Principles of 2-Sliding Mode Design ○

Arie Levant*a
This paper was not presented at any IFAC meeting. Corresponding author. Phone +972-3-6408812. Fax +972-3-6407543. a School of Mathematical Sciences, Tel-Aviv University, Ramat-Aviv, 69978 Tel-Aviv, Israel, E-mail:
* ○

Abstract Second order sliding modes are used to keep exactly a constraint of the second relative degree or just to avoid chattering, i.e. in the cases when the standard (first order) sliding mode implementation might be involved or impossible. Design of a number of new 2-sliding controllers is demonstrated by means of the proposed homogeneity-based approach. A recently developed robust exact differentiator being applied, robust output-feedback controllers with finite-time convergence are produced, capable to control any general uncertain single-input-single-output process with relative degree 2. An effective simple procedure is developed to attenuate the 1-sliding mode chattering. Simulation of new controllers is presented. Key words: Second-order sliding mode; Robustness; Homogeneity; Chattering; Output feedback. convergent HOSMs preserve the features of the standard (first order) sliding modes and improve their accuracy in the presence of switching delays and discrete measurements (Levant, 1993). While finite-time-convergent arbitrary-order slidingmode controllers are mostly still theoretically studied (Levant, 2001, 2003; Floquet, Barbot and Perruquetti, 2003), 2-sliding controllers with finite-time convergence have already been successfully implemented for solution of real problems (Bartolini, Ferrara & Punta, 2000; Bartolini, Pisano, Punta & Usai, 2003; Levant, Pridor, Gitizadeh, Yaesh & Ben-Asher, 2000; Sira-Ramírez, 2002; Orlov, Aguilar & Cadiou, 2003; Khan, Goh & Spurgeon, 2003; Shkolnikov & Shtessel, 2002; Shkolnikov, Shtessel, Lianos, & Thies, 2000; Shtessel & Shkolnikov, 2003; Shtessel, Shkolnikov & Brown, 2004). There are only few widely used 2-sliding controllers: the sub-optimal controller (Bartolini et al., 1998), the terminal sliding mode controllers (Man, Paplinski & Wu, 1994), and the twisting controller (Levant 1993). 2-3 more controllers are presented in (Levant, 1993, Khan et al., 2003), in particular, the super-twisting controller is used in differentiators (Levant, 1998, 2003). Generally speaking, 2-sliding controllers are used to keep at zero outputs of the relative degree 2 or to avoid chattering while zeroing outputs of the relative degree 1. The main difficulty of their implementation is the necessity to use the first time derivative of the output, which causes possible sensitivity of the approach to sampling noises. These three subjects are studied in the present paper, and some standard solutions are proposed. 1. The homogeneity-based approach (Levant, 2005a) is proposed in this paper to regularize the construction of new finite-time convergent 2-sliding controllers featuring the highest possible accuracy of 2-sliding control (Levant, 1993). The simplicity of such controller design is demon-

1. Introduction Sliding mode control (Utkin, 1992; Zinober, 1994; Edwards and Spurgeon, 1998) is considered to be one of the main methods effective under uncertainty conditions. The approach is based on keeping exactly a properly chosen constraint by means of high-frequency control switching, and is known as robust and very accurate. Unfortunately, its standard application is restricted: if an output s is to be zeroed, the standard sliding mode may keep s = 0 only if the output’s relative degree is 1 (i.e. if the control appears & explicitly already in s ). High frequency control switching may also cause the dangerous chattering effect (Fridman, 2001, 2003; Lee & Utkin, 2006; Boiko, Fridman & Castellanos, 2004; Boiko, 2005). Consider a smooth dynamic system with a smooth output function s, and let the system be closed by some possibly-dynamical discontinuous feedback. Then, provided (r-1) & are the successive total time derivatives s, s , ..., s continuous functions of the closed-system state-space vari(r-1) & ables, and the set s = s = ... = s = 0 is non-empty and consists locally of Filippov trajectories (Filippov, 1988), (r-1) & the motion on the set s = s = ... = s = 0 is said to exist in r-sliding mode (rth order sliding mode) (Levant, 1993, (r) 2003). The rth derivative s is mostly supposed to be discontinuous or non-existent. & The standard sliding mode is of the first order ( s is discontinuous). Higher-order sliding modes (HOSM) remove the above-mentioned relative-degree restriction and, being properly used, can practically eliminate the chattering. Asymptotically stable HOSMs appear in many systems with traditional sliding-mode control (Fridman, 2001, 2003) and are deliberately introduced in systems with dynamical sliding modes (Sira-Ramírez, 1993). Finite-time

