rti vorlesung 10 - eth z

33
22.11.2019 RTI Vorlesung 10 Jetzt geht’s los!

Upload: others

Post on 25-Oct-2021

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: RTI Vorlesung 10 - ETH Z

22.11.2019

RTI Vorlesung 10

Jetzt geht’s los!

Page 2: RTI Vorlesung 10 - ETH Z

2

Page 3: RTI Vorlesung 10 - ETH Z

Type and relative degree

3

Page 4: RTI Vorlesung 10 - ETH Z

Corresponding Nyquist Diagram

4

Re

Im

-1

Page 5: RTI Vorlesung 10 - ETH Z

Bode’s law

5

The gradient of the magnitude plot (! ⋅ 20 ⁄&' &()) determines the phase shift (! ⋅ ⁄* +).

Page 6: RTI Vorlesung 10 - ETH Z

FD system identification

6

Complete model description: Σ " ,$2(")

Page 7: RTI Vorlesung 10 - ETH Z

FD I/O measurement

7

Measure the frequency response ! times at each frequency "#:

magnitude $ and phase %at frequencies "# (& = 1,2, … , ,)and measurement . (. = 1,2, … , !)

Page 8: RTI Vorlesung 10 - ETH Z

8

FD I/O nominal system identification

Page 9: RTI Vorlesung 10 - ETH Z

FD I/O uncertainty identification

9

Here:decimal values, not in dB

Model uncertainty is ≥ 100% as of from %&

Plot the absolute value of the relative error and find an upper bound '& (% for it:

Page 10: RTI Vorlesung 10 - ETH Z

20.09. Lektion 1 – Einführung

27.09. Lektion 2 – Modellbildung4.10. Lektion 3 – Systemdarstellung, Normierung, Linearisierung

11.10. Lektion 4 – Analyse I, allg. Lösung, Systeme erster Ordnung, Stabilität18.10. Lektion 5 – Analyse II, Zustandsraum, Steuerbarkeit/Beobachtbarkeit

25.10. Lektion 6 – Laplace I, Übertragungsfunktionen1.11. Lektion 7 – Laplace II, Lösung, Pole/Nullstellen, BIBO-Stabilität8.11. Lektion 8 – Frequenzgänge (RH hält VL)

15.11. Lektion 9 – Systemidentifikation, Modellunsicherheiten 22.11. Lektion 10 – Analyse geschlossener Regelkreise 29.11. Lektion 11 – Randbedingungen

6.12. Lektion 12 – Spezifikationen geregelter Systeme13.12. Lektion 13 – Reglerentwurf I, PID (RH hält VL)20.12. Lektion 14 – Reglerentwurf II, „loop shaping“

Modellierung

Systemanalyse im Zeitbereich

Systemanalyse im Frequenzbereich

Reglerauslegung

10

Page 11: RTI Vorlesung 10 - ETH Z

Closed-loop system

12

Page 12: RTI Vorlesung 10 - ETH Z

Loop gain L(s)

13

The loop gain is the open-loop transfer function from ! → #.

Page 13: RTI Vorlesung 10 - ETH Z

Sensitivity S(s)

14

The sensitivity is the closed-loop transfer function from ! → #(and from $ → %).

Page 14: RTI Vorlesung 10 - ETH Z

Complementary sensitivity T(s)

15

The complementary sensitivity is the closed-loop transfer function from ! → # (and from $ → #).

Page 15: RTI Vorlesung 10 - ETH Z

Graphical interpretation

16

Page 16: RTI Vorlesung 10 - ETH Z

General properties

17

For very large and very small !(#), the following approximations hold:

Moreover, the two vectors (in the complex plane) % # and &(#)always add up to 1. Therefore, at a specific frequency, only one of them can be substantially smaller than 1:

Page 17: RTI Vorlesung 10 - ETH Z

Recap: Disturbance and Noise

18

The disturbance !(#) is an external signal, which causes a deviation of the actual plant output % # .

The noise '(#) is an external signal, which causes a deviation of the measurement of the plant output % # .

Typically, the disturbance is a low-frequency signal, whereas the noise is a high-frequency signal.

