delay robustness mirkin

Upload: johnv13

Post on 13-Apr-2018

237 views

Category:

Documents


1 download

TRANSCRIPT

  • 7/26/2019 Delay Robustness Mirkin

    1/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    An Opinionated View on Delay Robustness

    Leonid Mirkin

    Faculty of Mechanical EngineeringTechnion IIT

    May 18, 2006

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    2/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Single-delay system with uncertain delay

    Consider LTI system

    x(t) =A0x(t) + A1x(t h), x() =0, [h, 0]

    or, equivalently, in thes-domain:

    sx(s) =A0x(s) + A1esh

    x(s).

    Wed like to be able to check

    whether this system isstableh [0,h]for some h > 0.

    This clearly requires that

    A1: delay-free system is stable, i.e.,A0+ A1is Hurwitz.

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    3/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Precise methods

    Methods yielding exact stability intervals:

    Nyquist criterion(Tsypkin, 1946)

    Delay-sweeping arguments

    (Cooke & Grossman, 1982; Walton & Marshall, 1987)

    Schur-Cohn criterion inspirations(J. Chen, G. Gu, & Nett, 1995)

    Common pitfalls:

    not suitable for analytic controller design

    not readily extendible to multiple-delay systems

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    4/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Precise methods

    Methods yielding exact stability intervals:

    Nyquist criterion(Tsypkin, 1946)

    Delay-sweeping arguments

    (Cooke & Grossman, 1982; Walton & Marshall, 1987)

    Schur-Cohn criterion inspirations(J. Chen, G. Gu, & Nett, 1995)

    Common pitfalls:

    not suitable for analytic controller design

    not readily extendible to multiple-delay systems

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    5/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    The quest for alternatives

    Has intensified during the last decade:

    a zillion of papers published

    dominated by Lyapunov-Krasovski (LK) methods(LMI solutions derived via state-space LK technique)

    d l f ll h fi l d k

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    6/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Outline

    Lyapunov-Krasovski methods & model transformations

    Good ol (scaled) Small Gain Theorem

    Some comparisons

    Possible refinements

    Concluding remarks

    M d l f i S ll G i Th S i R fi C l di k

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    7/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Outline

    Lyapunov-Krasovski methods & model transformations

    Good ol (scaled) Small Gain Theorem

    Some comparisons

    Possible refinements

    Concluding remarks

    M d l t f ti S ll G i Th S i R fi t C l di k

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    8/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Lyapunov-Krasovski methods

    Analysis based on

    1. constructing a Lyapunov-Krasovski functional (storage function), like

    V=x(t)P1x(t) + 2x(t)0h

    P2()x(t+ )d

    +0h

    0h

    x (t+ )P3(, )x(t+ )dd+ ,

    2. calculating its derivative along system trajectory,

    3. completing squares via approximating some cross-terms,

    4. ending up with LMI conditions guaranteeing that V < 0.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    9/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Lyapunov-Krasovski methods: transformations

    Apparent obstacle in the use of this approach is that

    equation x(t) =A0x(t) + A1x(t h)is not quite compatible

    with Lyapunov-Krasovski techniques (not LK-friendly).

    Conventional way to circumvent this obstacle is to

    transform this model to more suitable form

    by rearranging its terms.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    10/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Lyapunov-Krasovski methods: transformations

    Apparent obstacle in the use of this approach is that

    equation x(t) =A0x(t) + A1x(t h)is not quite compatible

    with Lyapunov-Krasovski techniques (not LK-friendly).

    Conventional way to circumvent this obstacle is to

    transform this model to more suitable form

    by rearranging its terms.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    11/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    The first transformation

    Rewrite

    sx =A0x + A1eshx= (A0+ A1)x A1(1 e

    sh)x

    = (A0+ A1)x A11esh

    s sx

    = (A0+ A1)x A11esh

    s (A0x + A1e

    shx)

    or in the time domain:

    x(t) = (A0+ A1)x(t) A1

    h0

    A0x(t ) + A1x(t h )

    d.

    Turns out to be more LK-friendly, yet introducesadditional dynamics:

    (s) =det

    sI A0 A1esh

    det

    sI A1

    1esh

    s

    ,

    stability of which is hard to check (source of additionalconservatism).

