1 / 28

Lecture #3 What can go wrong? Trajectories of hybrid systems

Hybrid Control and Switched Systems. Lecture #3 What can go wrong? Trajectories of hybrid systems. Jo ã o P. Hespanha University of California at Santa Barbara. Summary. Trajectories of hybrid systems: Solution to a hybrid system Execution of a hybrid system Degeneracies

darva
Télécharger la présentation

Lecture #3 What can go wrong? Trajectories of hybrid systems

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Hybrid Control and Switched Systems Lecture #3What can go wrong?Trajectories of hybrid systems João P. Hespanha University of Californiaat Santa Barbara

  2. Summary • Trajectories of hybrid systems: • Solution to a hybrid system • Execution of a hybrid system • Degeneracies • Finite escape time • Chattering • Zeno trajectories • Non-continuous dependency on initial conditions

  3. Hybrid Automaton (deterministic) Q ´ set of discrete states Rn´ continuous state-space f : Q£Rn!Rn´ vector field  : Q£Rn!Q´ discrete transition r : Q£Rn!Rn´ reset map j(q1,x–) = q2 ? x›r(q1,x–) mode q1 mode q3 mode q2 x›r(q1,x–) j(q1,x–) = q3 ?

  4. Hybrid Automaton Q ´ set of discrete states Rn´ continuous state-space f : Q£Rn!Rn´ vector field F : Q£Rn!Q£Rn´ discrete transition (& reset map) F1(q1,x–) = q2 ? x› F2(q1,x–) mode q1 mode q3 mode q2 x› F2(q1,x–) F1(q1,x–) = q3 ? Compact representation of a hybrid automaton

  5. Solution to a hybrid automaton mode q1 mode q2 F1(q1,x–) = q2 ? x› F2(q1,x–) Definition: A solution to the hybrid automaton is a pair of right-continuous signals x : [0,1) !Rnq : [0,1) !Qsuch that 1. x is piecewise differentiable & q is piecewise constant 2. on any interval (t1,t2) on which q is constant and x continuous 3. continuous evolution discrete transitions

  6. Example #4: Inverted pendulum swing-up q Hybrid controller: 1st pump/remove energy into/from the system by applying maximum force, until E ¼ 0 2nd wait until pendulum is close to the upright position 3th next to upright position use feedback linearization controller u2 [-1,1] E2 [–e,e] ? remove energy E < –e ? stabilize wait E>e? pump energy |w| + |q| ·d ? E2 [–e,e] ?

  7. Example #4: Inverted pendulum swing-up stabilize wait pump energy E¸ – e ? |w| + |q| ·d ? Q = { pump, wait, stab } R2 = continuous state-space

  8. Example #4: Inverted pendulum swing-up stabilize wait pump energy E¸ – e ? |w| + |q| ·d ? Q = { pump, wait, stab } R2 = continuous state-space |w| + |q| d – e t q = wait q = stab E q = pump What if we already have |w| + |q| ·d when E reaches – e?

  9. Example #4: Inverted pendulum swing-up stabilize wait pump energy E¸ – e ? |w| + |q| ·d ? Q = { pump, wait, stab } R2 = continuous state-space |w| + |q| t* d – e t q = stab In general, we may need more than one value for the state at each instant of time q = wait E q( t* ) = ? q = pump

  10. Hybrid signals Definition: A hybrid time trajectory is a (finite or infinite) sequence of closed intervals t = { [ti,t'i] : ti·t'i, t'i = ti+1 , i =1,2, … } (if t is finite the last interval may by open on the right) T´ set of hybrid time trajectories Definition: For a given t = { [ti,t'i] : ti·t'i, ti+1 = t'i, i =1,2, … } 2 T a hybrid signal defined on t with values on X is a sequence of functions x = {xi : [ti,t'i]!X i =1,2, … } x : t!X´ hybrid signal defined on t with values on X E.g., t› { [0,1], [1,1], [1,1], [1,+1) }, x› { t/2 , 1/4 , 3/4, 1/t } 3/4 1/t t/2 1/4 t A hybrid signal can take multiple values for the same time-instant [1,1] [0,1] [1,1] [1,+1)

  11. Execution of a hybrid automaton mode q1 mode q2 F1(q1,x–) = q2 ? x› F2(q1,x–) Definition: An execution of the hybrid automaton is a pair of hybrid signals x : t!Rnq : t!Q t = { [ti,t'i] : i =1,2, … } 2 Tsuch that 1. on any [ti,t'i] 2t, qi is constant and 2. continuous evolution discrete transitions

  12. Example #4: Inverted pendulum swing-up stabilize wait pump energy E¸ – e ? |w| + |q| ·d ? Q = { pump, wait, stab } R2 = continuous state-space |w| + |q| t› { [0,t*], [t*, t*], [t*,+1) } q› { pump , wait , stab } x› { … , … , … } t* d – e t q = stab q = wait E • There are other concepts of solution that also avoid this problem [Teel 2005] • Could one “fix” this hybrid system to still work with the usual notion of solution? q = pump

  13. What can go wrong? • Problems in the continuous evolution: • existence • uniqueness • finite escape • Problems in the hybrid execution: • Chattering • Zeno • Non-continuous dependency on initial conditions

  14. Existence f(x) discontinuous x There is no solution to this differential equation that starts with x(0) = 0 Why? one any interval [0,e) x cannot: remain zero, become positive, or become negative. (x = 0 would make some sense) Theorem [Existence of solution] If f : Rn!Rn is continuous, then 8x0 2 Rn there exists at least one solution with x(0) = x0, defined on some interval [0,e)

