2.153 Adaptive Control Fall 2019 Lecture 1:...

23
2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy [email protected] September 4, 2019 ( [email protected]) September 4, 2019 1 / 11

Transcript of 2.153 Adaptive Control Fall 2019 Lecture 1:...

Page 1: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

2.153 Adaptive ControlFall 2019

Lecture 1: Introduction

Anuradha Annaswamy

[email protected]

September 4, 2019

( [email protected]) September 4, 2019 1 / 11

Page 2: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Adaptive Control

Adaptive Control, a thirty-year old field, is an advanced control method that isbecoming increasingly popular in various engineering applications. The ability toself-correct a controller in the presence of uncertainties using online information isits main and most compelling feature. This course will lay out the foundation ofadaptive control in continuous-time and discrete-time systems. Examples fromaerospace, propulsion, automotive, and energy systems will be used to elucidatethe underlying concepts.

( [email protected]) September 4, 2019 2 / 11

Page 3: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Definition of Adaptation

Biology: Advantageous Conformation of an Organism to Changes

Adaptive Control: The control of Uncertain Systems

Other Definitions:Truxal: An adaptive system is one designed from an adaptive viewpoint.

Zadeh: A is adaptive with respect to Sγ and Γ if it performs acceptablywell for every source in the family Sγ , γ ∈ Γ

( [email protected]) September 4, 2019 3 / 11

Page 4: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Definition of Adaptation

Biology: Advantageous Conformation of an Organism to Changes

Adaptive Control: The control of Uncertain Systems

Other Definitions:Truxal: An adaptive system is one designed from an adaptive viewpoint.

Zadeh: A is adaptive with respect to Sγ and Γ if it performs acceptablywell for every source in the family Sγ , γ ∈ Γ

( [email protected]) September 4, 2019 3 / 11

Page 5: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Definition of Adaptation

Biology: Advantageous Conformation of an Organism to Changes

Adaptive Control: The control of Uncertain Systems

Other Definitions:Truxal: An adaptive system is one designed from an adaptive viewpoint.

Zadeh: A is adaptive with respect to Sγ and Γ if it performs acceptablywell for every source in the family Sγ , γ ∈ Γ

( [email protected]) September 4, 2019 3 / 11

Page 6: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Definition of Adaptation

Biology: Advantageous Conformation of an Organism to Changes

Adaptive Control: The control of Uncertain Systems

Other Definitions:Truxal: An adaptive system is one designed from an adaptive viewpoint.

Zadeh: A is adaptive with respect to Sγ and Γ if it performs acceptablywell for every source in the family Sγ , γ ∈ Γ

( [email protected]) September 4, 2019 3 / 11

Page 7: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Definition of Adaptation (contd.)

Bellman:

Deterministic System: When the controller has complete informationabout the behavior of the inputs, and the system is completely specified.

Stochastic system: When unknown factors are present in the system whichappear mathematically as random variables with knwon distributionfunctions

Adaptive system: Even less is known about the system and the controllerhas to learn to improve its performance through the observation of theoutputs of the system as it evolves.

In this course:Adaptive Control: The control of plants with unknown parameters

( [email protected]) September 4, 2019 4 / 11

Page 8: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Definition of Adaptation (contd.)

Bellman:

Deterministic System: When the controller has complete informationabout the behavior of the inputs, and the system is completely specified.

Stochastic system: When unknown factors are present in the system whichappear mathematically as random variables with knwon distributionfunctions

Adaptive system: Even less is known about the system and the controllerhas to learn to improve its performance through the observation of theoutputs of the system as it evolves.

In this course:Adaptive Control: The control of plants with unknown parameters

( [email protected]) September 4, 2019 4 / 11

Page 9: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Control System

Open-loop Dynamic System

Feedback Control System

Performance Measures: Stability, speed, accuracy

( [email protected]) September 4, 2019 5 / 11

Page 10: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Control System

Open-loop Dynamic System Feedback Control System

Performance Measures: Stability, speed, accuracy

( [email protected]) September 4, 2019 5 / 11

Page 11: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Adaptive Control System

In the presence of uncertainties, using prior and on-line information, thecontroller adapts itself.