Preprint submitted to Automatica


9 October 2006

The & function ueq is defined from the equation s = 0 and is in& dependent of u . b. for the controller will not && be effective for the simple autonomous linear system s = c + ku. Hence. n s = s(t. s ) providing for s º 0. 2002). 1989) is assumed to be 2. it is sensitive to large noises or small sampling intervals. 2005a) in the simplest case of the sliding order 2. s . g = ¶ ¶u && s ¹0 (2) where the functions g. In particular. (1) where x Î R . being continuous everywhere except the 2& sliding set s = s = 0. && h = s u =0 . Indeed. KM]u. in the case when j is con& tinuous almost everywhere. such controllers are obviously robust with respect to any perturbations preserving (3). Korovin & Levant. 2. Km £ k £ KM.x) + g(t. |c| £ C. && | s u =0 | £ C. and can be also considered as a demonstration of the general homogeneity-based HOSM controller design (Levant. (4) Most 2-sliding controllers may be considered as control& lers for (4) steering s.. u = ueq(t. x). in other words. with j(0. s is the only measured output. the sliding-sector method (Furuta & Pan. x). since most of 2& sliding controllers explicitly use possibly unavailable s or & sign s . Since inclusion (4) does not “remember” the original system (1). 2003) are proved to be valid for general finite-time-convergent homogeneous 2-sliding controllers. x). A number of approaches were developed to avoid the dangerous chattering often featuring standard (1-sliding) modes. 1993. Thus.x)u. s and the dimension n are unknown. In other words. the problem is to find such a feedback & u = j(s. in the vicinity of the 2-sliding mode u is close to the so-called equivalent control ueq(t. producing robust output-feedback control. All contemporary applications are computer-based and use discrete-time sampling. 98. Assume now that (3) holds globally. x denote the time and the state variables. The problem statement Consider a dynamic system of the form & x = a(t. That approach is simple and valid for all 2-sliding homogeneous controllers..strated. the expression for s contains terms with u. s ) is substituted for u in (4). The corresponding generalcase asymptotic accuracies are also calculated. The main features of known 2sliding controllers (Levant. Since ueq is typically unknown.x)u. Fortunately. In particular. the so-called quasi-continuous controllers have better performance. and a few new controllers are developed. The relative degree of the system (Isidori. 1991). 0) = . Note that at least locally (3) is satisfied for any smooth system (1) with the well-defined relative degree 2. some of which feature new advantageous properties. while using its time derivative u as the actual control ((Emelyanov. KM. the system stability is nevertheless difficult to guarantee. The results of this paper were partially presented at a conference (Levant. Suppose that the input-output termed conditions 0 < Km £ ¶ ¶u && s £ KM. Obviously. s ). Following Levant (2005a). 1988). x)/g(t. u is to dominate over u itself.0) ¹ . this paper suggests to use a recently developed robust exact differentiator (Levant. h are some unknown smooth functions. The corresponding asymptotic accuracies are estimated. The controller robustness with respect to sampling noises is especially important. which looks problematic. In particular. s1 ) obtained when the continuity point Preprint submitted to Automatica 2 9 October 2006 . In that case u && has to dominate in the equation for s . in gen&& & eral. or. Differential inclusion (4). no continuous feedback controller can solve the stated problem. smooth functions a. but. In practice the sampled values of the output s are corrupted by noises. The resulting output-feedback controllers preserve the ultimate accuracy and finite-time convergence of the original controllers and do not require any information on the noises. the approach is always valid in & some vicinity of the 2-sliding set s = s = 0. the 2-sliding mode s = 0 is to be established. 3. Regretfully. which means that the right-hand vector set is enlarged in a special way in order to satisfy certain convexity and semi-continuity conditions (Levant. Thus. high-gain control with saturation is used to approximate the sign-function in a boundary layer around the switching manifold (Slotine & Li. any continuous control u = & j(s. u Î R is control.h(t. C] + [Km. Bartolini et al. C > 0. A simple effective procedure previously established only for the twisting controller is proved in this paper to resolve the problem by means of all known and newly-developed 2-sliding controllers. && Calculating the second total time derivative s along the trajectories of (1) achieve that under these conditions && s = h(t. The problem uncertainty prevents it. a set F(s. Historically the first way to get the lacking information was to use finite differences instead of the derivative. 1993. treating the input u of the system & as a new state variable. has to satisfy the equality & j(0. in a vicinity of the set s = 0. Simulation demonstrates the feasibility of new controllers.c/k. whenever s = s = 0 holds. Then (2). x) (Utkin. Levant. Bartolini et al. 1992). 2000) is suitable to control disturbed linear time-invariant systems. (3) hold globally for some Km. System trajectories are supposed to be infinitely extendible in time for any bounded input. (5) is understood here in the Filippov sense (Filippov. Another standard way is to avoid chattering by means of 2-sliding mode control. s to 0 in (preferably) finite time. F being the convex closure of all possible limit & values of j(s1. unfortunately. (3) imply the differential inclusion && s Î [-C. & 1993. 1993. 2005a). The system is understood in the Filippov sense. where t. 2. 2003) as a standard part of all 2-sliding homogeneous controllers. (5) that all the trajectories of (4). (5) converge in finite time to & & the origin s = s = 0 of the phase plane s.x) + b(t. The task is to make the output s vanish in finite time and to keep s º 0 by means of a discontinuous globally-bounded feedback control. 1998).

