lpv approaches for sampling period dependent control design: …o.sename/docs/lpv... · 2021. 1....

77
LPV approaches for sampling period dependent control design: application to AUVs Olivier Sename, Daniel Simon Master Course - January 2021 Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 1/73

Upload: others

Post on 09-Mar-2021

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

LPV approaches for sampling period dependent control design:application to AUVs

Olivier Sename, Daniel Simon

Master Course - January 2021

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 1/73

Page 2: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

1. Control/Computing Co-design overviewObjectivesNCS constraintsControl and Scheduling IntegrationComputing-aware control

2. A polytopic approach discrete-time control with varying samplingObjectiveDiscrete polytopic LPV modeling with variable samplingA polytopic method with reduced complexityLPV polytopic control design method

3. An LFT/LPV approachIntroduction and ObjectiveA new LFT Formulation for varying sampling systemsH∞/LFT control designExtension to LPV systems

4. Application to AUVsLTI modelsLPV modelsAUV controlGeneral control schemeLPV polytopic sampling dependent controller for an LTI system (case 1)LPV/LFR varying sampling controller for a LPV model (case 2)

5. ConclusionRemarks on LPV toolsFuture works

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 2/73

Page 3: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Objectives

Context

• Control/Computation Codesign• Development of sampling dependent controllers: PhD thesis of D. Robert (2007) and E.

Roche (2011)• application to AUVs (European IST FeedNetBack project)

Main contribution : LPV control of an AUV with the sampling interval as varying parameter

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 3/73

Page 4: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Objectives

Control/computing co-design objectives

Goal : reaching a specified control performance under execution resources constraints Stankovicet al., 1999; Cervin, 2003; Sename et al., 2003

• control systems implementation over networks, using digital computers and real-timeoperating systems.

• cross coupling between closed-loop control of continuous processes and the controllerimplementation;

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 4/73

Page 5: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing NCS constraints

NCS induced computing constraints

• Non-linear and uncertain models• Limited on board computing power• Scheduled U.S. sensors• Varying time/delays• Noise, data-loss

• limited computing power→ computing duration are not negligible• variable computing power→ they are not constant• distributed information sources→ delays, data loss or corruption• many sources of induced delays: computing, preemption, networking. . .• implementation induced uncertainties (in addition to the usual process uncertainties)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 5/73

Page 6: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Control and Scheduling Integration

Overview

Traditional design : separation of concerns between control and computing :• perfectly periodic, no jitter, no data-loss -> over-constrained• perfect implementation (worst case) -> under-used resources• dedicated hardware/software -> costly components and tools

=⇒ formally provable, but costly and inflexible/unrealisticIn Progress• dynamic environments -> flexible and adaptive implementation• NCS connects asynchronous nodes -> avoid useless synchronizations• cheap off-the-shelf hardware/software components

⇒ Need for a co-design approach

• holistic view : process + algorithm + execution resources• trade off : performance/robustness/energy/costs. . .

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 6/73

Page 7: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Control and Scheduling Integration

Pioneering works 1999/2000

Eker, Hagander and Arzen, 2000

control of n identical pendulums under CPU power constraint

n LQ controllers, period hi, exec. time Ci, single CPU

Cost function J(h) =1h

∫ h

0

(xT (t) uT (t)

)Q(

x(t)u(t)

)dt

Problem: minhi

J =n

∑i=1

Ji(hi) under constraintn

∑i=1

Ci/hi ≤Ud

theoretic feedback-scheduler : h(k+1) = f (∆c(k),Ure f (k))on-line solving of Riccati and Lyapunov equationstoo costly for real-time implementation. . .

. . . although quite simple : LTI process, no delay, no jitter, no data loss, no uncertainty

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 7/73

Page 8: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Computing-aware control

New perspective and recent works: Control & Computing Co-Design

Benefits of feedback• adaptive w.r.t. operating conditions• robustness w.r.t. process modeling uncertainties• robustness w.r.t. timing uncertainties

Control under computing constraints

Uk+

Process state estimates

Y

RTOS

Scheduler

Process objectives

(QoC)

Scheduling

Scheduling Policy

WCET

(priority, period)Parameters

SAMP

Controller

ProcessProcessZOH

• reinforced specific robustness issues• delay margin, jitter margin,. . .• time-delay systems : varying, bounded. . .

• adaptive control rate• effective actuator for CPU load/bandwidth• optimal period selection, gain scheduling. . .• stability during interval switching

see... Stankovic et al., 1999; Cervin, 2003; Sename et al., 2003; Sala, 2005...

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 8/73

Page 9: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Computing-aware control

Computer load aware control

Method : Design of a controller• Which period may vary in [hmin, ...,hmax]

• Preserving of the stability during timing change• Meeting a specified performance

Possible solutions :Bank of n correctors for n periods :• Tedious when n large• a posteriori study of the transitions between controllers• Cost of the implementation

Ojective :

Synthesis of a variable sampling rate controller using a Linear Parameters Varying (LPV) method.Robert, 2007; Robert et al., 2006, 2007 & 2010

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 9/73

Page 10: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Computing-aware control

LPV/H∞ with varying intervals

ctrl2ctrl3

CPU activity hcontrolinterval

Controlled output

reference

ProcessController

sensor schedule Controller

Process

SchedulingX(k+2)

X(k) X(k+1)

t

k

U[x(k)]

U[x(k+1)]

state

control

k+1 k+2

h(k+1)h(k)

X(k+1) = A(h(k)) X(k) + B(h(k)) U(k)

• LPV where h is the varying parameter• CPU load adaptation using varying control intervals• accommodation with scheduled sensors• stability whatever h change h ∈ [hmin,hmax]

• specified performance level

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 10/73

Page 11: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Computing-aware control

IEEE TCST Robert et al., 2010

Polytopic approach:• h as the single varying parameter• Taylor’s expansion of matrix exponentials• H∞ with sampling dependent performance templates

Application to a T-inverted pendulum:

0 10 20 30 40 50 60 70 80 90−0.5

0

0.5

θ [r

ad]

Pendulum angle

0 10 20 30 40 50 60 70 80 90−5

0

5

u []

Control input

0 10 20 30 40 50 60 70 80 900

1

2

3

4x 10

−3

Time [s]

h [s

]

h

u

0 5 10 15−0.5

0

0.5

θ [r

ad]

Pendulum angle

0 5 10 15−4

−2

0

2Control input

u []

0 5 10 151

2

3x 10

−3 Control interval

h [s

]

Time [s]

7.4 7.42 7.44 7.46 7.48 7.5 7.52 7.54 7.56 7.58 7.6−0.5

0

0.5

θ [

rad]

Pendulum angle

7.4 7.42 7.44 7.46 7.48 7.5 7.52 7.54 7.56 7.58 7.6−5

0

5Control input

u [

]

7.4 7.42 7.44 7.46 7.48 7.5 7.52 7.54 7.56 7.58 7.61

2

3x 10

−3 Control interval

h [

s]

Time [s]

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 11/73

Page 12: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Control/Computing Computing-aware control

This talk - AUV Motivation

Objective : to control the altitude z of an AUVMeasurement of z by an ultrasonic sensor : induced delay in the measurement interval→ variablesampling

sensorcontrollerreference output

variable sampling

system

Sensor acting on the sampling period of the controller

• Development of both main approaches for LPV control design: the polytopic approach andthe LFR one.

