Multivariable Continuous Integral Sliding Mode Control

Shyam Kamal, Asif Chalanga, Ramesh Kumar P. and B. Bandyopadhyay

Abstract— This paper proposes multivariable integral sliding systems the resulting control input become discontinuous in
mode controller for any Multi Input Multi Output (MIMO) nature which results in chattering. Chattering will lead to
system. The conventional integral sliding mode controller con- unnecessary wear and tear on the actuator components of
sists of a nominal control and a discontinuous feedback control
such that the overall control is discontinuous in nature. The the system and reduce the life of actuators [3].
discontinuity in feedback control is undesirable for practical Recently, continuous integral sliding mode is reported in
applications due to chattering. In this paper the existing integral the literature, which has been able to removed the drawback
sliding mode control law for MIMO systems with matched of discontinuous integral sliding mode [12]- [14]. However,
disturbances is modified by replacing the discontinuous part it is only able to generates the absolutely continuous signal
by a multivariable super twisting algorithm. This proposed
controller is continuous in nature due to the combinations of in the case of decoupled system. But in reality most of the
two continuous controls and also disturbance can be rejected engineering and natural systems are very much complicated
successfully. The effectiveness of the proposed control method and it is closely represented by the multi input multi output
has been demonstrated with the simulation results. (MIMO) system. Hence, it is essential to extend the concept
Index Terms— Continuous control, integral sliding mode, of continuous integral sliding mode for the case of multi-
higher order sliding mode, finite time stability.
variable systems.
I. I NTRODUCTION To fulfilling the above mentioned goal, we use the concept
of higher order sliding mode (HOSM) control which has
The sliding mode controller is a class of variable struc-
been developed to overcome the problem of chattering and
ture control, which has been successfully implemented in
relative degree restriction [15]. The HOSM controllers are
many electromechanical systems for improving their perfor-
effective in extending the good properties of classical sliding
mance and achieving robustness against disturbances [1],
mode controllers to systems with higher relative degree and
[2]. The main attractions of sliding mode controller are
keeping the main advantages of the standard sliding mode
invariance to matched uncertainties, model order reduction,
control. In the case of r th order sliding mode control the
simplicity in design and robustness against perturbations.
objective is to keep the sliding variable σ and its first r −
The classical sliding mode control is a powerful tool for
1 time derivatives to zero through a discontinuous function
the control of large classes of systems with matched un-
acting on the r th time derivative of the sliding variable [16].
certainty/perturbations. However this control scheme suffers
Among all the above mentioned higher order sliding
from some problems like relative degree restriction, no
mode controller, the super twisting can generate absolutely
robustness during reaching phase and chattering effect [2],
continuous control for relative degree one system using the
[3].
information of only output [17]- [20] . This exceptional
To overcome some above mentioned draw back of con-
feature of super twisting gives a lot of new dimensions in
ventional sliding mode control, a new class of control is
the literature of robust control, observation and estimation.
evolved which is called integral sliding mode control [4].
Recently well known super twisting has been extended for
The concept of integral sliding mode control concentrates on
the multivariable case [21]. However, it is only restricted
the robustness of the system in the entire phase rather than
to system with vector relative degree [22] one with respect
only in the sliding phase. Hence the robustness of the system
to output, to generates the absolutely finite time continuous
can be guaranteed through the entire response of the system
control signal.
starting from any initial conditions as explained in [4]. The
If we want to use multivariable super twisting for the
integral sliding mode control is a combination of two control,
higher vector relative degree system [22], then again we
a nominal control which decides the desired performance
have to define multidimensional sliding manifold. In this
of the system and a discontinuous control which rejects
case only sliding variables can be steered to zero in finite
the disturbance and model uncertainty. The nominal control
time, however states variable is still asymptotically stable.
methods are conventional proportional-integral plus deriva-
Above restrictions motivates us to further look in the area of
tive (PID) control, composite non linear feedback (CNF),
multivariable continuous higher order sliding mode.
optimal linear quadratic regulator (LQR) etc [4]- [11]. Even
Hence the main goal of this paper can be stated in two
though this method ensures a better performance of the
fold, first define the multivariable continuous integral sliding
Shyam Kamal, Asif Chalanga, Ramesh Kumar P. and B. Bandyopad- mode and if possible then use the concept of it to define
hyay is with Systems and Control Engineering Indian Institute of Tech- multivariable continuous higher order sliding mode. One can
nology Bombay Mumbai– 400076, India. shyam@[Link],
asif@[Link], rameshkumar@[Link] see the second goal of the paper can be easily achieved if
and bijnan@[Link] the nominal control of integral sliding mode is from either
finite time or fixed time control family [23]. MIMO system (1) in finite time. Substituting u from (2) into
(1), one can further write
Main Contributions x
ẋ = −k1 1/2
+ v − k2 x + d1 ||x|| + d2
This paper proposes a novel and simple solution to the x
chattering minimization problem, which has been frequently x
v̇ = −k3 − k4 x (3)
reported in the case of discontinuous integral sliding mode. x
To the best of the authors knowledge, the proposal of Defining z = v + d2 , one can rewrite (3)
continuous sliding mode control within the framework of x
ISMC for MIMO has not been reported in the literature. ẋ = −k1 1/2
− k2 x + d1 ||x|| + z
Here, the integral sliding-mode control law is modified for x
x
more generalized systems with matching disturbance by ż = −k3 − k4 x + ḋ2 (4)
replacing the discontinuous part of the feedback control by a x
multivariable super-twisting control. Replacement is possible Following theorem gives the finite time convergence of the
due to the unique feature of a disturbance observation state variables x and ẋ, as stated in [21] :
property of the multivariable super twisting algorithm. The Theorem 1: [21] For the system (4), there exists a range
proposed controller is continuous due to the combination of of values for the gains k i , i = 1, · · · , 4
two continuous controls. Hence, chattering is alleviated and  √

