• Aucun résultat trouvé

Flow Control: TD#3

N/A
N/A
Protected

Academic year: 2022

Partager "Flow Control: TD#3"

Copied!
44
0
0

Texte intégral

(1)

Flow Control: TD#3

Overview

- General issues, passive vs active…

- Control issues: optimality and learning vs robustness and rough model - Model-based control: linear model, nonlinear control

- Linear model, identification - Sliding Mode Control

- Delay effect

- Time-delay systems - Introduction to delay systems

- Examples

- Much a do about delay? Some special features + a bit of maths - Time-varying delay

- Model-based control: nonlinear model, nonlinear control - Overview of MF’s PhD: Sliding Mode Control - Application to the airfoil

- Application to the Ahmed body (MF and CC’PhDs)

- Machine Learning and model-free control: + 4h with Thomas Gomez

(2)

examples

Smith’s predictor

special features

a bit of mathematics

A brief introduction to time-delay systems

(3)

A telling example…

Natural loop of audio-phonatory control

TO BE OR NOT TO BE, THAT IS THE QUESTION: WHETHER ‘TIS

NOBLER IN THE MIND TO SUFFER THE SLINGS AND ARROWS OF OUTRAGEOUS FORTUNE. OR TO TAKE ARMS AGAINST A SEA OF TROUBLES, AND BY OPPOSING, END

THEM? TO DIE: TO SLEEP; NO MORE; AND BY A SLEEP

TO SAY WE END

Brief introduction to time-delay systems

(4)

A telling example…

Networked loop of audio-phonatory ctrl

G…

Brief introduction to time-delay systems

(5)

… and a palpable one.

Current interactive systems often take between 50 to 200ms to update the display in response to touch input.

Brief introduction to time-delay systems

(6)

as well as…

Brief introduction to time-delay systems

(7)

Strejc-Broïda models for process control

 frequently used in process engineering

(inertial phenomena)

 simple and generic approximation

(if there is no oscillation / instability)

 PID control? ... OK if t > 5T , poor effectiveness at t <

T

 Smith predictor or « Generalized PID »

(only if open-loop stability)

Exemple du GV LAGIS

T

heating +

thermic transfer

Victor Broïda (Fr) 1969 The determination of large time-constants by step-response extrapolation. Automatica 5(5): 677-683 [IFAC President 1969-1972]

Vladimír Strejc (Cz) 1965 The physical realizability of an optimum Ν-parameter, discrete, linear control system determined in Wiener's sense. Kybernetika 1(5): 399-409

Brief introduction to time-delay systems

Some classical example

(8)

... of control engineering classes (1900’s)

PID ok

T

Brief introduction to time-delay systems

Another classics

T

(9)

Tele surgery: the Lindbergh operation,

07/09/2001

Constant RTT

< 200 msec Distance:

17 000 km

Com. cost ≈ 160k$/month

« The only restriction to the development of long-distance tele-surgery has to do, still today, with its cost. For tele-surgery, you must use a transcontinental ATM line, that you have to book during 6 monthes, at the price of about 1 million dollars. » Prof. J. Marescaux, Le Monde, January 6, 2010

Brief introduction to time-delay systems

More spectacular…

(10)

Brief introduction to time-delay systems

Networked control and communication delays

RTT (40km) Mean = 82 ms Maxi = 857 ms Mini = 1 ms

France - France

one week of RTT…

(11)

Brief introduction to time-delay systems

Networked control and communication delays

RTT (1640km) Maxi = 415 ms Mini = 70 ms RTT (1640km) Maxi = 415 ms Mini = 70 ms

France – North-Africa

one week of RTT…

(12)

Brief introduction to time-delay systems

transmission time + access time + packet loss + sampling…

Measurement channel

1 - Controller

Actuating

2 - Plant

channel

Network

variable delay h

1

variable delay h

2

= 2 variable delays

can be estimated by Plant (time stamps + packet nb)

can be estimated

by Controller (time stamps + packet nb)

known / unknown ?

