Iterative System Identification & Control Design: an ...oodgeroo.ucsd.edu/~bob/approx/Approximate_Spring_2012/Lecture... · Identification & Control Design: an approach to ... quantum-mechanical

Post on 06-May-2018

214 Views

Category:

Documents

1 Downloads

Preview:

Click to see full reader

Transcript

Iterative System Identification & Control

Design:an approach to Adaptive

Control

Bob BitmeadMechanical & Aerospace Engineering

University of California, San Diego

MAE283B Approximate System Identification & ControlTuesday, April 3, 2012

Iterative System Identification & Control

Design:an approach to Adaptive

Control

Bob BitmeadMechanical & Aerospace Engineering

University of California, San Diego

MAE283B Approximate System Identification & Control

?

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big Picture

2Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big Picture

2Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big Picture

2Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big PictureThe interaction between modeling and model-based control design

when the objective is to achieve good feedback controland the model is derived from experimental data plus physics

premise: all models are necessarily approximatethe approximation properties are central

2Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big PictureThe interaction between modeling and model-based control design

when the objective is to achieve good feedback controland the model is derived from experimental data plus physics

premise: all models are necessarily approximatethe approximation properties are central

2

Aha!

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big PictureThe interaction between modeling and model-based control design

when the objective is to achieve good feedback controland the model is derived from experimental data plus physics

premise: all models are necessarily approximatethe approximation properties are central

2

Aha!The model fit and quality of fit affects the achieved controlThe presence of the feedback controller affects the model fit

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big PictureThe interaction between modeling and model-based control design

when the objective is to achieve good feedback controland the model is derived from experimental data plus physics

premise: all models are necessarily approximatethe approximation properties are central

2

Aha!The model fit and quality of fit affects the achieved controlThe presence of the feedback controller affects the model fit

This course is about formulating the connections

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big PictureThe interaction between modeling and model-based control design

when the objective is to achieve good feedback controland the model is derived from experimental data plus physics

premise: all models are necessarily approximatethe approximation properties are central

2

Aha!The model fit and quality of fit affects the achieved controlThe presence of the feedback controller affects the model fit

This course is about formulating the connectionsdo this so as to gain a synergism

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big PictureThe interaction between modeling and model-based control design

when the objective is to achieve good feedback controland the model is derived from experimental data plus physics

premise: all models are necessarily approximatethe approximation properties are central

2

Aha!The model fit and quality of fit affects the achieved controlThe presence of the feedback controller affects the model fit

This course is about formulating the connectionsdo this so as to gain a synergism

better models ⇒ better control ⇒ better models ⇒ ...

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

The Big PictureThe interaction between modeling and model-based control design

when the objective is to achieve good feedback controland the model is derived from experimental data plus physics

premise: all models are necessarily approximatethe approximation properties are central

2

Aha!The model fit and quality of fit affects the achieved controlThe presence of the feedback controller affects the model fit

This course is about formulating the connectionsdo this so as to gain a synergism

better models ⇒ better control ⇒ better models ⇒ ...

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximation

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantity

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative process

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

Deductive Physics, Inductive System Identification

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

Deductive Physics, Inductive System IdentificationRinse and repeat

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

Deductive Physics, Inductive System IdentificationRinse and repeat

Modeling has a purpose

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

Deductive Physics, Inductive System IdentificationRinse and repeat

Modeling has a purposePrediction, simulation, control

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

Deductive Physics, Inductive System IdentificationRinse and repeat

Modeling has a purposePrediction, simulation, control

Differing purposes make differing demands of model quality

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

Deductive Physics, Inductive System IdentificationRinse and repeat

Modeling has a purposePrediction, simulation, control

Differing purposes make differing demands of model qualityModeling from data depends on the experiment conducted