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    12/74

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    The first transformation

    Rewrite

    sx =A0x + A1eshx= (A0+ A1)x A1(1 e

    sh)x

    = (A0+ A1)x A11esh

    s sx

    = (A0+ A1)x A11esh

    s (A0x + A1e

    shx)

    or in the time domain:

    x(t) = (A0+ A1)x(t) A1

    h0

    A0x(t ) + A1x(t h )

    d.

    Turns out to be more LK-friendly, yet introducesadditional dynamics:

    (s) =det

    sI A0 A1esh

    det

    sI A1

    1esh

    s

    ,

    stability of which is hard to check (source of additionalconservatism).

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    13/74

    p g

    The second transformation

    Rewrite

    sx=A0x + A1eshx

    I + A1

    1esh

    s

    sx= (A0+ A1)x

    or in the time domain:

    d

    dt

    x(t) + A1 h0

    x(t )d

    = (A0+ A1)x(t).

    Turns out to be more LK-friendly, yet

    requires the stability of

    det

    I + A11esh

    s

    =0 or, equiv., of x(t) = A1

    h0

    x(t )d,

    which is hard to verify, so it might be source of additional conservatismtoo.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    14/74

    p g

    The second transformation

    Rewrite

    sx=A0x + A1eshx

    I + A1

    1esh

    s

    sx= (A0+ A1)x

    or in the time domain:

    d

    dt

    x(t) + A1 h0

    x(t )d

    = (A0+ A1)x(t).

    Turns out to be more LK-friendly, yet

    requires the stability of

    det

    I + A11esh

    s

    =0 or, equiv., of x(t) = A1

    h0

    x(t )d,

    which is hard to verify, so it might be source of additional conservatismtoo.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    15/74

    The third transformation

    Rewrite

    sx =A0x + A1eshx= (A0+ A1)x A1

    1esh

    s sx

    (in fact, this is midway toward first transformation) or in the time domain:

    x(t) = (A0+ A1)x(t) A1 h0

    x(t )d.

    Somehow is also LK-friendly, yet claimed to

    introduce additional terms to V

    i.e., leads to overdesign (might be source of additionalconservatismtoo).

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    16/74

    The third transformation

    Rewrite

    sx =A0x + A1eshx= (A0+ A1)x A1

    1esh

    s sx

    (in fact, this is midway toward first transformation) or in the time domain:

    x(t) = (A0+ A1)x(t) A1 h0

    x(t )d.

    Somehow is also LK-friendly, yet claimed to

    introduce additional terms to V

    i.e., leads to overdesign (might be source of additionalconservatismtoo).

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    17/74

    The fourth transformation

    Rewrite

    sx=A0x + A1eshx

    sx= y

    y= (A0+ A1)x A11esh

    s y

    or in the time domain asdescriptorsystem

    I 0

    0 0

    x(t)

    y(t)

    =

    0 I

    A0+ A1 I

    x(t)

    y(t)

    0

    A1

    h0

    y(t )d.

    Not surprise that it is also considered LK-friendly. Moreover, it is

    claimed to be less conservative.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    18/74

    The fourth transformation

    Rewrite

    sx=A0x + A1eshx

    sx= y

    y= (A0+ A1)x A11esh

    s y

    or in the time domain as descriptor system

    I 0

    0 0

    x(t)

    y(t)

    =

    0 I

    A0+ A1 I

    x(t)

    y(t)

    0

    A1

    h0

    y(t )d.

    Not surprise that it is also considered LK-friendly. Moreover, it is

    claimed to be less conservative.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    19/74

    What bothers me

    conservatism sources are hidden

    LK functional choice, cross-terms approximations, model transformation,. . .

    rationale behind model transformations is obscure (recondite?)(mysteriously, they all concentrated on 1e

    sh

    s , yet I found no hint why)

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    20/74

    Outline

    Lyapunov-Krasovski methods & model transformations

    Good ol (scaled) Small Gain Theorem

    Some comparisons

    Possible refinements

    Concluding remarks

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    21/74

    The Small Gain Theorem

    G(s)

    (s)