Explain the Theorem in few words.sign k s. s1 ) approaches (s. and dk D Ì D for any k < 1. D2 and T > 0 such that D2 lies in the interior of D1 and contains the origin. (5) is called further globally uniformly finite-time stable at 0 if it is Lyapunov stable and for any R > 0 exists T > 0 such that any trajectory start& ing within the disk ||(s. s and k > 0. 2005a). 2 A set D is called dilation-retractable if 0 Î D.s| sign(z0 . retracting to the origin. k s ). The corresponding system motion is actually shown to be a finite-time collapse towards the origin. Theorem 1. k s ) º j(s. (8) (9) (10) Preprint submitted to Automatica 3 9 October 2006 . k s. which can be transferred one into another by means 2 & & of a simple linear transformation dk(s. 3. s ) & a ( kt. g1 for any d>0. (5) for almost all t. & & u = (2s . & z1 = . The only difference is that the corresponding motion near the origin requires proportionally less time. & Controller (5) requires availability of s . The following theorems consider the robustness of homogeneous controllers with respect to sampling errors. & have the same solutions). 2003) as follows: u = . Let controller (5) be 2-sliding homogeneous. (5) and controller (5) itself are called 2sliding homogeneous (Levant. z1). such that & any trajectory starting within the disk ||(s. and all trajectories starting at the time 0 within D1 are localized in D2 at the time moment T. For example. The function j is assumed to be a locally bounded Borel-measurable function (actually all functions used in sliding-mode control satisfy that restriction). s (t)) satisfying (4). 2001). C] + [Km.l2 L | z0 . No features of the noises are assumed to be known. Corollary 1.s).. where they are formulated for general homogeneous Filippov differential inclusions with negative homogeneity degrees. Surely these controllers do not solve the stated problem. Obviously. a chain of embedded domains can be constructed. Most of known 2-sliding controllers (Levant. 1993. s ). 1°. s ) = 2 & (k s.j(z0. 1/2 & | s | < g1d with the same positive constants g0. s ) a small homogeneous perturbation of (5) if the difference f – j is a 2-sliding homogeneous function with the homogeneity degree 0. k > 0 (the dilation). The combined timecoordinate transformation Gk: & (t. according to (6). 1/2 1/2 & z0 = . Hence. For example any disk centered at the origin is dilation-retractable. The mapping dk(s.l1 L sign(z0 . any changes of j on such a set do not change the homogeneity properties of the controller. (5) is 2-sliding homogeneous if & & j(k s. s ) = (k s. Differential inclusion (4). The homogeneity of the system (4). s be less than b0d and b1d respectively with some positive constants b0 and b1. 2001). if it is Lyapunov stable and for any R > 0 and e > 0 exists T > 0. (5) is called further globally uniformly asymptotically stable at 0. Then any finite-time-stable 2-sliding-homogeneous controller (5) provides in finite time for keeping the inequalities |s| < g0d. Inclusion (4). s )|| < R enters & the disk ||(s.& & (s1. review of the corresponding theoretical results and numerous references can be found in (Bacciotti & Rosier. That information can be obtained by means of the real-time robust finite-time-convergent exact differentiator (Levant.. k > 0 is called the homogeneity dilation (Bacciotti et al. (5) into the solutions of the transformed inclusion && & s Î [-C. (5) is called further contractive if there are 2 compact sets D1. The homogeneous differential inclusion (4). D1 is dilation-retractable. 2° and 3° are equivalent. k s ). Since the Filippov solutions do not depend on the values of j on any set of the zero measure. s. using the contractivity property 3°. Indeed. Bartolini et al.s) + z1. 2005a). 3°. which are supposed to be some bounded Lebesgue-measurable functions of time of any nature. Let the noise magnitudes of 1/2 & measurements of s.(k s ) )/(|k s| + (k s ) ). (5). 2005a). 2-sliding homogeneity and finite-time stability The general notion of homogeneous differential equation. s )|| < R stabilizes at zero in the time T. (5) means that it is invariant with respect to the transformation (6). It is assumed in this paper that (7) holds for any & s. 1998. Global uniform finite-time stability of the 2-sliding homogeneous controller (5) is robust with respect to small homogeneous perturbations of the controller.s 2 )/(|s| + s 2 ) = 2 2 2 2 & & (2 k s .sign s = . 2°. the following controllers are 2-sliding homogeneous: u = . Note that (7) requires global boundedness of j (excluding possibly a zero-measure set). Differential inclusion (4). Solution is any absolutely continuous vector & function (s(t). it follows from the robustness of property 3°. small in some fixed vicinity of the origin. s )|| < e in the time T to stay there forever. 2003) satisfy these properties. otherwise it is unbounded in any vicinity of 0 and Filippov’s definition is not applicable to (1). k s ). Since 1° implies 2° and 3°. Such a function j is called 2-sliding homogeneous with the homogeneity degree 0. it is sufficient to show that 3° implies 1°. & It is natural to call the controller u = f(s. the behavior of the system trajectories is geometrically the same in all points. The following Theorem is actually true for any homogeneous differential inclusion of a negative homogeneity degree (Levant. Introduce a few auxiliary notions and theorems adopted from (Levant.e. then properties 1°. KM] j(k s. s and k > 0 (i. Thus. k s ) 2 (6) transfers solutions of (4). s ) 2 2 (7) holds for any k > 0. Theorem 2 (Levant. 2005a) if these two differen& tial inclusions are equivalent for any s.

Dsi/t) = j(t s. i.C ] > (r1 . r1 > r2 > 0. in the phase plane. is built. Design of 2-sliding controllers As follows from the previous Section it is sufficient to build a 2-sliding-homogeneous contractive controller. | s | < m1e are established with some positive constants m0.r2 sign(m2 Dsi + l2t |si| sign si). Nevertheless. The long-term motion is determined by the system properties.4 have obvious local analogues in the case when (3) holds only locally. L is the only differentiator parameter to be && tuned.(r1 sign s + r2 sign s ). The parameters of the differentiator li are to be chosen in advance.r1 sign(m1 Dsi + l1t |si| sign si) 1/2 .r2 sign(m2 s + l2|s| sign s). 2.e. mi. Note that in the case. (5) (respectively (1). 2005a). 2005a). achieve the drift controller (Levant. Results similar to Theorems 1 . | s | < g1t are established with some positive constants g0. && 1989). then in the absence of measurement noises controller (11) 2provides in finite time & for keeping the inequalities |s| < g0t . (r1 . Thus. A homogeneous form of the controller with prescribed convergence law (Levant 1993) arises when m1 = m2.r2)KM + C. & There is another way to estimate s with discrete sampling. Indeed. or the corresponding inequalities are established in the case of noisy measurements or discrete sampling. 1998. m1. one can readily check that r1.(10) provides in finite time for & keeping s = s = 0. 4 is the best possible with discontinuous control and the relative degree 2 (Levant. 1/2 (13) provides for the accuracy described in Theorem 4. but with the opposite sign. Its convergence condition is (r1 + r2)Km . Dsi = si.r1 sign (s . If for each i one of the coefficients is zero. Design of such 2-sliding controllers is greatly facilitated by the simple geometry of the 2-dimensional phase plane with & coordinates s.4 are formulated and proved in (Levant. 1993) from (13) (it does not converge with continuous measurements).& Here z0. & Usually the moments when s changes its sign are detected using finite differences. Theorem 4 (Levant. (11)) starting from some vicinity of a 2-sliding point with well-defined relative degree 2 converge in finite time to the 2-sliding mode s º 0.r2)Km > C. s : any smooth curve locally divides the plane in two parts. l2 = 1. Recall that 2-sliding point is a & point where s = s = 0. 1993) & u = . Floquet et al. The popular sub-optimal controller (Bartolini et al. The initial value of s* is 0. 1998). x(ti)). Let controller (5) be 2-sliding homogeneous and finite-time stable. g1. z1 are the estimations of s and s respectively.r1 sign(m1 s +l1|s| sign s) 1/2 & . (8) – (10) or (1).si-1. Theorem 3 (Levant.e. Due to the finite-time convergence of the controllers. g1.s*/2) + r2 sign s*. 2003). The accuracy described in Theorems 3. in particular l1 = 1. m1. 4. and does not have the feedback form (5). Sliding mode conditions being fulfilled at one point of the line. 3. 2 broad sense. l1 + l2 > 0. Also the statements of Theorems 2 . | s | < g1t with some positive constants g0. If s is sampled with a constant sam2 & pling interval t > 0. let u = j(s. then the output-feedback controller (8) . i. ti+1). r2 > 0. Suppose that controller (5) is 2-sliding homogeneous and finite-time stable. l1 = l2: Preprint submitted to Automatica 4 9 October 2006 .1/2 when the switching takes place on & parabolas m1 s + l1|s| sign s = 0 instead of the coordinate axes. 1/2 2 1/2 (11) where si = s (ti. and considering various possible cases. and it is natural to call it 2-sliding homogeneous in the (12) 2 & Drawing the two lines mi s + li|s| sign s = 0. r2 > r1 > 0. Controller (12) may be considered as a generalization of the twisting controller. and it has to satisfy the only condition | s | £ L. Then in the absence of noises all trajectories of (1). the twisting controller (Levant. the homogeneity transformation (6) preserves its trajectories. the 1/2 & inequalities |s| < m0e. i = 1. Indeed. if. L ³ C + KM sup|j| (recall that j is bounded). then a 1-sliding mode can easily be organ1/2 & ized on the line m1 s + l1|s| sign s = 0. Consider special interesting cases of controller (12). & A typical trajectory in the plane s. the inequalities |s| < g0t . Dsi).C > (r1 .r2)KM + C. t Î [ti. the noisy measurements lead to the accuracy provided by Theorems 2. 1993). when ¶¶u s is negative the same controller (5) is to be used. At the same time design of new r-sliding controllers is rather difficult with r > r 2 due to the complicated geometry of R (Levant. s ~ t. 2003) is defined by the formula u = .. r1.4 remain valid for this controller.1. (13). If s is sampled (continuously) with a noise being a Lebesguemeasurable function of time of the magnitude e > 0.. 2005a) for any relative degree r and r-sliding homogeneous controllers. 1a. 2003. s ~ 2 & t . m1 + m2 > 0.5 is a good choice (Levant. the resulting controller satisfies Theorem 1. they automatically hold along the whole line due to the homogeneity properties. (r1 . Similarly. Note that 1-sliding mode provides only for the accuracy proportional to t. especially by its zero dynamics (Isidori.r2) Km > C. A number of known 2-sliding controllers may be considered as particular cases of a generalized 2-sliding homogeneous controller & u = . and its discrete-sampling version u = . Theorems 2 . where s* is the value of s detected at the closest time in & the past when s was 0. 2005b. li ³ 0. mi + li 2 2 2 2 > 0. for example. The corresponding convergence conditions are 2[(r1 + r2)Km . l1 > 0. r2 can always be chosen so that controller (12) be finite-time stable. With m1 = l2 = 0. s is shown in Fig. The control u depends & actually on the whole history of measurements of s and s.

