Incremental stability of bimodal Filippov systems in Rn Davide Fiore Supervisor: Mario di Bernardo University of Naples "Federico II" Department of Electrical Engineering and Information Technology IEEE International Meeting on Analysis and Applications of Nonsmooth Systems 10-12 September 2014, Como, Italy D. Fiore and M. di Bernardo AANS 2014 1 / 21 Outline 1 Introduction and Motivation 2 Incremental stability of bimodal Filippov systems 3 An example 4 Conclusions D. Fiore and M. di Bernardo AANS 2014 2 / 21 Introduction A dynamical system is said to be incrementally stable if its solutions asymptotically converge towards each other In turn it implies that initial conditions or temporary disturbances are asymptotically forgotten in time This allows us to study convergence properties of a system without the need of having an exact knowledge of some attractors or nominal solutions Incremental stability plays an important role in many nonlinear control problems including tracking design of nonlinear state observers synchronization in networks of dynamical systems output regulation problems D. Fiore and M. di Bernardo AANS 2014 3 / 21 Incremental stability and Contraction Theory Contraction theory provides sufficient conditions for incremental exponential stability of a smooth dynamical system. Definition Consider the dynamical system x˙ = f (x) defined on a forward invariant and path-connected set C ⊆ Rn and let x(t) and z(t) be any two of its solutions with initial conditions x(t0 ) = x0 ∈ C and z(t0 ) = z0 ∈ C. The system is said to be incrementally exponentially stable (IES) on C if there exist constants K ≥ 1 and λ > 0 such that |x(t) − z(t)| ≤ K e−λ(t−t0 ) |x0 − z0 | D. Fiore and M. di Bernardo ∀t ≥ t0 , ∀x0 , z0 ∈ C AANS 2014 4 / 21 Incremental stability and Contraction Theory Definition The smooth dynamical system x˙ = f (x) is said to be contracting on a set C ⊆ Rn if there exists some norm in C, with associated matrix measure µ, such that, for some constant c > 0, it holds ∂f µ (x) ≤ −c ∀x ∈ C, ∀t ≥ t0 ∂x Note: For example, the matrix measure µ induced by Euclidean norm is the maximum ∂f . eigenvalue of the symmetric part of ∂x The basic result of nonlinear contraction analysis states that If a system is contracting, then all of its trajectories exponentially converge towards each other, i.e. the system is IES. ∂f µ (x) ≤ −c ⇒ ∆x → 0 ∂x D. Fiore and M. di Bernardo AANS 2014 5 / 21 Motivation Still, most of the results available in the literature on both incremental stability and contraction assume that the system vector field is continuously differentiable. Only relatively few results deal with the problem of investigating incremental stability of piecewise-smooth and switched dynamical systems. See for example Slotine et al. (1998) for constrained systems Pavlov et al. (2005) for PWA systems di Bernardo et al. (2011) for Carathéodory systems di Bernardo and Liuzza (2013) for planar Filippov systems The aim of our research is to extend the existing results to the case of n-dimensional Filippov systems. These theoretical results are useful in applications like design PWS observers proving convergence in networks of PWS systems D. Fiore and M. di Bernardo AANS 2014 6 / 21 Bimodal Filippov systems A common class of PWS dynamical systems particularly important in many applications is that of bimodal Filippov systems A bimodal Filippov system is given by ( f + (x) if h(x) > 0 x˙ = f − (x) if h(x) < 0 where f + (x) and f − (x) are two smooth vector fields h(x) is a smooth scalar function whose zero set defines the smooth discontinuity manifold Σ := {x ∈ Rn : h(x) = 0} We want to find sufficient conditions for a such class of systems able to guarantee incremental exponential stability on a subset C ⊂ Rn . D. Fiore and M. di Bernardo AANS 2014 7 / 21 The key idea We want that every pairs of trajectories converge towards each other in C independently of where they start from, on the switching manifold Σ or outside of it. To achieve this we require that: 1 2 ˆ is globally attractive for all system solutions evolving in C\Σ ˆ The sliding region Σ The sliding vector field is contracting, and therefore such that every pair of solutions within the sliding region will exponentially converge towards each other D. Fiore and M. di Bernardo AANS 2014 8 / 21 ˆ to be attractive 1. Conditions for Σ The switching manifold Σ is guaranteed to be attractive in C if the two vector fields f + (x) and f − (x) point towards it in every point on C. For example, assume Σ be a hyperplane with h(x) = hT (x − xh ) for some constant vectors h, xh ∈ Rn . If there exist two scalar l1 > 0 and l2 < 0 such that hT f − (x) ≥ l1 for all x ∈ C ∩ G − and hT f + (x) ≤ l2 for all x ∈ C ∩ G + , then Σ is attractive for every solution starting in C. Indeed for ξ ∈ C ∩ G − d h(x) = hT f − (x) ≥ l1 dt integrating we have that the flow reaches Σ at a time instant ts ≤ −h(ξ)/l1 Similarly for ξ ∈ C ∩ G + . D. Fiore and M. di Bernardo AANS 2014 9 / 21 2. Contracting sliding vector field In order for the sliding vector field f sl to be contracting we need to study the Jacobian of ( sl f (x) = α f + (x) + (1 − α) f − (x) with α ∈ [0, 1] f sl (x) · ∇h(x) = 0 As in general α(x) can have cumbersome expressions, we treat this system as a constrained system. We derive conditions for contraction in a constrained coordinate set z ∈ Rn−1 such that x = ϕ(z) D. Fiore and M. di Bernardo AANS 2014 10 / 21 Constrained coordinates The idea is to: 1 consider a projector P(x) on the hyperplane Tx Σ tangent to the switching manifold Σ 2 move to the set of n − 1 constrained coordinates on Σ, x = ϕ(z). We then require ∂f sl µ (z) ∂z where ! ≤ −c ∂f sl (z) ∂ sl = P(x) · f (x) ∂z ∂x · x=ϕ(z) ∂ϕ ∂z The derivation requires lots of algebra and can be found in di Bernardo, Fiore "Incremental stability of bimodal Filippov systems in Rn ", accepted to 53rd IEEE Conference on Decision and Control, 2014. D. Fiore and M. di Bernardo AANS 2014 11 / 21 Incremental stability of bimodal Filippov systems Theorem Assume h(x) = hT (x − xh ), for some constant vectors h, xh ∈ Rn . If there exists a forward invariant and path-connected subset C ⊂ Rn such that ˆ 6= ∅ and the following conditions hold: ΣC = C ∩ Σ 1 2 there exists two scalar l1 > 0 and l2 < 0 such that hT f − (x) ≥ l1 for all x ∈ C ∩ G − and hT f + (x) ≤ l2 for all x ∈ C ∩ G + ; sl sl µ ∂f∂z ≤ −c for all z such that ϕ(z) ∈ ΣC , with c > 0, where ∂f∂z is the Jacobian matrix of the sliding vector field restricted on Σ, and z ∈ Rn−1 are a set of constrained coordinates on Σ; then all trajectories of the system globally converge towards each other in C. D. Fiore and M. di Bernardo AANS 2014 12 / 21 Example Let us consider the relay feedback system in R2 ( f − (x) = Ax + B if h(x) = Cx < 0 x˙ = f + (x) = Ax − B if h(x) = Cx > 0 with 0 A= 1 D. Fiore and M. di Bernardo −1 , −1 1 B= , 1 C= 0 1 AANS 2014 13 / 21 Example - Region of incremental stability Applying the Theorem to this system we can obtain a region C of incremental stability. This region is in fact such that every trajectory starting inside will stay in it for all t > t0 (forward invariance) both vector fields are directed towards Σ for all x ˆ the region is connected, bounded and it intersects the sliding region Σ D. Fiore and M. di Bernardo AANS 2014 14 / 21 Example - Simulation 1 Let us consider two initial conditions inside the region of incremental stability, one in G + and the other in G − . D. Fiore and M. di Bernardo AANS 2014 15 / 21 Example - Simulation 1 Evolution in time of the distance between the trajectories. D. Fiore and M. di Bernardo AANS 2014 16 / 21 Example - Simulation 2 Let us consider two initial conditions both in G − or in G + . D. Fiore and M. di Bernardo AANS 2014 17 / 21 Example - Simulation 2 Evolution in time of the distance between the trajectories. D. Fiore and M. di Bernardo AANS 2014 18 / 21 Example - Remarks As can been seen from the previous numerical simulations, the system is incrementally exponentially stable in the green subset C. It is worth to notice that the constants K ≥ 1 and λ in the definition of IES |x(t) − z(t)| ≤ K e−λ(t−t0 ) |ξ − ζ| ∀t ≥ t0 , ∀ξ, ζ ∈ C appear to be related to the geometry of the region of incremental stability and to the contraction rate c of the sliding vector field. An open problem is how to analytically estimate the value of K and λ for a generic system. D. Fiore and M. di Bernardo AANS 2014 19 / 21 Conclusions Contraction theory is a powerful tool to assess incremental stability of a dynamical system. Unfortunately there are few results on this topic for PWS systems and, in particular, for Filippov systems. We have presented here a sufficient condition for incremental stability of Filippov systems based on treating the system as a constrained system. This condition has been defined in terms of the properties of the subset C the vector fields f + , f − and f sl the switching manifold Σ More work needs to be done to extend the theory to other classes of PWS systems, like constrained systems, in a more general framework. D. Fiore and M. di Bernardo AANS 2014 20 / 21