 k1 > 2δ2
there by the ISMCs scope for MIMO system gets widened 

k2 = 2δ1
and becomes useful for practical system.  
2δ22 9/16(k1 δ1 )2 1/2k12 δ1 −2k12 k2 +k2 δ2

 k3 > max 3δ 2 + 2 ,
k1 k2 (k2 −2δ1 )
+ (k2 −2δ1 )


Organization of Paper  2k22 δ1 +1/4k2 δ12
k4 > ββ12 + 2k22 + 32 k2 δ1 , α2 (kα2 −2δ
1
1)
+ (k2 −2δ1 )
The organization of the paper is as follows: In Section (5)
2
II discusses the disturbance observation property of the where β1 := 32 k12 k2 + 3δ2 k2 , β2 := k3 k12 − 2δ22 − 3δ2 k12 ,
2 2
multivariable super twisting algorithm. In Section III a new α1 := 9/16(k1 δ1 ) k(k
2
2 +1/2δ1 )
and α2 := k2 (k3 + 2k12 − δ2 ) −
2
proposal of continuous integral SMC for MIMO system 2

is presented. Section IV discusses the simulation results (2k3 + 1/2k12 )δ1 − 9/16 (k1kδ21 ) , such that the variables x, ẋ
followed by concluding remarks in Section V. and z are forced to zero in finite time and remain zero for
all subsequent time.
II. M ULTIVARIABLE S UPER TWISTING ALGORITHM AS Once the states reaches the origin then from (4), z = v+d 2 =
DISTURBANCE OBSERVER 0,