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

Deductive Physics, Inductive System IdentificationRinse and repeat

Modeling has a purposePrediction, simulation, control

Differing purposes make differing demands of model qualityModeling from data depends on the experiment conducted

including the presence of a feedback controller

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Modeling

Modeling involves approximationA general-relativistic, quantum-mechanical partial differential equation model for an ore crusher

Model approximation is a fungible but constrained quantityParsimony is needed

Modeling is an iterative processModel development

Deductive Physics, Inductive System IdentificationRinse and repeat

Modeling has a purposePrediction, simulation, control

Differing purposes make differing demands of model qualityModeling from data depends on the experiment conducted

including the presence of a feedback controller

3Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Control

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Control

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Control

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

Stabilization

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference tracking

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

the feedback amplifier; Black, Nyquist, Bode

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

the feedback amplifier; Black, Nyquist, BodeAdvanced Control is Model-based

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

the feedback amplifier; Black, Nyquist, BodeAdvanced Control is Model-based

Model approximation is an in-built issue

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

the feedback amplifier; Black, Nyquist, BodeAdvanced Control is Model-based

Model approximation is an in-built issueThe nominal model — the specific model used for design

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

the feedback amplifier; Black, Nyquist, BodeAdvanced Control is Model-based

Model approximation is an in-built issueThe nominal model — the specific model used for designThe model error — the range of possible deviation from nominal

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

the feedback amplifier; Black, Nyquist, BodeAdvanced Control is Model-based

Model approximation is an in-built issueThe nominal model — the specific model used for designThe model error — the range of possible deviation from nominal

Performance — nominal model behavior

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

the feedback amplifier; Black, Nyquist, BodeAdvanced Control is Model-based

Model approximation is an in-built issueThe nominal model — the specific model used for designThe model error — the range of possible deviation from nominal

Performance — nominal model behaviorRobustness — variation due to model error

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about ControlFeedback control is used for a number of simultaneous purposes

StabilizationDisturbance rejection

Reference trackingReduction in the effect of system variability

the feedback amplifier; Black, Nyquist, BodeAdvanced Control is Model-based

Model approximation is an in-built issueThe nominal model — the specific model used for designThe model error — the range of possible deviation from nominal

Performance — nominal model behaviorRobustness — variation due to model error

4Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

5Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

5Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

5Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

5

“Adaptive Control Theory has been a colossal waste of paper”

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

5

“Adaptive Control Theory has been a colossal waste of paper”Bob Bitmead

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

5

“Adaptive Control Theory has been a colossal waste of paper”Bob Bitmead, Adaptive Control Theorist

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

Adaptive Control seeks to combine on-line model adjustment and real-time controller design based on operational data taken during closed-loop plant operation

5

“Adaptive Control Theory has been a colossal waste of paper”Bob Bitmead, Adaptive Control Theorist

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

Adaptive Control seeks to combine on-line model adjustment and real-time controller design based on operational data taken during closed-loop plant operation

5

“Adaptive Control Theory has been a colossal waste of paper”Bob Bitmead

PlantController

IdentifierControlDesign

+

disturbance

output

input

model

+reference

, Adaptive Control Theorist

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

Adaptive Control seeks to combine on-line model adjustment and real-time controller design based on operational data taken during closed-loop plant operation

5

“Adaptive Control Theory has been a colossal waste of paper”Bob Bitmead

In spite of linear modeling and linear control design, this is a fiercely nonlinear problem

PlantController

IdentifierControlDesign

+

disturbance

output

input

model

+reference

, Adaptive Control Theorist

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Messages about Adaptive Control

Adaptive Control seeks to combine on-line model adjustment and real-time controller design based on operational data taken during closed-loop plant operation

5

“Adaptive Control Theory has been a colossal waste of paper”Bob Bitmead

In spite of linear modeling and linear control design, this is a fiercely nonlinear problem

PlantController

IdentifierControlDesign

+

disturbance

output

input

model

+reference

, Adaptive Control Theorist

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

More Adaptive Control Messages

6Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

More Adaptive Control Messages

6Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

More Adaptive Control Messages

6Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

More Adaptive Control MessagesAdaptive Control involves simultaneous modeling and control design

Nonlinear nasties can appear — escapes, chaos, oscillationThis greatly upsets our linear thinking for models and controllers

Is there a way forward?

6Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

More Adaptive Control MessagesAdaptive Control involves simultaneous modeling and control design

Nonlinear nasties can appear — escapes, chaos, oscillationThis greatly upsets our linear thinking for models and controllers

Is there a way forward?

Adaptation is a learning approach to dealing with uncertaintyIt is well understood that learning is often antithetical to control

Control is designed to disguise system variationLearning is focused on exposing system variation

6Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

More Adaptive Control MessagesAdaptive Control involves simultaneous modeling and control design

Nonlinear nasties can appear — escapes, chaos, oscillationThis greatly upsets our linear thinking for models and controllers

Is there a way forward?

Adaptation is a learning approach to dealing with uncertaintyIt is well understood that learning is often antithetical to control

Control is designed to disguise system variationLearning is focused on exposing system variation

Dual Adaptive ControlStrike a balance between learning and controlling

Fel’dbaum 1960s

6Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

More Adaptive Control MessagesAdaptive Control involves simultaneous modeling and control design

Nonlinear nasties can appear — escapes, chaos, oscillationThis greatly upsets our linear thinking for models and controllers

Is there a way forward?

Adaptation is a learning approach to dealing with uncertaintyIt is well understood that learning is often antithetical to control

Control is designed to disguise system variationLearning is focused on exposing system variation

Dual Adaptive ControlStrike a balance between learning and controlling

Fel’dbaum 1960s

6Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Modeling Approximation Formulalim

N!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=1

2!

$ !

#!

%

&

&

&P (ej"

! P (ej", ")&

&

&

2

!u(#) +&

&H(ej")&

&

2

'

&

&L(ej")&

&

2

&

&

&H(ej", ")

&

&

&

2d#

7

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Modeling Approximation Formulalim

N!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=1

2!

$ !

#!

%

&

&

&P (ej"

! P (ej", ")&

&

&

2

!u(#) +&

&H(ej")&

&

2

'

&

&L(ej")&

&

2

&

&

&H(ej", ")

&

&

&

2d#

7

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Modeling Approximation Formula

It existsIt is a discrete frequency domain formula

Connected to time domain prediction errors

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=1

2!

$ !

#!

%

&

&

&P (ej"

! P (ej", ")&

&

&

2

!u(#) +&

&H(ej")&

&

2

'

&

&L(ej")&

&

2

&

&

&H(ej", ")

&

&

&

2d#

7

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Modeling Approximation Formula

It existsIt is a discrete frequency domain formula

Connected to time domain prediction errorsIt is useful

Model approximation is the subjectPlant, noise, plant model, noise model involved

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=1

2!

$ !

#!

%

&

&

&P (ej"

! P (ej", ")&

&

&

2

!u(#) +&

&H(ej")&

&

2

'

&

&L(ej")&

&

2

&

&

&H(ej", ")

&

&

&

2d#

7

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Modeling Approximation Formula

It existsIt is a discrete frequency domain formula

Connected to time domain prediction errorsIt is useful

Model approximation is the subjectPlant, noise, plant model, noise model involved

There are some free variables - design handlesInput spectrumData filter

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=1

2!

$ !

#!

%

&

&

&P (ej"

! P (ej", ")&

&

&

2

!u(#) +&

&H(ej")&

&

2

'

&

&L(ej")&

&

2

&

&

&H(ej", ")

&

&

&

2d#

7

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Modeling Approximation Formula

It existsIt is a discrete frequency domain formula

Connected to time domain prediction errorsIt is useful

Model approximation is the subjectPlant, noise, plant model, noise model involved

There are some free variables - design handlesInput spectrumData filter

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=1

2!

$ !

#!

