of the equilibrium point Wear and failure in mechanical systems Loss of accuracy in regulation Parameters of the limit cicle can be used to discriminate between acceptable and dangerous oscillations oscillation frequency oscillation magnitude
Electronic oscillators can be based on limit cycles
Describing Function - Assumptions

The describing Function approach to the analysis of steady-state oscillations in non linear systems is an approximate tool to estimate the limit cycle parameters.
It is based on the following assumptions There is only one single nonlinear component The nonlinear component is not dynamical and time invariant The linear component has low-pass filter properties The nonlinear characteristic is symmetric with respect to the origin

There is only one single nonlinear component
The system can be represented by a lumped parameters system with two main blocks: The linear part The nonlinear part
r=const.
+ _
u(t)
N.L.
m(t)
G (j)
w(t)

The nonlinear component is not dynamical and time invariant The system is autonomous. All the system dynamics is concentrated in the linear part so that classical analysis tools such as Nyquist and Bode plots can be applied.
m +M
m=M
u u +
-M

The linear component has low-pass filter properties
This is the main assumption that allows for neglecting the higher frequency harmonics that can appear when a nonlinear system is driven by a harmonic signal
G ( j ) >> G ( jn ) n = 2, 3, K
The more the low-pass filter assumption is verified the more the estimation error affecting the limit cycle parameters is small

The nonlinear characteristic is symmetric with respect to the origin
This guarantees that the static term in the Fourier expansion of the output of the nonlinearity, subjected to an harmonic signal, can be neglected
r(t)=0 + _ u(t) m(t) w(t)
f(u)
G (j)
Such an assumption is usually taken for the sake of simplicity, and it can be relaxed
Fourier expansion - Recall

Consider a periodic function
y (t ) = f (t ),
y (t ) = y (t T ), T is a real constant
a0 y (t ) = + (ak sin (k 2 k =1
2 2 a k = f (t ) cos(k T T 2 2 2 bk = f (t )sin (k T T 2
T T
2 T
t ) + bk cos(k
t )d ( 2T t )
2 T
t ))
2 T
t )dt =
f (t )cos(k
2 2
2 T
k = 0,1,2,K b0 = 0
2 T
2 T
t )dt =
f (t )sin (k
t )d ( 2T t )
Describing Function Harmonic balance

r(t)=0 + _ u(t) f(u) m(t) G (j) w(t)
u (t ) = U sin ( t ) m(t ) = (a k cos(k t ) + bk sin (k t ))

k =1
w(t ) = Gk (ak cos(k t + k ) + bk sin (k t + k )) Gk = G ( jk )

k =1
k = G ( jk )

Consider the polar representation of a complex number associated with the exponential form of harmonic signals
u (t ) = Ue j t
2 M k = ak + bk2 a k = arctan k bk
m(t ) = M k e jk e jk t
k =1
Taking into account the low-pass property of the linear part of the system
w(t ) = Gk e j k M k e jk e jk t G1e j1 M 1e j1 e j t
k =1

r(t)=0 + _ u(t) N.L. m(t) G (j) w(t)
A permanent oscillation in the loop appears if
u(t)=-w(t)
j1
Ue
j t
= G1e M 1e e
j1
j1
j t
1 + G1e
M 1 j1 e =0 U
Harmonic balance equation
1 + G ( j )N (U , ) = 0
N (U , ) =
1 (b1 + ja1 ) U
is the Describing Function of the nonlinear term

The armonic balance equation is a necessary condition for the existence of limit cycles in the nonlinear system The approximate analysis gives good estimates if the low-pass filter hypothesis is strongly verified. It is a good tools for engineers The harmonic balance equation is similar to the characteristic polynomial function, i.e. it leads to the Nyquist condition for closed-loop stability The Describing Function is a linear approximation of the static nonlinearity limited to the first harmonic In most cases the Describing Function is not a function of the frequency and this simplifies the verification of the harmonic balance equation by means of the Nyquist plot of the transfer function
G ( j ) =
1 N (U , )

Nyquist Diagram
2
=+
0
-2
Imaginary Axis
-4
+ U
(U0 , 0 ) -\1/N(U)
U=0
-6
-8
G(j)
=0
-10 -1.4 -1.2 -1
+ -0.8 -0.6 -0.4 -0.2 0 0.2
Real Axis