    TheoremLet G(s)and (s)be stable and such that

    1 and G < 1.Then the closed-loop system is internally stable.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    22/74

    Delay as unstructured uncertaintyI

    Straightforward approach is to exploit the facts that

    esh 1, h.Then,

    sx=

    A0x+

    A1e

    sh

    x stableh > 0 if

    sx=A0x + A1 x stable

    1

    We then end up withdelay-independent(sufficient) condition

    (sI A0)1A1 < 1,

    which is easily solvable.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    23/74

    Delay as unstructured uncertaintyI

    Straightforward approach is to exploit the facts that

    esh 1, h.Then,

    sx=A0

    x + A1e

    sh x stable

    h > 0 if

    sx=A0x + A1 x stable

    1

    We then end up withdelay-independent(sufficient) condition

    (sI A0)1A1 < 1,

    which is easily solvable.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    24/74

    Delay as unstructured uncertaintyI (contd)

    Conservatism can be a bit reduced by noticing that

    Mesh =eshM, M Rnn

    Then,

    sx=A0

    x + A1esh x

    stableh > 0

    if

    sx=A0x + A1 x stable

    1such thatM=M

    We then end up withdelay-independentcondition

    M= M > 0such thatM(sI A0)1A1M

    1

    < 1,

    which is LMI-able.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    25/74

    Delay as unstructured uncertaintyI (contd)

    Advantages:

    easilyunderstandable easilytractable

    easily extendible tomultiple-delayproblems

    easily incorporable into controllerdesign(Hoptimization)

    Disadvantages:

    delay independent, hence too conservative

    (not so many, if any, problems, where delays can become arbitrarily large) all phase information about delay is neglected

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    26/74

    Delay as unstructured uncertaintyI (contd)

    Advantages:

    easily understandable easily tractable

    easily extendible to multiple-delay problems

    easily incorporable into controller design (Hoptimization)

    Disadvantages:

    delay independent, hence tooconservative

    (not so many, if any, problems, where delays can become arbitrarily large) allphase informationabout delay isneglected

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    27/74

    Delay as unstructured uncertaintyII

    Rewrite

    sx=A0x + A1eshx= (A0+ A1)x A1(1 e

    sh)x.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    28/74

    Delay as unstructured uncertaintyII

    Rewrite

    sx=A0x + A1eshx= (A0+ A1)x A1(1 e

    sh)x.

    Term1 esh is a better candidate for approximations because its

    size (norm) does depend on the phase lag ofejh.

    Re

    Im

    1

    1 ej1h

    Re

    Im

    1

    1 ej2h

    Re

    Im

    1

    1 ej3h

    Here1

    < 2

    < 3

    .

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    29/74

    Delay as unstructured uncertaintyII

    Rewrite

    sx=A0x + A1eshx= (A0+ A1)x A1(1 e

    sh)x.

    Term1 esh is a better candidate for approximations because its

    size (norm) does depend on the phase lag ofejh.

    Re

    Im

    1

    1 ej1h

    Re

    Im

    1

    1 ej2h

    Re

    Im

    1

    1 ej3h

    Here1

    < 2

    < 3

    .

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    30/74

    Covering1 esh

    Re

    Im

    1

    h

    lh(

    )

    Simple geometry yields that

    lh() =

    2 sin h

    2 ifh

    2 ifh >

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    31/74

    Delay as unstructured uncertaintyII (contd)

    Then,

    sx= (A0+ A1)x A1(1 esh)x stableh [0,h]

    if

    sx= (A0+ A1)x + A1 x stable/lh 1

    We then have delay-dependent (sufficient) condition

    (sI A0 A1)1

    A1 lh(s) < 1,which might not be easy to check though, because

    lh()is not rational.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    32/74

    Delay as unstructured uncertaintyII (contd)

    Then,

    sx= (A0+ A1)x A1(1 esh)x stableh [0,h]

    if

    sx= (A0+ A1)x + A1 x stable/lh 1

    We then havedelay-dependent(sufficient) condition

    (sI A0 A1)1

    A1 lh(s) < 1,which might not be easy to check though, because

    lh()is not rational.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    33/74

    Rational approximations oflh

    We need to construct stable and rationalW(s)such that

    |W(j)| lh(), .