The super twisting algorithm has been developed to con- d2 = −v (6)

trol systems with relative degree one with respect to output in The above property leads to use the multivariable super
order to avoid chattering. It generates absolutely continuous twisting algorithm as a controller as well as a disturbance
control signal which ensures all the main properties of first observer. In the next section we are going to propose
order sliding mode control, provided the first derivative of continuous integral sliding mode control for MIMO system
the matched disturbance should be bounded. using the above mentioned disturbance observation property.
The multivariable super twisting algorithm is an extension
of the well known super twisting algorithm. It can be used III. C ONTINUOUS I NTEGRAL SMC FOR MIMO S YSTEM
for the control of MIMO systems, without decoupling the The integral sliding mode control has two parts viz.
system states. Consider the following MIMO system, (i) nominal control (u nominal ) and (ii) a continuous control
(ucontinuous ) or discontinuous control (as suggested by Utkin).
ẋ = B(u + d) (1) The nominal control is designed for the system without
disturbance to have a desired trajectory. The design of
where x ∈ Rm×1 , B ∈ Rm×m is assumed to be known unominal and ucontinuous are completely independent. It may
and invertible, u ∈ R m represents control and d ∈ R m be noted that the once trajectories on the sliding surface
represents unknown but bounded uncertainties/perturbations, (i.e; x = 0 as discussed in disturbance observation property
also d = d1 ||x|| + d2 , where |d1 | ≤ δ1 and |ḋ2 | ≤ δ2 . The of multivariable super twisting), continuous control acts
multivariable super twisting control u can be written as like a disturbance observer and its value is the negative

 of the disturbance. So, when u = u nominal + ucontinuous is
x
u = B−1 −k1 1/2
+ v − k2 x applied to the system having disturbance, u continuous rejects
x the disturbance and the desired trajectory is obtained by
x the application of u nominal . The above procedure is explained
v̇ = −k3 − k4 x (2)
x mathematically as:
Consider the system
where ki > 0, i = 1, · · · , 4 are the properly chosen scalars
such that the above mentioned controller is able to stabilized ẋ = Ax + B(u + d) (7)
where x ∈ Rn×1 , A ∈ Rn×n , B ∈ Rn×m , u ∈ Rm×1 , variable. However, to start the second order sliding mode
d ∈ Rm×1 are the state, system matrix, input matrix, control from initial instant the initial value of s and ṡ must be zero.
and disturbance respectively. It might be possible that system It is clear from (12), initial value of transformed variable z
matrix and input matrix are linear/nonlinear depending upon contains the initial value of the disturbance d. It is always
system characteristic. Control input for the system (7) is hard to predict the initial condition of disturbance d, except
given by u = unominal + ucontinuous , where unominal is nominal some particular cases where initially there is no disturbance,
control and u continuous is multivariable super twisting control. for example fault detection problems where initially there is
The nominal control u nominal is designed to achieve the spec- no fault means d = 0. In this particular case, both s and ṡ are
ified performance, when system is free from disturbances. zero from the initial instant t ≥ 0 because z = 0. However,
For example: unominal can be any multivariable linear con- in the case of nonzero initial disturbance the sliding mode
trol, PID, LQR (linear quadratic regulator), state feedback, will start after some finite time t ≥ τ because z = 0, which
optimal control, time varying control, adaptive control etc. can be designed as small as possible by choosing appropriate
Required sliding surface for the system (7) is defined as values of ki , i = 1, · · · , 4 as given in Theorem 1. Once the
t s and transformed variable z are zero, then ṡ becomes zero
s = G x(t) − x(t0 ) − (Ax + Bunominal )dτ (8) and it remains zero forever in spite of the disturbance.
0 From (12), one can conclude that s = z = 0 in finite time
where G ∈ Rm×n , x(t0 ) are the projection matrix and for the proper value of the gains k i , i = 1, · · · , 4 as given
initial condition of the system respectively. Sliding surface in Theorem 1. Further, s = z = 0, implies ṡ = 0. So finally
is chosen such that the system trajectories start from the from (12) v = −d. Now from (10), u continuous = v = −d.
sliding surface and if disturbance comes into the picture This means that, when system is on the sliding mode, value
then ucontinuous becomes active and disturbances/perturbations of disturbance d = −v and it is canceled out.
become compensated. Mathematically, above procedure can Now when the system is on the sliding surface closed loop
be explained as system is governed by the nominal control and which is
stable by design.
ṡ = G [Ax + Bu + d − Ax − Bunominal ]
= G [Ax + B(unominal + ucontinuous + d) − Ax − Bunominal ] IV. S IMULATION R ESULTS
= GB [ucontinuous + d] Consider a planar two link robot with joint 1 and 2. It is
assumed that mass is concentrated only at the tips. Here, the
Without loss of generality let us consider GB = I m×m , following notations have been used, m 1 , m2 and l1 , l2 are
where I is m dimensional unit square matrix (otherwise masses and lengths of link 1 and 2, respectively. Similarly,
ucontinuous control is scaled by (GB) −1 ), one can get θ1 and θ2 are angles made by the link 1 with reference
ṡ = ucontinuous + d (9) frame and link 2 with link 1, respectively. The manipulator
is actuated by the control inputs u 1 and u2 at the joint under
Now suppose that ucontinuous is designed based on the multi- the action gravity g.
variable super twisting control The dynamical model is derived by Lagrangian approach,
s i.e, finding kinetic and potential energies associated with link
ucontinuous = −k1 1/2
+ v − k2 s 1 and link 2 and using Lagrange’s equation
s  
s d ∂L ∂L
v̇ = −k3 − k4 s (10) − =u
s dt ∂ q̇ ∂q
where ki > 0, i = 1, · · · , 4 are chosen scalars according to where L = T − U with T and U are the total kinetic and po-
 

