Dynamic sliding mode control design

A.J. Koshkouei, K.J. Burnham and A.S.I. Zinober Abstract: Dynamic sliding mode control and higher order sliding mode are studied. Dynamic sliding mode control adds additional dynamics, which can be considered as compensators. The sliding system with compensators is an augmented system. These compensators (extra dynamics) are designed for achieving and/or improving the system stability, hence obtaining desired system behaviour and performance. Higher order sliding mode control and dynamic sliding mode control yield more accuracy and also reduce and/or remove the chattering resulting from the high frequency switching of the control. It is proved that certain J-trajectories reach a sliding mode in a nite time. A sliding mode differentiator is also considered.

Introduction

Sliding mode control (SMC) has widely been extended to incorporate new techniques, such as higher-order sliding mode control (HOSMC) [1 3] and dynamic sliding mode control (DSMC) [4 9]. These techniques retain the main advantages of SMC and also yield more accuracy. These techniques can also be applied to design observers to differentiate signals achieving robustness in the absence of noise. SMC utilises a high frequency switching control signal to enforce the system trajectories onto a surface, the so-called sliding surface (or hyperplane), after a nite time and remain within the vicinity of the sliding surface towards the equilibrium point thereafter [8]. The sliding surface is designed to achieve desired specications. SMC is robust with respect to matched internal and external disturbances. However, undesired chattering produced by the high frequency switching of the control may be considered a problem for implementing the sliding mode control methods for some real applications. Methods have been presented to reduce the chattering, for instance the continuous approximation technique [8]. Another way is to use HOSMC control. A drawback of continuous approximation methods is the reduction of the accuracy of the system and the sliding mode stability. SMC techniques are applicable to any minimum phase system with relative degree less than the system order. There are some SMC techniques for stabilisation of non-minimum phase systems including use of DSMC, for stabilising the internal dynamics when the output tracking error tends asymptotically to zero in the sliding mode [6, 7]. DSMC has received attention in recent years [4, 6 9]. Introducing extra dynamics into a sliding surface helps to solve many difculties in practice, such as ight control design and timescale separation of control loops in a

multi-loop system [5, 6]; replacement of a state observer to achieve stability under incomplete information about actuator dynamics [10]; and, even accommodation of unmatched disturbances extending the system state space into the exogenous states of an unknown signal modelled by linear dynamics. This method can be applied to the nonminimum phase tracking problem, e.g. stabilisation in a dynamic sliding manifold of tracking error dynamics together with unstable internal dynamics, plus unmatched input exogenous dynamics with insufcient information about states of this composite system [6]. DSMC provides stability to the internal states and asymptotic stability to the states of the tracking error dynamics. Young and Ozguner [9] and Koshkouei and Zinober [4] have designed compensators using the optimal control and realisation methods for linear systems in the sliding mode. The sliding system with a compensator (extra dynamics) is an augmented system which is a higher-order system compared with the original system. However, the designed compensators may not only improve the stability of the sliding system but also yield desired performance and characteristics. In this paper a non-linear compensator for linear and non-linear systems is designed. 2 Higher-order sliding mode control

HOSMC is a way to improve the accuracy of the sliding mode and remove chattering. Consider a system of the form _ x Ax; t Bx; tu 1

where x 2 Rn is the state and the scalar control u 2 R: Ax; t and Bx; t are smooth vector elds. Dene the sliding function as s sx; t: Suppose the systems relative degree r with respect to s sx; t is constant and known. Then the system (1) is transferred to new coordinates _ _ sr f t; s; s; . . . sr1 ; z gt; s; s; . . . ; sr1 ; zu r1 _ _ _ z ft; s; s; . . . ; s ; z ct; s; s; . . . ; sr1 ; z 2 where z z1 ; . . . znr1 [11]. The rth-order sliding mode exists if there is a control u such that the zero dynamic equations _ s s sr1 0
IEE Proc.-Control Theory Appl., Vol. 152, No. 4, July 2005

q IEE, 2005 IEE Proceedings online no. 20055133 doi: 10.1049/ip-cta:20055133 Paper rst received 28th July and in revised form 19th October 2004 A.J. Koshkouei and K.J. Burnham are with the Control Theory and Applications Centre, Coventry University, Coventry, UK A.S.I. Zinober is with the Department of Applied Mathematics, University of Shefeld, Shefeld, UK 392

are satised. The control during the sliding mode, the so-called equivalent control, is [2] ueq f t ; z=gt; z 3

