Chapter 2

Integral Sliding Mode Control

Variable Structure Control Systems (VSCS) are a class of systems where the control
law, as a function of the system state, is deliberately changed (from one structure
to another) according to some predefined rules: for example a relay system. During
a sliding mode the closed-loop system response is constrained to evolve along a
sliding surface in the state-space to an equilibrium point. In sliding mode schemes,
a switching function typically dictates which structure of control law is to be used
at a particular time instant, depending on the position of the state from the sliding
surface. The set of points for which the switching function is zero is called the
sliding surface. SMC has now become an established tool to design controllers for
uncertain systems, and provides robustness properties against matched uncertainties
i.e. uncertainties that affect the plant dynamics acting in the input channels. However
this robustness against external disturbances and parameter variations matched to the
control can only be achieved after the occurrence of the sliding mode. Before the
occurrence of the sliding mode i.e. during the so-called reaching phase, the system
is affected by external disturbances—even matched ones. In order to eliminate the
reaching phase and to ensure robustness throughout the entire closed-loop system
response (i.e. to enforce a sliding mode for all time) the idea of Integral Sliding Modes
(ISM) was proposed. In this chapter a step-by-step design procedure is described for
the synthesis of sliding mode controllers; then these ideas are extended to integral
sliding modes in order to ensure robustness throughout the entire system response.
Necessary conditions for the existence of sliding modes are also given. The properties
of the system while in the sliding mode are also explained, and are examined through
simulations.

2.1 Introduction

SMC is a useful robust technique to handle sudden and large changes in the system
dynamics and has been applied to many areas—for example motor control, aircraft
and spacecraft control, process control and power systems. The realisation of a sliding
© Springer International Publishing Switzerland 2016 17
M.T. Hamayun et al., Fault Tolerant Control Schemes Using Integral
Sliding Modes, Studies in Systems, Decision and Control 61,
DOI 10.1007/978-3-319-32238-4_2
18 2 Integral Sliding Mode Control

mode controller comprises two steps. The first step is to design a sliding (switching)
surface on which the sliding motion will take place. The second step is to design
a control law, which depends on the choice of switching function and forces the
system state trajectories to reach and slide on the surface. An important condition
in the sliding mode literature is the reachability condition, which guarantees the
existence of the sliding mode. Once sliding is achieved and maintained, robustness
against matched uncertainties is guaranteed. Details of the design procedures are
given in the next sections.

2.2 Problem Statement and Equivalent Control

In order to explain the design procedure for a system where full state information is
available, consider an uncertain linear time invariant (LTI) system of the form

ẋ(t) = Ax(t) + Bu(t) + Mξ(t, x) (2.1)

where A ∈ IRn×n , B ∈ IRn×m .

Assumption 2.1 It is assumed that the matrix B has full rank i.e. rank(B) = m,
where 1 ≤ m < n and the pair (A, B) is controllable.

Assumption 2.2 The matrix M ∈ IRn×l is assumed to be known and lies in the range
space of the input distribution matrix B i.e. R(M) ⊂ R(B), therefore it is possible
to write M = BD for some D ∈ IRm×l .

The function ξ(t, x) represents an external disturbance or models uncertainty which

is unknown but has a known upper bound for all x and t. Therefore the uncertain
system in (2.1) can be rewritten as

ẋ = Ax(t) + Bu(t) + BDξ(t, x) (2.2)

Uncertainty of the form in (2.2), acting in the channel of the input distribution matrix,
is referred to as matched uncertainty. As a first step, define a sliding surface as

S = {x ∈ IRn : σ (t) = 0} (2.3)

where σ (t) is a linear switching function defined as

σ (t) = Gx(t) (2.4)

where G ∈ IRm×n is a design matrix and is of full rank. Furthermore by design it is

assumed that the square matrix GB is nonsingular i.e. det(GB) = 0. It is important
that the sliding motion on the sliding surface should be stable and robust against the
2.2 Problem Statement and Equivalent Control 19

uncertainty ξ(t, x). Therefore in order to analyse the sliding motion associated with
the sliding surface in (2.3), consider the time derivative of (2.4) given by

σ̇ (t) = Gẋ(t) (2.5)

Substituting the open-loop system equations from (2.2) into (2.5) gives

σ̇ (t) = G (Ax(t) + Bu(t) + BDξ(t, x)) (2.6)

It is assumed that the system states are forced to reach the sliding surface at time ts
say, so that for all t ≥ ts an ideal sliding motion can be obtained i.e.

