Sliding mode control (SMC) has proven to be an effective robust control strategy for incompletely modeled or nonlinear systems since its first appearance in the 1950s [70, 103, 197]. One of the most distinguished properties of SMC is that it utilizes a discontinuous control action which switches between two distinctively different system structures such that a new type of system motion, called sliding mode, exists in a specified manifold. The peculiar characteristic of the motion in the manifold is its insensitivity to parameter variations, and its complete rejection of external disturbances [260]. SMC has been developed as a new control design method for a wide spectrum of systems including nonlinear, time-varying, discrete, large-scale, infinite-dimensional, stochastic, and distributed systems [101]. Also, in the past two decades, SMC has successfully been applied to a wide variety of practical systems such as robot manipulators, aircraft, underwater vehicles, spacecraft, flexible space structures, electrical motors, power systems, and automotive engines [60, 77, 199, 259].
In this section, we will first present some preliminary background and fundamental theory of SMC, which will be helpful to some readers who have little or no knowledge on SMC, and then we will give an overview of recent development of SMC methodologies.
We first formulate the SMC problem as follows. For a general nonlinear system of the form
where x(t) ∈ Rn is the system state vector, u(t) ∈ Rm is the control input. We need to design a sliding surface
where s(x) is called the switching function, and the order of s(x) is usually the same as that of the control input, i.e. s(x) ∈ Rm, and
Then a sliding mode controller is designed in the form of
where u+i(t) ≠ ui+(t), such that the following two conditions hold:
Further consider (1.1) with single input, that is, u(t) ∈ R and s(x) ∈ R, and suppose that the sliding mode can be reached in a finite time, then the solutions of the equation
will approach s(x) = 0 and reach there in a finite time. During the approaching phase, . Similarly, the solutions of the equation
will also approach s(x) = 0 and reach there in a finite time, thus we have . To summarize the above analysis, we have
or, equivalently,
which is the so-called ‘reaching condition’. This is the condition under which the state will move toward and reach a sliding surface. The system state trajectories under the reaching condition is called the reaching phase [77, 101].
In summary, Condition 1 requires the reachability of a sliding mode, which is guaranteed through designing a sliding mode controller, while Condition 2 requires the sliding mode dynamics to be stable with some specified performances, which is assured by designing an appropriate sliding mode surface. Therefore, a conventional SMC design consists of two steps:
Step 1. Design a sliding surface s(x) = 0 such that the dynamics restricted to the sliding surface has the desired properties such as stability, disturbance rejection capability, and tracking;
In the following, we will briefly introduce some commonly used methods in the design of sliding surfaces and sliding mode controllers, and in the elimination/reduction of chattering. Readers can refer to various books on SMC theory for more details, for example, [60, 77, 197, 199].
In this section, three kinds of sliding surfaces, namely, linear sliding surface, integral sliding surface, and terminal sliding surface, are introduced.
The linear sliding surface, due to its simplicity of implementation, is commonly used in SMC design. There are two approaches to designing linear sliding surface. First, we introduce the ‘regular form’ model transformation approach. Consider the following nonlinear system:
where x(t) ∈ Rn and u(t) ∈ Rm are the system states and control inputs, respectively. f(x, t) ∈ Rn and B(x, t) ∈ Rn × m are assumed to be continuous with bounded continuous derivatives with respect to x. B(x, t) is bounded away from zero at any time.
By applying an appropriate diffeomorphic transformation , system (1.2) can be written in the following regular form [120]:
where z1(t) ∈ Rn − m and z2(t) ∈ Rm are the transformed system states. is nonsingular (to ensure this, the matrix B(x, t) should be of full column rank for all t for the existence of such a transformation).
Design a switching function as
where ℏ( · ) is a function to be defined. When the system state trajectories reach onto the sliding surface, we have s(z) = 0, thus z2(t) = −ℏ(z1(t)). Substituting this into the first equation of the regular form yields
which is a reduced-order system representing the sliding mode dynamics. The remaining work of the sliding surface design is to choose a function ℏ( · ) such that the above nonlinear sliding mode dynamics is stable and/or satisfies a specified performance.
For a linear time-invariant (LTI) system of the form
where x(t) ∈ Rn is the system state vector, u(t) ∈ Rm is the control input, and the matrices A ∈ Rn × n and B ∈ Rn × m. The matrix B is assumed to have full column rank and the pair (A, B) is assumed to be controllable.
It is well known that for the controllable system (1.3) there exists a nonsingular transformation, defined by
such that
Thus, by z(t) = Tx(t) system (1.3) can be transformed into the following regular form:
where z1(t) ∈ Rn − m and z2(t) ∈ Rm are the transformed system states. A11 ∈ R(n − m) × (n − m), A12 ∈ R(n − m) × m, A21 ∈ Rm × (n − m), A22 ∈ Rm × m, B1 ∈ Rm × m, and B1 is nonsingular.
Now, a sliding surface can be designed under the model of (1.4). For example, we can choose the following linear one:
where C is the design parameter to be designed. Similarly, when the system state trajectories reach onto the sliding surface, that is, s(z) = 0, it follows that
Substituting (1.6) into the first equation of (1.4) yields
The above reduced-order system is the so-called sliding mode dynamics (that is, the motion equation in the sliding surface), which is an autonomous system. Therefore, the design of sliding surfaces becomes choosing the matrix parameter C such that the sliding mode dynamics is stable. Furthermore, since it can be shown that, if the pair (A, B) is controllable, then the pair (A11, A12) is controllable as well, the problem of finding the design matrix C is in fact a classical state feedback problem with matrix C as a feedback gain and A12 as an input matrix. Therefore, all existing linear state feedback control design methods can be used to solve this problem, for example, the conventional eigenvalue allocation method and linear-quadratic regulator (LQR) design method.