C > b /2 controller (14) provides for the establishment of the finite-time stable 2-sliding mode s º 0. 2. Contraction property of some 2-sliding homogeneous controllers Proposition 2. s < 0. if aKm . & Consider the region We confined by the lines s = ± e && and the trajectories of the differential2equations s = -C + 2 & && Kma with initial conditions s = e /b . Since that moment the trajectory slides along the curve S = 0 towards the origin and reaches it in finite time. the convergence time being uniformly bounded in the disk. 0 £ k £ 1. s ) when j(s. C] .& u = .C < b /2. s ). on the way it inevitably hits the curve S = 0. n The 2-sliding stability analysis is greatly simplified by & the fact that all the trajectories in the plane s. s ) < 0. This controller is a 2-sliding homogeneous analogue of the terminal sliding mode controller originally featuring a singularity at s = 0 (Man et al. KM] sign S + 1 b s |s| . s ). but still aKm > C. Thus. Recall that a region is dila& tion-retractable iff. obtain -1/2 & & S Î [-C. (Fig.a [Km. The boundaries of the regions are the && trajectories of the differential equation s = -C + Kma with && S > 0 and s = C . Since e can be taken arbitrarily small. Define once more S = 1/2 & s + b|s| sign s.C > b /2.. Fig. s ) with & && & && j(s. s ) && & and s = ±C + Kmj(s. terminates sooner or later at the semi& axis s = 0. As follows from Corollary 1 a small violation of the conditions of Proposition 1 preserves the finite-time stabil2 ity of controller (14).Kma with initial conditions s = . Let 2 C < aKm £ C + b /2. These border trajectories cannot be crossed by other paths. s which & pass through a given continuity point of u = j(s. 1b). the trajectory cannot leave the origin. 2b).aKm controller (14) provides for the twisting-like convergence to the finite-time stable 2-sliding mode s º 0. The same reasoning proves the Lyapunov stability of the origin. s ) > 0. and may be often chosen as boundaries of appropriate dilationretractable regions. each trajectory starting from a disk centered at the origin comes to the origin in a finite time. 1/2 & Proof. Indeed. Proof. s = . each trajectory. Obviously. The rule is to take trajectories satisfy&& & && & ing s = C + KMj(s. 1/2 (14) where a = r1 + r2. 2 2 & Checking the condition S sign S < const < 0 in a vicin& ity of each point on the curve S = 0. The corresponding trajectories and dilationretractable sets are shown in Fig. and s = . 1. The same is true for the trajectory starting with S < 0. if aKm . b = l1/m1. s = e Fig. if j is locally Lipschitzian.C + Kmj(s. The trajectories of the inclusion inevitably hit the curve S = 0 due to geometrical reasons. s ) and s = C + & & Kmj(s. 1b). Then with sufficiently small C + b /2 . if u = -a sign S keeps its constant value -a (Fig. This requirement 2 Preprint submitted to Automatica 5 9 October 2006 .e /b . obtain using s = 1/2 -b|s| sign s that the 1-sliding-mode existence condition 2 holds at each point except of the origin. Consider the combined 2 curve 1-2-3-4 (Fig. s ) and s = . The complete elementary proof of the following simple proposition was actually never published. With aKm . No trajectory starting from the origin can leave We. With sufficiently small C + b /2 aKm point 4 is closer to zero than point 1. Differentiating the function S = s + b|s| sign s along the trajectory. k s ).Kma with S < 0. it contains 2 & all the points of the parabolic segment (k s. with each its point (s. Convergence of various 2-sliding homogeneous controllers Proposition 1.C + KMj(s.2e. 1994).a sign( s + b|s| sign s). and s = 2 & C . 2b. s ) are confined between the properly chosen trajectories of the && & homogeneous differential equations s = ±C + KMj(s. starting with S > 0.