Nyquist and DF Diagrams
1
0.5
=100
System: sys Real: -2.01 Imag: -0.021 Freq (rad/sec): 0.997
-0.5
=17.7828 =0.1
Imaginary Axis
-1
U
-1.5 System: sys Real: -1.14 Imag: -2.02 Freq (rad/sec): 0.651
-2
=3.1623
-2.5
-3
=0.56234
-3.5 -8
-7
-6
-5
-4
-3
-2
-1
Real Axis
Describing Function Computation

The DF computation can be performed by means of its definition
Usin(t) m(t)
f(u)
N (U , ) =
1 (b1 + ja1 ) U
2 2 a1 = m(t ) cos(t )dt T T 2 2 2 b1 = m(t )sin (t )dt T T 2

T
The evaluation of coefficients a1 and b1 can be performed by means of both analytical calculation and numerical integration, depending on the type of nonlinearity involved

Ideal relay
3
u m
+M
2
1.5
+M
1
0.5
m(t)
T
-0.5
-1
-M
-2
-1
-M
-3 -3 -2 -1
-1.5
u(t)
-2
0.5
1.5
2.5
Time
2 2 a1 = m(t ) cos(t )dt = 0 T T 2

T
Because of the odd symmetry of the m(t) signal
2 2 4 2 4M b1 = m(t )sin (t )dt = M sin ( )d = T T 2 0
N (U , ) =
4M U

Ideal relay
0.5
Nyquist and DF Diagrams U

0
The negative reciprocal of the DF is the negative real axis in backward direction
Imaginary Axis
-0.5
( ;4M/)
c
-1/N(U)
-1
-1.5
-2
A limit cycle can exist if the relative degree of G(j) is greater than two
-2.5
-3
G(j )
-3.5
-4 -1.6
-1.4
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0.2
Real Axis
The oscillation frequency is the critical frequency c of the linear system and the oscillation magnitude is proportional to the relay gain M

Pure hysteresis
2
2
1.5
M
1.5
u m
0.5
0.5
-0.5
-0.5
-1
-1
-M
-1.5
-1.5
-2 -2.5
-2
-1.5
-1
-0.5
0.5
1.5
2.5
-2
0.5
1.5
2.5
Time
2 2 b1 = m(t ) cos(t )dt = 0 T T 2

T
Because of the even symmetry of the m(t) signal
2 2 2 2 4M a1 = m(t ) cos(t )dt = M cos( )d = T T 2 0
4M N (U , ) = j U

Pure hysteresis
The negative reciprocal of the DF is the negative imaginary axis in backward direction A limit cycle can exist if the relative degree of G(j) is greater than one and G(j) is a type-0 system
0.2
-0.2
-0.4
Imaginary Axis
-0.6
-0.8
G(j )
-1
-1.2
-1/N(U)
-1.4
U
-1.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4
Real Axis
The oscillation frequency is lower than the critical frequency c of the linear system and the oscillations magnitude is proportional to the relay gain M and to the modulus of the transfer function at phase -/4

Hysteretic relay
2
2.5 u m
1.5
+M
1.5
1
1
0.5
0.5
-
0
t
-0.5
-0.5
-1
-1
-1.5
-1.5
-M
-2 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
-2
-2.5
0.5
1.5
2.5
Time
If U the hysteretic relay behaves as pure hysteresis

8 8 4 4M b1 = M sin (t )dt + M sin (t )dt = 1 T 0 T t U
t
T
8 8 a1 = M cos(t )dt + M cos(t )dt = T 0 T t
The imaginary part of N(U) is proportional to the hysteresis area
4M j U N (U , ) = 2 4M 4 M 1 j U U 2 U
U U >

Hysteretic relay
0.5
Imaginary Axis
The negative reciprocal of the DF is the parallel to the negative real axis and with constant negative imaginary part
-0.5
U
-1
-1/N(U)
-1.5
-2
If is larger than U the relay could not behave as a pure hysteresis
-2.5
-3
G(j )
-3.5
-1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2
-4
Real Axis
The oscillation frequency is lower than the critical frequency c of the linear system and the oscillations magnitude is proportional to the relay gain M

Saturation
4 3
0.8
u m
+M
0.6
+M
0.4
2
0.2
-1
-0.2
-2
-0.4
-M
-3
-0.6
-M
-4 -2.5
-0.8
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
u
-1 0 0.5 1 1.5 2 2.5 3
Time
If U the saturation behaves as pure gain