σ (t) = σ̇ (t) = 0 for all t ≥ ts

The control signal u(t) such that the time derivative σ̇ (t) along the state trajectories
is equal to zero can be obtained by equating Eq. (2.6) to zero which yields

ueq (t) = −(GB)−1 (GAx(t) + GBDξ(t, x)) for t ≥ ts (2.7)

where the square matrix GB is nonsingular by design. The expression ueq (t) in (2.7)
is termed the equivalent control and can be thought of as the average value which
the control signal must take to maintain the sliding motion on the sliding surface.
However, it is not the control law that is applied to the system to induce the sliding
mode. In order to obtain an expression for the sliding motion (i.e. the motion while
the system is in the sliding mode), substituting the value of ueq (t) from (2.7) into
(2.2), yields
 
ẋ(t) = Ax(t) + B −(GB)−1 (GAx(t) + GBDξ(t, x)) + BDξ(t, x)
= (In − B(GB)−1 G) Ax(t) + (In − B(GB)−1 G)BDξ(t, x) (2.8)
  
Γ

Note that the projection operator Γ has the property that

ΓB = 0 (2.9)

As a result, Eq. (2.8) reduces to

ẋ(t) = Γ Ax(t) for t ≥ ts (2.10)

From (2.10), it is clear that the effect of the uncertainty ξ(t, x) during the sliding
mode is completely rejected i.e. the reduced order system motion is insensitive to
matched uncertainties. Also the stability of the sliding motion depends on the choice
of sliding surface, i.e. the choice of switching matrix G.
20 2 Integral Sliding Mode Control

2.2.1 Sliding Mode Control Laws

The second step is to design a control law such that the sliding motion on the surface
S is guaranteed in finite time. A sliding mode controller for a system of the form in
(2.1) typically consists of two parts; a linear part and a nonlinear part so that

u(t) = ul (t) + un (t) (2.11)

where the nonlinear part contains a discontinuous component and is responsible

for inducing a sliding motion on S , whereas the linear part, which is normally the
nominal equivalent control, is responsible for helping to maintain sliding. Specifically

σ (t)
u(t) = −(GB)−1 GAx(t) − ρ(t, x)(GB)−1 for σ (t) = 0 (2.12)
σ (t)

where σσ (t)

(t)
is the unit vector component and ρ(t, x) is a scalar gain chosen large
enough (i.e. greater than the size of the uncertainty present in the system) to enforce
the sliding motion.

Remark 2.1. For single input systems, the sliding mode controller in (2.12) becomes

u(t) = −(GB)−1 GAx(t) − ρ(t, x)(GB)−1 sign(σ (t)) for σ (t) = 0 (2.13)

where sign(.) is the signum function and has the property that σ sign(σ ) = |σ |.

2.3 Reachability Problem

In the sliding mode literature the controller u(t) is designed so that the so-called
reachability condition is satisfied, which is a sufficient condition to ensure that at
each time instant, the system state trajectories will converge towards the sliding
surface. Mathematically this can be expressed for the case of single input systems as

lim σ̇ (t) < 0 lim σ̇ (t) > 0 (2.14)

σ (t)→0+ σ (t)→0−

or in a compact form as
σ (t)σ̇ (t) < 0 (2.15)

near the sliding surface σ (t) = 0. A stronger condition which ensures an ideal sliding
motion in finite time, even in the presence of external disturbances or uncertainty, is
given by
σ (t)σ̇ (t) ≤ −η|σ (t)| (2.16)
2.3 Reachability Problem 21

where η represents a positive design scalar. The expression in (2.16) is often called
the η-reachability condition.
For multi-input systems, a natural multivariable version of the reachability con-
dition in (2.16) is
σ T (t)σ̇ (t) ≤ −ησ (t) (2.17)

This is a sufficient condition to show that the sliding surface S is attractive.

In order to demonstrate that the controller designed in (2.12) satisfies the η-
reachability condition (2.17), substituting the value of (2.12) into (2.6) gives

σ (t)
σ̇ (t) = GAx(t) + GB −(GB)−1 GAx(t) − ρ(t, x)(GB)−1 + GBDξ(t, x)
σ (t)
σ (t)
= −ρ(t, x) + GBDξ(t, x) (2.18)
σ (t)

Pre-multiplying both sides of (2.18) by σ T (t) yields

σ T (t)σ (t)
σ T (t)σ̇ (t) = −ρ(t, x) + σ T (t)GBDξ(t, x) (2.19)
σ (t)

and using the property that σ T σ = σ 2 , Eq. (2.19) becomes

σ T (t)σ̇ (t) = −ρ(t, x)σ (t) + σ T (t)GBDξ(t, x)

≤ σ (t)(−ρ(t, x) + GBDξ(t, x)) (2.20)

For any particular choice of scalar gain ρ(t, x) such that

ρ(t, x) ≥ GBDξ(t, x) + η (2.21)

where η is a positive scalar, the inequality in (2.20) becomes

σ T (t)σ̇ (t) ≤ −ησ (t) (2.22)

From (2.22), it is clear that the η-reachability condition is satisfied, which ensures
the existence of an ideal sliding motion on the sliding surface S .