There is another approach to linear surface design, named the Lyapunov approach [186]. Let V(x) be a Lyapunov function for system (1.2), that is, V(x) > 0 and . The sliding surface can be chosen as
where
Lemma 1.1.1 [186] System (1.2) with sliding mode on the sliding surface (1.8) is asymptotically stable.
For linear system (1.3), since (A, B) is controllable we know that there exists a feedback matrix K such that is stabilizable. Thus, there exist matrices P > 0 and Q > 0 such that the following Lyapunov equation holds:
Now, design the sliding surface as
and rewrite system (1.3) as
where , and Kx is a fictitious feedback to system (1.3).
Let V(x) = xT(t)Px(t) > 0, and we have
When the system state trajectories are driven onto the sliding surface, that is, s(x) = BTPx(t) = 0, it follows that
for x(t) ≠ 0. Therefore, the system states are asymptotically stable on the sliding surface. Therefore, we have the following lemma.
Lemma 1.1.2 [186] System (1.10) with sliding mode on the sliding surface (1.9) is asymptotically stable.
We can see from Lemmas 1.1.1–1.1.2 that for the Lyapunov approach, the design of sliding surface is given by the positive definite matrix P.
In the above-mentioned linear sliding surface design, the order of the resulting sliding mode dynamics is (n − m), with n being the dimension of the state space and m being the dimension of the control input. Unlike in linear sliding surface design, in the integral sliding surface, the order of the sliding motion equation is the same as that of the original system, rather than being reduced by the number of the dimension of the control input. As the result, the robustness of the system can be guaranteed throughout an entire response of the system starting from the initial time instance [198].
Consider system (1.3) with a nonlinear perturbation included in the input channel (called matched perturbation), that is,
where d(x, t) is a nonlinear perturbation with known upper bound d0(x, t), that is, |d(x, t)| < d0(x, t). Design control u(t) = u0(t) + u1(t) for the above system, and suppose that there exists a feedback control law u(t) = u0(t) such that the perturbation-free system, that is, can be stabilized in a desired way. That is, the state trajectories of the closed-loop system follow pre-specified reference trajectories with a desired accuracy. Here, u0(t) may be designed through linear static feedback control, such as u0(t) = Kx(t) in which the feedback gain K can be determined by eigenvalue allocation or LQR methods.
Design the integral switching function as
and C is the parameter matrix to be designed such that CB is nonsingular. Notice that, at t = t0, the switching function , and hence the reaching phase is eliminated. By (1.11), the resulting sliding mode dynamics coincides with that of the ideal system , which means that the integral sliding surface is robust to the perturbation throughout the entire response of the system starting from the initial time instance.
The approaches to integral sliding surface design were then developed for uncertain systems with mismatched uncertainties/perturbations [27, 30, 32, 43, 170], higher order SMC systems [114, 118], stochastic systems [12, 155, 156, 157, 158], singular systems [219, 221, 223, 225], and switched hybrid systems [125]. For discrete-time SMC systems, the integral sliding surface design approaches were developed in [1, 107, 233].
The terminal SMC technique was first proposed in [201]. Compared to the conventional SMC, the terminal SMC has some superior properties such as fast and finite-time convergence and high steady-state tracking precision.
Consider the second-order linear system
where x1(t) and x2(t) are the system states and u(t) is the control input.
The following terminal switching function is designed:
where p and q are positive odd integers that satisfy p > q.
Similar to the conventional SMC technique, if the system state trajectories are driven onto the sliding surface, that is, s(x1, x2) = 0, then
Let the initial condition of x1(t) at t = 0 be x1(0)( ≠ 0), then the relaxation time t1 for a solution of above equation is
which means that on the terminal sliding surface, the system state trajectories converge to zero in a finite time.
For a high-order single-input single-output (SISO) linear system
the following terminal switching functions are designed:
where βi > 0 are constants and pi and qi are positive odd integers satisfying pi > qi, i = 1, 2, …, n − 1.
The terminal SMC technique for multiple-input multiple-output (MIMO) systems was proposed in [141], and then developed in [39, 72, 142].
Having designed the sliding mode via the design of switching functions, the next step is to design a sliding mode controller such that the system state trajectories can be driven onto the specified sliding surface in a finite time and maintained there for all subsequent time. The main requirement in this step is that the control should be designed to satisfy the reaching condition, thus guaranteeing the existence of a sliding mode on the sliding surface. Additional requirements in this reaching phase include some desired properties such as fast reaching and low chattering. In the following, we will introduce some commonly used methods to the sliding mode controller design.
Equivalent control is designed in the reaching phase, which can satisfy the reachability of the system state trajectories to the sliding surfaces if the system is free of parameter uncertainties and external disturbances. Consider system (1.2) with switching function being s(x). Suppose that the system state trajectories reach onto the sliding surface at time instant t1 and remain there in the subsequent time. We then have s(x) = 0 for all t > t1. Along sliding mode trajectories, s(x) is constant, and so sliding mode trajectories are described by the differential equation . Differentiating s(x) yields
where
is called the gradient of s(x). Here, we suppose that is nonsingular for all x and t, thus the equivalent control can be solved as follows:
Substituting the above equivalent control into the original system, it follows that on the sliding surface s(x) = 0 the system dynamics satisfy
The above differential equation represents the sliding mode dynamics, which is actually is a reduced-order model of order n − m. (Considering s(x) = 0, thus m of the system states can be eliminated from the equation.)
