DF PDF

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) Plant
r(t) + e(t) u(t) m(t) + y(t)

_ C A P
+
controller actuator process
z(t) ym(t) +
F T
+ trasmitter
filter
n(t)
Controller
r: reference variable ym: measure of the output

e: error signal z: feedback signal
u: control variable d: disturbance
m: manipulated input n: measurement noise
y: controlled variable
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
¾ Saturated actuators +M
¾ Relay control
¾ Gears backlash u
¾ Hysteresis in magnetic
materials
-M
¾ Dead zone in electro-
mechanical systems

e
+U
¾ Saturated actuators
¾ Relay control
-E
¾ Gears backlash e
+E
materials
¾ Dead zone in electro- -U

mechanical systems

m (driven gear)
¾ Relay control
-U u (driving gear)
¾ Gears backlash
+U
materials

mechanical systems

m
¾ Relay control +M0
¾ Gears backlash u
¾ Hysteresis in magnetic -M0

materials

mechanical systems

¾ Relay control
-U u
¾ Gears backlash
+U
materials

mechanical systems
Nonlinearities are not always a drawback, they can also a have a

“stabilising” effect
r(t) + e(t) u(t) m(t) y(t)

_ C A P
z(t)
H
P(s ) =
1
(
s 1 + 1 .4 s + s 2 )
N 1
H (s ) = A(s ) =
0.5
(1 + 0.1s ) O sat (± 2 )
C (s ) = 10
Step response
20
NOT saturated actuator
Saturated actuator
15
10
The saturated
5
system
System output
0
oscillates but
does not
-5 diverge
-10
-15
-20
0 2 4 6 8 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
0 System: sys
Magnitude (dB)

Magnitude (dB): 12 The NOT
-50
saturated
system is not
-100
stable since its
-150
stability margin
-90 are negatives
System: sys
Phase (deg): -180 • mg = -12 dB
Phase (deg)
-180
System: sys
• mϕ = - 47 deg
-270 Frequency (rad/sec): 1.68
Phase (deg): -227
-360
-1 0 1 2
10 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) u(t) m(t) y(t)

C(jω) f(u) P(jω)
_
z(t) H(jω)
r(t) r’(t) + u(t) m(t) w(t)

C(jω) f(u) P(jω)H (jω)C (jω)
_
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 high-
frequency behaviour due to small variations around such a working point
r(t)=R r’(t) + u(t) m(t) w(t)
C(jω) f(u) P(jω)H (jω)C (jω)
_
u (t ) = u0 + Δu (t )
r’(t)=kCR + u(t)
m(t) w(t) m(t ) = m0 + Δm(t )
f(u) G (jω)
w(t ) = w0 + Δw(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
Δm
m
kC
⎧⎪m0 = f (u0 )
R
⎨
Δu
kG
m0 ⎪⎩ u0 = kC R − kG m0
kC R u
u0
Δm0 = f ′(Δu )
f ′(Δu ) = f (u 0 + Δu ) − m0
1
−
kG
Δr’(t)=0 + Δu(t) Δm(t) Δw(t)

f’(u) G (jω)
_
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

f’(u) G (jω)
_
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 varia-
bles diverge to infinity or some periodic motion can appear?
Limit cycles
Limit cycle: a periodic oscillation around a constant working point
G (s ) =
10
r’=5δ−1(τ) + u(t)
f(u) m(t)
G (jω)
w(t) (
(s + 1) s 2 + 1.4s + 1 )
_
f (u ) = sat (1,− 116 , 16

11
)
2
Δm
1 1
1.5 m=− u+
10 2
1
5
Δu u0 =
11
m
0.5
m0
5
m0 =
0
u0 11
-0.5
-1
-6 -4 -2 0 2 4 6 8
u
Limit cycles
Limit cycle: a periodic oscillation around a constant working point
f’(u) G (jω)
_
7
5
u0 =
6 11
5
5 m0 =
11
output w(t)
w
0
4
50
1 w0 = = 4.545
11
0
0 5 10 15 20 25 30
time [s]
Limit cycles
Limit cycle can be a drawback in control systems:
9 Instability of the equilibrium point
9 Wear and failure in mechanical systems
9 Loss of accuracy in regulation
Parameters of the limit cicle can be used to discriminate between

acceptable and dangerous oscillations
¾ oscillation frequency
¾ 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
9 There is only one single nonlinear component
9 The nonlinear component is not dynamical and time

invariant
9 The linear component has low-pass filter properties
9 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) m(t) w(t)