2.4 A Simple Simulation Example

In this section, the design procedure for the typical sliding mode controller discussed
in the previous sections is applied to a simulation example, to offer insight into the
design procedure.
22 2 Integral Sliding Mode Control

2.4.1 Spring Mass Damper System

A simple example of a spring-mass-damper system (SMDS), driven by a force u(t), is

considered here as shown in Fig. 2.1. It is assumed that at t = 0 the mass m is pulled
down from the equilibrium position, such that y(0) = 0.1 m and ẏ(0) = 0.05 m/s.
The dynamical equation of the mechanical system (Fig. 2.1) can be written as

mÿ(t) + cẏ(t) + ky(t) = u(t) (2.23)

where k is the spring constant, c is the viscous-friction coefficient and m is the mass.
A disturbance signal a sin(y) is added to the control input channel to demonstrate
the invariance against a disturbance while in the sliding mode. The values of these
constants are chosen as m = 1 kg, c = 3 N · s/m, k = 2 N/m and a = 0.1. In order to
write the differential equation in (2.23) in state-space form, define the state variables
as x1 (t) = y(t) and x2 (t) = ẏ(t), which represent the position and velocity of the
mass m. Equation (2.23) can be written in terms of the state variables as

ẋ1 (t) = ẏ(t) = x2 (t) (2.24)

k c 1
ẋ2 (t) = ÿ(t) = − x1 (t) − x2 (t) + (u(t) + a sin(x1 (t)) (2.25)
m m m

Fig. 2.1 Spring mass

damper system
2.4 A Simple Simulation Example 23

By substituting values for the spring constant k, viscous friction coefficient c, and
mass m;

ẋ1 (t) 0 1 x1 (t) 0

= + (u(t) + 0.1 sin(x1 (t))) (2.26)
ẋ2 (t) −2 −3 x2 (t) 1
   
A B

2.4.2 Simulation Objective and SMC Design

In the simulation it is assumed that at t = 0 the mass m is pulled down from the
equilibrium position such that y(0) = 0.1 m and ẏ(0) = 0.05 m/s. The objective here
is to design a sliding mode controller to bring the system back to the equilibrium
position from the initial conditions without overshooting in terms of displacement,
and with a settling time of not more than 6 s.
The first step is to design a sliding surface. The switching function in (2.4) can be
written in terms of the states x1 (t) and x2 (t) as

x1 (t)
σ (t) = G1 G2 = G1 x1 (t) + G2 x2 (t) (2.27)
x2 (t)

where G1 ∈ IR and G2 ∈ IR. Here it is assumed G2 = 0. While sliding, the switching

function σ (t) = 0, and Eq. (2.27) can be written as

x2 (t) = −G−1
2 G1 x1 (t) (2.28)

It is clear from (2.28) that once x1 (t) is known, the state x2 (t) can be easily determined,
therefore substituting the value of (2.28) into (2.24), the sliding motion is given by

ẋ1 (t) = −G−1

2 G1 x1 (t) (2.29)

From (2.29) it is clear that during sliding the system behaves as a reduced order
system. Choosing the value of G2 = 1, the switching matrix G takes the form

G = G1 1

In this example the value of G1 = 0.9 is chosen. Using the fact that GB = 1, the
sliding mode control law defined in (2.13) becomes

u(t) = 2 2.1 x(t) − ρ(t, x) sign(σ (t)) (2.30)

Finally in order to verify that the control law u(t) in (2.30) satisfies the reachability
condition (2.16), by substituting (2.30) and (2.26) into the time derivative of (2.27):
24 2 Integral Sliding Mode Control

σ̇ (t) = G1 ẋ1 (t) + G2 ẋ2 (t)

= −ρ(t, x)sign(σ (t)) + 0.1 sin(x1 (t)) (2.31)

Multiplying (2.31) with σ (t) and choosing ρ(t, x) ≥ |0.1 sin(x1 (t))| + η = 0.1 + η,
it is clear the reachability condition in (2.16) has been established and

σ (t)σ̇ (t) ≤ −η|σ (t)| (2.32)

which ensures the existence of an ideal sliding mode.

2.4.3 Simulation Results

The sliding mode controller in (2.30) based on the nominal system (2.26) is now
tested in simulation using the MATLAB/SIMULINK environment. In the simula-
tions, the value of ρ(t, x) is selected as ρ(t, x) = 0.15. From Fig. 2.2 it is clear that
the disturbance has no effect on the system performance, which means that the design
requirements of the displacement reaching the equilibrium position with no overshoot

state state
0.12 0.06

0.1 0.04
displacement (m)

velocity (m/sec)

0.02
0.08
0
0.06
−0.02
0.04
−0.04
0.02 −0.06

0 −0.08
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec)
0.15 0.5
switching function σ(t)

0.1 (Reaching phase)

u(t)

0.05 (Sliding phase) 0

−0.05 −0.5
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec)

Fig. 2.2 Simulation results for the SMDS with disturbance

2.4 A Simple Simulation Example 25

and within 6 s is met. The switching function plot in Fig. 2.2 shows that the sliding
surface is attained in 1 s, i.e. the sliding motion starts after t ≥ 1 s. However the dis-
continuous control signal exhibits high frequency switching which is undesired in
most systems due to high wear on moving mechanical components.