• Application to AUVs: a LPV model to account for some non linearities. Design, of LPVcontrollers with two set of parameters, representing either the sampling interval or the modelnon linearities.

Roche et al., 2010, 2011

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 12/73

Page 13: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach Objective

Discrete polytopic modeling with variablesampling

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 13/73

Page 14: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach Objective

General discrete polytopic system

{xk+1 = A(θk)xk +B(θk)ukyk =C(θk)xk +D(θk)uk

• A(θk), B(θk), C(θk) et D(θk) affine in θk

• θk evolves inside a polytope Θ defined by vertices ω1, . . . ,ωN

θk ∈ Co{ω1, . . . ,ωN} (1)

• θk: convex combination of vertices ωi

θk =N

∑i=1

αi(k)ω1, αi(k)≥ 0,N

∑i=1

αi(k) = 1 (2)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 14/73

Page 15: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach Objective

From the continuous-time state space representation of the system :

G :

{x = Ax+Bu

y =Cx+Du(3)

The exact discretization of this system is :

Gd :

{xk+1 = Ad(h)xk +Bd(h)uk

yk =Cdxk +Dduk(4)

with Ad(h) ∈Rns×ns and

Ad(h) = eAh Bd(h) =∫ h

0eAτ dτB

Cd =C Dd = D(5)

Objective : schedule the discrete-time system with the variation of sampling interval

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 15/73

Page 16: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach Discrete polytopic LPV modeling with variable sampling

A polytopic model

Usual numerical method :(

Ad(h) Bd(h)0 I

)= exp(M×h) = exp

((A B0 0

)h)

(6)

With h ∈ [hmin,hmax] and defined by : h = h0 +δ

(Ad(h) Bd(h)

0 I

)= eMh0 eMδ =

(Ah0 Bh00 I

)(Aδ Bδ

0 I

)(7)

Finally the matrices of the discretized plant are :

Ad = Ah0 Aδ

Bd = Bh0 +Ah0 Bδ

(8)

Aδ and Bδ can be approximated by Taylor series expansion:

Aδ ≈ I +k

∑i=1

Ai

i!δ

i Bδ ≈k

∑i=1

Ai−1Bi!

δi (9)

The obtained model (combination 8 - 9) is affine in δ which is bounded (h ∈ [hmin,hmax])⇒ polytopic model scheduled by δ

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 16/73

Page 17: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach Discrete polytopic LPV modeling with variable sampling

Polytopic controller synthesis using LMIs

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 17/73

Page 18: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach A polytopic method with reduced complexity

Recall : LPV design methods

LPV system : discrete time system with θk ∈Θ⊂ Rn vector of varying parameters :

{xk+1 = A(θk)xk +B(θk)uk

yk =C(θk)xk +D(θk)uk

zw

K(z, δ)

P (z, δ)

u yδ

LPV method : synthesis of a controller parametrised by θk

[Apkarian et Gahinet (1995)]

• Consider the period h as the varying parameter;• Continuous linear system⇒ discrete time LPV model;• Synthesis of a h-dependent controller.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 18/73

Page 19: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach A polytopic method with reduced complexity

Recall : Polytopic model

Discrete time LPV polytopic model :{

xk+1 = A(θk)xk +B(θk)uk

yk =C(θk)xk +D(θk)uk

with :

• A(θk), B(θk), C(θk), D(θk) affine en θk ∈Θ

• Θ is a polytope with k vertices ωi

• ωi = θ mink ,θ max

k

• θk = ∑ki=1 αi(k)ωi with αi(k)≥ 0, ∑

ki=1 αi(k) = 1

(A(θk) B(θk)C(θk) D(θk)

)=

k

∑i=1

αi(k)(

Ai BiCi Di

)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 19/73

Page 20: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach A polytopic method with reduced complexity

Parametrised discretization

Continuous time system :{

x = Ax+Buy = Cx+Du

Discrete time system :{

xk+1 = Ad(h) xk +Bd(h)ukyk = Cd xk +Dd uk

withAd(h) = Ah0 Aδ

Bd(h) = Bh0 +Ah0 Bδ

h = h0 +δ

Aδ = eAδ

Affine form (via a Taylor’s expansion) in δ i, i = 1 . . .N :

Ad(h) = Ah0 Aδ = Ah0 eAδ ≈ Ah0 (N

∑i=1

Ai

i!δ

i)

Bd(h) = Bh0 +Ah0 Bδ = Bh0 +Ah0 (N

∑i=1

Ai−1Bi!

δi)

with h = h0 +δ

N parameters⇒ 2N vertices!

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 20/73

Page 21: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach A polytopic method with reduced complexity

Approximation error

Study of the order N for the series :• Error criterion JM(h) = ‖Gd(h,z)− GdN (h,z)‖∞

• Example on a stable 2nd order system

0.04

0.06

0.08

0.1

0.12 1

2

3

4

5

−300

−250

−200

−150

−100

−50

0

M

Erreur d’approximation, h0 = h

min

h (s)

Err

eur

J (d

B)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 21/73

Page 22: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach A polytopic method with reduced complexity

Polytope size reduction

Transform into a polytopic model with :

• θk = (δ ,δ 2, . . . ,δ N)T

• State matrices affine in θk

Exploiting the parameters dependency to reduce the polytope :

N=2

0 δh

δh

δ2h

0

δ2h

0 δh

δh

δ2h

0

δ2h

0 δh

δh

δ2h

0

δ2h �

��

��

ω1 ω2

ω3

Θ

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 22/73

Page 23: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach A polytopic method with reduced complexity

Polytope size reduction: general case

h = hmin +δ , 0≤ δ ≤ δmax, δmax = hmax−hmin

(0,0,0, . . . ,0)

(δmax,0,0, . . . ,0)

(δmax,δ2max,0, . . . ,0) (10)

. . .