( [email protected]) September 4, 2019 6 / 11

Page 12: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

History

Adaptive systems

Caught the imagination in the 60’s

Several symposia - to define adaptation

First use in flight control

Hypersonics program - highly successful

Five classes of adaptive systems

Passive Adaptation

Input Signal Adaptation

System variable adaptation

System characteristic adaptation

Extremum adaptation

( [email protected]) September 4, 2019 7 / 11

Page 13: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

History

Adaptive systems

Caught the imagination in the 60’s

Several symposia - to define adaptation

First use in flight control

Hypersonics program - highly successful

Five classes of adaptive systems

Passive Adaptation

Input Signal Adaptation

System variable adaptation

System characteristic adaptation

Extremum adaptation

( [email protected]) September 4, 2019 7 / 11

Page 14: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Adaptive Control: A Parametric Framework

Nonlinear, time-varying, with unknown parameter θ

x = f(x, u, θ, t) y = h(x, u, θ, t)

Nonlinear, with unknown parameter θ

x = f(x, u, θ) y = h(x, u, θ)

Linear Time-Varying (LTV) with unknown parameter θ

x = A(θ, t)x+B(θ, t)u y = C(θ, t)x+D(θ, t)u

Linear Time-Invariant (LTI) with unknown parameter θ

x = A(θ)x+B(θ)u y = C(θ)x+D(θ)u

Adapt to the unknown parameter

( [email protected]) September 4, 2019 8 / 11

Page 15: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Adaptive Control: A Parametric Framework

Nonlinear, time-varying, with unknown parameter θ

x = f(x, u, θ, t) y = h(x, u, θ, t)

Nonlinear, with unknown parameter θ

x = f(x, u, θ) y = h(x, u, θ)

Linear Time-Varying (LTV) with unknown parameter θ

x = A(θ, t)x+B(θ, t)u y = C(θ, t)x+D(θ, t)u

Linear Time-Invariant (LTI) with unknown parameter θ

x = A(θ)x+B(θ)u y = C(θ)x+D(θ)u

Adapt to the unknown parameter

( [email protected]) September 4, 2019 8 / 11

Page 16: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Direct and Indirect Adaptive Control

θp: Plant parameter - unknown; θc: Control parameter

Indirect Adaptive Control: Estimate θp as θp. Compute θc using θp.Also known as Explicit EstimationDirect Adaptive Control: Directly estimate θc as θc. Compute the plantestimate θp using θcAlso known as Implicit Estimation

( [email protected]) September 4, 2019 9 / 11

Page 17: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

Parametric Adaptation

Parameter perturbation methods

http://flyingv.ucsd.edu/krstic/talks/ExtremumSeeking.pdf

Sensitivity methods

( [email protected]) September 4, 2019 10 / 11

Page 18: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

In this course

Many of our discussions will involve

Control designs motivated by linear design methods

Parametrizations of the control structure

Parameter adjustment rules - Adaptive laws

What to measure?What to adjust? How many parameters?How often?How do we adjust?

A Stability Framework

( [email protected]) September 4, 2019 11 / 11

Page 19: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

In this course

Many of our discussions will involve

Control designs motivated by linear design methods

Parametrizations of the control structure

Parameter adjustment rules - Adaptive laws

What to measure?What to adjust? How many parameters?How often?How do we adjust?

A Stability Framework

( [email protected]) September 4, 2019 11 / 11

Page 20: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

In this course

Many of our discussions will involve

Control designs motivated by linear design methods

Parametrizations of the control structure

Parameter adjustment rules - Adaptive laws

What to measure?What to adjust? How many parameters?How often?How do we adjust?

A Stability Framework

( [email protected]) September 4, 2019 11 / 11

Page 21: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

In this course

Many of our discussions will involve

Control designs motivated by linear design methods

Parametrizations of the control structure

Parameter adjustment rules - Adaptive laws

What to measure?What to adjust? How many parameters?How often?How do we adjust?

A Stability Framework

( [email protected]) September 4, 2019 11 / 11

Page 22: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

In this course

Many of our discussions will involve

Control designs motivated by linear design methods

Parametrizations of the control structure

Parameter adjustment rules - Adaptive laws

What to measure?What to adjust? How many parameters?How often?How do we adjust?

A Stability Framework

( [email protected]) September 4, 2019 11 / 11

Page 23: 2.153 Adaptive Control Fall 2019 Lecture 1: Introductionaaclab.mit.edu/material/lect/2_153_Lecture_01.pdf · 2.153 Adaptive Control Fall 2019 Lecture 1: Introduction Anuradha Annaswamy

In this course

Many of our discussions will involve

Control designs motivated by linear design methods

Parametrizations of the control structure

Parameter adjustment rules - Adaptive laws

What to measure?What to adjust? How many parameters?How often?How do we adjust?

A Stability Framework

( [email protected]) September 4, 2019 11 / 11