. 2a). where a. s ¹ 0 & . aKm . l2 / m2 ).4. (19) & î. = s never fulfilled in practice with r ³ 2.a sign s. especially interesting are the cases when r1 = r2 . then controller (15) provides for the establishment of the finite-time stable 2-sliding mode s º 0. Let a.2 aKm (16) 2 b . The idea of that controller is close to that of the sliding sector control (Furuta et al. The definition of the control u with s = 0 is made by continuity and does not influence the system trajectories. gz(b) > a and g be sufficiently large.b)/(|r| + b) and & r Î ([-C. 2a. Furuta et al. where s M is the maximal value of | s | in the corresponding dilation-retractable set. 2000). b.. Hence. r2. thus there is such positive r1 < b that the trajectories enter the region r > r1. Fridman. Proposition 3. let z(y) be any monotonously growing positive continuous function of a non-negative argument. The corresponding proposition as well as its proof are obvious and are omitted. Hence. Let a. 4r1Km . Enlarging g one com& pels1/2 trajectories to get closer to the parabola s + the b|s| sign s = 0 without increasing the control magnitude (Fig. Let the relative degree of the system (1) be 1. Following is the 2sliding controller from such a family of arbitrary-order sliding controllers (Levant..1r >0 (17) r+b 2 hold for some positive r > b (which is always true with sufficiently large a). 2 | r | +b 1/2 Due to the symmetry of the problem. g > 0. b > 0. the convergence time to the smaller set is estimated by & & & 3 s M /(Kma . & ì u (s.. Fridman. Moreover.n Following is another example of a quasi-continuous 2-sliding controller: ì min{a. featuring control continuous everywhere except of the (r-1) & = 0 itself.. Thus. Therefore. That is exactly condition (17). since it only influences values on a zero-measure set. 2000.e.a. KM]a r -b 2 -1/2 + 1 r sign s) |s| . b > 0. As a result. 2003). Pisano & Usai. Preprint submitted to Automatica 6 9 October 2006 . Since the r-sliding HOSM s = s = . i. Iriarte.C). this geometric condition does not depend on & the placement of point 1 on the axis s .C > 0 and the inequality aKm . & | s | +b | s |1 / 2 1/ 2 2 2 2 2 & Proof. Direct calculation will replace it with an algebraic condition. s).s /|s| . The corresponding dilation-retractable sets are shown in Fig. The convergence time to the smaller set & & is estimated by 2 s M /(2Kmr1 . It is needed to show now that there is r2 > b such that in some vicinity of r = r2 the & inequality r < 0 holds.max[.a sign s.C . Note that no explicit conditions on the parameter choice are obtained in (Levant. The proof is very similar to that of Proposition 3. 1d). Due to the homogeneity of the system. the finite time stability is obtained due to Theorem 1. s = 0 Proposition 4. Boiko.C). this one also provides for the finite-time convergence to the 2-sliding mode and the accuracies corresponding to Theorems 2 .g ( s / | s |1 / 2 +b sign s)}. m1) ¹ (l2. Calculations show that u = a (r . g. Like other listed controllers. then controller (19) provides for the establishment of the finite-time stable 2-sliding mode s º 0. 2006). An important class of HOSM controllers comprises recently proposed so-called quasi-continuous controllers. and the control signal turns out to be continuous. Denote r = . Enlarging g one compels the trajectories to get closer to the 1/2 & parabola s + z(b) |s| sign s = 0 without increasing the control magnitude (Fig. The absolute && value | s | is separated from zero by Kma . m2) and all components are not zero. 2005.-gz(| s |/|s| )sign s +z(b)sign s)]}. with the sliding order r > 1 the general-case trajectory does never hit the(r-1) r-sliding & = 0 is manifold. 2005b): u=-a & s + b | s | sign s . s ¹ 0 u= í V . conditions (16). 1c. the condition s = s = . gb > a. aKm . Also here the discontinuity is concentrated at s = & s = 0.. Chattering attenuation. 2-sliding mode control provides effective tools for the reduction or even practical elimination of the chattering without compromising the benefits of the standard sliding mode (Boiko & Fridman. (2) is replaced by 2 1/2 2 (15) This control is continuous everywhere1/2 except of the ori& gin.C > z(b) /2 .. It vanishes on the parabola s + b|s| sign s = 0. With negative or & small positive r the rotation velocity r is always positive due to (16). 0 < r1 < b < r2 that all the trajectories enter the region between the 1/2 & curves s + ri|s| sign s = 0 and cannot leave it (Fig. Conditions of the proposition can be solved for a. (17) provide for the establishment and keeping of the inequality r1 < r < r2. max[ -a. With sufficiently large a there are such numbers r1.(18) u= í & s=0 î. -¥ < r < ¥. 1d). 2005b). 1991. Thus. 1c). The control vanishes in that region. where s M is the maxi& mal value of | s | in the corresponding dilation-retractable set. it is enough to consider the case s > 0. = s condition has the codimension r. but the resulting expressions are redundantly cumbersome. and the control remains continuous function of time all the time. n A new controller is obtained when (l1. the chattering is significantly reduced. in the presence of measurement errors the & motion takes place in some vicinity of the mode s = s = 0 without entering it. aKm . The standard problem of classical (first order) sliding-mode control is attenuation of the chattering effect (Slotine et al. In that case the trajectory is confined between two parabolic segments (Fig.2C > max(l1 / m1 . C] .C. then the following controller generalizes (18) & & uz = min{a.[Km.forms the convergence condition.C > b /2 .