2 2 a1 = m(t ) cos(t )dt = 0 T T 2
T
T t 2 M M 8 8 4 2kU M 2 arcsin b1 = kU sin (t )dt + M sin (t )dt = + 1 T 0 T t kU kU kU M U k k 2 N (U ) = M 2k M M M arcsin U> + 1 k kU kU kU

Saturation
0.5
Nyquist and DF Diagrams -1/N(U)
Imaginary Axis
The negative reciprocal of the DF is part of the negative real axis in backward direction
U
-0.5
-1/k
-1
-1.5
-2
A limit cycle can exist if the relative degree of G(j) is greater than two and gain k is sufficiently high
-2.5
-3
-3.5
G(j )
-4 -1.6
-1.4
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0.2
Real Axis
The oscillation frequency is the critical frequency c of the linear system and the oscillations magnitude depends on the saturation parameters M and k

2.5
Dead zone
3 2
u m
1.5
k
1
0.5
-
0
+
-0.5
-1
-1
-2
-1.5
-3
-2
-3 -2 -1 0 1 2 3
u
-2.5
If U the dead zone has no output

2 2 a1 = m(t ) cos(t )dt = 0 T T 2
T
0.5
1.5
2.5
Time

2 arcsin 1 2 U U U
8 4 2kU b1 = k (U sin (t ) )sin (t )dt = T t
= U sin (t )
0 2k N (U ) = k
U arcsin U
2 + 1 U > U U

Dead zone
The negative reciprocal of the DF is part of the negative real axis in forward direction
1
0.5
-1/N(U)
0
-1/k
Imaginary Axis
U=+
-0.5
-1
A limit cycle can exist if the relative degree of G(j) is greater than two and gain k is sufficiently high
G(j )
-1.5
-2
-2.5 -3.5
-3
-2.5
-2
-1.5
-1
-0.5
Real Axis
The oscillation frequency is the critical frequency c of the linear system and the oscillation magnitude depends on the dead zone parameters and k

3
Dead zone
2
The nonlinear characteristics of the Dead Zone can be computed by subtracting the Saturation characteristics from a linear one
saturation ( u)
dead zone ( u)
-1
(u ) = k (u )
-2
-3 -3
-2
-1
N (U ) = k N (U )

The Describing function of a nonlinear characteristics can be computed as the combination of the Describing Functions of the elementary constituting nonlinear characteristics Nt(U,)
r(t)=0 + u(t) _ N1(U,) N2(U,) m(t) G (j) w(t)
N t (U ) = N1 (U ) + N 2 (U )

A number of Describing Function can be computed by particularisation of the function
( ) =
[arcsin( ) +
2
1 2
in which the parameter defines a peculiar point of the nonlinear characteristics

m +M
k
- + u/U
1 k N ( ) = k ( ) < 1
-M

+M
k2 k1
u/U + m -M
1 k1 N ( ) = k 2 + (k1 k 2 ) ( ) < 1
m k - + u/U
1 0 N ( ) = k (1 ( )) < 1

m +M
k2
- +
k1 +
u/U
1 0 N ( ) = k (1 ( )) <1 k ( ( ) ( )) < 1
-M
m +M
k
u/U -M
N ( ) = k +
4M

m +M
u/U
N ( ) =
4M
-M
m +M
+
-M u/U
0 N ( ) = 4 M 1 2
1 <1

m k - + u/U
1 0 N ( ) = k [1 (2 1)] j 4k (1 ) < 1 2
m k - + u/U
1 0 N ( ) = k [1 (2 1)] + j 4k (1 ) < 1 2

+M m
- +
u/U
0 N (U , ) = 4 M 4 M 1 2 j U U
1 <1
-M +M m
- +
u/U
0 N (U , ) = 4 M 4 M 2 U 1 + j U
1 <1
-M

Nt(U,)
r=0 + u _ m 1/j G (j) w
Negative inverse of the Describing Function

0.5
1 N t (U ) = N 1 (U ) + N 1 (U ) j
N (U , ) = 4M 4M j U U
Imaginary axis
U
0 =100 -0.5 =17.7828
U
=0.1
-1
-1.5
-2
U
=3.1623
U 1 ( + j ) = 2 N (U , ) 4M 1 +
-2.5
-3 =0.56234 -3.5 -8 -7 -6 -5 -4 -3 -2 -1 0 1
Real axis
Stability of limit cycles

r(t)=0 + _ u(t) N.L. m(t) G (j) w(t)
If the block N.L. is a pure constant k, the stability of the feedback system can be performed by means of the Nyquist criterion, which gives the number of roots with positive real roots of the Harmonic Balance Equation
1 1 + G ( j ) = 0 k
The Nyquist criterion looks at the relative position of the transfer function G(j) with respect to the point (-1/k, 0) in the complex plain. By extension the criterion can be applied to any point of the complex plain with reference to the Harmonic Balance Equation
1 + G ( j ) = 0 , C