2.5 Practical Sliding Mode Control Law

The discontinuity associated with the nonlinear discontinuous part of the control law
in (2.12) is the main hurdle in a practical implementation—especially in mechanical
systems. Traditionally this has been circumvented by “smoothing” the discontinuity.
After doing this the state trajectories no longer slide on the sliding surface, and instead
they evolve in the vicinity of the sliding surface: this is termed as pseudo-sliding.
However this means total invariance against matched uncertainties is not guaranteed.
Nevertheless a good (point-wise) approximation of the discontinuous control term
ensures a certain level of robustness against matched uncertainties still remains.1
One possibility is to use a sigmoidal approximation, where the unit vector term in
(2.12) is replaced by

σ (t)
un (t) = −ρ(t, x)(GB)−1 (2.33)
||σ (t)|| + δ

where δ is a small positive design scalar. In this book the sigmoidal approximation
given in (2.33), as shown in Fig. 2.3, is used. Here, the value of δ is chosen as
δ = 0.0001, and the control law in (2.30) becomes

σ (t)
u(t) = 2 2.1 x(t) − 0.15
|σ (t)| + 0.0001

From Fig. 2.4, it is clear that the chattering or high frequency switching of the control
signal has been removed. Due to this approximation, the sliding motion will be in
the vicinity of the sliding surface and will be termed pseudo-sliding instead of ideal
sliding. The design requirements however are still met in the presence of the external
disturbance as can be seen in Fig. 2.4.

1 An alternative approach to smoothing the discontinuity which leads to chattering is to use a higher

order sliding mode control approach [2]. Now the sliding motion takes place on the constraint set
σ = σ̇ = · · · = σ r−1 = 0 and is called an rth order sliding mode. Furthermore if it is possible to
steer σ to zero using the discontinuous control based on u̇(t), then the actual control signal u(t) will
be continuous and the unwanted chattering effects can be alleviated [3].
26 2 Integral Sliding Mode Control

1.5
20 δ
1

0.5
sign(σ)

0 sign function
approx. sign function
−0.5

−1
20 δ
−1.5
−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.2 0.25
σ(t)

Fig. 2.3 An approximation of the sign function [1]

state
state
0.12 0.06

0.1 0.04
displacement (m)

velocity (m/sec)

0.02
0.08
0
0.06
−0.02
0.04
−0.04
0.02 −0.06

0 −0.08
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec)

0.15 0.5
switching function σ(t)

(Reaching phase)
0.1
u(t)

0.05 (Sliding phase) 0

−0.05 −0.5
0 2 4 6 8 10 0 2 4 6 8 10

Time (sec) Time (sec)

Fig. 2.4 Simulation results for the SMDS with modified control law

2.6 Properties of the Sliding Mode

The properties of conventional (1st order) siding modes are summarised below:
• during a sliding mode, the order of the sliding motion is n − m, where n and m
represent the number of states and the number of inputs respectively;
2.6 Properties of the Sliding Mode 27

• the stability of the closed-loop sliding motion depends only on these n − m non-
negative eigenvalues;
• the performance of the closed-loop sliding motion depends on the choice of sliding
surface;
• during sliding, the sliding motion is invariant to matched uncertainties;
It should be noted that the robustness to uncertainty is only achieved once sliding takes
place. In the sequel, Integral Sliding Mode Control (ISMC) schemes are discussed
which eliminate the reaching phase associated with the classical SMC approach
discussed in the previous sections, and induce a sliding mode for the entire closed-
loop system response.

2.7 Integral Sliding Mode Control (ISMC)

The basic idea of ISMC was initially proposed to enforce a sliding mode from the
beginning of the system response, which means a controller based on ISMC ideas
can provide compensation to matched uncertainties throughout the entire system
response. In this section, a step-by-step design procedure for Integral Sliding Mode
(ISM) controllers is explained, and the special features associated with ISMC design
are discussed. Again in this section, it is assumed that state information is available
for the controller design.

2.7.1 Introduction

In ISMC, it is assumed that there exists a nominal plant, for which a properly designed
state feedback controller has already been designed to ensure asymptotic stability
of the closed-loop system, and to satisfy predefined performance specifications. A
discontinuous controller is ‘added’ to the nominal state feedback controller to ensure
the nominal performance is maintained, and the system is insensitive to external dis-
turbances (faults/failures from a FTC perspective). This design philosophy provides
the opportunity to retro-fit an ISM to the existing baseline controller to compen-
sate for the matched uncertainties and external disturbances throughout the system
response. As demonstrated in this chapter, when using sliding mode based schemes,
the system state trajectories are insensitive to matched uncertainties while in the slid-
ing mode. However no discussion has been made regarding unmatched uncertainties
i.e. uncertainties which are not in the range space of the input distribution matrix.
This will be addressed in the remainder of the chapter.
28 2 Integral Sliding Mode Control

2.7.2 Problem Statement and ISM Controller Design

To explain the design procedure, consider an uncertain LTI system of the form

ẋ(t) = Ax(t) + Bu(t) + Mξ(t, x) + fu (t, x) (2.34)

subject to Assumptions 2.1 and 2.2, where ξ(t, x) is a bounded unknown disturbance
and the matrix M satisfies the matching condition and can be written as M = BD,
for some D ∈ IRm×l .

Assumption 2.3 The function fu (t, x) represents unmatched uncertainty i.e. it does
not lie within the range space of matrix B, but is assumed to be bounded with known
upper bound.