suppose that s does not change its sign. Indeed.C > 0 (21) & is assumed. which comprises these transient trajec& tory segments. Indeed. The remark that u = -k sign s & can be established only with s s < 0 proves the Lemma.C. which is considered as control. kKm .C. s.x. according & to (23). u moves towards u = -k sign s with | u | ³ min (a. With sufficiently large a any trajectory of the system (1). n Lemma 2. A load of a known mass m moves without friction along the pendulum rod (Fig. & Proof. n Lemma 4. 1998). the control u moves towards the value . s s < 0. Simulation results A number of new controllers from the previous Section are demonstrated here. s. which means & & that | s | ³ kKm . and on certain time intervals u º k or u º -k is kept in 1-sliding mode. (21) that the inequality | s | < & kKm . For example. 3 or 4. Within this region & u = -a sign s. 3a). s ). and the statements of Theorems 1 . (22) & Any listed controller u = f(a. Also the statements of Theorems 2 . Proof. where (22) is really kept. which still features |u| < k. (23) hits in fi& nite time the manifold s = 0 or enters the set s s < 0. There is a specific set Q for each controller. (18). In particular. |u| £ k.x)u. t x t x Assume that with |u| £ k1. It follows from (20).k sign s solve the problem of establishing and keeping s º 0. if this motion does not leave W.4 are true. Differentiating (20) achieve && & s = h1(t. |u| < k. and the corresponding differential inclusion. | u |£ k . h are some unknown smooth functions. in order to overcome the chattering. s ) (Filippov.g sin x + u. (23) hits in finite time the manifold s = 0. Consider u as a new control. it continues to converge to the 2-sliding mode according to the corresponding 2-sliding homogeneous dynamics. As follows from (20).C. The task is to track some function xc given in real time by the angular coordinate x of the rod. Its distance from O equals R(t) and is not measured. s ) can be used here in order to overcome the chattering and improve the sliding accuracy of the standard sliding mode.C1) takes place at the moment when s vanishes. With sufficiently large a that point inevitably belongs to W. Denote S the set defined by the inequalities | s | < & kKm . it is valid for all standard controllers (Levant. Hence. With sufficiently large a any trajectory starting on the manifold s = 0 with |u| £ k enters the invariant set W. the homogeneous dynamics is still in charge. Consider the auxiliary problem. n Once the trajectory enters W. Obviously. n 5.C. | s | < kKm . either converges in finite & time to s = s = 0.k sign s. |u| & £ k implies the global bound | s | £ kKM + C.& s = h(t. There is a vicinity W of the origin within the & strip | s | < kKm .x) + g(t. (15). converge to the origin in finite time.x. s. (19) and the controller parameters be chosen in accordance with the corresponding Propositions 1. s s < 0. Note that though Theorem is not formulated for arbitrary 2-sliding homogeneous controllers. Since all trajectories starting in a closed disk centered at the origin.C for controller (14). In both cases the trajectory hits s = 0. the function h1(t. Lemma 1. adjacent to the axis s = 0. or hits the axis s = 0. & Proof. 2. h1 = h¢ + h¢ ×(a + bu) + ( g¢ + g¢ × (a + bu)) u . then. there are 2 options: from some moment on a trajectory stays out of S.. 1993. s. Thus. or it enters Q.C & & the system is driven by the controller u = f(a. Suppose that s s ³ 0 holds. when (22) holds independently of the control value. Any trajectory of the system (1). This proves the convergence to the 2-sliding mode. Thus.C implies |u| < k.u)| £ C1 . Q is defined by the 1/2 & & inequalities ( s + b|s| sign s)s £ 0. and on the way the trajectory hits the set & s = 0. both if |u| > k or |u| £ k.x. such that any trajectory entering it. the statement of the & Lemma is true. is an invariant compact for the controller u & = f(a. Bartolini et al. s ). which is invariant with respect to the & & controller u = f(a. The system is described by the equation && = . the inequality |u| £ k is established in finite time. or enters Q. within the set | s | < kKm . 1988).u) is bounded: |u|£ k1 sup |h1(t. Consider an academic example of a variable-length pendulum with motions restricted to some vertical plane. Controller (23) keeps |u| £ k. (20) where the functions g. Lemma 3. Let f be anyone of controllers (14).u. An engine transmits a torque u. n The following Lemma is obviously true for any convergent 2-sliding controller. That restriction is true also at the initial point on the axis s = 0. s). Applying now the homogeneity transformation. Any trajectory starting with s = 0 and s ¹ 0 inevi& tably enters the region s s > 0. the set.u) + g(t. define ì. Then with sufficiently large a controller (23) provides for the establishment of the finite-time-stable 2-sliding mode s º 0.4 are valid with sufficiently small noises or sampling intervals.2 x & R 1 1 & x . Any trajectory starting in S either leaves it in finite time. | u |> k & u= í & îf( a. 0 < Km £ g £ KM. (23) Theorem 5. |h| £ C. and lying in the strip S. the set can be retracted into the strip & | s | < kKm . R R mR 2 (24) Preprint submitted to Automatica 7 9 October 2006 . also no trajectory can enter S outside of Q. k1 > k. If & the condition s s < 0 is attained. Simple 2 calculation shows that the inequality |s| £ 1 (kKM + C) 2 & /(aKm . & Proof. Let also the control u = . s.x) u . k). In the presence of small noises and sampling intervals the resulting motion will & take place in a small vicinity of the 2-sliding mode s = s = 0. Thus.