(δmax,δ2max,δ

3max, . . . ,δ

Nmax)

N +1 vertices rather than 2N !

k influences the conservatism and complexity• of the synthesis (2k + 1) LMIs, k = N +1 or 2N

• of the real-time implementation : convex combination of k controllers

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 23/73

Page 24: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach LPV polytopic control design method

Recall : Polytopic LPV controller

Apkarian et al., 1995; Scherer and Wieland, 2004:Synthesis of a controller K which guarantees :• Internal stability of the closed loop• ‖z‖2 < γ‖w‖2

for all trajectory h in H

zw

K(z, δ)

P (z, δ)

u yδ

Polytopic controller :

K(H) :

{xk+1 = AK(H)xk +BK(H)yk

uk =CK(H)xk +DK(H)yk

with : (AK(H) BK(H)CK(H) DK(H)

)=

N

∑i=1

αi(k)(

AKi BKiCKi DKi

)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 24/73

Page 25: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach LPV polytopic control design method

Performance specification

• H∞ framework• find a controller K such internal stability is achieved• ‖z‖2 < γ‖w‖2, where γ is the H∞ attenuation level

HWoWi

K

z

y

w

u

zw

P

• closed-loop performance strongly depends on the sampling period→ make the performancespecification vary with the period

• weighting functions are sampling rate dependent Wi(h), Wo(h)• smaller the period, faster the system : ωS = α/h

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 25/73

Page 26: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach LPV polytopic control design method

H∞ design

Kθ G

We Wu

-+

θref ε

β1β2

y

Structure chosen for the control design

• Wu constant: to account for actuators limitations• We: weight on the tracking error

Choice of sampling dependent weighting function (1st order, continuous) :

We(s, f ) =s

Ms +ωs( f )s+ εsωs( f )

(11)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 26/73

Page 27: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach LPV polytopic control design method

discretization of We

Parameter dependent weighting function :

We(s, f ) :

{x = (a× f )x+(a× f −b× f )u

y = x+u(12)

Discretisation→ constant discrete weight :

Wed :

{xk+1 = Adxk +Bdu

y = x+u(13)

with

Ad = ea f h = ea

Bd = (a f )−1(Ad − I)b f = a−1(Ad − I)b

Simplification between h and f → discrete LTI weight

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 27/73

Page 28: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach LPV polytopic control design method

Generalized plant P(z,δ ): gathering polytopic model Gd(δ ), and weighting function (Wu and Wed )

zw

K(z, δ)

P (z, δ)

u yδ

Figure: General control configuration

Controller computation: tools derived from the bounded real lemma applied to LPV polytopicsystems

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 28/73

Page 29: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach LPV polytopic control design method

Discrete LPV synthesis

• Solving (2N + 1) LMI to synthesise k fixed gains controllers vertex controllers (off-line)• On-line convex combination of the vertex controllers giving :

K(h) = K(α1, . . .αk) =N

∑j=1

α jK j

• αi(h) polytopic coordinates (explicit recursive solution for the reduced polytope)• size of K(h) = size of K j

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 29/73

Page 30: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

A polytopic approach LPV polytopic control design method

LMIs to solve

(NR 0

0 I

)T

AiRATi −R AiRCT

1i B1iC1iRAT

i −γI +C1iRCT1i D11i

BT1i DT

11i −γI

(

NR 00 I

)

< 0, i = 1 . . .2n (14)

(NS 0

0 I

)T

ATi SAi −S AT

i SB1i CT1i

BT1iSAi −γI +BT

1iSB1i DT11i

C1i D11i −γI

(

NS 00 I

)

< 0, i = 1 . . .2n (15)(

R II S

)≥ 0 (16)

NR and NS bases of null space of (BT2 , DT

12) and (C2, D21) respectivelyTo guaranty stability and performances: same R, S and γ at each vertex of the polytopeMatrices of the controller derived from R and S

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 30/73

Page 31: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach

LFT (Linear Fractional transformation) model forvarying sampling period systems

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 31/73

Page 32: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach Introduction and Objective

Objective

• Taking the sampling interval into account• Formulation of a discretized model

Tool : Coming from Robust control theoryFirst step : LFT transformation of the system :

N(z)

∆(k)

yu

y∆u∆

System under LFT representation

∆ : "uncertainties" depending on the sampling interval hAim : gain scheduling design w.r.t the sampling interval

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 32/73

Page 33: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach Introduction and Objective

From the continuous-time state space representation of the system :

G :

{x = Ax+Bu

y =Cx+Du(17)

The exact discretization of this system is :

Gd :

{xk+1 = Ad(h)xk +Bd(h)uk

yk =Cdxk +Dduk(18)

(Ad(h) Bd(h)

0 I

)=

(Ah0 Bh00 I

)(Aδ Bδ

0 I

)(19)

Finally the matrices of the discretized plant are :

Ad = Ah0 Aδ

Bd = Bh0 +Ah0 Bδ

(20)

with Aδ and Bδ obtained by Taylor series expansion :

Aδ ≈ I +k

∑i=1

Ai

i!δ

i Bδ ≈k

∑i=1

Ai−1Bi!

δi (21)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 33/73

Page 34: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach A new LFT Formulation for varying sampling systems

A new LFT Formulation for varying sampling systems

First, an LPV/LFT model of a system, considering the sampling period as varying parameter ispresented, as in [Robert,2007].Objective: transform the system under the LFT form where ∆ represents the uncertaintiesdepending on the sampling period h.