A straightforward method of sliding mode controller design is based on the reaching condition, that is, for i = 1, 2, …, m,
or, equivalently,
With the designed controller satisfying the above reachability condition, the tangent vectors of the state trajectories are guaranteed to point toward the sliding surface, hence, the reachability of the system state trajectories to the sliding surface can be guaranteed. Some more discussions of this approach can be found in [101].
The Lyapunov function approach is commonly used in sliding mode controller design. Choosing a Lyapunov function of the form
a sufficient condition for the sliding surface to be globally attractive is that the control u(t) is designed such that
Finite reaching time can be guaranteed by [103]
where ε > 0 is a constant.
For system (1.2), design the sliding mode controller as
where ueq(t) is the equivalent control which is designed in (1.12), and the discontinuous control uN(t) is to be chosen such that
Clearly, this approach leads to the global attraction of the system state trajectories to the sliding surface.
The reaching law is a differential equation which specifies the dynamics of a switching function, and by the choice of the parameters in the reaching law, the dynamic quality of SMC system in the reaching mode can be controlled [78]. A general form of the reaching law can be described by the differential equation
where
The functions gi(si(x)) satisfy gi(0) = 0 and
Therefore, using reaching law (1.13) directly to system (1.2) with
the sliding mode controller can be obtained as
Equation (1.13) is a general form of the reaching law, and some special cases are
The reaching law approach not only guarantees the reaching condition but also specifies the dynamic characteristics of the motion during the reaching phase [101, 78].
The chattering problem is one of the most common handicaps for applying SMC to real applications. Chattering in SMC systems is usually caused by 1) the unmodeled dynamics with small time constants, which are often neglected in the ideal model; and 2) utilization of digital controllers with finite sampling rate, which causes so called ‘discretization chattering’. Theoretically, the ideal sliding mode implies infinite switching frequency. Since the control is constant within a sampling interval, switching frequency can not exceed that of sampling, which also leads to chattering. From the control engineer’s point of view, chattering is undesirable because it often causes control inaccuracy, high heat loss in electric circuitry, and high wear of moving mechanical parts. In addition, the chattering action may excite the unmodeled high-order dynamics, which probably leads to unforeseen instability. Therefore, a good deal of research work has been reported in literature on the chattering elimination/reduction problem; see for example, [2, 11, 9, 10, 19, 36, 44, 83, 116, 117, 183, 200, 209, 258] and references therein. In the following, we will review some chattering elimination/reduction approaches.
Roughly speaking, an SMC law consists of two parts, that is, u(t) = ueq(t) + uN(t). The continuous control ueq(t), known as the equivalent control, controls the system when its states are on the sliding surface, and the discontinuous control uN(t) handles the system uncertainties. Since the discontinuous control uN(t) will switch between two structures during operation, the SMC system will undergo oscillation near the sliding surface. A commonly used method to alleviate chattering is to insert a boundary layer near the sliding surface so that a continuous control replaces the discontinuous one when the system is inside the boundary layer [52, 183, 196]. For this purpose, the discontinuous controller of
is often replaced by the saturation control of
or
for some, preferably small, δ > 0.
The boundary layer approach has been utilized extensively in practical applications. However, this method has some disadvantages such as: 1) it may give a chattering-free system but a finite steady-state error must exist; 2) the boundary layer thickness has a trade-off relation between control performance of SMC and chattering migration; and 3) within the boundary layer, the characteristics of robustness and accuracy of the system are no longer assured.
The second way to eliminate chatter is the dynamic SMC approach [9, 10, 36]. The main idea of this method is to insert an integrator (or any other strictly proper low-pass filter) between the SMC and the controlled plant, see Figure 1.1. The time derivative of the control input, , is treated as the new control input for the augmented system (including the original system and the integrator). Since the low-pass integrator in Figure 1.1 filters out the high frequency chattering in w, the control input to the real plant u = ∫wdt becomes chattering free [36].
Chattering reduction using the dynamic SMC approach is achieved by using an integrator, and the property of perfect disturbance rejection is guaranteed (no boundary layer is used in the controller). Such a method can eliminate chattering and ensure zero steady-state error; however, the system order is increased by one and the transient responses will be degraded [209].
Another way of reducing chattering is to decrease the amplitude of the discontinuous control. However, the robustness property of the controller is affected, and the transient performance of the system will also be degraded. There is a trade-off between the chattering reduction and the robustness property. A compromise approach is to decrease the amplitude of the discontinuous control when the system state trajectories are near to the sliding surface (to reduce the chattering), and to increase the amplitude when the system states are not near to the sliding surface (to guarantee the robustness to system uncertainties and unmodeled dynamics). This can be implemented by tuning the parameters of the reaching law
where ϵi and ki, i = 1, 2, …, m, are positive parameters to be tuned. When the system state trajectories are closed to the sliding surface, we have si(x, t) ≈ 0 and . Here, the parameter ϵi represents the reaching velocity. By choosing ϵi small, the momentum of the motion will be reduced as the system state trajectories approach the sliding surface. As a result, the amplitude of the chattering will be reduced. However, in this case, the transient performance of the system is also degraded. To guarantee the transient performance, a large value for the parameter ki should be chosen to increase the reaching rate when the state is not near the sliding surface.
Apart from the above-mentioned chattering elimination/reduction approaches, there have been some others, which can be found in [2, 11, 19, 44, 83, 116, 117, 200, 258].
Due to its simplicity and robustness against parameter variations and disturbances, SMC has been studied extensively for many kinds of systems such as uncertain systems, time-delay systems, stochastic systems, parameter-switching systems, and singular systems. Many important SMC methodologies have been reported in literature. Here, we review some recently developed results in this area.