t = 0.xc(0) = . not precise. The assumptions (3) are fulfilled here only locally. Km. being the only method valid for sliding-mode simulation.xc | £ 2.5. The Euler integration method was used. R .036. (28) and the differentiator performance in the presence of a noise with the magnitude 0. Consider a controller of the form (14) u = .01 are demonstrated in Figs. The tracking results of (25). s respectively. xc and &&c be bounded.0001.0×10 with t = 0. s = x-xc . Both main applications of 2-sliding modes are demonstrated: the straight-forward implementation of 2-slidingmode controllers leading to discontinuous control and possibly dangerous chattering. Note that it is the first time when such a twisting-type convergence is demonstrated for this controller. The noise was a periodic nonsmooth function with nonzero Preprint submitted to Automatica 8 9 October 2006 . xc = 0. f) and (25). d. KM is difficult and. Differentiator (25). and 3c respectively.xc | £ the -2 & & & 4.2×10 .c respectively. e) & where z0.0001.5t + 0. Let s = x-xc.xc | £ 0. The phase trajec& tories in the plane s. Note that the 3-sliding-mode controllers (Levant. | s | = | x . 2003. & The initial conditions x(0) = x (0) = 0 were taken.s)+ z1. z0(0) = x(0) . Discontinuous control The controllers include a real-time differentiator and have the form u = j(z0.s). (28) (c. z1). & | s | + | s |1 / 2 (28) sign z0 ).d respectively. The trajectory and 2-sliding tracking performance in the absence of noises are shown in Figs. (26) with noisy measurements Consider the quasi-continuous controller u = . Control (28) is demonstrated in Fig. 4c.55 L sign(z0 .xc | £ 1. (26) is exact for input signals s with second derivative not exceeding 50 in absolute value. inevitably.8×10 . the tracking accuracy being |s| = | x . 3b. (27) (b. | x . The relative degree of the system equals 2. nevertheless the 2-sliding mode s = s = 0 is established here in finite time according to Proposition 2. (26).3 cos 4t. 3f.10. Choosing the controller parameter a the convergence region can be made arbitrarily large.7×10 -5.4×10 . (27) The magnitude of the control is not sufficiently large here & to establish a 1-sliding mode on the curve s +11|s| sign s & = 0. s and the first 0. R . (26). The parameters of the controllers were found by simulation.0. 3: Pendulum and performance of controllers (25). and the standard practical removal of chattering by means of 2-sliding mode.s| sign(z0 . (26).5 cos t .10 & s + | s |1 / 2 sign s . 2005b) are probably more effective in that case.xc | £ 3.81 is the gravitational constant. z1 are real-time estimations of s. s = x-xc x be available. After the sampling step t -4 was reduced from 10 to 10 the resulting accuracies -7 -3 & & & changed to | x . and the controllers to be applied are effective only for some bounded set of initial conditions. but are out of the scope of this paper.1 seconds of the differentiator convergence are shown in Fig.10 sign( z1 + 11| z0| 1/2 Fig. 3e.61 | z0 . 4: Performance of the controller (28). 3d.5 sin 0. The differentiator convergence is demonstrated in Fig. z1(0) = 0.xc | £ 4. In the latter case a redefinition of the output and input are needed. since the direct calculation of the constants C.1×10 which corresponds to Theorem 4. (25). | s | = | x . (25) (26) Fig.8 + 0.xc| £ & & 5. the sampling step t and the integration steps being the same.1 sin 8t + 0. Let & && & 0 < Rm £ R £ RM.-6the corresponding-2 accuracies being |s| = |x .where m = 1 and g = 9. It is seen from the graph that the control remains continuous until the entrance into the 2& sliding mode s = s = 0. Following are the "unknown" functions R and xc considered in the simulation: R = 0.-5 corresponding accuracies being |s| = | x . & z1 = . 1/ 2 & z0 = .

. and they are globally applicable. The graphs of 3-sliding deviations s. when solving practical control problems. Chattering attenuation The controller includes a real-time second-order differentiator (Levant. z2 are real-time estimations of s. when the frequency of the noise varies from 10 to 100000. 2003. s . Substituting the estimations z0.xc -4 £ 4. | s | = | x . In the absence of noises the tracking accu2 racy proportional to t is provided.&&c | £ | x x 4. A real-time robust exact differentiator having been used as a standard part of the 2-sliding controllers. s0 = z0 + z1.26. the quasi-continuous controllers (15).0×10 . (23) which estab& lishes in finite time the 2-sliding mode S = S = 0. design of higher-order sliding controllers is much more difficult due to the higher dimension of the problem (Levant.0×10 . the differentiator is successfully replaced by the first finite difference (Theorem 4). (19) have probably better performance than the standard 2-sliding controllers popular today.330 sign(z2 . Differentiator (29) . where & S = s + s . | s | = | && . u(0) = 0. z2 of the de& && rivatives s. 1/2 1/2 Fig. s obtain the controller & u = min{30.v0)+ z2. z1.average. z1. if the boundedness conditions (3) hold globally. s .u with |u| ³ 10.s)+ z1. (18). Conclusions 2-sliding homogeneity and contractivity are shown to provide for all needed features of 2-sliding mode controllers (Theorems 1-4).v0| & z 2 = . In particular. Construction of controllers is not difficult due to the simplicity of the plane geometry. 1993). Preprint submitted to Automatica 9 9 October 2006 . 2000). whose derivative u remains continuous until the entrance into the 2-sliding & & mode S = S = 0. The differentiator is to be used whenever the sampling step can be taken small. The 2-sliding convergence in the plane S. (32) & u = -30 sign s1 with |u| < 10. . That result does not depend on the noise features. & u = . (33) is constructed so that the overflow be avoided during the computer simulation (or practical implementation). The performance does not significantly change. In the absence of noises the controller provides in finite & time for keeping s + s º 0. 5b. s1 = z1 + z2. s = x-xc .4 | z0 . 5d that s + s º 0 is kept in 2-sliding mode. Note that since (19) is quasi-continuous. Without changing the control values. The graph of u is very similar to Fig.xc | £ 4. The resulting continuous lipschitzian control can be used to keep auxiliary constraints. and. The resulting robust output-feedback controllers preserve the ultimate accuracy of the original 2-sliding controllers with direct measurements of the input derivative (Theorem 3). Due to this equality the accuracies achieved -4at t = 10 are of the same order: |s| = -4 & & & && | x . These results correspond to Theorem 3. 2005b). s respectively. (29) & z1 = v1 = .40 (ln(|s1|/|s0| + 1) × sign s1 + 1/2 ln 2 × sign s0)]} with |u| < 10. (33) (34) where the function z(*) = ln(* + 1) is chosen. (30) (31) & && where z0. the applied controller & (29) – (34) produces the control u. t being a sampling period. The resulting controllers are locally applicable to any uncertain smooth process of relative degree 2. 5: Chattering attenuation The number of such 2-sliding homogeneous controllers is obviously infinite. Unfortunately.v1). The main results known for the known 2-sliding controllers are extended to any finite-time-stable 2-slidinghomogeneous controllers. S is & demonstrated in Fig. significantly increasing the choice of known 2sliding controllers. s and the control u are demonstrated in Figs 5a. 1/ 2 sign(z0 . 2003) 2/ 3 & z0 = v0 = . z1(0) = z2(0) = 0 were taken. s . A simple efficient procedure was developed for the chattering attenuation in the systems with standard (first order) sliding modes based on their replacement by 2-sliding modes (Theorem 5). max[-30. |s1| ³ 100|s0| . In the presence of a measurement noise the tracking accuracy is proportional to the unknown noise magnitude. 3e & && and is omitted. the full single-input-single-output control is achieved based on the input measurements only. That is the best possible accuracy with discontinuous second output derivative (Levant. when the sampling step is sufficiently large compared with the noises.0 | z0 . New finite-time stable 2-sliding controllers were obtained in such a way.0×10 .s| sign(z0 . and one can adjust a controller to his needs. |s1| < 100|s0| . is readily combined with different control technique (Levant et al.c respec& tively. 6. The initial values z0(0) = s(0). As a result the asymptotically & && stable 3-sliding mode s = s = s = 0 is established. Consider a controller of the form (19).13.(31) is exact for input signals s with the third derivative not exceeding 300 in absolute value. It is seen from Fig. At the same time in the practically important case.