N(z)

∆(k)

yu

y∆u∆

Figure: System under LFT representation

Variation w.r.t the nominal system→ 2 unstructured uncertainties :

Ad(h)−Ah0 = Ah0 (Aδ − I)︸ ︷︷ ︸

∆1

, Bd(h)−Bh0 = Ah0 Bδ = ∆2 (22)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 34/73

Page 35: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach A new LFT Formulation for varying sampling systems

The matrices of the discrete time system (38) are rewritten as :

xk+1 = [Ah0 +Ah0 (Aδ − I︸ ︷︷ ︸∆1

)]xk +[Bh0 +Ah0 Bδ︸ ︷︷ ︸∆2

]u

y =Cdxk +Ddu

The system is transformed into the form described in Figure 2 by defining :

u∆ = [u∆1 , u∆2 ]T , y∆ = [y∆1 , y∆2 ]

T and ∆ =

(∆1 00 ∆2

)

Finally, the LFT form of the model with unstructured uncertainties (with no approximation) is asfollows :

xk+1 = A xk +B

(u∆

u

)

(y∆

y

)= C xk +D

(u∆

u

) ∆ =

(∆1 00 ∆2

)(23)

A = Ah0 B =(Ah0 I Bh0

),C =

I0

Cd

D =

0 0 00 0 I0 0 Dd

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 35/73

Page 36: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach A new LFT Formulation for varying sampling systems

The LPV/LFT model of the system is then given by the upper LFT interconnection

Tu,y = Fu(P(z),∆) = P22 +P21∆(I−P11∆)−1P12 (24)

This approach considers ∆ as a full block diagonal with full matrices ∆1 and ∆2. However matrices∆1 and ∆2 depend on a single parameter δ , so this uncertainty definition may be too conservative.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 36/73

Page 37: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach A new LFT Formulation for varying sampling systems

Structured uncertainty case

In order to reduce the conservatism, the Taylor series expansion of order k for matrices Aδ and Bδ

is used to obtain a structured uncertainty matrix.

∆1 ≈ Aδ − I =k

∑i=1

Ai

i!δ

i

≈ Aδ (I +Aδ

2(I +

3(I + . . .)) (25)

∆2 ≈ Ah0 Bδ = Ah0

k

∑i=1

Ai−1Bi!

δi

≈ Ah0 (I +Aδ

2(I +

3(I + . . .)B δ (26)

Polynomial dependence on δ → structured uncertainties (reduce the conservatism)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 37/73

Page 38: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach A new LFT Formulation for varying sampling systems

Structured uncertainty case

With this structure the LFT representation of the system becomes :

xk+1 = A xk +B

(u∆

u

)

(y∆

y

)= C xk +D

(u∆

u

) , ∆ = δ I2×k×ns ,A = Ah0

with B =(B∆ Bh0

), B∆ =

(AaA 0ns∗[ns∗(k−1)] Aa 0ns∗[ns∗(k−1)]

)

C =

(C∆Cd

), C∆ =

(C1C2

)withC1 =

Ins...

.

.

.Ins

k times ; C2 =

0ns...

.

.

.0ns

k times

D =

(D D20 Dd

), D =

(D 00 D

)with D =

0nsA2 . . . 0ns

.

.

.. . .

. . ....

.

.

.. . .

. . . Ak

0ns . . . . . . 0ns

D2 =

0B

.

.

.B

This allows to express the parameter block ∆ as a direct dependence on the varying samplingperiod δ .

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 38/73

Page 39: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach A new LFT Formulation for varying sampling systems

A new LFT representation

∆ = δ I2×k×ns

A = Ah0 B =(B1 B2 Bh0

)(27)

C =

C1C2Cd

D =

D 0 00 D D2u0 0 Dd

(28)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 39/73

Page 40: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach H∞ /LFT control design

H∞/LFT control design

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 40/73

Page 41: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach H∞ /LFT control design

H∞ / LFT control design

Objective : LPV controller function of the set of parameter ∆ that ensures internal stability and‖Tzw‖∞

≤ γ∞, withTzw = Fl(K,∆) := K11 +K12K(I−K22∆)−1K21 (29)

The LFT control scheme is presented in Figure 41, where the dependency on the parameters ∆ ofthe generalized plant P and of the controller K is described in an LFT form.

z∆w∆

z

y

w

u

z∆ w∆

∆(k)

P (z)

K(z)

∆(k)

LPV process

LPV controller

∆(t)zw

∆(k) 00 ∆(k)

P (z)

K(z)

P ′(z)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 41/73

Page 42: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach H∞ /LFT control design

Problem statement

Kθ G

We Wu

-+

θref ε

β1β2

y

Structure chosen for the control design

• Wu = I2 (normalized)• We : weight on the tracking error

Choice of sampling dependent weighting function (1st order, continuous) :

We(s, f ) =s

Ms +ωs( f )s+ εsωs( f )

(30)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 42/73

Page 43: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach H∞ /LFT control design

discretization of We

Parameter dependent weighting function :

We(s, f ) :

{x = (a× f )x+(a× f −b× f )u

y = x+u(31)

Discretisation→ constant discrete weight :

Wed :

{xk+1 = Adxk +Bdu

y = x+u(32)

with

Ad = ea f h = ea

Bd = (a f )−1(Ad − I)b f = a−1(Ad − I)b

zw

∆(k) 00 ∆(k)

P ′(z)

K(z)

Wed -WedGd

0 Wu

I Gd

Simplification between h and f → LTI system

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 43/73

Page 44: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach H∞ /LFT control design

discretization of We

Parameter dependent weighting function :

We(s, f ) :

{x = (a× f )x+(a× f −b× f )u

y = x+u(31)

Discretisation→ constant discrete weight :

Wed :

{xk+1 = Adxk +Bdu

y = x+u(32)

with

Ad = ea f h = ea

Bd = (a f )−1(Ad − I)b f = a−1(Ad − I)b

zw

∆(k) 00 ∆(k)

P ′(z)

K(z)

Wed -WedGd

0 Wu

I Gd

Simplification between h and f → LTI system

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 43/73

Page 45: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach H∞ /LFT control design

H∞ / LFT control design

The problem is now to find a discrete controller K for which the system described in Figure 41 isstable and respect some specifications defined by weighting functions set in the control loop(H∞ design).The methodology used here for the discrete-time controller synthesis is based on theH∞ framework, by the resolution of LMIs (derived from the bounded real Lemma), as described inApkarian and Gahinet, 1995 and Gauthier, 2007.To reduce the conservatism due to the presence of uncertainties, the problem is reformulatedusing some scaling variables L∆.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 44/73

Page 46: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach H∞ /LFT control design

H∞ / LFT control design

The solution of this problem relies on the bounded real Lemma, as follow :Lemma Consider a parameter structure Θ, the associated scaling setLΘ = {L > 0 : L∆ = ∆L,∀∆ ∈Θ}, and a discrete-time square transfer function T (z) of realizationT (z) = Dcl +Ccl (zI−Acl)

−1 Bcl , then the following statements are equivalent.i) Acl is stable and there exists L ∈ LΘ such that ||L1/2

(Dcl +Ccl (zI−Acl)

−1 Bcl

)L−1/2||∞ < 1.

ii) There exist positive definite solutions Xcl and L ∈ LΘ to the matrix inequality

−X−1cl Acl Bcl 0

ATcl −Xcl 0 CT

clBT

cl 0 −L DTcl

0 Ccl Dcl −L−1

< 0 (33)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 45/73

Page 47: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach H∞ /LFT control design

H∞ / LFT control design

Theorem

Consider a discrete-time LPV/LFT plant P∆(z). Let Nr and Ns denote bases of null spaces of(BT

2 ,DTθ2,D

T12,0) and (C2,D2θ ,D21,0) . The gain-scheduled H∞ LFT control problem is solvable if

and only if there exist pairs of symmetric matrices (R,S) ∈ Rna×na , (L3,J3) ∈ Rnθ×nθ and γ > 0 s.t

N Tr

ARAT −R+Bθ J3BTθ

?