  15. Uniqueness f(x) 1 derivative at 0 There are multiple solutions that start at x(0) = 0, e.g., x x t t Definitions: A function f : Rn!Rn is Lipschitz continuous if in any bounded subset of S of Rn there exists a constant c such that (f is differentiable almost everywhere and the derivative is bounded on any bounded set) Theorem [Uniqueness of solution] If f : Rn!Rn is Lipschitz continuous, then 8x0 2 Rn there a single solution with x(0) = x0, defined on some interval [0,e)

  16. Finite escape time x unbounded derivative t Any solution that does not start at x(0) = 0 is of the form T-1 T finite escape ´ solution x tends to 1 in finite time Definitions: A function f : Rn!Rn is globally Lipschitz continuous if there exists a constant c such that f grows no faster than linearly Theorem [Uniqueness of solution] If f : Rn!Rn is globally Lipschitz continuous, then 8x0 2 Rn there a single solution with x(0) = x0, defined on [0,1)

  17. Degenerate executions due to problems in the continuous evolution t = { [0,1] , [1,2] } q = {q1(t) = 1 , q2(t) = 2 } x = {x1(t) = t , x2(t) = 2 – t } x¸ 1 ? q = 2 q = 1 x1(t) x x2(t) 1 t 2 t = { [0,1] , [1,2) } q = {q1(t) = 1 , q2(t) = 2 } x = {x1(t) = t , x2(t) = 1/(2 – t) } x¸ 1 ? q = 2 q = 1 x1(t) x x2(t) 1 t 2

  18. Chattering q = 2 q = 1 There is no solution past x = 0 x(t) t 1 x¸ 0 ? but there is an execution x < 0 ? t = { [0,1] , [1], [1], [1], … } q = {q1(t) = 2 , q2(t) = 1, q1(t) = 2, …} x = {x1(t) = 1 – t , x2(t) = 0, x2(t) = 0, …} q2k+1 (t)k¸ 1 q1 (t) q2k(t)k¸ 1 x1 (t) Chattering execution ´t is infinite but after some time, all intervals are singletons xk(t)k¸ 2 t 1

  19. Example #3: Semi-automatic transmission g = 1 g = 2 g = 3 g = 4 v(t) 2 { up, down, keep } ´ drivers input (discrete) v = up or w¸w2 ? v = up or w¸w3 ? v = up or w¸w4 ? v = down or w·1 ? v = down or w·2 ? v = down or w·3 ? w2 g = 1 w3 1 g = 2 w4 2 g = 3 g = 4 3 If the driver sets v(t) = up8t¸t* and w(t*)·1 one gets chattering. For ever?

  20. Example #1: Bouncing ball (Zeno execution) Free fall ´ g g Collision ´ y y c2 [0,1) ´ energy absorbed at impact x1 = 0 & x2 <0 ? t x2› – c x2 –

  21. Zeno solution x1 = 0 & x2 <0 ? t1 t2 t3 t x2› – c x2 –

  22. Zeno execution x1 = 0 & x2 <0 ? t1 t2 t3 t x2› – c x2 – Zeno execution ´t is infinite but the execution does not extend to t = +1 Zeno time´ t1 › sups2t s

  23. Example #9: Water tank system (regularization) inflow ´l goal ´ prevent both tanks from becoming emptying x2 · h2 ? q = 2 q = 1 x1 h2 h1 x2 outflow ´m1 outflow ´m2 x1 · h1 ? x2 x1 h1 = h2 q = 1 q = 2 q = 1

  24. Temporal regularizationExample #9: Water tank system inflow ´l x2 · h2 ? q = 2 x1 h2 h1 x2 x1 · h1 ? outflow ´m1 outflow ´m2 x2 · h2, t > e ? t › 0 q = 2 q = 1 only jump after a minimum time has elapsed t › 0 x1 · h1, t > e ?

  25. Spatial regularizationExample #9: Water tank system inflow ´l x2 · h2 ? q = 2 x1 h2 h1 x2 x1 · h1 ? outflow ´m1 outflow ´m2 x2 · h2, |x1–x3|+|x2–x4|> e ? x3›x1, x4›x2 only jump after a minimum change in the continuous state has occurred q = 2 q = 1 x3›x1, x4›x2 x1 · h1, |x1–x3|+|x2–x4|> e ?

  26. Continuity with respect to initial conditions Theorem [Uniqueness & continuity of solution] If f : Rn!Rn is Lipschitz continuous, then 8x0 2 Rn there a single solution with x(0) = x0, defined on some interval [0,e) Moreover, given any T < 1, and two solutions x1, x2 that exist on [0,T]: 8e > 0 9d > 0 : ||x1(0) – x2(0)|| ·d) ||x1(t) – x2(t)|| ·e8t2 [0,T] value of the solution on the interval [0,T] is continuous with respect to the initial conditions

  27. Discontinuity with respect to initial conditions x1 ¸ 0 & x2 > 0 ? x1 ¸ 0 & x2 · 0 ? mode q1 mode q2 mode q3 1 x2 x2 e x1 e – 1 x1 no matter how close to zero x2(0) = e > 0 is, x2(2) = e – 1 if x2(0) = 0 then x2(2) = 1 problem arises from discontinuity of the transition function

  28. Next class… • Numerical simulation of hybrid automata • simulations of ODEs • zero-crossing detection • Simulators • Simulink • Stateflow • SHIFT • Modelica • Follow-up homework • Find conditions for the existence of solution to a hybrid system

More Related