Filtering, Control, and Optimization of Distributed Networked SystemsView this Special Issue
Research Article | Open Access
Liu Yaqiang, Ren Zhigang, Jin Zengwang, "Robust Feedback Compensator Design for Linear Parabolic DPSs with Pointwise/Piecewise Control and Pointwise/Piecewise Measurement", Complexity, vol. 2021, Article ID 6660857, 14 pages, 2021. https://doi.org/10.1155/2021/6660857
Robust Feedback Compensator Design for Linear Parabolic DPSs with Pointwise/Piecewise Control and Pointwise/Piecewise Measurement
In this paper, a robust control problem of a class of linear parabolic distributed parameter systems (DPSs) with pointwise/piecewise control and pointwise/piecewise measurement has been investigated via the robust feedback compensator design approach. A unified Lyapunov direct approach is proposed in consideration of the pointwise/piecewise control and point/piecewise measurement based on the distributions of the actuators and sensors. A new type of Luenberger observer is developed on the continuous interval of space domain to track the state of the system, and an performance constraint with prescribed attenuation levels is proposed in this paper. By utilizing Lyapunov technique, mathematical inequalities, and integration theory, a sufficient condition based on LMI for the exponential stability of the corresponding closed-loop coupled system under an performance constraint is presented. Finally, the effectiveness of the proposed design method is verified by numerical simulation results.
Distributed parameter systems (DPSs) are infinite-dimensional in nature and are generally modeled by partial differential equations (PDEs). DPSs are widely used in engineering systems [1–4], such as thermodynamics, chemical engineering, missile, aerospace, aviation, and nuclear fission and fusion. Control problem of DPSs has attracted extensive attention due to the important applications in engineering systems, such as the vibration control of flexible structures that the vibration process can be described by Euler–Bernoulli equations, the diffusion control of oil spill that the diffusion phenomena can be described by diffusion equations, and the temperature control of heating furnace that the thermal conduction process can be described by heat equations. In recent decades, fruitful achievements in the design of DPSs control have been achieved from many scholars all over the world [5–17].
Generally, control forms of DPSs can be divided into boundary control and in-domain control based on the actuators’ location. Fruitful achievements on boundary control of DPSs that the actuators are located at the boundary area have been published already. For example, the boundary control problem of flexible robot manipulators has been developed to solve the DPSs with flexible structures [18–20]. This technique has been extended to the boundary antidisturbance control and boundary adaptive robust control for flexible DPSs [21–23]. Boundary control of nonlinear parabolic DPSs has been studied in , in which the continuum backstepping method is utilized. Boundary feedback control of DPSs has been addressed in , and a novel combination of feedback idea and backstepping approach is presented in . Sampled-data boundary control and sliding mode boundary control of DPSs have been studied in [27, 28], where a sampled-data strategy for the boundary control problem is proposed. Fuzzy boundary control based on the T-S fuzzy DPS model is shown in [29, 30]. boundary control has been proposed in  that a linear matrix inequality (LMI) approach has been utilized. Meanwhile, there are also some achievements on in-domain control of DPSs. For example, pointwise control of DPSs with T-S fuzzy DPS model has been developed in , where a fuzzy state feedback controller is designed. Furthermore, this technique has been extended to the [33, 34]. Robust sampled-data control has been proposed in [35, 36], where the sampled-data pointwise controller method is applied. Mobile piecewise control of DPSs has been studied in  that a mobile actuator-plus-sensor network is developed, and this technique has been extended to solve the DPSs in . More recently, collocated control and noncollocated control of in-domain control in DPSs have been studied deeply. Static collocated feedback control has been presented in [39, 40] that the static collocated pointwise and piecewise feedback controller has been designed for parabolic DPSs. For the noncollocated control that the actuators and sensors can never be placed at the same location exactly, the static feedback control has been studied in [32, 36, 41, 42], and the observer-based dynamic feedback control has been designed in [43–47]. The estimation problems in controller design of DPSs have been studied in [48–53], and for some DPSs with unknown parameters, parameter estimation methods have been applied in [54–58].The design and analysis methods have also been extended to switched control systems and filtering technique in [59–62]. Although there have been many promising efforts, there are still many control problems of DPSs need to be studied in the future.
In general, disturbance problems of DPSs are unavoidable because of the errors from model calculations and equipments. Thus, an approach of robust control is proposed to deal with the control problem of DPSs with external disturbances. The robust control has attracted much attention from many scholars over the past few decades. For example, an static output feedback boundary controller for semilinear parabolic and hyperbolic DPSs is proposed in . This idea has extended to solve the sampled-data distributed control problem for a class of parabolic DPSs in . An fuzzy observer-based controller is proposed for a class of nonlinear parabolic DPSs in , and this technique has developed to the observer-based sampled-data fuzzy control design in [46, 64] and mixed fuzzy observer-based feedback control design in . In this paper, we will extend the works in [66, 67] to design the output feedback compensator for linear parabolic DPSs with external disturbances by using a unified Lyapunov approach. A sufficient condition for the static feedback compensator can stabilize the DPSs under an performance constraint with the collocated observation case which is first proposed in terms of standard linear matrix inequalities (LMIs); then, another sufficient condition for the observer-based dynamic feedback compensator can stabilize the DPSs under an performance constraint with the noncollocated observation case which is developed by using the Lyapunov direct method, Poincaré–Wirtinger inequality’s variants, Cauchy–Schwartz inequality, integration by parts, and first mean value theorem for definite integrals.