Cθ RAT +Dθθ J3BTθ

Cθ RCTθ+Dθθ J3DT

θθ− J3

C1RAT +D1θ J3BTθ

C1RCTθ+D1θ J3DT

θθ

BT1 DT

θ1

? ?? ?

C1RCT1 +D1θ J3DT

1θ− γI ?

DT11 −γI

Nr < 0

Ns

AT SA−S+CTθ

L3Cθ ?

BTθ

SA+DTθθ

L3Cθ BTθ

SBθ +DTθθ

L3Dθθ −L3BT

1 SA+DTθ1L3Cθ BT

1 SBθ +DTθ1L3Dθθ

C1 D1θ

? ?? ?

BT1 SB1 +DT

θ1L3Dθ1 − γI ?

D11 −γI

Ns < 0

(R II S

)> 0

L3Θ = ΘL3, J3Θ = ΘJ3,(

L3 II J3

)> 0

This LMI is then solved using Yalmip interface (Loftberg, 2004) and Sedumi solver (SEDUMI,1999).

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 46/73

Page 48: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach Extension to LPV systems

Extension to LPV models

Σ(ρ(.)) :

xzy

=

A(ρ(.)) B1(ρ(.)) B2(ρ(.))C1(ρ(.)) D11(ρ(.)) D12(ρ(.))C2(ρ(.)) D21(ρ(.)) D22(ρ(.))

xwu

(34)

ρ(.) is a varying parameter vector that takes values in the parameter space Pρ such that:

Pρ :={

ρ(.) :=[

ρ1(.) . . . ρn(.)]T ∈Rn

ρi(.) ∈[

ρi

ρ i]∀i = 1, . . . ,n

}

where n is the number of varying parameters. Pρ is a convex set.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 47/73

Page 49: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach Extension to LPV systems

LPV to LFR model

z∆w∆

zw P

u y

Figure: System under LFT form

The equation of the system under LFR representation is asfollows:

xz∆

zy

=

A B∆ B1 B2C∆ D∆∆ D∆1 D∆2C1 D1∆ D11 D12C2 D2∆ D21 D22

xw∆

wu

(35)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 48/73

Page 50: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach Extension to LPV systems

Parameterized discretisation

The LFR model in figure 3 can be discretised, assuming that the uncertain input/outputs (w∆ andz∆) are sampled and hold at the sampling frequency. The previous methodology proposed canthen be applied in that context to get a sampling dependent LFR discrete-time model.Let us consider the continuous LTI part of the LFR given in equation 35, and define:

A = A B = [B∆ B1 B2]

C =

C∆

C1C2

D =

D∆∆ D∆1 D∆2D1∆ D11 D12D2∆ D21 D22

(36)

The inputs and outputs are gathered in a single vector as:

u =

w∆

wu

and y =

z∆

zy

(37)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 49/73

Page 51: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach Extension to LPV systems

Parameterized discretisation

The exact discretisation of the LTI system (A,B,C,D) is described in equation (38)

xk+1 = Adxk +Bd uk

yk =Cdxk +Dd uk(38)

with:

Ad = eAh Bd =∫ h

0eAτ dτB

Cd = C Dd = D(39)

Matrices Ad and Bd depend both on the constant part h0 and on the varying part δ of the samplingperiod:

Ad = Ah0 Aδ (40)

Bd = Bh0 +Ah0 Bδ (41)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 50/73

Page 52: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach Extension to LPV systems

Following the previous method with Taylor expansion ...

Pd :

xk+1 = A xk +B

w∆

u

z∆

y

= C xk +D

u∆

u

(42)

with:

Θ =

[∆ 00 δ I2×k×ns

](43)

A = Ah0 B =(B1 B2 Bh0

)(44)

C =

C1C2Cd

D =

D 0 00 D D2u0 0 Dd

(45)

where k is the order of the Taylor serie expansion in δ and ns is the number of states.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 51/73

Page 53: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

An LFT/LPV approach Extension to LPV systems

z∆w∆

z

y

w

u

z∆ w∆

LPV process

LPV controller

∆ 00 δ2kns

Pd

K

∆ 00 δ2kns

Wi Wo

Figure: LFR control structure

The dependence between the discrete-time LTIplant Pd , the controller K, the parameter block(including the system and sampling parameters)and the weighting functions Wi and Wo is pre-sented in figure 4.

Finally, the LFR controller is computed using the previous Theorem.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 52/73

Page 54: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs

AUV models

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 53/73

Page 55: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LTI models

Description of the AUV

c© Copyright Ifremer

YX

Z

φ

ψ

θ

propellerone horizontal fin : δ

2 inclined fins : β2 and β′2

2 fins : β1 and β′1

12 state variables :• 6 positions : η = [x y z φ θ ψ]T

• 6 speeds : ν = [u v w p q r]T

6 control inputs : β1, β ′1, δ , β2, β ′2 and Qc

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 54/73

Page 56: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LTI models

Nonlinear model

Dynamical equations representing the physical system :

Mν = G(ν)ν +D(ν)ν +Γg +Γu (46)

η = Jc(η2)ν (47)

See Fossen, 1994)

• M the inertial matrix• G(ν) Coriolis and centrifugal forces• D(ν) hydrodynamics damping coefficients• Γg gravity effort and hydrostatic forces• Γu forces and moments due to vehicle actuators

• Jc(η2) change of referential

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 55/73

Page 57: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LTI models

Linearized and reduced model

Tangential linearization (around equilibrium point : νe = [ueq, 0, 0, 0, 0, 0]T because constantcruising speed needed for the cartography) yields to :

{x = Ax(t)+Bu(t)y =Cx(t)+Du(t) (48)

Motion in the vertical plane⇒ model reduction :• 4 state variables [z, w, θ , q]T

• 2 control inputs [β1, β2]T (pair of fins actuated together : β1 = β ′1 and β2 = β ′2)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 56/73