    Some examples:

    W0(s) = hs,

    W1(s) = 2

    3hs

    hs+2

    3

    W3(s) = 2.007hs

    hs+2s2+1.567s+2

    s2+1.283s+2 with =2.358/h.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    34/74

    Rational approximations oflh

    We need to construct stable and rationalW(s)such that

    |W(j)| lh(), .

    Some examples:

    W0(s) = hs,

    W1(s) = 23hshs+2

    3

    (note that|W1(j)| 0),

    W3(s) = 2.007hs

    hs+2s2+1.567s+2

    s2+1.283s+2 with =2.358/h.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    l f

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    35/74

    Rational approximations oflh

    We need to construct stable and rationalW(s)such that

    |W(j)| lh(), .

    Some examples:

    W0(s) = hs,

    W1(s) = 23hshs+2

    3

    W3(s) = 2.007hs

    hs+2s2+1.567s+2

    s2+1.283s+2 with =2.358/h.

    We then end up with delay-dependent (sufficient) condition

    (sI A0 A1)1A1W(s) < 1,

    which is easily calculable. . .

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    l f l

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    36/74

    Rational approximations oflh

    We need to construct stable and rationalW(s)such that

    |W(j)| lh(), .

    Some examples:

    W0(s) = hs,

    W1(s) = 23hshs+2

    3

    W3(s) = 2.007hs

    hs+2s2+1.567s+2

    s2+1.283s+2 with =2.358/h.

    . . . or, exploitingM(1 esh) = (1 esh)M, with (sufficient) condition

    M=M > 0such thatM(sI A0 A1)1A1W(s)M

    1 < 1,which is LMI-able.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    D l d i II ( d)

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    37/74

    Delay as unstructured uncertaintyII (contd)

    The idea is rather old (pre-LK), it

    can be traced back to (Owens & Raya, 82; Morari & Zafiriou, 89) and extensively exposed in (Wang, Lundstrom & Skogestad, 94).

    Advantages:

    easily understandable

    easily tractable

    easily extendible to multiple-delay problems

    easily incorporable into controller design (Hoptimization) conservatism sources, unlike LK approach, clearly seen.

    Disadvantages:

    seems to be too conservative in general

    conservatism sources, unlike LK approach, clearly seen1.

    1This appears to harm the acceptance of the method.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    D l t t d t i t II ( td)

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    38/74

    Delay as unstructured uncertaintyII (contd)

    The idea is rather old (pre-LK), it

    can be traced back to (Owens & Raya, 82; Morari & Zafiriou, 89) and extensively exposed in (Wang, Lundstrom & Skogestad, 94).

    Advantages:

    easilyunderstandable

    easilytractable

    easily extendible tomultiple-delayproblems

    easily incorporable into controllerdesign(Hoptimization) conservatism sources, unlike LK approach, clearly seen.

    Disadvantages:

    seems to be too conservative in general

    conservatism sources, unlike LK approach, clearly seen1.

    1This appears to harm the acceptance of the method.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    D l t t d t i t II ( td)

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    39/74

    Delay as unstructured uncertaintyII (contd)

    The idea is rather old (pre-LK), it

    can be traced back to (Owens & Raya, 82; Morari & Zafiriou, 89) and extensively exposed in (Wang, Lundstrom & Skogestad, 94).

    Advantages:

    easily understandable

    easily tractable

    easily extendible to multiple-delay problems

    easily incorporable into controller design (Hoptimization) conservatism sources, unlike LK approach,clearly seen.

    Disadvantages:

    seems to be tooconservativein general

    conservatism sources, unlike LK approach,clearly seen1.