N.L. G (jω)
_
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
u
u m=M
u +ε
-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 ( jω ) >> G ( jnω ) 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

f’(u) G (jω)
_
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 2π
T t ) + bk cos(k 2π
T t ))
2 k =1
T π
a k = ∫ f (t ) cos(k t )dt = ∫ f (t )cos(k t )d ( 2Tπ t )

2 2 2π 1 2
2π
T −T 2 T
π −π 2
T
k = 0,1,2,K b0 = 0
T π
bk = ∫ f (t )sin (k t )dt = ∫ f (t )sin (k t )d ( 2Tπ t )

2 2 2π 1 2
2π
T −T 2 T
π −π 2
T
Describing Function – Harmonic balance

f’(u) G (jω)
_
Δu (t ) = U sin (ω t )
∞
Δm(t ) = ∑ (a k cos(kω t ) + bk sin (kω t ))
k =1
∞
Δw(t ) = ∑ Gk (ak cos(kω t + ϕ k ) + bk sin (kω t + ϕ k ))
k =1
Gk = G ( jkω ) ϕ k = ∠G ( jkω )
Consider the polar representation of a complex number associated

with the exponential form of harmonic signals
Δu (t ) = Ue jω t
∞ M k = ak2 + bk2
m(t ) = ∑ M k e jϑk e jkω t a
k =1 ϑk = arctan k
bk
Taking into account the low-pass property of the linear part of the
system
∞
Δw(t ) = ∑ Gk e jϕ k M k e jϑk e jkω t ≅ G1e jϕ1 M 1e jϑ1 e jω t
k =1

N.L. G (jω)
_
A permanent oscillation in the loop appears if Δu(t)=-Δw(t)

jω t jϕ1 jϑ1 jω t jϕ1 M 1 jϑ1
Ue = −G1e M 1e e 1 + G1e e =0
U
Harmonic balance equation
1 + G ( jω )N (U , ω ) = 0
N (U , ω ) = (b1 + ja1 )
1
is the Describing Function of the nonlinear term
U
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 ( jω ) = −
1
N (U , ω )
Nyquist Diagram
2
ω=+ ∞
0
-2
Imaginary Axis
(U0 ,ω 0 )
-4 +∞← U U=0
-\1/N(U)
-6
-8
G(jω)
+
ω=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 System: sys

Real: -2.01
Imag: -0.021
Freq (rad/sec): 0.997
0 ω=100
-0.5
ω=17.7828
Imaginary Axis
ω=0.1
-1
U
-1.5 System: sys
Real: -1.14
Imag: -2.02
-2
ω=3.1623
-2.5
-3 ω
ω=0.56234
-3.5
-8 -7 -6 -5 -4 -3 -2 -1 0 1
Real Axis
Describing Function – Computation
The DF computation can be performed by means of its definition
Usin(ωt) Δm(t)
f’(u)
T
2 2
a1 = ∫ Δm(t ) cos(ωt )dt
T −T 2
N (U , ω ) = (b1 + ja1 )
1
U T
2 2
b1 = ∫ Δm(t )sin (ωt )dt
T −T 2
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 2
Δu
Δm
1.5
+M
2
+M 1
1
0.5
Δ m(t)
0 0
T
-0.5
-1
-M -1
-2
-M -1.5
-3
-3 -2 -1 0 1 2 3 -2
0 0.5 1 1.5 2 2.5 3
Δ u(t) Time
T
2 2
a1 = ∫ Δm(t ) cos(ωt )dt = 0 Because of the odd symmetry of the Δm(t) signal
T −T 2
T π
N (U , ω ) =
2 2 4 2
b1 = ∫ Δm(t )sin (ωt )dt = ∫ M sin (ϑ )dϑ =
4M 4M
π 0 π
T −T 2
Uπ
Ideal relay Nyquist and DF Diagrams

0.5
U
The negative reciprocal 0
of the DF is the (ω ;4M/π) -1/N(U)

-0.5 c
negative real axis in
backward direction -1
Imaginary Axis
-1.5
-2 ω
A limit cycle can exist
if the relative degree of -2.5
G(jω) is greater than -3

G(jω )
two
-3.5
-4
-1.6 -1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2
Real Axis
The oscillation frequency is the critical frequency ωc of the linear system and the
oscillation magnitude is proportional to the relay gain M
Pure hysteresis
2
2
Δu
M Δm
1.5
1.5
1 1
0.5 0.5
Δm
0 0
-0.5 -0.5
-1 -1
-M
-1.5 -1.5
-2 -2
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5 3
Δu Time
T
2 2
b1 = ∫ Δm(t ) cos(ωt )dt = 0 Because of the even symmetry of the Δm(t) signal
T −T 2
T π
2 2 2 2
a1 = ∫ Δm(t ) cos(ωt )dt = − ∫ M cos(ϑ )dϑ = −
4M
π 0 π
N (U , ω ) = − j
T −T 2 4M
Uπ
Pure hysteresis 0.2
The negative reciprocal