The nominal linear system associated with Eq. (2.34) can be written as

ẋ(t) = Ax(t) + Buo (t) (2.35)

where uo (t) is a nominal control law which can be designed by any suitable state
feedback paradigm to achieve desired nominal performance. Since it is assumed that
the pair (A, B) is controllable, then there exists a state feedback controller of the form

uo (t) = −Fx(t) (2.36)

where F ∈ IRm×n is a state feedback gain to be designed, so that the state trajectories
of the nominal system (2.35), say xo (t), are stable and meet the performance specifi-
cations. The matrix F can be designed using any state feedback design approach. The
objective is to design a control law u(t), such that the state trajectories x(t) of (2.34),
while in the sliding mode satisfy the condition x(t) ≡ xo (t) for all time if fu (.) = 0,
starting from the initial time instant i.e. when x(0) ≡ xo (0). To achieve x(0) ≡ xo (0)
the order of the sliding dynamics should be the same as the nominal system.

2.7.3 Design Principles

Define a control law u(t) of the form

u(t) = uo (t) + un (t) (2.37)

where uo (t) is the nominal controller and un (t) is a nonlinear injection to induce a
sliding mode. Then using (2.37), (2.34) can be written as

ẋ(t) = Ax(t) + Buo (t) + Bun (t) + BDξ(t, x) + fu (t, x) (2.38)

2.7 Integral Sliding Mode Control (ISMC) 29

where uo (t) is the state feedback controller defined in (2.36), and where un (t) is
chosen to reject the disturbance term ξ(t, x) while in the sliding mode. Here the
switching function is defined as

σ (t) = Gx(t) + z(t) (2.39)

where G ∈ IRm×n is design freedom and z(t) is to be specified. Since the matrix B is of
full rank, the switching matrix G can be chosen so that the matrix GB is nonsingular
i.e. det(GB) = 0. During sliding σ (t) = σ̇ (t) = 0 and therefore

σ̇ (t) = Gẋ(t) + ż(t) = 0 (2.40)

In order to ensure that the equivalent control term associated with un (t) rejects the
effect of the matched disturbance term ξ(t, x) (in the case when fu (t, x) = 0), so that
the condition x(t) ≡ xo (t) is satisfied for all t > 0, substituting the value of (2.38)
into (2.40) gives:

σ̇ (t) = G (Ax(t) + Buo (t) + Bun (t) + BDξ(t, x)) + ż(t) = 0 (2.41)

During sliding it is expected that uneq (t) = −Dξ(t, x), i.e. it should compensate for
the uncertainty, then selecting

ż(t) = −G(Ax(t) + Buo (t)), z(0) = −Gx(0) (2.42)

ensures
σ̇ (t) = GBun (t) + GBDξ(t, x) (2.43)

and so during sliding uneq (t) = −Dξ(t, x). Substituting the value of uneq (t) into (2.38)
means the integral sliding mode is governed by

ẋ(t) = Ax(t) + Buo (t) (2.44)

which confirms that the condition x(t) ≡ xo (t) is satisfied if fu (t, x) = 0 and x(0) =
xo (0). In the case when fu (t, x) = 0 in (2.38), the equivalent control obtained from
(2.41) can be written as

uneq (t) = −(GB)−1 GBDξ(t, x) − (GB)−1 Gfu (t, x)

= −Dξ(t, x) − (GB)−1 Gfu (t, x) (2.45)

Substituting the value of equivalent control uneq (t) from (2.45) into (2.38) and sim-
plifying, the expression for the integral sliding mode dynamics can be written as

ẋ(t) = Ax(t) + Buo (t) + (I − B(GB)−1 G) fu (t, x) (2.46)

  
Γ
30 2 Integral Sliding Mode Control

From Eq. (2.46), it is clear that the effect of the matched uncertainty has been com-
pletely rejected while in the sliding mode. However the matrix Γ in (2.46) can
amplify the effect of unmatched uncertainty fu (t, x). Therefore the objective in the
next section is to design the integral sliding surface design parameter G to avoid any
amplification of the unmatched uncertainty.

2.7.4 Integral Switching Surface

Using Eqs. (2.39) and (2.42), an integral switching function which eliminates the
reaching phase is
 t
σ (t) = Gx(t) − Gx(0) − G (Ax(τ ) + Buo (τ )) d(τ ) (2.47)
0

The term −Gx(0) ensures that σ (0) = 0, so the reaching phase is eliminated. The
sliding mode will exist from time t = 0 and the system will be robust throughout the
entire closed-loop system response against matched uncertainties.
From the previous analysis, it is clear that in the case of only matched uncertainty,
then any choice of G which ensures GB is invertible is sufficient for the ISM design,
but for unmatched uncertainty, a specific choice of G is needed. Here it will be argued
that
G = B+ = (BT B)−1 BT (2.48)

is an appropriate choice. Note G in (2.48) is the Moore–Penrose left pseudo-inverse

of the input distribution matrix B. The particular choice of G in (2.48) brings two
advantages:
1. the modulation gain associated with un (t) in (2.37) is minimised which means
the amplitude of the chattering can be reduced;
2. it avoids amplifying the effect of the unmatched disturbance.
This choice of G also has the simplifying property that

GB = (BT B)−1 BT B = Im
  
G

and ensures that the square matrix GB is nonsingular. With the choice of G in (2.48),
the projection operator Γ in (2.46) becomes

Γ = In − B(BT B)−1 BT (2.49)

Notice that the projection operator Γ in (2.49) is symmetric and idempotent i.e.
Γ 2 = Γ . The properties of symmetry and idempotency imply that Γ  = 1, which
means that the effect of fu is not amplified since Γ fu  ≤ fu . In fact, it can be
2.7 Integral Sliding Mode Control (ISMC) 31

proved that I − B(GB)−1 G ≥ 1 for any G, and so the choice of G in (2.48) is an

optimal one in the sense of non-amplification of the unmatched uncertainty.