Hyp: Clock Synchro NTP, GPS

Our approach (2008-2012)

(13)

Brief introduction to time-delay systems

Our approach (2008-2012)

(14)

Brief introduction to time-delay systems

Networked control and communication delays

unshared CAN 2m: 200 µsec

bluetooth 2m: 40 msec

Internet: 100-400 msec

orbital stations: 0.4-7 sec underwater 1.7km: >2 sec

Other RTT approximated values:

(15)

Brief introduction to time-delay systems

Alternative to PID: the Smith Predictor

Example of a delayed measurement (sensor)

How could you implement a PID?

Use a simulation model?

Could you

make it robust?

(16)

Brief introduction to time-delay systems

Alternative to PID: the Smith Predictor

Equivalent scheme:

Controller structure:

?

(17)

Brief introduction to time-delay systems

Alternative to PID: the Smith Predictor

Equivalent scheme:

Controller structure:

Pros:

 Very simple

o H

1

= PID tuning

o case of a PI controller H

1

« PIR »

Cons:

 Restricted use!

o needs Open-Loop stability

o constant delay, known

(18)

predict = « advance » time using the model

Brief introduction to time-delay systems

Interpretation of « Predictor »

 Various techniques are stemming from this idea, including Smith’s

(19)

 such as the « Artstein’s transformation » Brief introduction to time-delay systems

Interpretation of « Predictor »

equivalence of controllability

(20)

Much a do about delay ?

delay

t x

h

t x

Brief introduction to time-delay systems

Much a do about delay?

(21)

drive

voltage u measured angle x

+ -

gap e = 0 – x

speed target angle

x

c

= 0

Brief introduction to time-delay systems

A crude example

(22)

drive u (t )

measured angle x

+ -

e

(

t

)

speed target angle

x

c

= 0

Brief introduction to time-delay systems

A crude example

ctrl.channel.

delay ~ h/2

meas.channel

delay~ h/2

received angle

x(t-h/2)

received control

e (t-h/2)

(23)

t x

x

0

0

?

?

t

0

t

1

t

2

t

3

t

4

t

5

t

6

t

7

t

8

t

9

t

10

Exercise…

Exercise… for my students, don’t worry ;-)

Brief introduction to time-delay systems

A crude example

(24)

?

w.r.t.

Brief introduction to time-delay systems

A crude example

(25)

(parenthèse...)

Brief introduction to time-delay systems

… Last, note that delay may also stabilize

(26)

 notion of « state » ?

initial variable

X(t)

generating a unique solution from time

t

Brief introduction to time-delay systems

Back to the crude example

(27)

Brief introduction to time-delay systems

Back to the crude example

(Shimanov’s notation, 1960)

function x

t

= state at time t vector x ( t ) = x

t

(0) solution at t

t

72

infinite dim. syst.

 notion of « state » ?

initial variable

X(t)

generating a unique solution from time

t

Functional state x

t

(28)

Brief introduction to time-delay systems

Back to the crude example

Re(s) Im(s)

poles?

infinite dim. syst.

infinite number of poles

(29)

© J.P. RICHARD 2011 74

frequency behaviour?

infinite dim.

phasis - 

phasis

+-

(BO)

+-

h

t x

h

t

(Bode, open loop)

x

log w gain

log w phasis

 = -p/2-jhw

Brief introduction to time-delay systems

Back to the crude example

(30)

Let’s sum up...

delay  strong influence on stability functional state

infinite number of eigenvalues (Hurwitz OK, no Routh)

important dephasing (

- )

… and, until now, it was the most simple:

constant delay

scalar, linear system 1

rst

order derivative

What about variable delays h ( t ) ?

a counter-example...

Brief introduction to time-delay systems

« Crude », but not that simple?