Gitizadeh R. International Journal of Control.A. & Wu H.. 586594. 76 (9/10). of Control. 76 (9/10). International Journal of Control. Fossas C. 10 . B.. Ferrara A. Edwards C. & Levant A. (2005). (L. Lianos D.. (1993). Sliding Mode Control: Theory and Applications. Second order sliding mode control of a diesel engine. 906-912. Higher-order sliding mode stabilization for a class of nonholonomic perturbed systems. the Netherlands: Kluwer. Applied Nonlinear Control. Man Z. Shkolnikov I. Boiko I. (2004). & Usai E. editors. Chattering analysis in sliding mode systems with inertial sensors. Fridman L. 946-950. L.V. 5(4). J. W. Control 5(4).. (2003). Levant. (1998). Dynamic Second-Order Sliding Mode Control of the Hovercraft Vessel. Las-Vegas. & Fridman. (2005). (2003). Control and Dynamics. Analysis of second order sliding mode algorithms in the frequency domain. & Rosier.. (2001).) (1994). 43(2). Yaesh I. G. Variable structure control with sliding sector. differentiation and output-feedback control. Navigation. Aircraft pitch control via second order sliding technique. on Automatic Control. IEEE Transactions on Automatic Control. L. & Usai E. IEEE Transactions on Automatic Control.. Switched chattering control vs. 498-504.R..K. An Asymptotic Second-Order Smooth Sliding Mode Control. International Journal of Control. Sira-Ramírez H.A.I. E. K... (1991). Shkolnikov. Inc. Boiko I. 76(9/10).B. I. (2001).. (2005b). Shtessel Y. & Li.I. of Control. 49(6).. Chatterring Analysis. & Perruquetti W. Automat. E. sc. Sc.. backlash/friction phenomena in electrical servo-motors.. Taylor & Francis. 10(6). Slotine.1017. (1998). 211-228.-J. Boiko I. Furuta. Levant A. 29(11). 823-830. IEEE Transactions on Automatic Control. Analysis of Chattering in Continuous Sliding-Mode Controllers. Dynamics and Control. 833839. (2000). (1993). Fridman L. Isidori A. Automatica. E. 1247-1263. A. 1812-1816.. Ferrara. L. Asian J. Bartolini. & Usai. 241-246. Dordrecht. Iriarte R. In Edwards. Levant A. AIAA paper No. Proceedings of AIAA Guidance.A.. 379-384. & Spurgeon S. and Control Conference. Utkin V. Lect.. Intern.. Khan M. Differential Equations. Higherorder sliding modes in control systems. & Shkolnikov. Robust Missile Autopilot Design via High-Order Sliding Mode Control. Aeronautical and space vehicle control in dynamic sliding manifolds. Y. Lecture notes in control and inf. 39(12). Fridman L.-P. & Utkin V. Proc.. of the 42th IEEE Conference on Decision and Control. Nevada.. Construction principles of output-feedback 2sliding mode design.. 1260 –1265. Boiko I. 50(9). (2005a). International Journal of Control. Homogeneity approach to high-order sliding mode design. 860-865.K. 50(11).I. Chattering avoidance by second-order sliding mode control. Fridman. K. & Castellanos M. Bartolini. 924-941. Liapunov functions and stability in control theory. 875-892. Variable Structure and Lyapunov Control. (1994). (2003). Emelyanov S.D. Sliding Modes in Optimization and Control Problems. Analysis of sliding modes in the frequency domain. Robust exact differentiation via sliding mode technique. Control. 76 (9/10).K. Levant A.277 –296. New York: Springer Verlag. (1998).. Orlov Y. Denver... Berlin: Springer-Verlag. (2006). An averaging approach to chattering. Springer Verlag. IEEE Transactions On Control Systems Technology... 23(4). A survey of applications of second-order sliding mode control to mechanical systems. Automatica. 614-619. Nonlinear Control Systems. (2003). A. (1988). (2003). (2000). Preprint submitted to Automatica 10 9 October 2006 . New York: Springer Verlag. (2002). C. CO. IEEE Trans. 1442-1446. London. A. Advances in Variable Structure and Sliding Mode Control.References Bacciotti.Z. (1992). Fridman. 34(3). A robust MIMO terminal sliding mode control for rigid robotic manipulators. Barbot J. 1000 . (2002). of Guidance. Automatica 36.. IEEE Trans. 267.967. International Journal of Control. (2003). 1039-1061. Multi-input secondorder sliding-mode hybrid control of constrained manipulators. (Ed. second edition.. & Ben-Asher J. 334. T. (1989). Goh K.. Automatica. E.B. Pridor A.. (2000). & Thies. 46(8). I. 76(9/10). December 10-13. Floquet T. Aguilar.F. Levantovsky) (1993).V. 1627-1647. & Punta. 41(5). Punta. (2004).. J. A.J. (2003).. 57(5). Levant A. Springer. J. Lee H. On the dynamical sliding mode control of nonlinear systems. 1077 –1083. C. Levant A. Control. 107-121. Filippov A. 2000-3968. International Journal of Control. Shtessel.S. International Journal of Control. 39. & Pan Y. Quasi-continuous high-order sliding-mode controllers. Differential Equations with Discontinuous Right-Hand Side. Parameter tuning of second-order sliding mode controllers for linear plats with dynamic actuators.B. 58(6). (2000). Korovin S. & Spurgeon S. Sliding order and sliding accuracy in sliding mode control.. E. 2464-2468. L.. 959 . Pisano A. Asian J. 78(10). Notes in Control and Inf. Pisano.. J. Automat. London: Prentice-Hall.. Higher-order sliding modes. (2006). 42(5). Automatica.P. G. Shtessel Y. L.. Bartolini G. 969-981.. Zinober A. & Brown M. IEEE Trans. Paplinski A. Levant A. Sira-Ramírez H.. & Cadiou. A..

Sign up to vote on this title
UsefulNot useful