Uncertainties exist in all practical physical systems, and the robust control, as a branch of control theory, is invented to explicitly deal with system uncertainties and to achieve robust performance and/or stability for controlled systems. SMC, as one of the robust control strategies, is well known for its strong robustness to system uncertainties in sliding motion. However, the uncertainties should satisfy the so-called ‘matching’ condition, that is, the uncertainties act within channels implicit in the control input. If a system has mismatched uncertainties in the state matrix or/and the input matrix, the conventional SMC approaches are not directly applicable. Therefore, in the past two decades, many researchers have investigated the SMC of uncertain systems with mismatched uncertainties/disturbances – see for example [31, 40, 41, 43, 112, 193] and references therein. To mention a few, in [112], the SMC of uncertain second-order single-input systems with mismatched uncertainties, was considered; in [40, 193], the authors investigated the SMC design for uncertain systems, in which the uncertainties are mismatched and exist only in state matrix. The related approaches were then developed in [41, 43] to deal with a more complicated case that the mismatched uncertainties are involved in not only the state matrix but also the input matrix. In addition, the integral SMC techniques were extensively used to deal with uncertain systems with mismatched uncertainties – see for example, [27, 30, 43, 170, 233] – and some other SMC approaches to deal with uncertain systems can be found in [65, 108, 172, 194, 229].
It is well known that time delays appear commonly in various practical systems, such as communication, electronic, hydraulic, and chemical processes. Their existence can introduce instability, oscillation, and poor performance [168]. Time-delay systems have continuously been receiving considerable attention over the past decades. The main reason is that many processes include after-effect phenomena in their inner dynamics, and engineers need their models to approximate the real processes more accurately due to the ever-increasing expectations of dynamic performance. Stability analysis is a fundamental and vital issue in studying time-delay systems, and the conservativeness of a stability condition is an important index to evaluate a stability result. Several methods have been proposed to develop delay-dependent stability conditions (which have less conservativeness compared to delay-independent ones), such as the model transformation approach (based on Newton–Leibniz formula) [121, 110], the descriptor system approach [74], the slack matrix approach [228, 243], the delay partitioning approach [86], and the input-output method (based on the small gain theorem) [88]. There have been a number of excellent survey papers on the stability analysis of time-delay systems – see for example, [168, 246]. SMC of time-delay systems have also been receiving considerable attention over the past decades – see for example, [69, 75, 85, 91, 120, 123, 160, 162, 212, 234, 250] and the references therein. To mention a few, El-Khazali in [69] proposed an output feedback robust SMC for uncertain time-delay systems, and the delay variables were considered as external perturbation when designing the sliding surface; Fridman et al. in [75] presented a descriptor approach to SMC of systems with time-varying delays; Xia and Jia in [234] considered the SMC of time-delay systems with mismatched parametric uncertainties by using a delay-independent approach and the LMI technique; Yan in [250] studied the SMC of uncertain time-delay systems with a class of nonlinear inputs by using a delay-dependent approach; Wu et al. in [212] investigated a sliding mode observer design and an observer-based SMC for a class of uncertain nonlinear neutral delay systems; Han et al. in [91] addressed the SMC design for time-varying input-delayed systems by using a singular perturbation approach.
Stochastic systems and processes have come to play an important role in many fields of science, engineering, and economics. Thus, stochastic systems have received considerable attention, in which the stochastic differential equations are the most useful stochastic models with extensive applications in aeronautics, astronautics, chemical or process control system, and economic systems. A great number of methods and techniques have been developed for stochastic systems governed by Itô stochastic differential equations – see for example, [144, 145, 240, 241]. SMC design scheme for stochastic systems has also been developed – see for example, [8, 12, 13, 33, 98, 99, 155, 156, 158] and references therein. In [33], based on the concept of SMC, the steady-state covariance assignment problem was investigated for perturbed stochastic multivariable systems. The robust integral SMC and the robust sliding mode observer were designed for uncertain stochastic systems with time-varying delay in [155, 156], respectively. In [98], SMC of nonlinear stochastic systems was addressed by using a fuzzy approach. In [158], by utilizing the disturbance attenuation technique, a novel SMC method was proposed for nonlinear stochastic systems. In [8], a covariance control scheme was proposed for stochastic uncertain multivariable systems via SMC strategy. In [99], a robust SMC design scheme was developed for discrete-time stochastic systems with mixed time delays, randomly occurring uncertainties, and randomly occurring nonlinearities.
The parameter-switching hybrid system, which is the main plant considered in this book, consists of two types: Markovian jump systems and switched hybrid systems. Parameter-switching systems have received considerable research attention in the past two decades – see for example, [53, 131, 136] – since such systems are capable of modeling a wide range of practical systems that are subject to abrupt variations in their structures, owing to random failures or repairs of components, sudden environmental disturbances, changing subsystem interconnections, abrupt variations, and so on. An overview of the development of uncertain parameter-switching hybrid systems is presented in Section 1.12, from which we can see that the study on such systems, including the problems of stability analysis, stabilization, optimal control, filtering and model reduction, have been fully developed. However, SMC of parameter-switching hybrid systems, as a relatively new problem, has had only limited attention, and further research in this area is needed. There have been some results reported in the literature – see for example, [34, 35, 125, 135, 157, 178, 210, 224, 227] and references therein. More recently, for Markovian jump systems, Shi et al. in [178] presented an SMC design scheme by designing a linear mode-dependent sliding surface; Niu et al. in [157] investigated the SMC of Markovian stochastic systems by designing an integral mode-dependent sliding surface; Ma and Boukas in [135] proposed a singular system approach to robust sliding mode control for uncertain Markovian jump systems; Chen et al. in [35] developed an adaptive SMC for stochastic Markovian jump systems with actuator degradation. For switched hybrid systems, Wu and Lam in [210] proposed a linear mode-independent sliding surface in designing SMC for switched hybrid systems with time delay, and then the results were developed to deal with the SMC design problem for switched stochastic systems in [224]. Wu et al. in [227] investigated the dissipativity-based SMC design for switched stochastic systems, in which an integral mode-dependent sliding surface was designed such that the sliding motion is strictly dissipative.