the Theorem 1. After substituting u continuous from (10) into tential energy, respectively and the vectors q = θ1 θ2 ,
(9), one can further write  

u = u1 u2 . The complete dynamical model is given
s below as
ṡ = −k1 1/2
+ v − k2 s + d
s
s M (q)q̈ + V (q, q̇) + G(q) = u (13)
v̇ = −k3 − k4 s (11)
s where
Defining z = v + d, one can rewrite (11) Ξ m2 l22 + m2 l1 l2 cos θ2
M (q) =
s m2 l22 + m2 l1 l2 cos θ2 m2 l22
ṡ = −k1 − k2 s + z
 s 1/2 where Ξ = (m1 + m2 )l12 + m2 l22 + 2m2 l1 l2 cos θ2 .
s
ż = −k3 − k4 s + ḋ (12) −m2 l1 l2 (2θ̇1 θ̇2 + θ̇22 ) sin θ2
s V (q, q̇) = and
m2 l1 l2 θ̇12 sin θ2
Remark 1: For the solution of system Eqn. (11) initial
conditions of s and v are required. We have already design (m1 + m2 )gl1 cos θ1 + m2 gl2 cos(θ1 + θ2 )
G= .
s such that its initial condition is zero and one can always m2 gl2 cos(θ1 + θ2 )
choose v = 0 as an initial condition because v is a fictitious The robot dynamics possess following properties,
 i) The inertia matrix M (q) satisfy the property
0.6
b1 x2 ≤ x
M (q)x ≤ b2 (q)x2 ∀x ∈ R2
0.4
(14)
0.2
for any real constant b 1 > 0 and real valued
function b 2 (q) : R → R>0 . This ensures inertia 0
matrix is invertible. x

States
−0.2 1
ii) For any x ∈ R2 ,the following matrix satisfy the x
2
skew symmetry property, i.e., −0.4
  x
3
x
Ṁ (q) − 2V (q, q̇) x = 0. (15) −0.6 x
4
−0.8
The system considered here is an two input two output
system. The state space model for the system (13) is given −1
with the states being x1 = θ1 , x2 = θ2 , x3 = θ̇1 and x4 = θ̇2
−1.2
as 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (s)
ẋ1 = x3
ẋ2 = x4 Fig. 1. Evolution of states
 (16)
ẋ3
= M −1 (q) (u − V (q, q̇) − G(q))
ẋ4
0.5
 