-0.2
of the DF is the
negative imaginary -0.4
axis in backward
Imaginary Axis
direction -0.6
-0.8
G(jω )
A limit cycle can exist -1
if the relative degree -1/N(U) ω

-1.2
of G(jω) is greater
than one and G(jω) is -1.4
U
a type-0 system
-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 ωc of the linear
system and the oscillation’s magnitude is proportional to the relay gain M and to
the modulus of the transfer function at phase -π/4
Hysteretic relay
2.5
2
Δu
Δm
2
+M
1.5
1.5
1
1
0.5 0.5
-β β
Δm
0 0
t
γ
-0.5
-0.5
-1
-1
-1.5
-1.5
-M -2
-2 -2.5
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5 3
Δu Time
If U≤β the hysteretic relay behaves as pure hysteresis

⎛β ⎞
tγ T 2
8 4
b1 = − M ∫ sin (ωt )dt + M ∫ sin (ωt )dt =
8 4M
1− ⎜ ⎟
T 0 T tγ π ⎝U ⎠
⎧ 4M
tγ T
4
⎪− j Uπ U ≤β
a1 = − M ∫ cos(ωt )dt + M ∫ cos(ωt )dt =
8 8
⎪
T 0 T tγ N (U , ω ) = ⎨
β 4 Mβ
2
⎪ 4 M ⎛ ⎞
1− ⎜ ⎟ − j U >β
The imaginary part of N(U) is proportional to ⎪⎩ πU ⎝U ⎠ πU 2
the hysteresis area
Hysteretic relay Nyquist and DF Diagrams

0.5
of the DF is the parallel -0.5
to the negative real axis U -1/N(U)

and with constant -1
Imaginary Axis
negative imaginary part -1.5
-2
If β is larger than U
-2.5
the relay could not -3

G(jω )
behave as a pure
-3.5 ω
hysteresis
-4
-1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2
Real Axis
The oscillation frequency is lower than the critical frequency ωc of the linear
system and the oscillation’s magnitude is proportional to the relay gain M
Saturation 1
Δu
Δm
0.8
4
+M 0.6
3
+M
0.4
2
k 0.2
1
0
Δm
0
t
γ
-1 -0.2
-2 -0.4
-M
-3 -0.6
-M
-4 -0.8
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Δu
-1
0 0.5 1 1.5 2 2.5 3
Time
If U≤Μ the saturation behaves as pure gain

T
2 2
T −T 2
2kU ⎡ ⎛ M ⎞ ⎤⎥
tγ T 2
8 4 ⎛M ⎞ ⎛M ⎞
b1 = ∫ kU sin (ωt )dt + M ∫ sin (ωt )dt =
8 2
⎢ arcsin⎜ ⎟+⎜ ⎟ 1− ⎜ ⎟
T 0 T tγ π ⎢ ⎝ kU ⎠ ⎝ kU ⎠ ⎝ kU ⎠ ⎥
⎣ ⎦
⎧ M
⎪k U≤
⎪ k
N (U ) = ⎨ ⎡
⎛ M ⎞ ⎤⎥
2
2k ⎛M ⎞ ⎛M ⎞ M
⎪ ⎢arcsin⎜ ⎟+⎜ ⎟ 1− ⎜ ⎟ U>
⎪ π ⎢⎣ ⎝ kU ⎠ ⎝ kU ⎠ ⎝ kU ⎠ ⎥⎦ k
⎩
Saturation Nyquist and DF Diagrams

0.5
-1/N(U)
U -1/k
of the DF is part of the -0.5
negative real axis in

-1
backward direction
Imaginary Axis
-1.5
-2
A limit cycle can exist -2.5
if the relative degree

-3
of G(jω) is greater
ω
than two and gain k is -3.5 G(jω )
sufficiently high
-4
-1.6 -1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2
Real Axis
oscillation’s magnitude depends on the saturation parameters M and k
2.5
Δu
Dead zone 2 Δm
3
1.5
2
1
k
1 0.5
-β 0
Δm
0
t
+β γ
-0.5
-1
-1
-2
-1.5
-2
-3
-3 -2 -1 0 1 2 3
Δu
-2.5
0 0.5 1 1.5 2 2.5 3
If U≤β the dead zone has no output Time
T
2 2
T −T 2
8 4
T
⎡π ⎛ β ⎞ ⎛β ⎞ ⎛ β⎞ ⎤
2
b1 = ∫ k (U sin (ωt ) − β )sin (ωt )dt =