Substituting (3) into (1) yields r-sliding mode motion, which is the zero dynamics of the system (1). Moreover, the system is stable if the system is minimum phase. The results of second-order SMC are interesting; in particular, it can be successfully applied to real problems. Various methods for designing second-order SMC and applications have been presented [1, 2]. Attention is now focused on the second-order sliding mode systems. Suppose the system (1) has relative degree one with respect to s st; x: Then the system (2) is _ _ s f t; s; s gt; s; su 4

In fact, for a sliding mode linear system (i.e. the system during the sliding mode, s 0) the desired system response and performance can be achieved by selecting a set of prespecied eigenvalues. Non-linear sliding mode dynamics are now introduced. By dening such dynamics, two different sliding surfaces are obtained. However, there is a close relationship between them. A sliding mode control is designed using the new component. Dene the J non-linear dynamic sliding mode _ w ajsj0:5 sgns bjw sj0:5 sgnw s J ws 6

The additional dynamics yields greater accuracy of the control design and suitable response behaviour and performance. These dynamics act as a compensator for the system. The resulting controller has the features of traditional SMC such as insensitivity to matched disturbances and nonlinearities, and a classical dynamic compensator with accommodation of unmatched disturbances. For simplicity, consider a system with the following sliding dynamics s f t; s u 5

For the existence of sliding mode control it is sufcient to show that the trajectories reach the sliding surface s 0 in a nite time and remain on it thereafter. 3 Dynamic sliding mode control

with a > 0; b > 0 and a 6 b [5, 6]. w is an error variable of two sliding mode variables s and J. Using the second-order sliding mode, a sliding mode control u is designed such that _ s s 0: The problem is to nd a suitable sliding mode control to guarantee a J-sliding mode (or s-sliding mode). The following theorem guarantees the existence of the sliding mode with sliding mode control u r sgnJ where r is a suitably large positive real number. In many practical problems, one needs to know the relationship between the J- and s-dynamics. In fact, it is desired that the J-sliding mode reaches the sliding surface J 0 faster than s-dynamics. In this case, lim J=s 0; where tJ is the t!tJ reaching time to the sliding surface J 0: The following theorem yields this relationship. Theorem 1: Consider the sliding dynamics (5) and (6). Let a > 0; b > 0 and a 6 b: Then the following statements are implied. (i) The J-dynamics sliding mode exists. (ii) The J-sliding mode occurs if and only if the s-sliding mode exists. (iii) The J-dynamics reaches and remains on the sliding surface J=0 before the trajectories of the s-dynamics hit the sliding surface s 0 if and only if  2 b js0j jJ0j a

The system in the sliding mode may need some additional dynamics to improve the system stability and the sliding mode stability as well as obtaining the desired system response and behaviour. It may also require a controller to be designed such that the output of an uncertain SISO dynamic system tracks some real-time measured signal. When the output is measurable, the convergence time is required to be nite so that the tracking is robust with respect to measurement errors and exact in their absence. In order to solve such a problem some additional assumptions may still be needed.

3.1 Dynamic sliding surface

A dynamic sliding function s is dened as a linear operator, which has a realisation as a linear time-invariant dynamic system _ w Fw G1 e G2 x s Cw He Kx where x is the state of the original system and e x xd is the error variable with xd as a desired state. w is a state resulting from realisation of the operator s: F, G1 ; G2 ; C, H and K are matrices, with compatible dimensions, which show the relationship between the states. For minimum-phase systems G2 0 and K 0 whilst for non-minimum-phase systems they are non-zero [7].

Proof: (i) From (5) and (6) one can obtain   b a _ 1 J F u J 2 jJj0:5 jsj0:5 where    2 a ab  J 0:5 F sgns   sgnJ f t; s: 2 s 2

Assume that jFj L: Consider the sliding mode control u r sgnJ with r > L: Then (7) yields   b a _ 1 J F r sgnJ 8 J 2 jJj0:5 jsj0:5 Let R r FsgnJ: Then R 2 r L r LsgnJ and substituting into (8) implies   b a 1 _ J J R sgnJ 2 jJj0:5 jsj0:5

3.2 Dynamic sliding mode

The dynamic sliding mode (DSM) in the dynamically extended state space is dened as s 0: The system output tracks the desired value if s 0 so that matched uncertainties and disturbance do not affect the tracking.
IEE Proc.-Control Theory Appl., Vol. 152, No. 4, July 2005