2.7.5 Integral Sliding Mode Control Laws

An integral sliding mode controller will now be designed based on the nominal
system in (2.35). The control law has a structure given by

u(t) = uo (t) + un (t) (2.50)

where uo (t) is the linear part of the controller, and un (t) is the discontinuous part to
enforce a sliding mode along the sliding surface in (2.47). One choice of u(t) is

σ (t)
u(t) = −Fx(t) − ρ(t, x)(GB)−1 for σ (t) = 0 (2.51)
σ (t)

where F is the state feedback controller which is responsible for the performance of
the nominal system and ρ(t, x) is the modulation gain to enforce the sliding mode—
whose precise value is given in the next subsection.

2.7.6 The Reachability Condition

To justify that the controller designed in (2.51) satisfies the η-reachability condition
(2.22), which is a sufficient condition to ensure the existence of an ideal sliding
motion, it can be shown from (2.34) and (2.36) that

σ̇ (t) = G (Ax(t) + Bu(t) + BDξ(t, x) + fu (t, x)) − GAx(t) + GBFx(t)

then substituting from (2.51), and after some simplification

σ̇ (t) = GAx(t) + GB(−Fx(t) + un (t)) + GBDξ(·) + Gfu (·) − GAx(t) + GBFx(t)

σ (t)
= −ρ(t, x) + GBDξ(t, x) + Gfu (t, x) (2.52)
σ (t)

Then

σ T (t)σ̇ (t) = −ρ(t, x)σ (t) + σ T (t)Dξ(t, x) + σ T (t)Gfu (t, x)

≤ σ (t)(−ρ(t, x) + Dξ(t, x) + Gfu (t, x)) (2.53)
32 2 Integral Sliding Mode Control

where the fact that GB = Im , has been used. In order to enforce a sliding mode
the value of the modulation gain ρ(t, x) should be greater than any disturbance or
uncertainty in the system, and therefore for any choice of ρ(t, x) which satisfies

ρ(t, x) ≥ Dξ(t, x) + Gfu (t, x) + η (2.54)

where η is some positive scalar, the η-reachability condition

σ T (t)σ̇ (t) ≤ −η||σ (t)|| (2.55)

is satisfied.

Remark 2.2. Inequality (2.55) can also be interpreted from a Lyapunov perspec-
tive. Define V (t) = 21 σ T (t)σ (t), then V̇ (t) = σ T (t)σ̇ (t) and from the inequalities in
(2.53)–(2.55) it follows

V̇ (t) ≤ −η||σ (t)|| = −η 2V (t) (2.56)

Integrating both sides of (2.56) yields

 
2V (t) − 2V (0) ≤ −ηt

which implies V (t) ≡ 0 in less than √ η units of time.

2V (0)

2.7.7 Properties of Integral Sliding Mode

The properties of integral sliding modes can be summarised as follows:

• there is no reaching phase and a sliding mode is enforced throughout the entire
system response;
• during sliding, the order of the motion is the same as the original system;
• by a suitable choice of sliding surface, the effect of unmatched uncertainty can be
ameliorated;
• during the sliding mode, the system motion is invariant to matched uncertainties;
• the ISM approach has the ability to be retro-fitted to an existing feedback controller;

2.7.8 Simulation Example

Here in this section, to make a direct comparison, the simulation scenario of the
spring-mass-damper system from Sect. 2.4 will be simulated. Recall the system was
represented as
2.7 Integral Sliding Mode Control (ISMC) 33

ẋ1 (t) 0 1 x1 (t) 0

= + (u(t) + 0.1 sin(x1 (t)))
ẋ2 (t) −2 −3 x2 (t) 1
   
A B

The objective is to design an ISM controller to bring the system back to the equilib-
rium position from the initial conditions without overshooting in terms of displace-
ment, and with a settling time not more than 6 s. The integral switching function from
Eq. (2.47) is  t
σ (t) = Gx(t) − Gx(0) − G (A − BF) x(τ )d(τ )
0

where the value of G is chosen as in (2.48), and here is equal to

G = (BT B)−1 BT = 0 1 (2.57)

The gain F in this example has been designed using the linear quadratic regulator
(LQR) method,2 and aims to regulate the system states to the origin by minimising
the cost function  ∞
J= (x(t)T Qx(t) + uo (t)T Ruo (t))dt (2.58)
0

where Q is a symmetric positive definite (s.p.d.) matrix and R is a positive scalar.