%

&

&

&P (ej"

! P (ej", ")&

&

&

2

!u(#) +&

&H(ej")&

&

2

'

&

&L(ej")&

&

2

&

&

&H(ej", ")

&

&

&

2d#

!u(!)

7

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Modeling Approximation Formula

It existsIt is a discrete frequency domain formula

Connected to time domain prediction errorsIt is useful

Model approximation is the subjectPlant, noise, plant model, noise model involved

There are some free variables - design handlesInput spectrumData filter

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=1

2!

$ !

#!

%

&

&

&P (ej"

! P (ej", ")&

&

&

2

!u(#) +&

&H(ej")&

&

2

'

&

&L(ej")&

&

2

&

&

&H(ej", ")

&

&

&

2d#

!u(!)L(ej!)

7

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Closed-loop Modeling Formulalim

N!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=

1

2!

$ !

#!

%

&

'

&

(

)

)

)P (ej") ! P (ej", ")

)

)

)

2

|1 + P (ej")C(ej")|2)

)C(ej"))

)

2!r(#)

+

)

)

)1 + P (ej", ")C(ej")

)

)

)

2

|1 + P (ej")C(ej")|2)

)H(ej"))

)

2

*

&

+

&

,

)

)L(ej"))

)

2

)

)

)H(ej", ")

)

)

)

2d#

8Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Closed-loop Modeling Formulalim

N!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=

1

2!

$ !

#!

%

&

'

&

(

)

)

)P (ej") ! P (ej", ")

)

)

)

2

|1 + P (ej")C(ej")|2)

)C(ej"))

)

2!r(#)

+

)

)

)1 + P (ej", ")C(ej")

)

)

)

2

|1 + P (ej")C(ej")|2)

)H(ej"))

)

2

*

&

+

&

,

)

)L(ej"))

)

2

)

)

)H(ej", ")

)

)

)

2d#

8Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Closed-loop Modeling Formulalim

N!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=

1

2!

$ !

#!

%

&

'

&

(

)

)

)P (ej") ! P (ej", ")

)

)

)

2

|1 + P (ej")C(ej")|2)

)C(ej"))

)

2!r(#)

+

)

)

)1 + P (ej", ")C(ej")

)

)

)

2

|1 + P (ej")C(ej")|2)

)H(ej"))

)

2

*

&

+

&

,

)

)L(ej"))

)

2

)

)

)H(ej", ")

)

)

)

2d#

8Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Closed-loop Modeling Formula

The same story

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=

1

2!

$ !

#!

%

&

'

&

(

)

)

)P (ej") ! P (ej", ")

)

)

)

2

|1 + P (ej")C(ej")|2)

)C(ej"))

)

2!r(#)

+

)

)

)1 + P (ej", ")C(ej")

)

)

)

2

|1 + P (ej")C(ej")|2)

)H(ej"))

)

2

*

&

+

&

,

)

)L(ej"))

)

2

)

)

)H(ej", ")

)

)

)

2d#

8Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Closed-loop Modeling Formula

The same storyA bit more complicated

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=

1

2!

$ !

#!

%

&

'

&

(

)

)

)P (ej") ! P (ej", ")

)

)

)

2

|1 + P (ej")C(ej")|2)

)C(ej"))

)

2!r(#)

+

)

)

)1 + P (ej", ")C(ej")

)

)

)

2

|1 + P (ej")C(ej")|2)

)H(ej"))

)

2

*

&

+

&

,

)

)L(ej"))

)

2

)

)

)H(ej", ")

)

)

)

2d#

8Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Closed-loop Modeling Formula

The same storyA bit more complicated

But not much more

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=

1

2!

$ !

#!