If the transfer function G(j) represents a stable system, the reduced Nyquist criterion can be applied, i.e. the closed loop stability can be stated if the reference point in the complex plain remains on the left-side when running along the Nyquist plot of G(j) from =0+ to =+.
Nyquist plot of G(j )
1
0.5
G ( j ) =
=+
1 2 j ( j ) + 0.5 j + 1
-1/k
-0.5
Imaginary Axis
-1
-1.5
The closed loop is not stable with respect to the point (-1/k,0).
-
-2
-2.5
-3
The closed loop is stable with respect to the point -

=0
+
0 -2 -1.5 -1 -0.5
-3.5
-4 -3
-2.5
Real Axis

Can the generalization of the Nyquist criterion be used to analyse the stability of a limit cycle?
YES, it can
What does it mean that a limit cycle is stable? If a perturbation of the magnitude of the periodic oscillation occurs, it tends to the original value as time passes. How can a limit cycle be considered from the Nyquist criterion point of view? It is a marginally stable condition. How can the magnitude of a limit cycle be represented? By a point in the Describing Function plot.

Nyquist and DF plots
2
Imaginary Axis
Points B, B, B, A and A represent possible magnitudes of the oscillation.
1.5
0.5
-0.5
U
B B B
-1
A and B represent two possible limit cycles
-1.5
-2
G(j )
-2.5
-2.5 -2 -1.5 -1 -0.5 0
-3 -3
Real Axis

Applying the reduced Nyquist criterion with respect to: point B: oscillations tend to decrease (stable system) point B: oscillations tend to increase (unstable system) point A: oscillations tend to decrease (stable system) B: Stable limit cycle
2 1.5
0.5
Imaginary Axis
-0.5
-1
U
B
G(j )
-1.5
-2
-2.5
-3 -2.5 -2 -1.5 -1 -0.5 0
-3
Real Axis
A: Unstable limit cycle
The linear system G(j) is assumed to be stable in order to apply the reduced Nyquist criterion
DF Analysis Example of application

Consider a DC motor with permanent magnets
+24 V
M
0V -24 V
d.c.
The position of the motor shaft is measured by means of a rotational variable resistance. The voltage on the rotational resistance drives the position of a relay that switches the motor supply voltage between +/- 24 V d.c.

The linear approximate model of a DC motor is the following di (t ) vr (t ) = Rr ir (t ) + Lr r + e(t ) dt Rr: rotor resistance d (t ) Lr: rotor inductance = Cem (t ) + B (t ) J dt ke: voltage feedback constant d (t ) kt: torque constant = dt Jm: motor inertia e(t ) = ke (t ) Jl: load inertia
Cem (t ) = kt ir (t ) J = Jm + Jl B = Bm + Bl
Bm: motor friction coefficient Bl: load friction coefficient vr: rotor supply voltage ir: rotor wound current Cem: electromagnetic torque : rotational speed : shatf angular position
+ 24 < d vr (t ) = 24 d

The motor transfer functions are the following
kt (s ) W (s ) = = Vs (s ) (sLa + Ra )(sJ + B ) + k t k e
kt ( j ) = = ( j ) + j( j ) W ( j ) = Vr ( j ) j (( jLa + Ra )( jJ + B ) + k t k e ) =-
2 (Ra J + La B )2 + (Ra B + k t k e 2 La J )
j
k t ( R a J + La B )
2 (Ra J + La B )2 + (Ra B + k t k e 2 La J )
k t R a B + k t k e 2 La J