    1This appears to harm the acceptance of the method.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    O tli

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    40/74

    Outline

    Lyapunov-Krasovski methods & model transformations

    Good ol (scaled) Small Gain Theorem

    Some comparisons

    Possible refinements

    Concluding remarks

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    SG isnt more conservative than LK

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    41/74

    SG isnt more conservative than LK

    Two papers in the early 2000s showed that in many cases

    LK conditions might actually be more conservative than SG conditions.These are

    (Huang & Zhou, 00), who cast delay robustness problem as -problem andclaimed that LK-based solutions are much more conservative

    (LK derivations mostly useW0-bound and static scaling);(Zhang, Knospe, & Tsiotras, 01), who proved that several LK-based results

    are equivalent to (statically scaled) SG-based results, which

    useW0-bound on|1 ejh|.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    SG isnt more conservative than LK

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    42/74

    SG isn t more conservative than LK

    Two papers in the early 2000s showed that in many cases

    LK conditions might actually be more conservative than SG conditions.These are

    (Huang & Zhou, 00), who cast delay robustness problem as -problem andclaimed that LK-based solutions are much more conservative

    (LK derivations mostly useW0-bound and static scaling);(Zhang, Knospe, & Tsiotras, 01), who proved that several LK-based results

    are equivalent to (statically scaled) SG-based results, which

    useW0-bound on|1 ejh|.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    SG isnt more conservative than LK

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    43/74

    SG isn t more conservative than LK

    Two papers in the early 2000s showed that in many cases

    LK conditions might actually be more conservative than SG conditions.These are

    (Huang & Zhou, 00), who cast delay robustness problem as -problem andclaimed that LK-based solutions are much more conservative

    (LK derivations mostly useW

    0-bound and static scaling);(Zhang, Knospe, & Tsiotras, 01), who proved that several LK-based results

    are equivalent to (statically scaled) SG-based results, which

    useW0-bound on|1 ejh|.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Descriptor approach vs Small Gain Theorem

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    44/74

    Descriptor approach vs. Small Gain Theorem

    Consider covering of|1 ejh|withW0 =sh. We have that our system is

    stableh [0,

    h]if

    sx(s) = (A0+ A1)x(s) A1(s)hsx(s)

    is stable for all

    1such thatM=Mfor allM. In principle, this

    is guaranteed ifM= M> 0such that to

    M

    I + (A0+ A1)(sI A0 A1)1

    A1M1 < 1h ,

    yet we may want to rewrite it as

    0 M

    s

    I 0

    0 0

    0 I

    A0+ A1 I

    1

    0

    A1M1

    < 1

    h

    and end up with the problem of

    calculatingHnorm of adescriptorsystem.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Descriptor approach vs Small Gain Theorem

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    45/74

    Descriptor approach vs. Small Gain Theorem

    Consider covering of|1 ejh|withW0 =sh. We have that our system is

    stableh [0,

    h]if

    sx(s) = (A0+ A1)x(s) A1(s)hsx(s)

    is stable for all

    1such thatM=Mfor allM. In principle, this

    is guaranteed ifM= M> 0such that to

    M

    I + (A0+ A1)(sI A0 A1)1

    A1M1 < 1h ,

    yet we may want to rewrite it as

    0 M

    s

    I 0

    0 0

    0 I

    A0+ A1 I

    1

    0

    A1M1

    < 1

    h

    and end up with the problem of

    calculatingHnorm of adescriptorsystem.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    H norm of descriptor systems

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    46/74

    H norm of descriptor systems

    Theorem (Rehm & Allgover, 99)

    Letdet(sE A)0, thenC(sE A)1B < iff

    X such that E X=X E 0and

    A X + X A X B C

    B X I 0C 0 I

    < 0.

    In our case,

    E X= X E 0 I 0

    0 0 X11 X12

    X21 X22

    =X 11 X 21

    X 12 X 22 I 0

    0 0 0.

    HenceX12 =0 and X11 =X

    11 0.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    H norm of descriptor systems

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    47/74

    H norm of descriptor systems

    Theorem (Rehm & Allgover, 99)

    Letdet(sE A)0, thenC(sE A)1B < iff

    X such that E X=X E 0and

    A X + X A X B C

    B X I 0C 0 I

    < 0.

    In our case,

    E X= X E 0 I 0

    0 0 X11 X12

    X21 X22

    =X 11 X 21

    X 12 X 22 I 0

    0 0 0.

    HenceX12 =0 and X11 =X

    11 0.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    H norm of descriptor systems

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    48/74

    H norm of descriptor systems

    Theorem (Rehm & Allgover, 99)

    Letdet(sE A)0, thenC(sE A)1B < iff

    X such that E X=X E 0and

    A X + X A X B C

    B X I 0C 0 I

    < 0.