The main contributions and novelty of this paper compared with the existing works before are summarized as follows:(i)Different from the results in [32–34, 43, 44] that the pointwise/piecewise control and collocated (or noncollocated) pointwise/piecewise measurement are discussed separately, all the forms of control and observation are considered in this paper by a unified Lyapunov direct approach.(ii)In contrast to the works in  that a unified Lyapunov-based compensator design for linear parabolic DPSs with free disturbance, this paper presents the static and dynamic robust output feedback compensator design for linear DPSs with external disturbances. Meanwhile, a new type of Luenberger observer is designed for noncollocated observation in space with continuous observation functions.(iii)An performance constraint in the sense of is proposed to deal with the external disturbance of the model and measurement disturbance in the measurement output.
The organizational structure of the remaining parts of this paper is arranged as follows: first, the problem formulation of this paper and some preliminary knowledge are presented in Section 2. Then, the static output feedback compensator design and observer-based dynamic output feedback compensator design in terms of collocated and noncollocated observation in space satisfying the performance constraint are shown in Section 3. Section 4 provides some numerical simulation results of the corresponding closed-loop systems to show the effectiveness of the proposed design method. Finally, brief conclusions are followed in Section 5.
Notation: , , and denote the set of all real numbers, -dimensional Euclidean space, and the set of all matrices, respectively. is a real Hilbert space of square integrable functions with the inner product induced norm . is a real Hilbert space of square integrable functions with . is a real Hilbert space of square integrable functions with . stands for the partial derivative of with respect to , i.e., , , and stands for the first-order and second-order partial derivative of with respect to , i.e., , , respectively. and denote two sets of natural numbers, i.e., , .
2. Problem Formulation and Preliminaries
In this paper, we consider a class of one-dimensional linear parabolic DPSs with external disturbances of the following form:subject to the Robin boundary conditions in one dimension,and the initial condition,where denotes the spatial position between , and denotes time, respectively. , denotes the state variable. is a known constant. is an unknown bounded external disturbance and satisfies . is a known integrable vector function of , and the element describes the distribution of -th actuator on the spatial domain . is a set of control inputs from actuators, expressed as . is a set of measurement outputs from sensors, expressed as . is a known integrable vector function of , and the element describes the distribution of -th sensor on the spatial domain . is the -dimensional bounded measurement disturbance vector. It should be pointed out that when , the one-dimensional linear parabolic DPSs is unstable.
Remark 1. It is worth noting that equation (1) is equivalent to the following general form :through the conversion of the following state variables and control variables:where is a known scalar function and continuously differentiable in time .
In practical applications of DPSs, the number of actuators and sensors is usually limited and actives at specified point or part thereof in the spatial domain, respectively. Therefore, the in-domain control forms of DPSs are generally divided into pointwise control and local piecewise control according to the distribution of actuators. In this paper, two forms of in-domain control are both considered; the actuators’ spatial distribution functions are described as follows:where is the Dirac delta function . The points , and local subdomains , satisfy and , which imply the chosen functions , produce pointwise control at the points and local piecewise uniform control over , respectively. Meanwhile, the spatial domain can be divided into parts by a spatial domain decomposition approach that . The locations of the actuators for pointwise control and local piecewise control satisfy the relationships , and .
Similar to the actuators’ distribution, the in-domain observation forms are generally divided into pointwise measurement and local piecewise measurement; the sensors’ spatial distribution functions are described in this paper as follows:
The points , and local subdomains satisfy and , which imply the chosen functions produce pointwise observation at the points and local piecewise uniform observation over , respectively. At the same time, the spatial domain can be divided into parts by a spatial domain decomposition approach that . The locations of the sensors for pointwise measurement and local piecewise measurement satisfy the relationships and .
Definition 2. The linear parabolic DPS (1)–(3) with the designed output feedback compensator is exponentially stable in the sense of under an performance constraint, when the corresponding closed-loop DPS with and is exponentially stable in the sense of ; meanwhile, the performance constraint in (10) is ensured when the initial value of is zero () and all .
The following lemmas are very useful for the development of the robust compensator design in this paper.
Lemma 1 (Poincaré–Wirtinger inequality’s variants). For a scalar function , we havewhere , , , , and . For more detailed details, please refer to the lemma in .
Lemma 2. (Cauchy–Schwartz inequality ). For any two scalar functions and , there exists constant which makes the following inequality hold:
3. Robust Feedback Compensator Design
Based on the distributions of actuators and sensors that , in (6) (or (7)) and , in (8) (or (9)), the observation obtained from sensors can be divided into collocated observation in space (i.e., ) and noncollocated observation in space (i.e., ). In other words, the collocated observation in space is a special case of noncollocated observation. Meanwhile, the noncollocated observation in space (i.e., ) consists of the following cases: pointwise control and noncollocated pointwise observation case, pointwise control and noncollocated piecewise observation case, piecewise control and noncollocated pointwise observation case, and piecewise control and noncollocated piecewise observation case. In this section, all the noncollocated observation cases will be considered to study the robust dynamic output feedback compensator design for the DPS (1)–(3).