 

where q = x1 x2 and q̇ = x3 x4 . In concise
form, we write the dynamics as ẋ = f (x, u). The control
objective here is to design MIMO continuous integral sliding 0
mode control for the two link rigid manipulator to drive
the states to equilibrium points in finite time with a smooth
States

control effort. x1
−0.5
Now consider the state variables as z 1 = [x1 x2 ]T and
z2 = [x3 x4 ]T using these variable a nominal control [23] x2
can be designed for the above system as, x3
−1
z1 z2 x
unominal = −η1 2 − η2 1 . (17) 4
z1  3 z2  2
where η1 = 30 and η2 = 11. The sliding surface for the −1.5
controller can be defined as, 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (s)
t
s = z2 − z20 − unominal dτ (18) Fig. 2. Evolution of states with applied disturbance
0

and hence the super twisitng control can be written as,

s 150
ucontinuous = −k1 1/2
+ v − k2 s u1
s
u2
s 100
v̇ = −k3 − k4 s (19)
s
Control Force (N)

50
where ki , i = 1, · · · , 4 are chosen according to the Theorem
. Now selecting a control u = M (q)ν +V (q, q̇)+G(q) where
the control ν = unominal + ucontinuous and this control will able 0
to bring the states to equilibrium point with a continuous
control in finite time even in the presence of disturbance. −50
For simulation, initial conditions for planer two link robot
is chosen as x1 = 1, x2 = 0.2, x3 = 0.2, x4 = 0.5 and −100
gain parameters are selected as η 1 = 30, η2 = 11, k1 =
2, k2 = 40, k3 = 2.2, k4 = 60. For checking the robustness
−150
property of the proposed controller matched uncertainty d = 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
[2sin(4t), 2sin(4t)]
has been injected through the input Time (s)
channel. Evolution of the states with respect to time of the
unperturbed and perturbed systems are shown in the Figs. Fig. 3. Evolution of control
 [3] A. Levant, “Chattering analysis ,” IEEE Trans. Autom. Control, vol. 55,
150 no. 6, pp. 1380–1389, 2010.
u1 [4] V. I. Utkin and J. Shi, “Integral sliding mode in systems operating
under uncertainty conditions,” in Proc. IEEE Conf. on Decision and
100 u2
Control, Kobe(Japan), pp. 4591–4596, 1996.
[5] W. -J. Cao and J. -X. Xu, “Nonlinear integral-type sliding mode
surface for both matched and unmatched uncertain systems,” IEEE
Control Force (N)