The conventional implementation of SMC schemes is usually based on state feedback, which requires the assumption that all the state variables of the controlled systems are completely accessible for feedback. Such an assumption, however, is not always valid in practice since some state components cannot be measured. Roughly, there are two commonly used methods to deal with the controller design in the case that the system state components are not fully accessible. One approach is first to design an observer or a filter to estimate the immeasurable state components, and then synthesize an observer-based sliding mode controller – see for example, [154, 184, 212, 252]. However, the observer-based SMC scheme will require more hardware and will increase system dimension. The other approach is to design a feedback controller by using the measurable output information, which is called the output feedback SMC approach.
During the past two decades, output feedback SMC approaches have been intensively studied, and many important results have been reported in the literature – see for example, [32, 42, 48, 61, 62, 63, 64, 68, 90, 113, 163, 253, 263] and references therein. To mention a few, output feedback SMC design for uncertain dynamic systems was investigated in [263], and an algorithm for output-dependent hyperplane design was proposed based upon eigenvector methods. The eigenvalue assignment approach was proposed in [68] to design the sliding surface of the output feedback SMC scheme. The LMI technique was applied to output feedback SMC design in [62, 63]. Output feedback SMC design for state-delayed systems was investigated in [90, 253]. The above-mentioned results are all for static output feedback (SOF) SMC problems. In fact, output feedback control has two different forms: the SOF control and dynamic output feedback (DOF) control. Generally speaking, DOF control is more flexible than SOF control since the additional dynamics of the controller is introduced. Although DOF control involves more design parameters, for linear systems the closed-loop system can usually be written in a more compact form where certain parameters can be embedded into augmented matrix variables. Compared to the SOF SMC design, the DOF SMC design problem has received less attention, and only a few results have been reported, for example, in [32, 163] the DOF SMC was studied for MIMO linear systems with mismatched norm-bounded uncertainties and matched nonlinear disturbances.
Switched systems form a class of hybrid systems consisting of a family of subsystems described by continuous- or discrete-time dynamics, and a rule specifying the switching among them [129, 191]. The switching rule in such systems is usually considered to be arbitrary. Switched systems have received increasing attention in the past few years, since many real-world systems such as, chemical processes, transportation systems, computer-controlled systems, and communication industries can be modeled as switched systems [131]. More importantly, many intelligent control strategies are designed based on the idea of controllers switching to overcome the shortcomings of the traditionally used single controller and to improve their performance, thus making the corresponding closed-loop systems into switched systems.
Switched hybrid systems with all subsystems described by linear differential or difference equations are called switched linear hybrid systems. A continuous-time switched linear system can be modeled as
where x(t) ∈ Rn is the state vector; u(t) ∈ Rm is the control input; {(A(α(t)), B(α(t))): . is a family of matrices parameterized by an index set and is a piecewise constant function of time t called a switching signal. At a given time t, the value of α(t), denoted by α for simplicity, might depend on t or x(t), or both, or may be generated by any other hybrid scheme. Therefore, the switched hybrid system effectively switches among N subsystems with the switching sequence controlled by α(t). It is assumed that the value of α(t) is unknown a priori, but its instantaneous value is available in real time.
Similarly, a discrete-time switched linear hybrid system can be described by
where x(k) ∈ Rn is the state vector; u(k) ∈ Rm is the control input; {(A(α(k)), B(α(k))): . is a family of matrices parameterized by an index set , and is a piecewise constant function of time, called a switching signal, which takes its values in the finite set . At an arbitrary discrete time k, the value of α(k), denoted by α for simplicity, might depend on k or x(k), or both, or may be generated by any other hybrid scheme.
The stability analysis of switched hybrid systems is a fundamental issue for the synthesis of such systems. Note that there are two facts related with the stability of switched hybrid systems: 1) a switched hybrid system may have divergent trajectories even when all the subsystems are stable; and 2) a switched hybrid system may have convergent trajectories even when some of the subsystems are unstable. These two facts show that the stability of a switched hybrid system depends not only on the dynamics of each subsystem but also on the properties of the switching signals.
When focusing on stability analysis of switched hybrid systems, there are many valuable results that have appeared in the past two decades, and interested readers may refer to survey papers, such as [53, 128, 131, 148], and books, such as [129, 191]. In the following, we will briefly overview some recently developed results.
We first consider the stability analysis of the switched hybrid systems without any restrictions on switching signal, that is, the switching is arbitrary. Several approaches have been reported on the stability analysis of switched hybrid systems with arbitrary switching, for example:
The above stability analysis result based on the SQLF approach will turn out to be the above-mentioned one with the CQLF approach if . Obviously, the SQLF approach is less conservative than the CQLF approach. Some results on the SQLF approach to stability analysis and control synthesis for switched hybrid systems can be found in [46, 71].