%

&

'

&

(

)

)

)P (ej") ! P (ej", ")

)

)

)

2

|1 + P (ej")C(ej")|2)

)C(ej"))

)

2!r(#)

+

)

)

)1 + P (ej", ")C(ej")

)

)

)

2

|1 + P (ej")C(ej")|2)

)H(ej"))

)

2

*

&

+

&

,

)

)L(ej"))

)

2

)

)

)H(ej", ")

)

)

)

2d#

8Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Closed-loop Modeling Formula

The same storyA bit more complicated

But not much moreThe controller appears in the picture explicitly

limN!"

1

N

N!

k=1

"

L(z)[yk ! yk|k#1]#2

=

1

2!

$ !

#!

%

&

'

&

(

)

)

)P (ej") ! P (ej", ")

)

)

)

2

|1 + P (ej")C(ej")|2)

)C(ej"))

)

2!r(#)

+

)

)

)1 + P (ej", ")C(ej")

)

)

)

2

|1 + P (ej")C(ej")|2)

)H(ej"))

)

2

*

&

+

&

,

)

)L(ej"))

)

2

)

)

)H(ej", ")

)

)

)

2d#

8Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Robust Control Formulae!

!

!

!

!

P (ej!) ! P (ej!, !)

P (ej!, !)"

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)

!

!

!

!

!

< 1

!

!

!

!

!

H(ej!) !P (ej!) " P (ej!, !)

P (ej!, !)!

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)!

1

1 + C(ej!)P (ej!)

!

!

!

!

!

< "

9Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Robust Control Formulae!

!

!

!

!

P (ej!) ! P (ej!, !)

P (ej!, !)"

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)

!

!

!

!

!

< 1

!

!

!

!

!

H(ej!) !P (ej!) " P (ej!, !)

P (ej!, !)!

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)!

1

1 + C(ej!)P (ej!)

!

!

!

!

!

< "

9Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Robust Control Formulae

Formulae for robust stability and performance exist

!

!

!

!

!

P (ej!) ! P (ej!, !)

P (ej!, !)"

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)

!

!

!

!

!

< 1

!

!

!

!

!

H(ej!) !P (ej!) " P (ej!, !)

P (ej!, !)!

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)!

1

1 + C(ej!)P (ej!)

!

!

!

!

!

< "

9Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Robust Control Formulae

Formulae for robust stability and performance existThey are in the discrete frequency domain

!

!

!

!

!

P (ej!) ! P (ej!, !)

P (ej!, !)"

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)

!

!

!

!

!

< 1

!

!

!

!

!

H(ej!) !P (ej!) " P (ej!, !)

P (ej!, !)!

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)!

1

1 + C(ej!)P (ej!)

!

!

!

!

!

< "

9Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Robust Control Formulae

Formulae for robust stability and performance existThey are in the discrete frequency domain

They involve designed properties

!

!

!

!

!

P (ej!) ! P (ej!, !)

P (ej!, !)"

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)

!

!

!

!

!

< 1

!

!

!

!

!

H(ej!) !P (ej!) " P (ej!, !)

P (ej!, !)!

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)!

1

1 + C(ej!)P (ej!)

!

!

!

!

!

< "

9Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Robust Control Formulae

Formulae for robust stability and performance existThey are in the discrete frequency domain

They involve designed properties They involve model errors

!

!

!

!

!

P (ej!) ! P (ej!, !)

P (ej!, !)"

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)

!

!

!

!

!

< 1

!

!

!

!

!

H(ej!) !P (ej!) " P (ej!, !)

P (ej!, !)!

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)!

1

1 + C(ej!)P (ej!)

!

!

!

!

!

< "

9Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Robust Control Formulae

Formulae for robust stability and performance existThey are in the discrete frequency domain

They involve designed properties They involve model errors

The controller is the design variable

!

!

!

!

!

P (ej!) ! P (ej!, !)

P (ej!, !)"

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)

!

!

!

!

!

< 1

!

!

!

!

!

H(ej!) !P (ej!) " P (ej!, !)

P (ej!, !)!

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)!

1

1 + C(ej!)P (ej!)

!

!

!

!

!