2kU
⎢ − arcsin⎜ ⎟ − ⎜ ⎟ 1 − ⎜ ⎟ ⎥
T tγ π ⎢2 ⎝U ⎠ ⎝U ⎠ ⎝ U ⎠ ⎥⎦
⎣
β = U sin (ωtγ )
⎧0 U ≤β
⎪ ⎡
N (U ) = ⎨ β ⎞ ⎛β ⎞ ⎛ β ⎞ ⎤⎥
2
⎢arcsin⎛⎜
2k
k− ⎟ + ⎜ ⎟ 1− ⎜ ⎟ U >β
⎪ π ⎢ ⎝U ⎠ ⎝ ⎠U ⎝ ⎠ ⎥⎦
U
⎩ ⎣
Dead zone 1
The negative reciprocal 0.5
of the DF is part of the -1/N(U) -1/k
negative real axis in 0
forward direction U=+∞
Imaginary Axis
-0.5
-1
A limit cycle can exist G(jω )

if the relative degree -1.5
of G(jω) is greater ω
than two and gain k is -2
sufficiently high
-2.5
-3.5 -3 -2.5 -2 -1.5 -1 -0.5 0
Real Axis
oscillation magnitude depends on the dead zone parameters β and k
3
Dead zone
2 k
The nonlinear
saturation φ (Δ u)
characteristics of 1
the Dead Zone

can be computed dead zone ψ (Δ u)
Δm
0
by subtracting the
Saturation
characteristics -1
from a linear one
Ψ (Δu ) = k − Φ (Δu )
-2
-3
-3 -2 -1 0 1 2 3
Δu
N Ψ (U ) = k − N Φ (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,ω)
N1(U,ω)
G (jω)
_
N2(U,ω)
N t (U ) = N1 (U ) + N 2 (U )
A number of Describing Function can be computed by particularisation of the

function
Φ (α ) =
2
π
[arcsin (α ) + α 1−α 2 ]
in which the parameter α defines a peculiar point of the nonlinear characteristics
m
+M
k
⎧⎪k α ≥1
-α
u/U N (α ) = ⎨
+α ⎪⎩k ⋅ Φ (α ) α < 1
-M
+M k2
⎧⎪k1 α ≥1
-α
k1
N (α ) = ⎨
⎪⎩k 2 + (k1 − k 2 ) ⋅ Φ (α ) α < 1
u/U
+α
m
-M
k
⎧⎪0 α ≥1
-α u/U N (α ) = ⎨
+α ⎪⎩k (1 − Φ (α )) α < 1
m
k2
+M
⎧0 α ≥1
⎪⎪
-β -α k1
u/U
N (α ) = ⎨k ⋅ (1 − Φ (α )) α <1≤ β
+α +β ⎪
⎪⎩k ⋅ (Φ (β ) − Φ (α )) β < 1
-M
+M k
N (α ) = k +
u/U
4M
-M π
m
+M
N (α ) =
4M
u/U
π
-M
m
+M
−α ⎧0 α ≥1
⎪
u/U N (α ) = ⎨ 4 M
+α
⎪⎩ π 1 − α 2
α <1
-M
k ⎧0 α ≥1
⎪
N (α ) = ⎨ k 4kα
[ ( )] (1 − α ) α < 1
-α u/U
⎪⎩ 2 1 − Φ 2α − 1 − j
+α
π
k
⎧0 α ≥1
⎪
-α N (α ) = ⎨ k 4kα
+α
u/U
⎪⎩ 2 [1 − Φ (2α − 1)] + j (1 − α ) α < 1
π
m
+M
⎧0 α ≥1
⎪
u/U N (U , ω ) = ⎨ 4 M 4 Mα
-α
⎪⎩ πU 1 − α 2
− j α <1
+α πU
-M
m
+M
u/U ⎧0 α ≥1
-α ⎪
+α N (U , ω ) = ⎨ 4 M 4 Mα
⎪⎩ πU 1 − α + j πU α <1
2
-M
Nt(U,ω)
Δr’=0 + Δu Δm Δw
_ G (jω)
1/jω
Negative inverse of the Describing Function