    In our case,

    E X= X E 0 I 0

    0 0 X11 0

    X21 X22

    =X 11 X 21

    0 X 22 I 0

    0 0 0.

    HenceX12 =0 and X11 =X

    11 0.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Descriptor approach vs. Small Gain Theorem (contd)

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    49/74

    Descriptor approach vs. Small Gain Theorem (contd)

    Proceedings further, we get LMI solvability condition

    A X21+ X 21A X11 X21+ AX22 X

    21A1M

    1 0

    X11 X21+ X22A

    X22 X 22 X22A1M

    1 M

    M1A 1X21 M1A 1X22

    1h

    I 0

    0 M 0 1h

    I

    < 0

    or, equivalently (via Schur complement of the (4, 4)term), LMI

    A X21+ X 21A X11 X

    21+

    AX22 X21A1h

    X11 X21+ X22A

    hY X22 X 22 X22A1h

    hA 1X21

    hA 1X22

    hY

    < 0,

    where A .=A0+ A1andY

    .=M2.

    This is

    exactly the condition of (Fridman, 01) derived via LK technique.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Descriptor approach vs. Small Gain Theorem (contd)

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    50/74

    Descriptor approach vs. Small Gain Theorem (contd)

    Proceedings further, we get LMI solvability condition

    A X21+ X 21A X11 X21+ AX22 X

    21A1M

    1 0

    X11 X21+ X22A

    X22 X 22 X22A1M

    1 M

    M1A 1X21 M1A 1X22

    1h

    I 0

    0 M 0 1h

    I

    < 0

    or, equivalently (via Schur complement of the (4, 4)term), LMI

    A X21+ X 21A X11 X

    21+

    AX22 X21A1h

    X11 X21+ X22A

    hY X22 X 22 X22A1h

    hA 1X21

    hA 1X22

    hY

    < 0,

    where A .=A0+ A1andY

    .=M2.

    This is

    exactly the condition of (Fridman, 01) derived via LK technique.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Descriptor approach is a version of SGT too

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    51/74

    p pp

    Thus, we have that

    descriptor transformation leads to conditions, which are equivalent toapplication of scaled Small Gain Theorem under covering jh >|1 ejh|(in a sense, the weakest covering) bringing in some redundancy into state-space realization via

    A1+ (A0+ A1)sI A0 A11A1

    0 I

    s

    I 0

    0 0

    0 I

    A0+ A1 I

    1 0

    A1

    ,

    which does not introduce any additional dynamics.

    In other words, descriptor transformation appears to be

    smart solution to problem one should not have gotten into in the first

    place.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Descriptor approach is a version of SGT too

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    52/74

    p pp

    Thus, we have that

    descriptor transformation leads to conditions, which are equivalent toapplication of scaled Small Gain Theorem under covering jh >|1 ejh|(in a sense, the weakest covering) bringing in some redundancy into state-space realization via

    A1+ (A0+ A1)sI A0 A11A1

    0 I

    s

    I 0

    0 0

    0 I

    A0+ A1 I

    1 0

    A1

    ,

    which does not introduce any additional dynamics.

    In other words, descriptor transformation appears to be

    smart solution to problem one should not have gotten into in the first

    place.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Example

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    53/74

    p

    Consider the system (Kolmanovski & Richard, 99):

    x(t) =

    1 0.5

    0.5 1

    x(t) +

    2 2

    2 2

    x(t h)

    The following stability bounds are available:

    Method IV SGT+W0 SGT+W1 SGT+lhhmax 0.271 0.2716 0.3042 0.3047

    whereunscaled SGT was used.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Some arguments in favor of SGT.

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    54/74

    g

    Lyapunov-Krasovski functional:

    1. Pros and cons obscure2. Hinges uponW0(s) = hs(?)

    3. Hinges upon static scalings

    4. Design limited to FD controllers(hence nominal delay has to be h

    0=0)

    Small Gain Theorem:

    1. Pros and cons transparent2. Can use tighter coverings

    3. Can use dynamic scalings ()

    4. Design can use Smith predictors(hence nominal delay may be h

    0=

    h

    2)

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Some arguments in favor of SGT.