< "

9Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Robust Control Formulae

Formulae for robust stability and performance existThey are in the discrete frequency domain

They involve designed properties They involve model errors

The controller is the design variable

!

!

!

!

!

P (ej!) ! P (ej!, !)

P (ej!, !)"

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)

!

!

!

!

!

< 1

!

!

!

!

!

H(ej!) !P (ej!) " P (ej!, !)

P (ej!, !)!

C(ej!)P (ej!, !)

1 + C(ej!)P (ej!, !)!

1

1 + C(ej!)P (ej!)

!

!

!

!

!

< "

P (ej!) ! P (ej!, !)

9Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course Outline

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course Outline

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course Outline

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophy

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopter

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief review

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loop

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structure

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loop

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust control

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control design

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control designLectures 10, 11 - Industrial example — Victoria Sugar Mill iterative design

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control designLectures 10, 11 - Industrial example — Victoria Sugar Mill iterative designLectures 12, 13 - Cautious controller tuning, Controller certification

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control designLectures 10, 11 - Industrial example — Victoria Sugar Mill iterative designLectures 12, 13 - Cautious controller tuning, Controller certificationLecture 14 - Iterative versus Adaptive Control

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control designLectures 10, 11 - Industrial example — Victoria Sugar Mill iterative designLectures 12, 13 - Cautious controller tuning, Controller certificationLecture 14 - Iterative versus Adaptive ControlLecture 15 - Observability in a stochastic context

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control designLectures 10, 11 - Industrial example — Victoria Sugar Mill iterative designLectures 12, 13 - Cautious controller tuning, Controller certificationLecture 14 - Iterative versus Adaptive ControlLecture 15 - Observability in a stochastic contextLecture 16 - Adaptive control in action — TCP network congestion control

10Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control designLectures 10, 11 - Industrial example — Victoria Sugar Mill iterative designLectures 12, 13 - Cautious controller tuning, Controller certificationLecture 14 - Iterative versus Adaptive ControlLecture 15 - Observability in a stochastic contextLecture 16 - Adaptive control in action — TCP network congestion control

10

Modeling

mostly

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control designLectures 10, 11 - Industrial example — Victoria Sugar Mill iterative designLectures 12, 13 - Cautious controller tuning, Controller certificationLecture 14 - Iterative versus Adaptive ControlLecture 15 - Observability in a stochastic contextLecture 16 - Adaptive control in action — TCP network congestion control

10

Modeling

mostly

Modeling

and control

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

Course OutlineLecture 1 - Big picture, outline. You are here ➷Lecture 2 - Modeling from data, some philosophyLecture 3 - Two examples of modeling for control; jet engine, helicopterLecture 4 - Prediction error modeling, a brief reviewLecture 5 - Approximation, bias formulæ in open and closed loopLecture 6 - Influencing model fit, experiment design, filtering, model structureLecture 7 - Influencing model fit in closed loopLecture 8 - Designing controllers with approximate models, robust controlLecture 9 - Iterative approaches to modeling and control designLectures 10, 11 - Industrial example — Victoria Sugar Mill iterative designLectures 12, 13 - Cautious controller tuning, Controller certificationLecture 14 - Iterative versus Adaptive ControlLecture 15 - Observability in a stochastic contextLecture 16 - Adaptive control in action — TCP network congestion control

10

Modeling

mostly

Modeling

and control

Adaptation

Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

A quick note

11Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

A quick note

11Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

A quick note

11Tuesday, April 3, 2012

MAE 283B, Lecture 1, Slide Approximate System Identification & Control

A quick noteThe presentation in this course will be mostly focused on scalar systems; SISO or single-input/single-output

This is because the prediction-error approach to MIMO systems is not really well set out

There are central parameterization issuesTools such as subspace methods work well but the tying together of modeling and control design is still needed

The extension of the methods of iterative modeling and control design to the MIMO framework would be quite worthwhile

802.11g ⇒ 802.11n

11Tuesday, April 3, 2012

top related