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

N.L. G (jω)
_
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 + G ( jω ) = 0
1
k
The Nyquist criterion looks at the relative position of the transfer function
G(jω) with respect to the point (-1/k, 0) in the complex plain.
By extension the criterion can be applied to any point –α of the complex
plain with reference to the Harmonic Balance Equation
1 + G ( jω )α = 0 , α ∈C
If the transfer function G(jω) represents a stable system, the reduced

Nyquist criterion can be applied, i.e. the closed loop stability can be stated
if the reference point –α in the complex plain remains on the left-side when
running along the Nyquist plot of G(jω) from ω=0+ to ω =+∞.
Nyquist plot of G(jω )

1
G ( jω ) =
1
0.5
0
ω=+∞ ( 2
)
jω ( jω ) + 0.5 jω + 1
-1/k
-0.5
The closed loop is not stable

Imaginary Axis
-1
-1.5
with respect to the point (-1/k,0).
-2
-2.5 -α
The closed loop is stable with
respect to the point -α
-3
-3.5
+
ω=0
-4
-3 -2.5 -2 -1.5 -1 -0.5 0
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
1.5
Points B, B’,
B”, A and A’ 1 A’
A
represent 0.5
possible Imaginary Axis

0
magnitudes of
the oscillation. -0.5 U
-1 B B”
B’
A and B re- -1.5
present two -2
possible limit G(jω )

-2.5 ω
cycles
-3
-3 -2.5 -2 -1.5 -1 -0.5 0
Real Axis
Applying the reduced 2
Nyquist criterion with 1.5
respect to: A’
1
A
0.5
point B’: oscillations tend

0
Imaginary Axis
to decrease (stable system)
-0.5
U
B B”
point B”: oscillations tend -1
B’
to increase (unstable -1.5
system) -2
G(j ω )
-2.5 ω
point A’: oscillations tend -3
-3 -2.5 -2 -1.5 -1 -0.5 0
to decrease (stable system) Real Axis
B: Stable limit cycle A: Unstable limit cycle
The linear system G(jω) 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 d.c.
-24 V
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
dω (t )
Rr: rotor resistance
J = Cem (t ) + Bω (t ) Lr: rotor inductance
dt
dϑ (t )
ke: voltage feedback constant
ω= kt: torque constant
dt Jm: motor inertia
e(t ) = keω (t ) Jl: load inertia
Bm: motor friction coefficient
Cem (t ) = kt ir (t ) Bl: load friction coefficient
J = Jm + Jl
vr: rotor supply voltage
B = Bm + Bl ir: rotor wound current
Cem: electromagnetic torque
ω: rotational speed
θ: shatf angular position
⎧⎪+ 24 ϑ < ϑd
vr (t ) = ⎨
⎪⎩− 24 ϑ ≥ ϑd
The motor transfer functions are the following
Ω(s ) kt
Wω (s ) = =
Vs (s ) (sLa + Ra )(sJ + B ) + k t k e
Θ ( jω ) kt
Wϑ ( jω ) = = = ℜ( jω ) + jℑ( jω )
Vr ( jω ) jω (( jωLa + Ra )( jωJ + B ) + k t k e )
k t ( R a J + La B )
=-
ω 2 (Ra J + La B )2 + (Ra B + k t k e − ω 2 La J )
2
(
k t R a B + k t k e − ω 2 La J )
−j
[
ω ω 2 (Ra J + La B )2 + (Ra B + k t k e − ω 2 La J )
2
]
Taking into account the parameters of the motor and of the load
R=0.4; % rotor resistance 0
L=0.001; % rotor inductance U

-2
ke=0.3; % voltage feedback constant -4
kt=0.3; % torque constant -6
Imaginary Axis
-8
Jm=0.01; % motor inertia
-10
Jl=0.09 % load inertia
-12
Bm=0.05; % motor friction coefficient

-14
Bl=0.05; %load friction coefficient

-16
J=Jm+Jl; -18
ω
B=Bm+Bl; -20
-2 -1.5 -1 -0.5 0 0.5 1
Real Axis
x 10
-4 Nyquist and DF plots
5
R=0.4; % rotor resistance
4
L=0.001; % rotor inductance
3
ke=0.3; % voltage feedback constant
2
kt=0.3; % torque constant System: Wp
Imaginary Axis
1 Real: -0.00568
Jm=0.01; % motor inertia Imag: -8.73e-007

0
Jl=0.09 % load inertia U

-1
Bm=0.05; % motor friction coefficient -2
Bl=0.05; %load friction coefficient -3
J=Jm+Jl; -4
ω
B=Bm+Bl; -5
-20 -15 -10 -5 0