If this frequency separation is not present then one cannot guarantee simultaneously disturbance rejection and noise attenuation.

Page 18: RTI Vorlesung 10 - ETH Z

19

Page 19: RTI Vorlesung 10 - ETH Z

20

QC: What is a “reasonable” loop gain L(s)?

Page 20: RTI Vorlesung 10 - ETH Z

Controller design rule # 1: Don’t cancel unstable poles with non-minimum phase zeros

22

Transfer function from ! → #?

Transfer function from ! → $?

% & = 1& + * + 1

Example:

Page 21: RTI Vorlesung 10 - ETH Z

Internal closed-loop stability

24

For «internal» (complete) stability, all nine transfer functions

must be asymptotically stable.

Alternative: closed-loop completely asymptotically stable iff1 + # $ has only zeros with negative real parts (all poles of % $ or &($) have negative real parts), and no cancellations of unstable poles with non-minimum phase zeros occur in the loop gain.

From now on this is always assumed.

Page 22: RTI Vorlesung 10 - ETH Z

Nyquist theorem

25

Nyquist Theorem: Let !" be the number of poles with positive real part and !# the number of poles with zero real part of the open-loop transfer function $(&), and let !( be the number of encirclements of the point −1 by $()*) counted positively in the counter-clockwise direction when * is changed from −∞ to +∞.

Then the closed-loop system is asymptotically stable iffthe number !( = ⁄!# 0 + !".

Page 23: RTI Vorlesung 10 - ETH Z

Wow!

26

1. We now can test a time-domain property (asymptotic stability) by working only in the frequency domain with transfer functions.

2. We can decide if the closed-loop system ! " (nonlinear function of #(")) is asymptotically stable by analyzing the open-loop transfer function & " (linear function of #(")).

3. Even better: we don’t need & " , knowing the frequency response of & () is sufficient.

4. The Nyquist theorem provides quantitative information about “how stable” a closed-loop system is and with that one can design (“optimize”) controllers #(").

Page 24: RTI Vorlesung 10 - ETH Z

Number of encirclements

27

0"

0#∞−∞

Page 25: RTI Vorlesung 10 - ETH Z

28

−∞∞

0$0%

Number of encirclements

Page 26: RTI Vorlesung 10 - ETH Z

29

Number of encirclements

0"

0#

+∞−∞

Page 27: RTI Vorlesung 10 - ETH Z

Example: Stability

30

! " = $ " ⋅ & " = 100") + 12" + 100 ⋅ 0.1

0.1" + 10.1"

Is the corresponding closed-loop system asymptotically stable?

+∞−∞

0/

00

Page 28: RTI Vorlesung 10 - ETH Z

Proof of Nyquist theorem

32

Using the “Principle of Arguments”, details: see Appendix B.3

A few remarks:

• The Nyquist theorem is valid for all SISO LTI systems, even those with non-rational transfer functions (delays).

• The proof of the Nyquist theorem is based on fundamental results of complex analysis; it does not rely on the frequency response interpretation (which works only for BIBO stable systems).

• Accordingly, even if L(s) is not BIBO stable, its frequency response L(jw) and the Nyquist theorem can be used.

Page 29: RTI Vorlesung 10 - ETH Z

33

The phase margin ! is the distance from -180° where " #$ enters the unit circle in the Nyquist diagram (magnitude 1).

The gain margin % is the the inverse of the magnitude of " #$ at -180°

The minimum return difference &min is the minimum distance from -1.

Robustness measures

Page 30: RTI Vorlesung 10 - ETH Z

34

Robustness measuresUncertainty model 1:

Closed loop system remains asymptotically stable as long as

!nom =0

Page 31: RTI Vorlesung 10 - ETH Z

35

Robustness measures

!nom = 1

Uncertainty model 2:

Closed loop system remains asymptotically stable as long as

Page 32: RTI Vorlesung 10 - ETH Z

Example

36

! =

# =

Page 33: RTI Vorlesung 10 - ETH Z

Robust Nyquist theorem

37

Robust Nyquist Theorem: The uncertain closed-loop system is asymptotically stable if the nominal closed-loop system is asymptotically stable and the following inequality is satisfied