Page 58: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV models

Towards a pitch oriented LPV model

Objective: keep θ as a varying parameter.Considered NL model with 2 state variables (the pitch angle θ an d velocity q):

θ =cos(φ)q− sin(φ)r

Mq =− p× r(Ix− Iz)−m[Zg(q×w− r× v)]− (Zqm−Z f µV )gsin(θ)

− (Xqm−X f µV )gcos(θ)cos(φ)+Mwqw|q|+Mqqq|q|+Ff ins (49)

where M interia matrix, m AUV mass, V volume and µ mass density. The other parameters(Ix, Iz,Zg,Zq,Z f ,Xq,X f ,Mwq) appear in the dynamical and hydrodynamical functions.Ff ins : forces and moments due to the fins.Tangent linearization around Xeq = [0 ueq 0 0 0 0 0 0 θeq 0 0 0]:

˙θ = q

M ˙q = [−(Zgm−Z f µV )gcos(θeq)+(Xgm−X f µV )gsin(θeq)]θ +Ff inseq

where θ and q are the variations of θ and q.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 57/73

Page 59: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV models

A LFR model

An LFR model of the vehicle described in the form:

z∆w∆

zkwk P (z)

∆(ρ)

uk yk

The ∆ block contains the varying part of the model, which depends on the linearization point (θeq).The LFR form is defined by:

xk+1 = A xk +[

B∆ B1](u∆

u

)

(y∆

y

)=

[C∆

C1

]xk +

[D∆ 00 D1

](u∆

u

) (50)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 58/73

Page 60: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV models

A LFR model

with

A =

[0 10 0

], B∆ =

[0 0

−(Zqm−Z f ρV )g (Xqm−X f ρV )g

]

B1 =

[0

Ff ins

], C∆ =

[1 10 0

], C1 =

[1 00 1

]

D∆ =

[0 00 0

], D1 =

[0 00 0

](51)

z∆ =

[∆θ

∆θ

]; u∆ = ∆z∆; ∆ =

[ρ1 00 ρ2

](52)

with: {ρ1 = cos(θeq)ρ2 = sin(θeq)

(53)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 59/73

Page 61: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV models

A LFR model

Following the previous described method, an LFR discrete-time model can be obtained,depending on ρ1, ρ2, and on the sampling period h as well:

z

y

w

u

δIkns 0 0 00 δIkns

0 00 0 ρ1 00 0 0 ρ2

P

wΘ =

wδ1wδ2wρ1wρ2

zδ1zδ2zρ1zρ2

= zΘ

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 60/73

Page 62: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs AUV control

AUV control

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 61/73

Page 63: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs General control scheme

Simulation scenario

Ku

KθKz(h)θref

zref β1, β2∑

NL

z

u

θ+

+

+

uref Qc

Figure: Simulation scheme

• Complete non-linear model of the AUV• Follow the sea bottom at constant speed (Ku is an LTI discrete-time H∞ controller with

h = 0.1 sec.)• Sinusoïdal variation of the sampling interval for the altitude ∈ [0.05, 0.3]s• Altitude reference: filtered step of 5m

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 62/73

Page 64: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs General control scheme

Depth cascade varying sampling control structure

Decoupling of the altitude and the pitch angle

∑NLKθKz(h)

θrefzref β1, β2

θ

z

Figure: Depth cascade control structure

Synthesis of 2 controllers for the vertical axis:

Case 1: Pitch angle controller Kθ : based on the reduced linearised model of the AUV

Altitude controller Kz(h): based on a geometrical model, and scheduled by h

Case 2: Pitch angle controller Kθ (h,ρ): based on the LPV/LFR pitch model, withh ∈ [0.005, 0.03] sec.

Altitude controller Kz(h): LFR controller scheduled by h, with h ∈ [0.05, 0.3] sec.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 63/73

Page 65: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV polytopic sampling dependent controller for an LTI system (case 1)

Altitude control: an simple LPV model

Z

l

θ

~u

Figure: Relation z θ

z = l sinθ

u = l(54)

z ' uθ

Gz = zθ= 1

p(55)

Inner loop approximated by an integratorDiscrete LPV polytopic model derived from this geometrical relations:• Discretisation leads to: xk+1 = xk +h.uk with h ∈ [hmin,hmax]

• By considering h = h0 +δ ⇒ xk+1 = xk +(h0 +δ ).uk

⇒ LPV model in δ (easy to put in polytopic or LFR forms)

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 64/73

Page 66: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV polytopic sampling dependent controller for an LTI system (case 1)

Altitude control: LPV polytopic synthesis (case 1)

Application of the methodology previously described

K G

We(h) Wu

-+

r ε y

Figure: Structure chosen for the control design

• We(s,h) =s

Ms +ω×hs+ω×h×ε

• Wu = 0.1

• sampling period h ∈ [hmin, hmax] = [0.05, 0.3]s, (h0 = hmin)• as shown before We(z) is a discrete-time LTI system.• Polytopic approach

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 65/73

Page 67: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV polytopic sampling dependent controller for an LTI system (case 1)

Polytopic Sensitivity functions (case 1)

10-4

10-3

10-2

10-1

100

101

102

-140

-120

-100

-80

-60

-40

-20

0

20

Mag

nitu

de (

dB)

Bode Diagram

Frequency (rad/sec)

weight We

h=0.03

h=0.05s

Figure: S sensitivity function

10-8

10-6

10-4

10-2

100

102

-30

-20

-10

0

10

20

30

40

50

Mag

nitu

de (

dB)

Bode Diagram

Frequency (rad/sec)

h=0.3s

h=0.05s

Figure: Controller Bode diagram

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 66/73

Page 68: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV polytopic sampling dependent controller for an LTI system (case 1)

Simulation results (case 1)

0 50 100 150 200 250 300 350 400 450 5000.05

0.1

0.15

0.2

0.25

0.3

Sampling period

0 50 100 150 200 250 300 350 400 450 5000

0.2

0.4

0.6

0.8

1

1.2

1.4

Longitudinal speed

0 50 100 150 200 250 300 350 400 450 500-1

0

1

2

3

4

5

6

reference

altitude

Altitude

0 50 100 150 200 250 300 350 400 450 500-0.2

-0.1

0

0.1

time(s)

pitc

h an

gle

(rad

)

pitch angle

reference

Pitch angle

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 67/73

Page 69: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV/LFR varying sampling controller for a LPV model (case 2)

LFR altitude controller (case 2)

Using the previous simple LPV model, the LFR approach is considered with the followingperformance weights:• Wze (h) defined in continuous-time, and function of the sampling period:

Wze (h) =s

MSze+wSze h

s+wSze hεSze

(56)

where MSze = 2, wSze = 5 and εSze = 0.001. After discretization, the corresponding discrete-timeLTI weight is part of the generalized plant.