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    55/74

    g

    Lyapunov-Krasovski functional:

    1. Pros and cons obscure2. Hinges uponW0(s) = hs(?)

    3. Hinges upon static scalings

    4. Design limited to FD controllers(hence nominal delay has to be h0 =0)

    Small Gain Theorem:

    1. Pros and cons transparent2. Can use tighter coverings

    3. Can use dynamic scalings ()

    4. Design can use Smith predictors(hence nominal delay may be h0 =

    h

    2

    )

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Some arguments in favor of SGT.

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    56/74

    Lyapunov-Krasovski functional:

    1. Pros and cons obscure2. Hinges uponW0(s) = hs(?)

    3. Hinges upon static scalings

    4. Design limited to FD controllers(hence nominal delay has to be h0 =0)

    Small Gain Theorem:

    1. Pros and cons transparent2. Can use tighter coverings

    3. Can use dynamic scalings ()

    4. Design can use Smith predictors(hence nominal delay may be h0 =

    h

    2

    )

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Some arguments in favor of SGT.

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    57/74

    Lyapunov-Krasovski functional:

    1. Pros and cons obscure2. Hinges uponW0(s) = hs(?)

    3. Hinges upon static scalings

    4. Design limited to FD controllers(hence nominal delay has to be h0 =0)

    Small Gain Theorem:

    1. Pros and cons transparent2. Can use tighter coverings

    3. Can use dynamic scalings ()

    4. Design can use Smith predictors(hence nominal delay may be h0 =

    h

    2

    )

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Some arguments in favor of SGT.

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    58/74

    Lyapunov-Krasovski functional:

    1. Pros and cons obscure2. Hinges uponW0(s) = hs(?)

    3. Hinges upon static scalings

    4. Design limited to FD controllers(hence nominal delay has to be h0 =0)

    Small Gain Theorem:

    1. Pros and cons transparent2. Can use tighter coverings

    3. Can use dynamic scalings ()

    4. Design can use Smith predictors(hence nominal delay may be h0 =

    h

    2

    )

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Some arguments in favor of SGT.

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    59/74

    Lyapunov-Krasovski functional:

    1. Pros and cons obscure2. Hinges uponW0(s) = hs(?)

    3. Hinges upon static scalings

    4. Design limited to FD controllers(hence nominal delay has to be h0 =0)

    Small Gain Theorem:

    1. Pros and cons transparent2. Can use tighter coverings

    3. Can use dynamic scalings ()

    4. Design can use Smith predictors(hence nominal delay may be h0 =

    h

    2

    )

    The question is

    what makes LK methods so dominating ?

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Outline

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    60/74

    Lyapunov-Krasovski methods & model transformations

    Good ol (scaled) Small Gain Theorem

    Some comparisons

    Possible refinements

    Concluding remarks

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Trading off DIS-DDS

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    61/74

    We can also rewrite equationsx=A0x + A1eshxas

    sx= (A0+ A1)x A1(1 esh)x A1( I)eshx,

    whereis arbitrary. If

    =0, delay-independent conditions recovered

    =I, delay-dependent conditions recoveredIn general,brings more freedom and this freedom is LMI-able.

    This freedom is exploited in LK methods too,

    either explicitly (parametrized model transformation) or implicitly (via so-called Parks inequality for bounding cross-terms)

    (connection not quite transparent, though Zhang, Knospe & Tsiotras (01) showed it via

    equivalence of resulting LMIs)

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Trading off DIS-DDS

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    62/74

    We can also rewrite equationsx=A0x + A1eshxas

    sx= (A0+ A1)x A1(1 esh)x A1( I)eshx,

    whereis arbitrary. If

    =0, delay-independent conditions recovered

    =I, delay-dependent conditions recoveredIn general,brings more freedom and this freedom is LMI-able.

    This freedom is exploited in LK methods too,

    either explicitly (parametrized model transformation) or implicitly (via so-called Parks inequality for bounding cross-terms)

    (connection not quite transparent, though Zhang, Knospe & Tsiotras (01) showed it via

    equivalence of resulting LMIs)

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Trading off DIS-DDS: example

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    63/74

    An (almost classical) example of (Li & de Souza, 97) considers the system

    x(t) =

    2 00 0.9

    x(t)

    1 01 1

    x(t h).