These matrices penalise the magnitude of the control signal uo (t) and the deviation
of the system states from the origin. Here the values of Q and R are chosen as
Q = diag(1, 0.5) and R = 1, which results in the matrix

F = 0.2361 0.1579

The ISM control law is

u(t) = −Fx(t) − ρsign(σ (t)) for σ (t) = 0 (2.59)

since the choice of G in (2.57) makes GB = 1. Here ρ is a fixed scalar satisfying

ρ = 0.1 + η where η > 0. It is easy to check that the control law u(t) satisfies the
reachability condition σ (t)σ̇ (t) ≤ −η|σ (t)|. Here the sigmoidal approximation given
in (2.33) and shown in Fig. 2.3 is used, and therefore the ISM control law in (2.59)
is modified to become
σ (t)
u(t) = −Fx(t) − ρ (2.60)
|σ (t)| + δ

where the value of the small positive scalar δ is chosen as δ = 0.0001, to eliminate
chattering. The control law u(t), after substituting for the value of F, can be written as

2 This is a well-known ‘classical’ state-space technique: for details see for example [4].
34 2 Integral Sliding Mode Control

state state
0.12 0.06

0.1 0.04
displacement (m)

velocity (m/sec)
0.08 0.02

0.06 0

0.04 −0.02

0.02 −0.04

0 −0.06
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec)

0.01 0
switching function σ(t)

−0.005
0.005
−0.01

−0.015
u(t)
0
−0.02

−0.025
−0.005
−0.03

−0.01 −0.035
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec)

Fig. 2.5 Simulation results for the SMDS nominally (with ISMC)

σ (t)
u(t) = −0.2361x1 (t) − 0.2361x2 (t) − ρ
|σ (t)| + δ

The displacement plots in Figs. 2.5 and 2.6 show that the design requirements are
met both nominally (without any disturbance) and in the presence of a disturbance
term. From Fig. 2.6, it is clear that the effect of the disturbance 0.1 sin(x1 (t)) has
been completely rejected. From the switching function plots in Figs. 2.5 and 2.6, it
is clear there is no reaching phase i.e. the sliding mode starts from time t = 0.

2.8 Sliding Modes as a Candidate for FTC

Sliding mode based control schemes are a strong candidate for fault tolerant con-
trol because of their inherent robustness to matched uncertainties. As argued in
Sect. 1.3.1, actuator faults can be effectively modelled as matched uncertainties and
therefore sliding mode based control schemes have an inherent capability to directly
deal with actuator faults. However actuator failures cannot be handled directly by slid-
ing modes schemes because the complete loss of effectiveness in a channel destroys
2.8 Sliding Modes as a Candidate for FTC 35

state state
0.12 0.06

0.1 0.04
displacement (m)

velocity (m/sec)
0.08 0.02

0.06 0

0.04 −0.02

0.02 −0.04

0 −0.06
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec)

0.01 0

−0.005
switching function σ(t)

0.005
−0.01

−0.015
u(t)
0
−0.02

−0.005 −0.025

−0.03

−0.01 −0.035
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec)

Fig. 2.6 Simulation results for the SMDS with disturbance (with ISMC)

the regularity of the sliding mode, and a unique equivalent control signal can no
longer be determined.
In the subsequent chapters, Control Allocation (as discussed in Sect. 1.3.3) is con-
sidered as a potential candidate to be combined with ISM control to deal with actuator
faults or failures due to its ability to effectively manage the actuator redundancy and
to redistribute the control signals to the healthy actuators in the case of an actuator
failure. The use of integral sliding modes ensures robustness for all time by eliminat-
ing the reaching phase associated with ‘classical’ SMC based methods. Furthermore
integral sliding modes have the capacity to be retro-fitted to the existing controller
design to introduce fault tolerance without changing or altering the existing control
loops, which is advantageous from an industrial perspective.

2.9 Notes and References

The term sliding mode was first used in the literature in the context of relay systems
[5]. Sliding mode control (SMC) is a particular class of variable structure control
systems (VSCS) [6]. VSCS evolved from work in Russia in the early 1960s and
spread around the world in the late 1970s after the publication of the survey paper by
36 2 Integral Sliding Mode Control