-3
Real Axis x 10
Ra B + k t k e
⋅ ℜ(W p ( jω cr )) = 0.1759 rad
4M
ω ℑ(W = ω cr = = 36.056 rad/s U=-
p ( jω ))= 0
La J π
x 10
-4 Nyquist and DF plots
5
R=0.4; % rotor resistance
4
L=0.001; % rotor inductance
ke=0.3; % voltage feedback constant
3 STABLE limit cycle!!!!
2
kt=0.3; % torque constant System: Wp
Imaginary Axis
1 Real: -0.00568
Jm=0.01; % motor inertia Imag: -8.73e-007

0
Jl=0.09 % load inertia U

-1
Bm=0.05; % motor friction coefficient -2
Bl=0.05; %load friction coefficient -3
J=Jm+Jl; -4
ω
B=Bm+Bl; -5
-20 -15 -10 -5 0

-3
Real Axis x 10
Ra B + k t k e
⋅ ℜ(W p ( jω cr )) = 0.1759 rad
4M
ω ℑ(W = ω cr = = 36.056 rad/s U=-
p ( jω ))= 0
La J π
80
angle
The system 60
speed
presents a current
periodic 40
steady-state
oscillation 20
-20
-40
-60
-80
0 1 2 3 4 5 6 7 8 9 10
Time
The system 2.5

presents a
periodic
steady-state 2
oscillation
Angle
1.5
0.5
0
0 1 2 3 4 5 6 7 8 9 10
Time
1.75 Tc=2π/ωc
The system
1.7
presents a
periodic 1.65
steady-state
Angle [rad]
oscillation 1.6
1.55
2U=8M|G(jωc)|/π
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 V
Step Sign Gain w
Scope
Cr
i
DC Motor
Wω (s ) =
kt
(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 ω=+∞ (corresponding to U=0 in the
DF plot)

U=0
ω =+∞
0
U ω =0 The Nyquist plot of the
-0.2 linear system is tangent
to the negative
-0.4 reciprocal of the DF at
Imaginary Axis
the origin, i.e., U=0 and

-0.6
ω=+∞
-0.8
Maybe a sliding mode
-1
behaviour is established
asymptotically!?
-1.2
ω
-1 -0.5 0 0.5 1 1.5 2 2.5 3
Real Axis
7
angle
speed
Possibly a sort of 5
dead-bit control,
or finite time
4
stabilisation
seams to appear
3
Can it be a 2
sliding mode
behaviour?
1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time
7
By reducing the 6
integration step
of the simulation 5
it is apparent that
the stabilisation
Speed [rad/s]
4
is asymptotic
Can it be a 2
sliding mode
behaviour?
1
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Time [s]
60
50
By reducing the
integration step 40
of the simulation
it is apparent that 30
the stabilisation
20
is asymptotic
Rotor current [A]
10
The rotor current 0
tends to a
constant value, -10
i.e. an
-20
asymptotic (2nd
order) sliding -30
sliding mode
appears -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
dω (t ) B k ⎡ B kt ⎤
= ω (t ) + t ir (t ) ⎢ ⎥ ⎡ω (t )⎤ ⎡ 0 ⎤
dt J J = ⎢
J J
⎥ ⎢ ⎥ + ⎢ 1 ⎥ v (t )
dir (t ) ⎢ ⎥
− ⎥ ⎢⎣ir (t )⎥⎦ ⎢ L ⎥
ke Rr ke Rr
= − ω (t ) − ir (t ) + v r (t ) ⎢−
1 r
⎣ r⎦
dt Lr Lr Lr ⎣⎢ Lr Lr ⎦⎥
kt ⎡ B k t ⎤ ⎡ω (t )⎤
s (t ) = ω (t ) + ir (t ) + 2ω (t ) = ⎢2 +
B
⎥⎢ ⎥
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 τ=0.5
A MATLAB-Simulink scheme is the following
Sign
ang
24 V
r=2π Manual Switch Gain w
Relay Scope
Cr
i
DC Motor
du/dt
Derivative
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
3
It is apparent that the shaft speed tends
to the steady state value π as a first
2.5
order system with τ = 0.5 s
Speed [rad/s]
1.5
0.5
2.5
0
0 0.5 1 1.5 2 2.5 3 3.5
Time [s]
2
Apart from a very short transient, the
Rotor current [A]

1.5
state trajectory tends to the steady-state

values sliding on a linear manifold of 1
the state space

0.5
0
0 0.5 1 1.5 2 2.5 3 3.5
Speed [rad/s]