The observation functions are choosen assuch that .
Then, we design an observer-based dynamic output feedback compensator of the following form:where is the compensator gain in the form of diagonal matrix, and the compensator functions are choosen assuch that .
The estimation error state is defined as
Hence, the closed-loop coupled DPS is represented by the estimation error system (18) and the closed-loop system (19) with expressions (6) (or (7)) and (8) (or (9)). The objective of this subsection is to seek an effective method to design an observer-based dynamic output feedback compensator such that the resulting closed-loop coupled DPS is exponentially stable under an performance constraint in the sense of with prescribed attenuation levels and .
Theorem 1. Consider a class of linear parabolic DPSs (1)–(3) with for noncollocated observation (i.e., ). If there exist compensator gains , , and scalars , ,satisfying the following LMI constraints:in whichand then there exists an observer-based dynamic output feedback compensator (15) ensuring the exponential stability of the closed-loop coupled DPS (18), (19), (6) (or (7)), and (8) (or (9)) with in the sense of .
Proof. Assume the LMIs (20) and (21) are fulfilled. Consider the following Lyapunov function candidate:in whichwhere is a Lyapunov parameter to be determined.
The time derivative of for the closed-loop system (19) isBased on the Cauchy–Schwartz inequality, the following inequality is fulfilled for any scalar :Due to and , by Poincaré–Wirtinger inequality’s variants in Lemma 1, the following inequality is fulfilled for each :Substituting the inequalities (27) and (26) into (25), we getwhereThe time derivative of iswhere . The following expression is utilized by applying integration by parts and considering the boundary conditions in (18):As , and . By Lemma 1, we get for each ,Thus, the expression (30) can be rewritten aswhere , andFrom (28) and (33), the time derivative of defined in (23) is given asBy the Schur complement and the LMI constraint (20), we haveIt can be deduced from the LMI (21) and the inequality (36) that there exists an appropriate constant , satisfying the following inequalities:Substituting the inequalities (37) to (35), we obtainwhere . Integrating from 0 to for the inequality (38), we can get , which implies , . Thus, we can obtain . In other words, the closed-loop coupled DPS (18), (19), (6) (or (7)), and (8) (or (9)) with is exponentially stable in the sense of . The proof is complete.
Next, the performance analysis is developed for the closed-loop coupled DPS (18), (19), (6) (or (7)), and (8) (or (9)) with the initial value and all . The following theorem provides a sufficient condition for exponential stability in the sense of of the closed-loop coupled DPS (18), (19), (6) (or (7)), and (8) (or (9)) under a prescribed performance constraint (10) based on the LMI conditions.
Theorem 2. Consider a class of linear parabolic DPSs (1)–(3), given attenuation levels and constant , if there exist compensator gains , and scalars , , satisfying the following LMI constraints:in whichand then there exists an observer-based dynamic output feedback compensator (15) ensuring the exponential stability of the closed-loop coupled DPS (18), (19), (6) (or (7)), and (8) (or (9)) under a prescribed performance constraint (10).
Proof. From expressions (18), (19), and (35), the time derivative of is rewritten asConsidering the performance constraint in (10) and constant , we havewhere , , andBy the Schur complement and the LMI constraint (40), we haveIt is easily obtained from (43) in consideration of the inequality (45) and the LMI (40) thatIntergrating (46) from 0 to and considering the zero initial value , we get the expression (10). In other words, the performance constraint (10) with the given prescribed attenuation levels and is guaranteed for the closed-loop coupled DPSs (18), (19), (6) (or (7)), and (8) (or (9)) with the initial value and all . Meanwhile, the LMI constraints (20) and (21) in Theorem 1 can be derived from the LMI constraints (39) and (40) in Theorem 2. Therefore, the closed-loop coupled DPS (18), (19), (6) (or (7)), and (8) (or (9)) is exponential stable in the sense of under a prescribed performance constraint (10). The proof is complete.
4. Numerical Simulation
In this section, we will provide some numerical simulations to demonstrate the effectiveness of the proposed design strategy. For the linear parabolic DPS (1)–(3), given the constants , and , set the initial value , , the open-loop profile of evolution of , and the open-loop trajectory of , as shown in Figure 1. It is easily seen from Figure 1 that the linear parabolic DPS (1)–(3) with is unstable.
The numerical simulations of the corresponding closed-loop coupled DPS (18), (19), (7), and (9) for local piecewise control and noncollocated local piecewise measurement case in space are then shown. Assume three actuators and two sensors are applied for implementation of the proposed design method, that is, , and . Three actuators are respectively distributed over , and over the spatial domain , and two sensors are respectively distributed over over the spatial domain , i.e., , , , , , and