Taking into account the parameters of the motor and of the load
R=0.4; % rotor resistance L=0.001; % rotor inductance ke=0.3; % voltage feedback constant kt=0.3; % torque constant
Imaginary Axis
U
-2 -4
-6
Jm=0.01; % motor inertia Jl=0.09 % load inertia Bm=0.05; % motor friction coefficient Bl=0.05; %load friction coefficient
-8
-10
-12
-14
-16
J=Jm+Jl; B=Bm+Bl;
-18
-1.5 -1 -0.5 0 0.5 1
-20 -2
Real Axis

x 10
-4
R=0.4; % rotor resistance L=0.001; % rotor inductance ke=0.3; % voltage feedback constant
Imaginary Axis
kt=0.3; % torque constant Jm=0.01; % motor inertia Jl=0.09 % load inertia Bm=0.05; % motor friction coefficient Bl=0.05; %load friction coefficient J=Jm+Jl; B=Bm+Bl;
System: Wp Real: -0.00568 Imag: -8.73e-007 Freq (rad/sec): 36.8
U
-1
-2
-3
-4
-5 -20 -15 -10 -5 0 x 10

-3
Real Axis
(W
= cr = p ( j ))= 0
Ra B + k t k e = 36.056 rad/s La J
U=-
4M
(W p ( j cr )) = 0.1759 rad

x 10
-4
R=0.4; % rotor resistance L=0.001; % rotor inductance ke=0.3; % voltage feedback constant
Imaginary Axis
STABLE limit cycle!!!!

System: Wp Real: -0.00568 Imag: -8.73e-007 Freq (rad/sec): 36.8
kt=0.3; % torque constant Jm=0.01; % motor inertia Jl=0.09 % load inertia Bm=0.05; % motor friction coefficient Bl=0.05; %load friction coefficient J=Jm+Jl; B=Bm+Bl;
U
-1
-2
-3
-4
-5 -20 -15 -10 -5 0 x 10

-3
Real Axis
(W
= cr = p ( j ))= 0
Ra B + k t k e = 36.056 rad/s La J
U=-
4M
(W p ( j cr )) = 0.1759 rad

80
The system presents a periodic steady-state oscillation
60
angle speed current
40
20
-20
-40
-60
-80
10
Time

3

Angle
2.5
1.5
0.5
10
Time

Tc=2/c
1.75
Angle [rad]
1.7
1.65
1.6
1.55
2U=8M|G(jc)|/
1.5
1.45
1.4
8.4
8.45
8.5
8.55
8.6
8.65
8.7
8.75
8.8
Time [s]

What does it happen if the same control law is applied to the speed control problem?
ang
24 Step Sign Gain
V w Cr
Scope
i
DC Motor
kt W (s ) = (sLa + Ra )(sJ + B ) + k t k e
The linear plant is characterised by an all-pole transfer function with relative degree two, therefore there is no contact point between the Nyquist plot and the real negative axis, but the origin at =+ (corresponding to U=0 in the DF plot)

0
U
-0.2
U=0 =+
=0
-0.4
Imaginary Axis
-0.6
The Nyquist plot of the linear system is tangent to the negative reciprocal of the DF at the origin, i.e., U=0 and =+ Maybe a sliding mode behaviour is established asymptotically!?
-0.8
-1
-1.2
-0.5 0 0.5 1 1.5 2 2.5 3
-1
Real Axis

7 angle speed 6
Possibly a sort of dead-bit control, or finite time stabilisation seams to appear
Can it be a sliding mode behaviour?
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Time

7
Speed [rad/s]
By reducing the integration step of the simulation it is apparent that the stabilisation is asymptotic
Can it be a sliding mode behaviour?
0 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Time [s]

60
Rotor current [A]
By reducing the integration step of the simulation it is apparent that the stabilisation is asymptotic
50
40
30
20
10
The rotor current tends to a constant value, i.e. an asymptotic (2nd order) sliding sliding mode appears
-10
-20
-30
-40 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time [s]

The DC motor is a second order system whose state variables are the rotor speed and current
kt k d (t ) B B 0 = (t ) + t ir (t ) J (t ) J J dt J = k + 1 v r (t ) k dir (t ) R R 1 = e (t ) r ir (t ) + v r (t ) e r ir (t ) L r Lr dt Lr Lr Lr Lr kt B B k t (t ) s (t ) = (t ) + ir (t ) + 2 (t ) = 2 + J J J J ir (t )
The system transfer function has a zero in 2, and if the system output is steared to zero in a finite time, than the system behaves as a first order system with time constant =0.5

A MATLAB-Simulink scheme is the following
Sign
ang
24
V w
r=2
Relay
Manual Switch
Gain
Cr
Scope
i
DC Motor du/dt Derivative
2 Gain1
Take care that because of the feedback, the new closed loop gain (assuming the rotor velocity as the output) will be 1/2