9
393

In the Fillipov sense,   b a _ 2 1 J r L r LsgnJ J 2 jJj0:5 jsj0:5

10

This process can be applied a nite number of times to obtain the desired higher-order differentiation. Shtessel and Shkolnikov [5, 6] have introduced the following differentiator _ w ajej0:5 sgne bjJj0:5 sgnJ J we ^ eyx ^ x K sgnJ 16

where the right-hand side is a differential inclusion J: Suppose ajs0j0:5 6 bjJ0j0:5 : Consider the trajectory G1 & _ r L if bjs0j0:5 ajJ0j0:5 J0 > 0 11 0:5 0:5 _ r L if bjs0j ajJ0j J0 < 0 and the trajectory G2 & _ r L if bjs0j0:5 ajJ0j0:5 J0 > 0 _ r L if bjs0j0:5 ajJ0j0:5 J0 < 0

12

_ ^ ^ where x is an estimate of x. Let y1 x: The new differentiator can be dened as _ w1 a1 je1 j0:5 sgne1 b1 jJ1 j0:5 sgnJ1 J 1 w1 e 1 ^ e1 y1 y1 ^ y1 K1 sgnJ1 17

Any trajectory G between the two trajectories G1 and G2 ; _ crosses the J-axis (i.e. when bjs0j0:5 ajJ0j0:5 ) at _ _ _ _ J1 J2 ; . . . so that for any i ! 0; j Ji1 j j Ji j: Therefore, _ _ i1 fj Ji jg1 is a non-negative decreasing sequence. So lim J 0 t!1 and the second-order J-sliding mode occurs. (ii) From (6) _ _ s ajsj0:5 sgns J bjJj0:5 sgnJ _ _ Therefore, s ajsj0:5 sgns 0 if and only if J bjJj0:5 1:5 _ sgnJ 0: This is equivalent to ss ajsj if and only _ _ _ if JJ bjJj1:5 : So ss 0 if and only if JJ 0: _ _ (iii) Assume that s ajsj0:5 sgns 0 then s ! 0 and s ! 0 in a nite time. So ts js0j0:5 2a 13 0 one can see that

_ which can be considered as a lter for the estimation of x: In this way, a nite series of lters can be produced. Theorem 1 implies that for any i ! 0; the ei - and Ji -sliding mode trajectories converge to the sliding surfaces ei 0 and Ji 0 in nite time and if an appropriate condition is satised, the Ji -dynamics converge faster than ei -dynamics. 5 Example

Consider a system that can be converted to the sliding dynamics (5) s f t; s u where f t; s 2s2 s 2 sin2t 0:5 and u K sgnJ This may arise from a system with tracking signal 2 sin2t 0:5: Consider the non-linear dynamic sliding mode (6) _ w ajsj0:5 sgns bjw sj0:5 sgnw s J ws with a > 0; b > 0 and a 6 b: According to theorem 1, the sliding mode J- and s-dynamics exist. Assume that a > 0; b > 0 a 6 b and jJ0j b=a2 js0j: Then the J-dynamics trajectories reach the sliding surface J 0 before the s-dynamics trajectories hit the sliding surface s 0: For simulation select a 1; b 2; J0 0:1; _ s0 1 and s0 0:5: The condition (iii) of theorem 1 is satised:  2 b js0j 4 jJ0j 0:1 a Therefore, the J-dynamics are faster than s-dynamics. The simulations also show this result. The reaching time of the J-sliding mode is less than 0.4 whilst the s-sliding mode reaching time is larger than 1.7 (see Fig. 1). Figure 1 also _ shows that s converges to zero at nite time ts % 2; and also _ illustrates the behaviour of the sliding mode control and the sliding mode reaching time; thereafter the s-trajectories remaining on the sliding surface s 0: Figure 2 illustrates the behaviour of the non-linear function f t; s with respect to time, which shows that the non-linear function f t; s
IEE Proc.-Control Theory Appl., Vol. 152, No. 4, July 2005

_ Consequently, from J bjJj0:5 sgnJ _ ! 0 and J ! 0 in a nite time and J tJ jJ0j0:5 2b

This completes the proof of (iii). Furthermore, if for the system (4), the conditions _ (i) s ! 0 and s ! 0 when t ! 1 (ii) lim f t; 0; 0 lim gt; 0; 0 0
t!1 t!1

are satised, then lim s 0:

t!1

Estimation of differentiation using DSMC

A DSMC differentiator is presented using the theory in Section 3. A SMC differentiator was introduced by Levant [3]. The Levant differentiator is as follows: let y x  where  is a Gaussian noise and y is a measurable variable. _ _ An estimation of x is required. An estimate of x is dened by Z t _ ^ sgnesds xt ajetj0:5 sgnet b 14 0 ^ et yt xt _ ^ ^ where x is an estimate of x. Select x1 x: It is possible to estimate a differentiator for x1 as _ ^ x1 t aje1 tj0:5 sgne1 t Z t b sgne1 sds
0

15

^ e1 t y1 t x1 t ^ where x1 is an estimate of x1 : In fact, (15) yields an estimation of the second-order differentiation of x.