Stability analysis approaches for arbitrary switching have been developed, but a natural question may still be raised, that is, can switched hybrid systems be stable under some restricted switchings in spite of the fact that they fail to preserve stability under arbitrary switching? If so, what kinds of restrictions should be put on the switching signals to guarantee the stability of switched hybrid systems? To answer such questions, there have been some stability analysis approaches for switched hybrid systems under restricted switching, for example,
For any T2 > T1 ≥ 0, let Nα(T1, T2) denote the number of switching of α(t) over (T1, T2). Here, Ta is called an average dwell time and N0 is the chatter bound. It has been proved in [92] that if all the subsystems are exponentially stable then the switched hybrid system remains exponentially stable provided that the average dwell time Td is sufficiently large. By using the average dwell time approach, Zhai et al. in [264] investigated the disturbance attenuation properties of continuous-time switched hybrid systems, and then the exponential stability and ℓ2 gain properties for discrete-time switched hybrid systems was investigated in [265]; Sun et al. in [188] studied the exponential stability and weighted -gain for switched delay systems; Wu and Lam in [215] considered the filtering problem of switched hybrid systems with time-varying delay. As well as the above-mentioned results, the model reduction problem for switched hybrid systems with time-varying delay was addressed in [219] by using the average dwell time approach incorporated with a piecewise Lyapunov function; and the DOF controller design problem was considered in [216, 217].
Over the past several decades, considerable interest has been devoted to synthesis problems of switched hybrid systems, including stabilization, robust/optimal control, state estimation/ filering, fault detection, model approximation, and so on. Here, we will review some relevant literature on the synthesis of switched hybrid systems. First, we introduce two important properties of switched hybrid systems, namely, the controllability and the observability. Roughly speaking, the concept of controllability denotes the ability to move a system around in its entire configuration space using only certain admissible manipulations. Observability is a measure for how well internal states of a system can be inferred by knowledge of its external outputs. Observability and controllability are dual aspects of the same problem. Some results on the controllability and the observability analysis for switched hybrid systems were reported in [16, 37, 105, 132, 166, 185, 189, 206, 235, 236, 270] and references therein.
In the previous section, we discussed the stability properties of switched hybrid systems. As mentioned earlier, the stability of a switched hybrid system depends not only on the dynamics of each subsystem but also on the properties of the switching signals, thus the synthesis problems include two strategies for implementation. The first is based on the subsystems’ dynamics with given switching signals, and the second is based on the switching signals. The stabilization problem for switched hybrid systems was investigated in [4, 6, 14, 38, 45, 46, 79, 84, 95, 100, 102, 124, 130, 140, 143, 190, 192, 222, 237, 247, 249, 267] and references therein.
Over the past decade, considerable attention has been paid to robust and optimal control problems for switched hybrid systems, and many important results have been reported – see for example, [17, 47, 51, 67, 80, 81, 106, 126, 138, 153, 161, 171, 217, 216, 232, 248, 264, 268] and references therein. To mention a few, Geromel et al. in [81] considered the passivity analysis and controller design problems; Kamgarpour and Tomlin in [106] studied the optimal control problem for non-autonomous switched systems with a fixed mode sequence; Lian and Ge in [126] addressed robust output tracking control for switched systems under asynchronous switching; Mahmoud in [138] proposed a generalized control design approach for discrete-time switched systems with unknown delays; Niu and Zhao in [153] used the average dwell time approach to the robust control problem for a class of uncertain nonlinear switched systems; Orlov in [161] presented finite time stability analysis and robust control synthesis methods for uncertain switched systems; Seatzu et al. in [171] studied the optimal control problem for continuous-time switched affine systems. The above-mentioned results are all based on state feedback control, and the output feedback control problem was also investigated – see for example, [51, 67, 80, 216, 217]. In addition, SMC design methodologies for switched hybrid systems were proposed in [210, 224, 227, 262].
It is well known that one of the fundamental problems in control systems and signal processing is the estimation of the state variables of a dynamical system through available noisy measurements, which is referred to as the filtering problem. The celebrated Kalman filter has been considered as the best possible (optimal) estimator for a large class of systems; it is an algorithm that uses a series of measurements observed over time, containing noise (random variations) and other inaccuracies, and produces estimates of unknown variables that tend to be more precise than those based on a single measurement alone. The Kalman filter for switched discrete-time linear systems was designed in [3]. However, the application of the Kalman filter is subject to two initial assumptions: the underlying system is linear with complete knowledge of the dynamical model, and the noise concerned is white/colored with known spectral density. Thus the Kalman filtering scheme is no longer applicable when a priori information on the external noises is not precisely known. Therefore, the past two decades have witnessed significant progress on robust filtering involving various approaches such as filtering, filtering, - filtering, and mixed / filtering. The robust filtering problem for switched hybrid systems has also been developed over the past decade – see for example, [50, 137, 139, 167, 202, 215, 216, 272] and references therein. To mention a few important robust filtering results for switched hybrid systems, Deaecto et al. in [50] developed a trajectory-dependent filter design approach for discrete-time switched linear systems; Mahmoud in [137] presented a delay-dependent filter design approach for a class of discrete-time switched systems with state delay; Qiu et al. in [167] investigated the robust mixed / filtering design for discrete-time switched polytopic linear systems; Wang et al. in [202] addressed the filtering problem for discrete-time switched systems with state delays via switched the Lyapunov function approach; Wu and Lam in [215] proposed an average dwell time approach to the weighted filter design for switched systems with time-varying delay; Wu and Ho in [216] developed a reduced-order - filter design scheme for a class of nonlinear switched stochastic systems.