3.5
2.5
It is apparent that the shaft speed tends to the steady state value as a first order system with = 0.5 s
Speed [rad/s]
1.5
0.5
2.5
0
0.5
1.5
2.5
3.5
Time [s]
2
Apart from a very short transient, the state trajectory tends to the steady-state values sliding on a linear manifold of the state space
Rotor current [A]
1.5
0.5
0.5
1.5
2.5
3.5
Speed [rad/s]

2
U
0
Imaginary Axis
The Nyquist plot of the linear system cross the negative reciprocal of the DF at the origin, i.e., U=0 and =+ A sliding mode behaviour is established in a finite time
-1
-2
-3
-4 -2
-1
Real Axis
DF Analysis & Sliding Modes
Sliding modes are characterised by infinite frequency of proper ideal switching devices Most sliding mode controllers use a sign function in the controller, i.e., an ideal relay, which can be approximately represented by its Describing Function By the example it is apparent that a sliding mode behaviour is established in a finite time if the Nyquist plot of cross the inverse negative of the describing function at the point (U=0,=+) It can also be derived that a sliding mode behaviour is established asymptotically if the Nyquist plot of is tangent to the inverse negative of the describing function at the point (U=0,=+) TAKE CARE: the Describing function is an approximate tool

Substitute the ideal relay with a hysteretic one with =1
Sign
ang
24 r=0 Relay Manual Switch Gain
V w
Scope
Cr i
2 Gain1
It is apparent that an ideal sliding mode cannot appear because of the switching delay
Supplu voltage [V]
The limit cycle is apparent, with a period Tlc0.6 ms The magnitude is very small because of the low-pass filter property of the motor transfer function
20 10 0 -10 -20 0.4 0.4001 0.4002 0.4003 0.4004 0.4005 0.4006 0.4007 0.4008 0.4009 0.401
x 10
-5
Speed [rad/s]
0.5
-0.5
-1 0.4
0.4001
0.4002
0.4003
0.4004
0.4005
0.4006
0.4007
0.4008
0.4009
0.401
Time [s]

The Nyquist plot of the linear system has no common point with the negative inverse of the Describing Function of the hysteretic relay Describing function analysis can be useful in sliding mode systems when a common point is present
2
-1/N(U) U=1
0
Imaginary Axis
=+
-1
=0
-2
G(j)
-3
-4 -2
-1
Real Axis
Most effective use of the Describing Function approach to sliding modes.
Analysis of the characteristics of a chattering behaviour due to unmodelled dynamics of sensors and/or actuators
What is chattering? It appears as oscillations of the system variables, whose magnitude is related to the influence of the neglected dynamics on the system bandwidth
It is very close to a limit cycle

Consider the motor drive as a hysteretic switching device plus a time constant a=0.1 s
Scope2 Sign
Scope3
24 r=2*pi Relay Manual Switch Gain
1 0.1s+1 Transfer Fcn
ang V w Cr
Scope
i
2 Gain1

2
-1/N(U)
0
Imaginary Axis
-1
-2
-3
G(j)
-4 -2
-1
-1/N(U)
System: untitled1 Real: -0.0507 Imag: -0.0328 Freq (rad/sec): 662
Real Axis
Nyquist plot parameters lc=662 rad/s Ulm=1.1833 rad/s2 Uls represents the magnitude of the oscillation of the sliding variable
G(j )

Steady-state regime
2
1.5
Sliding variable [rad/s
Simulation results Tls=9.7 ms lc=646 rad/s Ulm=1.8544 rad/s2
]
0.5 0 -0.5 -1 -1.5 -2 2.9
Nyquist plot parameters lc=662 rad/s Ulm=1.1833 rad/s2
Differences are due to: Numerical solution of the simulation Approximation of the DF approach
2.91
2.92
2.93
2.94
2.95
2.96
2.97
2.98
2.99
Time [s]

3.5
The rotor velocity tends to its steady-state value as a first order system with time constant =0.5s Rotor wound current presents large variations An approximate sliding mode is established
2.5
Velocity Current
1.5
0.5
0.5
1.5
2.5
3.5
Time [s]

kt B 0 & J J = k + 1 vr R & ir e r ir L r Lr Lr B k t = 2 + J J ir
Full system dynamics
kt B ir = 2 + J J
Internal reduced-order dynamics
& = 2 + s
& = 2 + r + 2 + r 2 e t + t v r L J J Lr J Lr JLr r B R B R k k k
Input-output dynamics