50 Trans. Autom. Control, vol. 49, no. 8, pp. 1355–1360, 2004.

[6] F. Castanos and L. Fridman, “Analysis and design of integral sliding
manifolds for systems with unmatched perturbations,” IEEE Trans.
0 Autom. Control, vol. 51, no. 5, pp. 853–858, 2006.
[7] A. Levant and L. Alelishvili, “Integral high-order sliding modes ,”
−50 IEEE Trans. [Link], vol. 52, no. 7, pp. 1278–1282, 2007.
[8] J. H. Lee, P. E. Allaire, G. Tao, and X. Zhang. “Integral sliding-
mode control of a magnetically suspended balance beam: analysis,
−100 simulation, and experiment.” IEEE/ASME Trans. Mechatronics, vol. 6,
no. 3, pp. 338–346, 2001.
[9] S. Laghrouche, F. Plestan, and A. Glumineau, “Higher order sliding
−150 mode control based on integral sliding mode,” Automatica, vol. 43,
0 1 2 3 4 5 6 7 8 9 10 no. 3, pp. 531–537, 2007.
Time (s) [10] M. Defoort, T. Floquet, A. Kokosy, and W. Perruquetti, “A new novel
higher order sliding mode control scheme,” Systems & Control Letters,
Fig. 4. Evolution of control with applied disturbance vol. 58, pp. 102–108, 2009.
[11] S. Kamal, A. Raman, and B. Bandyopadhyay, “ Finite time stabi-
lization of fractional order uncertain chain of integrator: An Integral
sliding mode approach” IEEE Trans. Autom. Control, vol. 58, no. 6,
1 and 2 respectively. Figs.3 and 4 show that the evolution pp. 1597-1602, 2013.
of the absolutely continuous control signal in the case of [12] A. Chalanga, S. Kamal and B. Bandyopadhyay, “A New Algorithm for
Continuous Sliding Mode Control with Implementation to Industrial
unperturbed and perturbed system respectively. Emulator Setup ”, IEEE/ASME Transactions on Mechatronics, DOI-
10.1109/TMECH.2014.2368717, 2014.
V. C ONCLUSION [13] H. Ros, S. Kamal, L. Fridman, and A. Zolghadri, “Fault tolerant
control allocation via continuous integral sliding-modes: a HOSM-
This paper proposes multivariable continuous integral slid- observer approach”, Automatica, vol. 51, no. 1, pp. 318-325, 201.
ing mode control for MIMO systems with uncertainties/ [14] A. Chalanga, S. Kamal, and B. Bandyopadhyay, “Continuous integral
disturbances. In this proposal the discontinuous part of sliding mode control: A chattering free approach,” Proc. of the 22nd
IEEE International Symposium on Ind. Electron., Taiwan on May 28-
conventional integral sliding mode controller is replaced by 31, pp. 1-6, 2013.
a multivariable super twisting algorithm due to its unique [15] A. Levant, “Sliding order and sliding accuracy in sliding mode
disturbance estimation property. This methodology can also control,” International Journal of Control, vol. 58, no. 6, pp. 1247–
1263, 1993.
be useful for the defining continuous higher order sliding [16] A. Levant, “Higher-order sliding modes, differentiation and output-
mode controller for the MIMO system, if the nominal feedback control,” Int. J. Control, vol. 76, no. 9/10, pp. 924–941,
control is selected from the finite or fixed time family of 2003.
[17] J. A. Moreno, and M. Osorio, “Strict Lyapunov functions for the super-
controllers. During the simulation study of planer two link twsiting algorithm,” IEEE Trans. Autom. Control, vol. 57, no. 4, pp.
robot, nominal controller has been chosen from the finite 1035–1040, 2012.
time controller family. Therefore, the proposed controller [18] A. Chalanga, S. Kamal, L. Fridman, B. Bandyopadhyay, Moreno,
J.A., “How to implement Super-Twisting Controller based on sliding
has been basically satisfied the criteria of continuous MIMO mode observer? ,” Proc. of 13th IEEE Workshop on Variable Structure
higher order sliding mode. Also, proposed control strategy Systems, Nantes, 2014, DOI:10.1109/VSS.2014.6881145.
generates the absolutely continuous control signals, which [19] S. Kamal, A. Chalanga , J.A. Moreno, L., Fridman, Bandyopad-
hyay, B. ”Higher order super-twisting algorithm,” Proc. of 13th
is able to attenuate the chattering. The effectiveness of the IEEE Workshop on Variable Structure Systems, Nantes, 2014, DOI:
proposed controller has been demonstrated with an example 10.1109/VSS.2014.6881129.
of two link robot. [20] L. Fridman, “Sliding mode enforcement after 1990: main results and
some open problems.” Sliding Modes after the First Decade of the
21st Century. Springer Berlin Heidelberg, pp. 3-57, 2012.
R EFERENCES [21] I. Nagesh, C. Edwards, “A multivariable super twisting sliding mode
[1] Y. Shtessel, C. Edwards, L. Fridman, and A. Levant, Sliding mode approach”, Automatica, 50, pp. 984-988, 2014.
control and observation. Control Engineering, Birkhuser, New York, [22] A. Isidori, Nonlinear control systems, vol. 1. Springer, 1995.
NY, USA, 2014. [23] S. P. Bhat, D. S. Bernstein,“Geometric homogeneity with applications
[2] C. Edwards and S. Spurgeon, “Sliding Mode Control: Theory and to finite-time stability”, Journal of Math. Control Signals Systems, 17,
Applications”, London: Taylor and Francis, 1998. pp 101 - 127, 2005.