2
1 The Nyquist plot of the

U
linear system cross the
0
negative reciprocal of
the DF at the origin, i.e.,
Imaginary Axis
U=0 and ω=+∞

-1
A sliding mode
-2
behaviour is established
in a finite time
-3
-4
-2 -1 0 1 2 3 4 5 6 7 8
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,ω=+∞)
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,ω=+∞)
TAKE CARE: the Describing function is an approximate tool
Substitute the ideal relay with a hysteretic one with β=1
Sign
ang
24 V
r=0 Manual Switch Gain w
Relay Scope
Cr
i
DC Motor
du/dt
Derivative
Gain1
It is apparent that an ideal sliding mode cannot appear because of the switching delay
20
The limit
Supplu voltage [V]

cycle is 10
apparent,
0
with a period
Tlc≅0.6 ms -10
-20
The 0.4 0.4001 0.4002 0.4003 0.4004 0.4005 0.4006 0.4007 0.4008 0.4009 0.401
magnitude is
very small -5
x 10
1
because of
the low-pass 0.5
Speed [rad/s]
filter property
of the motor 0
transfer
-0.5
function
-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 2
plot of the linear

system has no
common point 1
with the -1/N(U)

negative inverse 0
U=1
of the ω =+∞ ω =0
Imaginary Axis
Describing
Function of the -1
hysteretic relay
Describing -2
function
analysis can be -3
G(jω)
useful in sliding
mode systems
when a common -4
-2 -1 0 1 2 3 4 5 6 7 8
point is present 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 τa=0.1 s
Scope2 Scope3
Sign
1 ang
24 V
0.1s+1
r=2*pi Manual Switch Gain Transfer Fcn w
Relay Scope
Cr
i
DC Motor
du/dt
Derivative
Gain1
2
-1/N(U)
0
Imaginary Axis
-1
-2
-3 G(jω)
-4
-2 -1 0 1 2 3 4 5 6 7 8
-1/N(U)
Real Axis
System: untitled1
Real: -0.0507
Nyquist plot parameters Imag: -0.0328
Freq (rad/sec): 662
ωlc=662 rad/s
Ulm=1.1833 rad/s2
G(j ω)
Uls represents the magnitude of
the oscillation of the sliding
variable
Steady-state regime
2
1.5
Nyquist plot parameters

ωlc=662 rad/s 1
Ulm=1.1833 rad/s2
]
2
Simulation results 0.5
Tls=9.7 ms
Sliding variable [rad/s
ωlc=646 rad/s 0
Ulm=1.8544 rad/s2
-0.5
-1
Differences are due to:

• Numerical solution of
-1.5
the simulation
• Approximation of the
-2
DF approach 2.9 2.91 2.92 2.93 2.94 2.95 2.96 2.97 2.98 2.99 3
Time [s]
3.5
The rotor velocity tends 2.5

Velocity
to its steady-state value Current
as a first order system
2
with time constant
τ=0.5s
1.5
Rotor wound current
presents large variations 1
An approximate sliding
0.5
mode is established
0
0 0.5 1 1.5 2 2.5 3 3.5
Time [s]
⎡ B kt ⎤
⎡ω& ⎤ ⎢ J ⎥ ⎡ω⎤ ⎡ 0 ⎤
⎢ ⎥=⎢ k
J
⎥ ⎢ ⎥ + ⎢ 1 ⎥v
&
⎢⎣ir ⎥⎦ ⎢−
Rr ⎢ ⎥ r
e
− ⎥ ⎢⎣ir ⎥⎦ ⎢ L ⎥
⎢⎣ Lr Lr ⎥⎦ ⎣ r⎦ Full system dynamics
⎡ B k t ⎤ ⎡ω ⎤
σ = ⎢2 + ⎥ ⎢ ⎥
⎣ J J ⎦ ⎢⎣ir ⎥⎦
kt ⎛ B⎞
ir = σ − ⎜ 2 + ⎟ω
J ⎝ J⎠
Internal reduced-order dynamics ω& = −2ω + s
⎛ B R ⎞ ⎛⎛ B ⎞⎛ R ⎞ k k ⎞ k
Input-output dynamics σ& = ⎜⎜ 2 + − r ⎟⎟σ + ⎜⎜ ⎜ 2 + ⎟⎜⎜ r − 2 ⎟⎟ − e t ⎟⎟ω + t v r
⎝ J Lr ⎠ ⎝⎝ J ⎠⎝ Lr ⎠ Lr J ⎠ JLr
Ω(s ) = Σ(s )
1
Internal reduced-order dynamics ω& = −2ω + σ
s+2
At steady-state ⎛ ⎛ ⎞⎞
ω (t ) = ω 0 + U ⎜ 1 1
sin ⎜ ω lc t + arg⎜⎜ ⎟⎟ ⎟
σ (t ) = σ 0 + U sin (ω lc t ) jω lc + 2 ⎝ ⎝ j ω lc + 2 ⎠
⎟
⎠
1
≅ 0.0015
jω ls + 2 3.088
3.086
Nyquist plot parameters 3.084
ωlc=662 rad/s
3.082
Speed [rad/s]
Uls=1.1833 rad/s2
Δωls=0.0018 rad/s 3.08
3.078
Simulation results
3.076
Tls=9.7 ms
ωlc=646 rad/s 3.074
Uls=1.8544 rad/s2 3.072
Δωls=0.0029 rad/s 3.07