Utkin [7]. SMC design paradigms [1, 2, 5] have become mature techniques for the
control of uncertain systems and provide effective solutions against matched uncer-
tainties; however to provide compensation against matched uncertainties throughout
the entire system response, the idea of integral sliding mode control was initially
proposed in [5, 8–10]. In [11–13], integral sliding mode control ideas were used
for uncertain systems considering both matched and unmatched uncertainties and
demonstrated that the system dynamics while on the sliding surface meet the per-
formance specifications in the presence of matched uncertainties. In [13], it was
first demonstrated that the effect of mismatched uncertainties can be minimised by
the suitable choice of an integral sliding surface (the specific choice of G is given in
Sect. 2.7.4). Details of the integral sliding mode approach in the context of robust LQ
output control (but not in the context of FTC) can be found in [14]. Different methods
have been used in the literature to smooth the transition near the sliding surface to
remove chattering—see for example Chap. 3 in [1, 15]. An alternative approach to
smoothing the discontinuous switching control law (which leads to chattering) is to
use a higher order sliding mode control approach [3]. Many researchers have iden-
tified SMC as a potential candidate for FTC, see for example [16–20]. Researchers
in [21] have focused on fault reconstruction and fault tolerant control schemes for
aerospace applications using traditional SMC approaches. In [16, 17], it was argued
that SMC could deal with significant and sudden changes in the system dynamics due
to actuator faults and has the capability to become an alternative to reconfigurable
control systems. In [18], a ‘hedging’ based SMC design is used to reduce the effect
of neglected parasitic dynamics in a longitudinal control system for an aircraft. In
[16, 19] the authors have demonstrated the combination of SMC with control allo-
cation for FTC purposes. Recently in [22], a continuous integral sliding mode FTC
scheme was proposed using a higher order sliding mode observer by incorporating
fixed control allocation.

References

1. Edwards, C., Spurgeon, S.K.: Sliding Mode Control, Theory and Applications. Taylor and
Francis, London (1998)
2. Shtessel, Y., Edwards, C., Fridman, L., Levant, A.: Sliding Mode Control and Observation.
Birkhauser, New York (2013)
3. Bartolini, G., Ferrara, A., Usai, E.: Chattering avoidance by second order sliding mode control.
IEEE Trans. Autom. Control 43(2), 241–246 (1998)
4. Anderson, B.D.O., Moore, J.B.: Optimal Control: Linear Quadratic Methods. Prentice Hall
International Edition, Upper Saddle River (1989)
5. Utkin, V., Guldner, J., Shi, J.: Sliding Mode Control in Electromechanical Systems. Taylor and
Francis, London (1999)
6. Utkin, V.: Sliding mode control. In: Sabanovic, A., Fridman, L.M., Spurgeon, S.K. (eds.)
Variable Structure Systems: From Principles to Implementation. IEE Control Series 66, pp.
3–17. IEE, London (2004)
7. Utkin, V.: Variable structure systems with sliding mode. IEEE Trans. Autom. Control 22(2),
212–222 (1977)
References 37

8. Matthews, G.P., DeCarlo, R.A.: Decentralized tracking for a class of interconnected nonlinear
systems using variable structure control. Automatica 24(2), 187–193 (1988)
9. Utkin, V., Shi, J.: Integral sliding mode in systems operating under uncertainty conditions. In:
Proceedings of the 35th IEEE Conference on Decision and Control (1996)
10. Wang, J., Lee, T., Juang, Y.: New methods to design an integral variable structure controller.
IEEE Trans. Autom. Control 41(1), 140–143 (1996)
11. Cao, W., Xu, J.: Nonlinear integral type sliding surface for both matched and unmatched
uncertain systems. IEEE Trans. Autom. Control 49(8), 1355–1360 (2004)
12. Xu, J., Pan, Y., Lee, T.: Analysis and design of integral sliding mode control based on Lya-
punov’s direct method. In: Proceedings of the American Control Conference (2003)
13. Castanos, F., Fridman, L.: Analysis and design of integral sliding manifolds for systems with
unmatched perturbations. IEEE Trans. Autom. Control 51(5), 853–858 (2006)
14. Fridman, L., Poznyak, A., Bejarano, F.J.: Robust LQ Output Control: Integral Sliding Mode
Approach. Springer, Heidelberg (2013)
15. Burton, J.A., Zinober, A.S.: Continuous approximation of variable structure control. Int. J.
Syst. Sci. 17(6), 875–885 (1986)
16. Wells, S.R., Hess, R.A.: Multi-input/multi-output sliding mode control for a tailless fighter
aircraft. J. Guid., Control, Dyn. 26(3), 463–473 (2003)
17. Hess, R.A., Wells, S.R.: Sliding mode control applied to reconfigurable flight control design.
J. Guid., Control Dyn. 26, 452–462 (2003)
18. Vetter, T.K., Wells, S.R., Hess, R.A.: Designing for damage-robust flight control design using
sliding mode techniques. Proc. Inst. Mech. Eng., Part G: J. Aerosp. Eng. 217, 245–261 (2003)
19. Shtessel, Y., Buffington, J., Banda, S.: Tailless aircraft flight control using multiple time scale
re-configurable sliding modes. IEEE Trans. Control Syst. Technol. 10, 288–296 (2002)
20. Shin, D., Moon, G., Kim, Y.: Design of reconfigurable flight control system using adaptive
sliding mode control: actuator fault. Proc. Inst. Mech. Eng., Part G: J. Aerosp. Eng. 219,
321–328 (2005)
21. Alwi, H., Edwards, C., Tan, C.P.: Fault Detection and Fault Tolerant Control Using Sliding
Modes. Advances in Industrial Control Series. Springer, Heidelberg (2011)
22. Ríos, H., Kamal, S., Fridman, L.M., Zolghadri, A.: Fault tolerant control allocation via con-
tinuous integral sliding-modes: a HOSM-observer approach. Automatica 51, 318–325 (2014)
http://www.springer.com/978-3-319-32236-0