(31)

1

p/

2

constant :  h  [0,1] iff  grey zone

variable : asymptot. stable iff  yellow zone :

a b

(T=1)

2

2

0

-2

-2

2

1 2

stable h ( t )

<1

- unstable h =cte

<1

unstable h ( t )

<1

- stable h =cte

<1

=1

for

if if

1 2 3 4 5 6 7 8 91

Brief introduction to time-delay systems

… and mind the variable delays!

Note that such a delay is very – very – classic, guess what it represents?

(32)

Brief introduction to time-delay systems

Yes! It corresponds to any sampling effect, even in non-periodic situation

Idea and stability analysis initiated in

[Fridman-Seuret-Richard Automatica 2004]

Then, improved in:

[Fridman Automatica 2010]

[Seuret Automatica 2012]

[Karafyllis, Krstić IEEE TAC 2012]

[Mazenc, Malisoff, Dinh Automatica 2013]

See

https://scholar.google.fr/scholar?hl=fr&as_sdt=0%2C5

&q=input+delay+approach&btnG=

x(t

k

) = x(t -[t - t

k

] ) = x(t - h(t) )

(33)

Thus… another statement of the packet loss problem

maximum nb of successively lost packets  h

max

• piecewise-continuous delay with

delay

1 lost packet

2 lost packets  

0 ≤ h(t) ≤ h max

h ( t ) ≤ 1

dt d

Brief introduction to time-delay systems

Yes! It corresponds to any sampling effect, even in non-periodic situation

(34)

Time-varying sampling: any consequence?

[Zhang, Branicky, Phillips. - IEEE Ctrl.Syst.Mag. 2001]

Brief introduction to time-delay systems

(35)

Brief introduction to time-delay systems

[Gu, Kharitonov, Chen - Birkhauser 2003]

Time-varying sampling: any consequence?

(36)

Stability of TDS in the linear time-invariant case

Instable (et « dégénéré »)

Exemple 1:

Exemple 2:

Re(s) Im(s)

Brief introduction to time-delay systems

(37)

Instable (et « dégénéré »)

Exemple 1:

Exemple 2:

Re(s) Im(s)

Stability of TDS in the linear time-invariant case

Brief introduction to time-delay systems

(38)

Instable (and « degenerate »)

+ Exemple 1:

Exemple 2:

+

Stability of TDS in the linear time-invariant case

Brief introduction to time-delay systems

(39)

Méthode de Walton et Marshall (1987)

Extrait de Borne, Dauphin, Richard, Rotella, Zambettakis

Analyse et régulation des processus industriels - Régulation continue. 495 pages, Edt. Technip 1993

tcroissant stabilise 𝑞(𝜔2)

𝜔2

𝜔𝑗2

(40)
(41)

est instable ;

(42)
(43)

Stability : 1rst Lyapunov’s method

« small movements approximation »

(44)

Stability: case of small delays

« small delays approximation »

Références

Documents relatifs

Thin films of titanium oxynitride were successfully prepared by dc reactive magnetron sputtering using a titanium metallic target, argon, nitrogen and water vapour as reactive

Perimeter gating control scheme using the Nonlinear Model Predictive Controller (NMPC) is developed to track the optimal green routing coefficient which will indirectly track

We have proposed and simulated with TACITE some efficient schemes to prevent severe slugging: riser pressure control, installation of a pump (at the riser base or topside),

Using this new class, a universal homogeneous arbitrary HOSM controller is developed and it is shown that the homogeneity degree can be manipulated to obtain additional advantages

Robust Nonlinear Model Predictive Control based on Constrained Saddle Point Op- timization : Stability Analysis and Application to Type 1 Diabetes.M. THESE

In this work, we consider the application of the Particle Swarm Optimization algorithm to the NMPC optimisation problem model applied for the quadcopter tracking trajectory

In this work, we consider the application of the Particle Swarm Optimization algorithm to the NMPC optimisation problem model applied for the quadcopter tracking trajectory

Moreover, every time a new Agent, say agent N , dies (i.e., an in-progress application terminates), the Reduced Joint Action Space is updated by eliminating the actions involving