Internal reduced-order dynamics At steady-state
& = 2 +
1
(s ) =
1 (s ) s+2
(t ) = 0 + U sin ( lc t )
1 j ls + 2 0.0015
1 lc t + arg sin (t ) = 0 + U j + 2 j lc + 2 lc
3.088
3.086
Speed [rad/s]
Nyquist plot parameters lc=662 rad/s Uls=1.1833 rad/s2 ls=0.0018 rad/s Simulation results Tls=9.7 ms lc=646 rad/s Uls=1.8544 rad/s2 ls=0.0029 rad/s
3.084
3.082
3.08
3.078
3.076
3.074
3.072
3.07 2.9
2.91
2.92
2.93
2.94
2.95
2.96
2.97
2.98
2.99
Time [s]
The Describing Function approach to the analysis of the chattering phenomenon in sliding mode control systems can be used to have an estimate of the chattering parameters, i.e., frequency and magnitude The estimates are affected by an error which depends on the low-pass properties of the linear part of the plant The sliding variable must be considered as the output of the nonlinear feedback system The actual system output behaviour can be estimated by considering the reduced order dynamics In the presence of a constant reference value, the nonlinear function can be not symmetric, therefore an equivalent gain of the nonlinearity has to be considered to estimate the constant mean steady-state value of the output

Describing Function Analysis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Describing Function Analysis

Uploaded by

Copyright:

Available Formats

University of Cagliari Dept of Electrical and Electronic Eng.

Describing Function analysis

System definition and problem statement

System definition and problem statement

Saturated actuators Relay control Gears backlash Hysteresis in magnetic materials

Dead zone in electromechanical systems

System definition and problem statement

System definition and problem statement

System definition and problem statement

System definition and problem statement

System definition and problem statement

System definition and problem statement

The saturated system oscillates but does not diverge

System definition and problem statement

System: sys Frequency (rad/sec): 0.94 Magnitude (dB): 12

System: sys Frequency (rad/sec): 1.68 Phase (deg): -227

System definition and problem statement

P(j)H (j)C (j)

System definition and problem statement

m(t ) = m0 + m(t ) w(t ) = w0 + w(t ) u0 = kC R kG m0

System definition and problem statement

System definition and problem statement

Electronic oscillators can be based on limit cycles

Describing Function - Assumptions

Describing Function - Assumptions

Describing Function - Assumptions

Describing Function - Assumptions

Describing Function - Assumptions

Fourier expansion - Recall

Describing Function Harmonic balance

u (t ) = U sin ( t ) m(t ) = (a k cos(k t ) + bk sin (k t ))

w(t ) = Gk (ak cos(k t + k ) + bk sin (k t + k )) Gk = G ( jk )

Describing Function Harmonic balance

Describing Function Harmonic balance

A permanent oscillation in the loop appears if

Harmonic balance equation

is the Describing Function of the nonlinear term

Describing Function Harmonic balance

Describing Function Harmonic balance

+ -0.8 -0.6 -0.4 -0.2 0 0.2

Describing Function Harmonic balance

System: sys Real: -2.01 Imag: -0.021 Freq (rad/sec): 0.997

Describing Function Computation

2 2 a1 = m(t ) cos(t )dt T T 2 2 2 b1 = m(t )sin (t )dt T T 2

Describing Function Computation

2 2 a1 = m(t ) cos(t )dt = 0 T T 2

Because of the odd symmetry of the m(t) signal

2 2 4 2 4M b1 = m(t )sin (t )dt = M sin ( )d = T T 2 0

Describing Function Computation

Nyquist and DF Diagrams U

Describing Function Computation

2 2 b1 = m(t ) cos(t )dt = 0 T T 2

Because of the even symmetry of the m(t) signal

2 2 2 2 4M a1 = m(t ) cos(t )dt = M cos( )d = T T 2 0

Describing Function Computation

Describing Function Computation

If U the hysteretic relay behaves as pure hysteresis

8 8 a1 = M cos(t )dt + M cos(t )dt = T 0 T t

The imaginary part of N(U) is proportional to the hysteresis area

Describing Function Computation

Nyquist and DF Diagrams

If is larger than U the relay could not behave as a pure hysteresis

-1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2

Describing Function Computation