• Wzu = 5

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 68/73

Page 70: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV/LFR varying sampling controller for a LPV model (case 2)

LFR design for Kθ (h,ρ) (case 2)

The previous LFR model is used, that depends on the sampling δ , and on the system parameterscos(θeq) and sin(θeq).As previously the mixed sensitivity problem is considered with:• Wθu on the control inputs [β1, β2]:

Wθu =

[0.8 00 0.8

](57)

• Wθe on the tracking error θ −θre f , defined as previously:

Wθe (h) =

sMSθe

+wSθeh

s+wSθehεSθe

(58)

where MSθe= 2, wSθe

= 10, εSθe= 0.01 and h varies in [h, h] = [0.05, 0.3]

leads to γopt = 38

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 69/73

Page 71: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV/LFR varying sampling controller for a LPV model (case 2)

Simulation results (case 2)

0 50 100 150 200 250 300 350 400 450 5000.05

0.1

0.15

0.2

0.25

0.3

Sampling period0 50 100 150 200 250 300 350 400 450 500

0

0.2

0.4

0.6

0.8

1

1.2

Longitudinal speed

0 50 100 150 200 250 300 350 400 450 500-1

0

1

2

3

4

5

6

zref

et z

Altitude

0 50 100 150 200 250 300 350 400 450 500-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

Pitch angle

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 70/73

Page 72: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Application to AUVs LPV/LFR varying sampling controller for a LPV model (case 2)

0 50 100 150 200 250 300 350 400 450 500-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

Control Inputs

0 50 100 150 200 250 300 350 400 450 500-0.2

0

0.2

0.4

0.6

0.8

1

1.2

cos(θeq

)

hsin(θ

eq)

Parameters variations

• Good tracking of zre f and θre f with fast responses.• Longitudinal speed: almost always equals 1m/s with light oscillations during the altitude

change (t = 210s). The coupling between the state variables is mitigated.• The control inputs are saturated when the altitude reference is changing (at 10 and 210

secondes), but are becoming quickly normal. The remaining oscillations could be attenuatedto improve the energy efficiency.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 71/73

Page 73: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Conclusion Remarks on LPV tools

Interest of the LPV framework

• Sampling dependent discrete-time model• Use of affine and LFT models• Performance adaptation function of the sampling period (available CPU resources)• A single framework when the system is non linear and represented as a LPV model.• Brings an interesting methodology for the design of H∞ controllers dor non linear systems,

under computational constraints

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 72/73

Page 74: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Conclusion Future works

Perspectives

• Reduce the conservatism of the method to improve the problem solvability for complexsystem. Use of other multipliers for the LFT method, parameter-dependent Lyapunov functionfor the polytopic approach

• To be consider: controller with reduced complexity in order to cope with other computationalrequirements: low order, stable controller, quantification errors....

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 73/73

Page 75: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Conclusion Future works

• Apkarian, 1997 Apkarian, P. (1997). On the discretization of LMI-synthesized linear parameter-varying controllers. Automatica, 33(4):655 – 661.• Apkarian and Adams, 2008 Apkarian, P. and Adams, R. (2008). Advanced gain-scheduling techniques for uncertain systems. In IEEE Transaction on

Automatic Control, volume 6, pages 21–32.• Apkarian and Gahinet, 1994 Apkarian, P. and Gahinet, P. (1994). A linear matrix inequality approach to H∞ control. International journal of robust and

nonlinear control, 4:421–448.• Apkarian and Gahinet, 1995 Apkarian, P. and Gahinet, P. (1995). A convex characterisation of gain-scheduled H∞ controllers. IEEE Transaction on

Automatic Control, 40:853 – 864.• Apkarian et al., 1995 Apkarian, P., Gahinet, P., and Becker, G. (1995). Self-scheduled H∞ control of linear parameter-varying systems : a design

example. Automatica, 31(9):1251–1261.• Åström and Wittenmark, 1997 Åström, K. J. and Wittenmark, B. (1997). Computer-Controlled Systems. Information and systems sciences series.

Prentice Hall, New Jersey, 3rd edition.• Biannic, 1996 Biannic, J. (1996). Commande robuste des systèmes à paramètres variables - Application en aéronautique. PhD thesis, SUPAERO

Toulouse.• Boyd et al., 1994 Boyd, S., El-Ghaoui, L., Feron, E., and Balakrishnan, V. (1994). Linear Matrix Inequalities in System and Control Theory. SIAM

Studies in applied mathematics.• Bruzelius, 2004 Bruzelius, F. (2004). Linear Parameter-Varying Systems : an approach to gain scheduling. PhD thesis, Department of Signals and

Systems – CHALMERS UNIVERSITY OF TECHNOLOGY.• Cervin, 2003 Cervin, A. (2003). Integrated Control and Real-Time Scheduling. PhD thesis, Department of Automatic Control, Lund Institute of

Technology, Sweden.• Cuenca et al., 2009 Cuenca, A., Garcia Gil, P. J., Årzén, K.-E., and Albertos, P. (2009). A predictor-observer for a networked control system with

time-varying delays and non-uniform sampling. In Proceedings of the European Control Conference, Budapest, Aug 2009.• Eker et al., 2000 J. Eker, P. Hagander, and K. Årzén, “A feedback scheduler for real-time controller tasks,” Control Engineering Practice, vol. 8, no. 12,

pp. 1369–1378, Dec. 2000.• Feng. and Allen, 2004 Feng., Z. and Allen, R. (2004). Reduced order H∞ control of an autonomous underwater vehicle. Control Engineering

Practice, 12:1511 –1520.• Fossen, 1994 Fossen, T. I. (1994). Guidance and Control of Ocean Vehicles. John Wiley & Sons.• Gauthier, 2007 Gauthier, C. (2007). Commande multivariable de la pression d’injection dans un moteur diesel common rail. PhD thesis, Institut

National Polytechnique de Grenoble.• Gauthier et al., 2007 Gauthier, C., Sename, O., Dugard, L., and Meissonnier, G. (2007). An H∞ linear parameter-varying (lpv) controller for a diesel

engine common rail injection system. In Proceedings of the European Control Conference, Budapest, Hungary.• Jalving and Storkersen, 1994 Jalving, B. and Storkersen, N. (1994). The control system of an autonomous underwater vehicle. In Proceedings of the

Third IEEE Conference, volume 2, pages 851 – 856.• Lofberg, 2004 Lofberg, J. (2004). YALMIP : A toolbox for modeling and optimization in matlab. In Proceedings of the CACSD Conference.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 73/73

Page 76: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Conclusion Future works

• Miyamaoto et al., 2001 Miyamaoto, S., Aoki, T., Maeda, T., Hirokawa, K., Ichikawa, T., Saitou, T., Kobayashi, H., Kobayashi, E., and Iwasaki, S. (2001).Maneuvering control system design for autonomous underwater vehicle. MTS/IEEE Conference and Exhibition, 1:482 – 489.