The issues of fault detection and isolation are increasingly required in various kinds of practical complex systems for guaranteeing reliability and pursuing performance. Hence, how to develop effective methods for timely and accurate diagnosis of faults becomes a crucial problem. To combat this, many significant schemes have been introduced, such as model-based approaches and knowledge-based methods. Among them, the model-based approach is the most favored. The basic idea of model-based fault detection is to construct a residual signal and, based on this, determine a residual evaluation function to compare with a predefined threshold. When the residual evaluation function has a value larger than the threshold, an fault alarm is generated. Since accurate mathematical models are not always available, the unavoidable modeling errors and external disturbances may seriously affect the performance of model-based fault detection systems. Thus, the designed fault detection systems should be both sensitive to faults and suppressive to external disturbances. Fortunately, the fault detection filter or observer is known to be able to do a good job of achieving the above-mentioned requirements. The fault detection problem for switched hybrid systems was studied in [203, 261], and some other approaches can be found in [15, 49, 119]. Fault-tolerant control is a related issue that makes it possible to develop a control feedback that allows the required system performance to be maintained in the case of faults. The fault-tolerant control problem for switched hybrid systems has also been investigated: for example, Du et al. in [59] proposed an active fault-tolerant controller design scheme for switched systems with time delay; Li and Yang in [119] developed a simultaneous fault detection and control technique for switched systems under asynchronous switching; Wang et al. in [205] designed a robust fault-tolerant controller for a class of switched nonlinear systems in lower triangular form.
Mathematical modeling of physical systems often results in complex high-order models, which bring serious difficulties to analysis and synthesis of the systems concerned. Therefore, in practical applications it is desirable to replace high-order models by reduced ones with respect to some given criterion, which is the model reduction problem. Over the past decades, the model reduction problem has been the concern of many researchers. Many important results have been reported, which involve various efficient model reduction approaches, such as the balanced truncation approach [89], the Hankel-norm approach [82], Krylov projection approach [87], the Padé reduction approach [7], the approach [251], and the approach [217, 219]. Readers can refer to [5] for a detailed survey of model reduction. The model reduction problem for switched hybrid systems has also received considerable attention – see for example, [18, 133, 150, 173, 179, 204, 217, 219] and references therein. To mention a few important results, Birouche et al. in [18] investigated the model order-reduction for discrete-time switched linear systems by the balanced truncation approach; Monshizadeh et al. in [150] developed a simultaneous balanced truncation approach to model reduction of switched linear systems; Shi et al. in [179] studied the model reduction problem for discrete-time switched linear systems over finite frequency ranges; Wang et al. in [204] developed a delay-dependent model reduction approach for continuous-time switched state-delayed systems; and Wu and Zheng in [219] proposed a weighted model reduction approach for linear switched systems with time-varying delay.
Markovian jump linear systems (MJLSs) are another typical class of parameter-switching systems, and they are modeled by a set of linear systems with the transitions between the models determined by a Markov chain, taking values in a finite set [136]. MJLSs can also be considered as a special case of switched hybrid systems with the switching signals governed by a Markov chain. Applications of MJLSs may be found in many processes, such as target tracking problems, manufactory processes, solar thermal receivers, fault-tolerant systems, and economic problems. From a mathematical point of view, MJLSs can be regarded as a special class of stochastic system with system matrices changed randomly at discrete time points governed by a Markov process and remaining LTI between random jumps. Over the past decades, owing to a large number of applications in control engineering, MJLSs have received increasing interest. Many results in this field can be found in the literature, and in the following, we will review some recently published results on MJLSs.
The stability analysis and stabilization problems for MJLSs were addressed in [20, 22, 28, 56, 76, 104, 144, 174, 176, 187, 207, 238, 239]. Specifically, Cao and Lam in [28] investigated the stochastic stabilizability and control for discrete-time jump linear systems with time delay; de Souza in [56] studied the robust stability and stabilization problems for uncertain discrete-time MJLSs; Gao et al. in [76] considered the stabilization and control problems for two-dimensional MJLSs; Sun et al. in [187] dealt with the robust exponential stabilization of MJLSs with mode-dependent input delay; Xiong et al. in [238] studied the robust stabilization problem for MJLSs with uncertain switching probabilities. In addition, there have been some results on the stability and stabilization for Markovian jump stochastic systems. For example, Boukas and Yang in [20] proposed an exponential stabilizability condition for stochastic systems with Markovian jump parameters; Wang et al. in [207] solved the stabilization problem for bilinear uncertain time-delay stochastic systems with Markovian jump parameters; and some other results on Markovian jump stochastic systems can be found in [144].
The control for MJLSs was investigated in [21, 24, 28, 29, 76, 115, 230, 245]; robust control of MJLSs with unknown nonlinearities was studied in [21]. control was addressed in [24] for discrete-time MJLSs with bounded transition probabilities; the robust control problem was considered in [29] for uncertain MJLSs with time delay; the robust control of descriptor discrete-time Markovian jump systems is covered in [115]; delay-dependent control for singular Markovian jump systems with time delay appears in [230]; delay-dependent control and filtering for uncertain Markovian jump systems with time-varying delays are found in [245].
The filtering problem for MJLSs was considered in [54, 55, 57, 134, 175, 177, 208, 211, 231, 242, 254, 256]. To mention a few, de Souza and Fragoso studied the filter design problem for continuous- and discrete-time MJLSs in [54, 55], respectively; Ma and Boukas in [134] investigated robust filtering for uncertain discrete Markovian jump singular systems with mode-dependent time delay; Shi et al. in [175] considered Kalman filtering for continuous-time uncertain MJLSs; Wu et al. in [211] addressed the filtering problem for Markovian jump two-dimensional systems; Yao et al. in [254] dealt with robust filtering of Markovian jump stochastic systems with uncertain transition probabilities; and then they studied quantized filtering for Markovian jump LPV systems with intermittent measurements in [256].
The fault detection problem for MJLSs was investigated in [147, 152, 226, 255, 273, 274]. Specifically, Meskin and Khorasani in [147] investigated fault detection and isolation problems for discrete-time MJLSs with application to a network of multi-agent systems having imperfect communication channels; Nader and Khashayar proposed a geometric approach to fault detection and isolation of continuous-time MJLSs in [152]; Wu et al. in [226] studied generalized fault detection for Markovian jump two-dimensional systems; Yao et al. in [255] considered fault detection filter design for Markovian jump singular systems with intermittent measurements; and Zhong et al. in [273, 274] addressed robust fault detection problem for continuous- and discrete-time MJLSs, respectively.