2.9 2.91 2.92 2.93 2.94 2.95 2.96 2.97 2.98 2.99 3
Time [s]
9 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
9 The estimates are affected by an error which depends on the low-pass properties of
the linear part of the plant
9 The sliding variable must be considered as the output of the nonlinear feedback
system
9 The actual system output behaviour can be estimated by considering the reduced
order dynamics
9 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

DF PDF

Uploaded by

Document Information

Original Title

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

DF PDF

Uploaded by

Copyright:

Available Formats

University of Cagliari

Dept of Electrical and Electronic Eng.

Describing Function analysis

r(t) + e(t) u(t) m(t) + y(t)

r: reference variable ym: measure of the output

In many cases the system presents a nonlinear phenomenon which is fully

In many cases the system presents a nonlinear phenomenon which is fully

¾ Dead zone in electro- -U

In many cases the system presents a nonlinear phenomenon which is fully

¾ Dead zone in electro-

In many cases the system presents a nonlinear phenomenon which is fully

¾ Relay control +M0

¾ Hysteresis in magnetic -M0

¾ Dead zone in electro-

In many cases the system presents a nonlinear phenomenon which is fully

¾ Dead zone in electro-

Nonlinearities are not always a drawback, they can also a have a

r(t) + e(t) u(t) m(t) y(t)

Bode Diagram of the difference transfer function

Frequency (rad/sec): 0.94

r(t) + e(t) u(t) m(t) y(t)

r(t) r’(t) + u(t) m(t) w(t)

If constant (or very slowly varying) reference signals are considered,

Δr’(t)=0 + Δu(t) Δm(t) Δw(t)

Δr’(t)=0 + Δu(t) Δm(t) Δw(t)

f (u ) = sat (1,− 116 , 16

Limit cycle can be a drawback in control systems:

9 Instability of the equilibrium point

9 Wear and failure in mechanical systems

9 Loss of accuracy in regulation

Parameters of the limit cicle can be used to discriminate between

Electronic oscillators can be based on limit cycles

The describing Function approach to the analysis of steady-state oscillations

It is based on the following assumptions

9 There is only one single nonlinear component

9 The nonlinear component is not dynamical and time

9 The linear component has low-pass filter properties

9 The nonlinear characteristic is symmetric with respect

There is only one single nonlinear component

The system can be represented by a lumped parameters

r’=const. + u(t) m(t) w(t)

The nonlinear component is not dynamical and time invariant

The system is autonomous.

The linear component has low-pass filter properties

This is the main assumption that allows for neglecting the

The more the low-pass filter assumption is verified the

The nonlinear characteristic is symmetric with respect to the origin

Δr’(t)=0 + Δu(t) Δm(t) Δw(t)

Such an assumption is usually taken for the sake of simplicity, and

Consider a periodic function

a k = ∫ f (t ) cos(k t )dt = ∫ f (t )cos(k t )d ( 2Tπ t )

bk = ∫ f (t )sin (k t )dt = ∫ f (t )sin (k t )d ( 2Tπ t )

Δr’(t)=0 + Δu(t) Δm(t) Δw(t)

Consider the polar representation of a complex number associated

Δr’(t)=0 + Δu(t) Δm(t) Δw(t)

A permanent oscillation in the loop appears if Δu(t)=-Δw(t)

The armonic balance equation is a necessary condition for the existence of

The approximate analysis gives good estimates if the low-pass filter

The harmonic balance equation is similar to the characteristic polynomial

The Describing Function is a linear approximation of the static nonlinearity

Nyquist and DF Diagrams

0.5 System: sys

The DF computation can be performed by means of its definition