• Oliveira and Peres, 2006 Oliveira, R. C. and Peres, P. L. (2006). Lmi conditions for robust stability analysis based on polynomiallyparameter-dependent lyapunov functions. Systems and Control Letters, 55(1):52 – 61.

• Poussot-Vassal, 2008 Poussot-Vassal, C. (2008). Commande Robuste LPV Multivariable de Châssis Automobile. PhD thesis, Grenoble INP.

• Robert et al., 2006 D. Robert, O. Sename, and D. Simon, “Synthesis of a sampling period dependent controller using LPV approach,” in 5th IFACSymposium on Robust Control Design ROCOND’06, Toulouse, France, july 2006.

• Robert et al., 2007 D. Robert, O. Sename, and D. Simon, “A reduced polytopic LPV synthesis for a sampling varying controller : experimentation witha t inverted pendulum,” in European Control Conference ECC’07, Kos, Greece, July 2007.

• Robert, 2007 Robert, D. (2007). Contribution à l’interconnection commande / ordonnancement. PhD thesis, Institut National Polytechnique deGrenoble.

• Robert et al., 2010 Robert, D., Sename, O., and Simon, D. (2010). An h∞ LPV design for sampling varying controllers: experimentation with a tinverted pendulum. IEEE Transactions on Control Systems Technology, 18(3):741–749.

• Roche et al., 2010 Roche, E., Sename, O., and Simon, D. (2010). LPV / H∞ varying sampling control for autonomous underwater vehicles. InProceedings of the IFAC SSSC, Ancona, Italie.

• Roche et al., 2011 Roche, E., Sename, O., and Simon, D. (2011). A hierarchical varying sampling H∞ control of an AUV. In Proceedings of the IFACWorld Congress, Milan, Italie.

• Sala, 2005 Sala, A. (2005). Computer control under time-varying sampling period:an lmi gridding approach. Automatica, 41(12):2077–2082.

• Santos, 1995 Santos, A. S. (1995). Contribution à la conception des sous-marins autonomes : architecture des capteurs d’altitude, et commanderéférencées capteurs. PhD thesis (in french), Ecole nationale supérieure des Mines de Paris.

• Scherer et al., 1997 Scherer, C., Gahinet, P., and Chilali, M. (1997). Multiobjective output-feedback control via lmi optimization. Automatic Control,IEEE Transactions on, 42(7):896 –911.

• Scherer and Wieland, 2004 Scherer, C. and Wieland, S. (2004). Linear Matrix inequalities in Control. lecture support, DELFT University.

• Scherer, 2001 Scherer, C. W. (2001). Lpv control and full block multipliers. Automatica, 37(3):361 – 375.

• Sename et al., 2003 Sename, O., Simon, D., and Robert, D. (2003). Feedback scheduling for real-time control of systems with communication delays.In Proceedings of the 2003 IEEE Conference on Emerging Technologies and Factory Automation, pages 454–61, Lisbon, Portugal.

• Silvestre and Pascoal, 2004 Silvestre, C. and Pascoal, A. (2004). Control of the infante auv using gain scheduled static output feedback. ControlEngineering Practice, 12(12):1501 – 1509.

• Silvestre and Pascoal, 2007 Silvestre, C. and Pascoal, A. (2007). Depth control of the infante auv using gain-scheduled reduced order outputfeedback. Control Engineering Practice, 15(7):883 – 895.

• Simon et al., 2005 D. Simon, D. Robert, and O. Sename, “Robust control / scheduling co-design: application to robot control,” in Proceedings of the11th IEEE Real-Time and Embedded Technology and Applications Symposium, San Francisco, USA, March 2005.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 73/73

Page 77: LPV approaches for sampling period dependent control design: …o.sename/docs/LPV... · 2021. 1. 21. · Control/Computing Objectives Context Control/Computation Codesign Development

Conclusion Future works

• Simon et al., 2009 Simon, D., Seuret, A., Hokayem, P., Lygeros, J., and Camacho, E. (2009). FeedNetBack-D04.01- State of the art incontrol/computing co-design. Technical report.

• Skogestad and Postlethwaite, 1996 Skogestad, S. and Postlethwaite, I. (1996). Multivariable Feedback Control: analysis and design. John Wiley andSons.

• Stankovic et al., 1999 Stankovic, J. A., Lu, C., Son, S. H., and Tao, G. (1999). The case for feedback control real-time scheduling. In Proceedings ofthe 11th Euromicro Conference on RealTime Systems, pages 11–20, York, England.

• Tóth, 2010 Tóth, R. (2010). Modeling and Identification of Linear Parameter-Varying Systems. Springer Berlin / Heidelberg.

• Tóth et al., 2010 Tóth, R., Heuberger, P., and den Hof, P. V. (2010). Discretisation of linear parameter-varying state-space representations. IETControl Theory and Applications, 4(10):2082–2096.

• Tóth et al., 2009 Tóth, R., Lovera, M., Heuberger, P., and den Hof, P. V. (2009). Discretisation of linear fractional representations of lpv systems. InProceedings of the IEEE CDC, Shangai, China.

• Varrier, 2010 Varrier, S. (2010). Robust control of autonomous underwater vehicles. Master’s thesis, ENSE3 Master in Automatic Control, Systems &Information Technology, Grenoble INP, France.

• Wu, 2001 Wu, F. (2001). A generalized lpv system analysis and control synthesis framework. International Journal of Control, 74(7):745–759.

• Wu and Dong, 2006 Wu, F. and Dong, K. (2006). Gain-scheduling control of lft systems using parameter-dependent lyapunov functions. Automatica,42(1):39 – 50.

• Zanoli and Conte, 2003 Zanoli, S. M. and Conte, G. (2003). Remotely operated vehicle depth control. Control Engineering Practice, 11(4):453 – 459.

• Zhou et al., 1996 Zhou, K., Doyle, J., and Glover, K. (1996). Robust and Optimal Control. Prentice hall.

Olivier Sename, Daniel Simon () LPV AUV Master Course - January 2021 73/73