The SMC design problem was also addressed for MJLSs in [34, 35, 135, 157, 178, 221, 223, 225]. Chen et al. studied SMC of MJLSs with actuator nonlinearities in [34], and adaptive SMC design for Markovian jump stochastic systems with actuator degradation in [35]; Ma and Boukas in [135] proposed a singular system approach to robust SMC for uncertain MJLSs; Shi et al. in [178] considered the SMC design problem for MJLSs; Wu and Ho in [221] solved the SMC problem for Markovian jump singular stochastic hybrid systems; Wu et al. in [223] investigated state estimation and SMC of Markovian jump singular systems; and then they considered SMC design with bounded gain performance for Markovian jump singular time-delay systems in [225].
Apart from the above-mentioned synthesis problems for MJLSs, the model reduction problem for such systems was also investigated – see for example, [111, 266]. Kotsalis et al. in [111] studied the model reduction problem for discrete-time MJLSs; and Zhang et al. in [266] considered model reduction for both continuous- and discrete-time MJLSs.
This book represents the first of a number of attempts to reflect the state-of-the-art of the research area for handling the SMC problem for uncertain parameter-switching hybrid systems (including Markovian jump systems, switched hybrid systems, singular systems, stochastic systems, and time-delay systems). The content of this book can be divided into three parts. The first part is focused on SMC of Markovian jump singular systems. Some necessary and sufficient conditions are derived for the stochastic stability, stochastic admissibility, and optimal performances by developing new techniques for the considered Markovian jump singular systems. Then, a set of new SMC methodologies are proposed, based on the analysis results. In the second part, the problem of SMC of switched delayed hybrid systems is investigated. A unified framework under ‘average dwell time’ is established for analyzing the considered switched delayed hybrid systems. Then some sufficient conditions are derived for the stability, stabilizability, existence of the desired DOF controllers, and existence of the sliding mode dynamics in the SMC issue. More importantly, a set of SMC methodologies under a unique framework are proposed for the considered hybrid systems. In the third part, the parallel theories and techniques developed in the previous part are extended to deal with switched stochastic hybrid systems. Specifically, in this third part, the main attention will be focused on stochastic stability analysis, stabilization, control, and SMC of switched stochastic hybrid systems. Sufficient conditions are established first for the stochastic exponential stability and optimal performances (such as and dissipativity) of the continuous- and discrete-time switched stochastic systems. Based on the obtained analysis results, the synthesis issues, including control and SMC design, are solved.
The features of this book can be highlighted as follows. 1) A unified framework is established for SMC of Markovian jump singular systems, where the parameters are jumping from one mode to another stochastically, and at the same time there are time delays in existing system states. 2) A series of problems are solved with new approaches for analysis and synthesis of continuous- and discrete-time switched hybrid systems, including stability analysis and stabilization, DOF control, and SMC. 3) Three correlated problems, control (state feedback control and DOF control), SMC, and state estimation problems, are dealt with for switched stochastic systems. 4) A set of newly developed techniques (e.g. average dwell time method, piecewise Lyapunov function approach, parameter-dependent Lyapunov function approach, cone complementary linearization (CCL) approach, slack matrix approach, and sums of squares technique) are exploited to handle the emerging mathematical/computational challenges.
This book is a timely reflection on the developing area of system analysis and SMC theories for systems with uncertain switching parameters, typically resulting from varying operation environments. It is a collection of a series of latest research results and therefore serves as a useful textbook for senior and/or graduate students who are interested in knowing: 1) the state of the art of the SMC area; 2) recent advances in Markovian jump systems; 3) recent advances in switched hybrid systems; and 4) recent advances in singular systems, stochastic systems and time-delay systems. Readers will also benefit from new concepts, models and methodologies with theoretical significance in system analysis and control synthesis. The book can also be used as a practical research reference for engineers dealing with SMC, optimal control, and state estimation problems for uncertain parameter-switching hybrid systems. The aim of this book is to close the gap in literature by providing a unified, neat framework for SMC of uncertain parameter-switching hybrid systems.
In general, this book aims at third- or fourth-year undergraduates, postgraduates and academic researchers. Prerequisite knowledge includes linear algebra, matrix analysis, linear control system theory, and stochastic systems. It should be described as an advanced book.
More specifically, the readers should include: 1) control engineers working on nonlinear control, switching control, and optimal control; 2) system engineers working on switched hybrid systems and stochastic systems; 3) mathematicians and physicists working on hybrid systems and singular systems; and 4) postgraduate students majoring on control engineering, system sciences, and applied mathematics. This book could also serve as a useful reference to: 1) mathematicians and physicists working on complex dynamic systems; 2) computer scientists working on algorithms and computational complexity; and 3) third- or fourth-year students who are interested in knowing about advances in control theory and applications.
The organization structure of this book is shown in Figure 1.6. The general layout of this book is divided into three parts: Part One: SMC of Markovian jump singular systems; Part Two: SMC of switched hybrid systems with time-varying delay; and Part Three: SMC of switched stochastic hybrid systems. The main contents of this book are shown in Figure 1.7.
Part One presents the analysis and SMC design procedure for Markovian jump singular systems. It begins with Chapter 2, and consists of three chapters as follows.
Part Two presents the analysis and SMC design procedure for switched state-delayed hybrid systems. It begins with Chapter 5, and consists of four chapters as follows.
Part Three presents the analysis and SMC design procedure for switched stochastic hybrid systems. It begins with Chapter 9, and consists of four chapters as follows.