    This system can be presented as the cascade

    1s+2+esh

    1s+0.9+esh

    esh x1x2

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Trading off DIS-DDS: example

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    64/74

    An (almost classical) example of (Li & de Souza, 97) considers the system

    x(t) =

    2 00 0.9

    x(t)

    1 01 1

    x(t h).

    This system can be presented as the cascade

    1s+2+esh

    1s+0.9+esh

    esh x1x2

    Delay-independent stableDelay-dependent

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Trading off DIS-DDS: example

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    65/74

    An (almost classical) example of (Li & de Souza, 97) considers the system

    x(t) =

    2 00 0.9

    x(t)

    1 01 1

    x(t h).

    This system can be presented as the cascade

    1s+2+esh

    1s+0.9+esh

    esh x1x2

    Delay-independent stableDelay-dependent

    This means that the choice = 0 00 I

    yieldssx=

    2 0

    0 1.9

    x +

    0 0

    0 1

    (1 esh)x

    1 0

    1 0

    eshx

    and effectively reduces this system to sx2

    = 1.9x2

    +(1 esh)x2

    .

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Trading off DIS-DDS: example (contd)

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    66/74

    Thus

    sx=

    2 00 0.9

    x

    1 01 1

    eshx stability sx2 = 0.9x2e

    shx2.

    It then becomes clear why some methods are less conservative than otherson this particular example (and alike).

    Method I II III+PI I+ II+ IV+PI SGT+W3 Exact

    hmax .99 .99 4.36 4.35 4.35 4.47 4.84 6.17

    More successful methods get rid ofx1, either explicitly (via) or implicitly

    (via Parks inequality).

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Beyond SGT

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    67/74

    Small Gain Theorem isnottheonlyfrequency-domain robustness tool. Wemay try to

    combine small gain and passivity arguments

    to end up with less conservative results. This can be done in the

    IQC framework

    for example, as shown in (Megretski & Rantzer, 97).

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Shifted covering

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    68/74

    Clearly,

    sx=A0x + A1eshx= (A0+ A1V(s))x A1(V(s)esh)x.

    We may then try to

    chooseV(s)to reduce conservatism of covering |V(j) ejh|.

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Shifted covering: how to chooseV(s)

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    69/74

    Re

    Im

    1

    h

    lh(

    )

    Re

    Im

    V(j)

    h

    Simple geometry yields then:

    V(j) =

    cos h

    2 ej

    h2 ifh

    0 ifh > and lh,V() =

    12

    lh().

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Shifted covering: how to chooseV(s)

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    70/74

    Re

    Im

    1

    h

    lh(

    )

    Re

    Im

    V(j)

    h

    Simple geometry yields then:

    V(j) =

    cos h

    2 ej

    h2 ifh

    0 ifh > and lh,V() =

    12

    lh().

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Shifted covering: rational approximationh j

    h

    b l d b

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    71/74

    Frequency responsecos h2 ej

    h2 can be quite accurately approximated by

    V1(s) = 2hs + 2

    .

    Covering radii are then:

    102

    101

    100

    25

    20

    15

    10

    5

    0

    5

    lh

    lh,V1

    lh,V

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Outline

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    72/74

    Lyapunov-Krasovski methods & model transformations

    Good ol (scaled) Small Gain Theorem

    Some comparisons

    Possible refinements

    Concluding remarks

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Concluding remarks

    http://find/http://goback/
  • 7/26/2019 Delay Robustness Mirkin

    73/74

    Advantages of LK-methods are in no proportion to their popularity

    Relations between LK and SGT are yet to be understood(it looks like that all we can show is that they result in the same LMIs; it would be of

    great value to have clear correspondence between intermediate steps of each method)

    Model transformations Small Gain Theorem Some comparisons Refinements Concluding remarks

    Concluding remarks

    http://find/
  • 7/26/2019 Delay Robustness Mirkin

    74/74

    Advantages of LK-methods are in no proportion to their popularity

    Relations between LK and SGT are yet to be understood(it looks like that all we can show is that they result in the same LMIs; it would be of

    great value to have clear correspondence between intermediate steps of each method)

    http://find/