394

Fig. 1 Response of sliding dynamics with discontinuous control, u 15 sign (J) when a 1 and b 2
a b c d Sliding function s Sliding function J Function s _ Action of the discontinuous control u

Fig. 3 Response of sliding dynamics with discontinuous control, u 15 sign (J) when a b ( 2), and the initial conditions are s (0) J(0)
a b c d Sliding function s Sliding function J Function s Action of the discontinuous control u

Fig. 2 Behaviour of function f (s, t) for a 1 and b 2

does not tend to zero when t ! 1: However, since s 0; for t > 0:4; f t; s 2 sin2t 0:5: When a b; and the initial conditions are s0 J0; then the behaviour of the J-dynamics and s-dynamics coincide and the sliding system is marginally stable (see Fig. 3). In Fig. 3, the lower plot of the second column, depicts the control action when a b and s0 J0: The switching between two control values is not very fast in comparison with the case when the discontinuous control, u 15 signJ with a 1; b 2 is applied (see Fig. 1). In fact, the control is retained as a constant value, K; for a while and then switches to another value, K: This process repeatedly occurs. For example, if the control is K for a certain time, then it is switched to K and after another certain period of time, the control is switched to K again. The control is a rectangular signal with constant amplitude and a different width. Its width depends on the J-dynamics behaviour periods. The behaviour of the
IEE Proc.-Control Theory Appl., Vol. 152, No. 4, July 2005

Fig. 4 Behaviour of function f (s, t) for a b, and the initial conditions s(0) J(0)

function f t; s with these conditions is shown in Fig. 4, which is completely different from Fig. 2. In this case f t; s 2s2 s 2 sin2t 0:5 6 Conclusions

Dynamic and higher-order sliding mode controls have been studied in this paper. DSMC is a technique for improving and=or achieving the system stability or desired behaviour, by designing compensators. This paper has presented some conditions for reaching trajectories to the appropriate sliding surfaces. The prediction of the behaviour of different sliding mode dynamics is important for designing a sliding
395

mode control and for achieving the sliding mode stability and furthermore, the system stability. Using the main theorem in this paper, the DSMC differentiator has also been studied. 7 References

1 Bartolini, G., Pisano, A., Punta, E., and Usai, E.: A survey of applications of second-order sliding mode control to mechanical systems, Int. J. Control, 2003, 76, pp. 875892 2 Levant, A.: Higher-order sliding modes, differentiation and outputfeedback control, Int. J. Control, 2003, 76, pp. 924 941 3 Levant, A.: Sliding order and sliding accuracy in sliding mode control, Int. J. Control, 1993, 58, pp. 12471263 4 Koshkouei, A.J., and Zinober, A.S.I.: Robust frequency shaping sliding mode control, IEE Proc. Control Theory Appl., 2000, 147, pp. 312320

5 Shtessel, Y., and Shkolnikov, I.: Aeronautical and space vehicle control in dynamic sliding manifolds, Int. J. Control, 2003, 76, pp. 10001017 6 Shkolnikov, I., and Shtessel, Y.: Tracking a class of non-minimum phase systems with nonlinear internal dynamics via sliding mode control using method of system center, Automatica, 2002, 38, pp. 837 842 7 Shtessel, Y., Zinober, A.S.I., and Shkolnikov, I.: Sliding mode control of boost and buck-boost power converters using the dynamic sliding manifold, Int. J. Robust Nonlinear Control, 2003, 13, pp. 1285 1298 8 Zinober, A.S.I. (Ed.): Variable Structure and Lyapunov Control, Lecture notes in control and information sciences (Springer-Verlag, Berlin, 1994) 9 Young, K.D., and Ozguner, U.: Frequency shaping compensator design for sliding mode, Int. J. Control, 1993, 57, pp. 10051019 10 Krupp, D., and Shtessel, Y.: Chattering-free sliding mode control with unmodeled dynamics. Proc. American Control Conf., San Diego, CA, 1999, pp. 530534 11 Isidori, A.: Nonlinear control systems (Springer Verlag, New York, 1995, 3rd edn.)

396

IEE Proc.-Control Theory Appl., Vol. 152, No. 4, July 2005