Research Article | Open Access

Raouaa Tayari, Ali Ben Brahim, Fayçal Ben Hmida, Anis Sallami, "Active Fault Tolerant Control Design for LPV Systems with Simultaneous Actuator and Sensor Faults", *Mathematical Problems in Engineering*, vol. 2019, Article ID 5820394, 14 pages, 2019. https://doi.org/10.1155/2019/5820394

# Active Fault Tolerant Control Design for LPV Systems with Simultaneous Actuator and Sensor Faults

**Academic Editor:**Jean Jacques Loiseau

#### Abstract

The present paper addresses the problem of robust active fault tolerant control (FTC) for uncertain linear parameter varying (LPV) systems with simultaneous actuator and sensor faults. First, fault estimation (FE) scheme is designed based on two adaptive sliding mode observers (SMO). Second, using the information of simultaneous system state, actuator, and sensor faults, two active FTC are conceived for LPV systems described with polytopic representation as state feedback control and sliding mode control. The stability of closed-loop systems is guaranteed by mean of performance; sufficient conditions of the proposed methods are derived in LMIs formulation. The performance effectiveness of FTC design is illustrated using a VTOL aircraft system with both sensor and actuator faults as well as disturbances. In addition, comparative simulations are provided to verify the benefits of the proposed methods.

#### 1. Introduction

In recent decades, the performance of industrial equipment has been increasing significantly. The gain in performance was accompanied by an increasing in the complexity of the installations causing a higher demand of strong availability and security. However, by realising some tasks a degradation of performances may occur; these degradations are caused by abnormal phenomena which are called faults. The faults may lead to performance deterioration or even instability of the system. To preserve robust system performance, fault-tolerant control designs have attracted significant attention like in [1–3].

In the literature, two types of FTC approaches have been distinguished: passive ([4, 5]) and active ([6–8]). The passive control uses a fixed controller to deal with any fault free and faulty cases, which is based on robust control techniques. Also, it consists of ensuring that the closed-loop system remains insensitive to the occurrence of certain defects. In the case of active approaches, depending on the severity of the fault impact on the system, a new control law is applied. Moreover, to ensure the highest possible performance of the controlled system in presence of a fault, an active FTC strategy is necessary. Thus, the FTC unit receives the signal from the FE module identifying the type of the fault; an appropriate decision must be made in order to preserve the system performance.

Compared with a linear system, FTC for nonlinear system is much more challenging because of its complexity. Linear parameter varying models offer a potent tool to research the nonlinear characteristics of systems, where the nonlinear system can be approximated by several local linear models and the global stability over the entire working space can also be easily guaranteed. Consequently, LPV systems have attracted considerable attention in [9–11].

Active FTC is often dedicated to linear systems or the linearization of nonlinear systems, but rarely to LPV systems. However, few studies take into account the FTC of LPV systems. In [12] an observer based FTC design for LPV descriptor systems with actuator faults is developed. A proportional derivative extended state observer (PDESO) based FE and FTC design is proposed in [13] for LPV descriptor systems with both actuator and sensor faults. In [14] a Low Order (LO) observer and FTC strategy for LPV systems subject to sensor faults are proposed. Finally, virtual actuator FE/FTC are developed in [15].

In the last decade, sliding mode (SM) concepts have been the focus of research because of their robustness. The SM control methodology has the advantage of producing low complexity control laws compared to other robust control approaches [16, 17]. One of the FTC approaches which has received increasing attention in recent years is SMC. Despite the popularity of SMC, there has been little work on LPV systems.

This active approach comprises FE observer and FTC control modules. The FE module determines the size or even the dynamic behavior of the fault. So, FE approach gives the detailed information of the fault signal which is necessary to be developed. In fact, many practical applications may have multiple single faults that appear simultaneously, which is known as “simultaneous faults” [18]. Several FE strategies based on LPV systems have been proposed, e.g., using learning observer [19], SMO [20, 21], and reduced-order observer [22]. These approaches are based on robustness concepts and are thus good candidates to include in active FE based FTC system analysis and design.

In this paper, the FE is obtained using a sliding mode observers-based actuator and sensor faults estimation for linear parameter varying systems by [23]. FTC is then realized through a state feedback and sliding mode controller, which can compensate for both actuator and sensor faults based on fault estimation.

The main contributions of this paper compared with the existing control based FTC work are summarized as follows:(i)This paper investigates new methods based on a robust active FTC for polytopic LPV system with simultaneous actuator and sensor faults as well as disturbances. Hence, the FTC strategy is developed to avoid the effects of simultaneous actuator and sensor faults on polytopic LPV systems. In many research works, the FTC design is only used for polytopic LPV systems either for actuator faults or sensor faults, but does not consider simultaneous actuator and sensor faults. This paper deals with an FTC synthesis in the presence of simultaneous faults.(ii)A novel FTC based fault estimation (FE) design using sliding mode for each one is studied for polytopic system. In the literature, the SM was only considered for either FTC or FE like in [24, 25]. The proposed method transforms the LPV system into two subsystems where the first one includes the effects of actuator faults but is free from sensor faults and the second subsystem has only sensor faults. Furthermore, two sliding mode observers are designed where each one estimate a class of faults. Based on the updated values of these estimations, the proposed sliding mode controller is constructed for polytopic LPV system with simultaneous faults.(iii)A robust design of FTC is against unknown input disturbances, especially against measurement noise.(iv)A comparative study is developed using the Integral Absolute Error (IAE) to measure the performance of fault tolerant sliding mode control which gives exact comparisons between different control schemes highlighting the effectiveness of the proposed sliding mode control.

The remaining part of this paper is organized as follows. Formulation of the problem is in Section 2. In Section 3, a simultaneous actuator and sensor faults estimation design are presented. FTC design using SMC and state feedback control are considered in Section 4. Section 5, provide an illustrative example. Finally, some concluding remarks are made in Section 6.

The following Lemma is presented, which is rather standard, to facilitate the proofs of the main results of this paper.

Lemma 1. *For matrices X and Y with appropriate dimensions, the following condition holds: where is a positive scalar.*

#### 2. Problem Formulation

Consider an uncertain LPV system subject to both actuator and sensor faults as follows:where the parameter varying matrices , , , , , and . In (2), the state vectors , represent the controlled input and is the output measurements. and denote actuators and sensors faults, respectively. The signal includes the unknown external disturbances vector.

In the following, the parameter vector is considered bounded in the hypercube such that The LPV system (2) can be written in a polytopic form:where , and are real time matrices.

denotes the weighting function and satisfies the convex set properties: Before starting the main results of the present paper, we will make the following assumptions which are typically required in the sliding mode observer and control design for LPV systems.

*Assumption 2. *The actuator fault matrix is assumed to verify

*Assumption 3. *The invariant zeros of the system triplet , are all in the open left-hand complex plant, such that holds for all complex number where

*Assumption 4. *Faults and external disturbances are bounded by some constants such as and .

*Assumption 5. *The pair are controllable.

*Paper objective*. Given the polytopic LPV system (4) subject to actuator faults , sensor faults , and external disturbances .

This paper gives a robust fault estimation based sliding mode control design, i.e, the main objective reside primarily to solve two problems as follows:(i)Estimate simultaneous actuator and sensor faults as well as system states using sliding mode observers.(ii)Conceive sliding mode controller to stabilize the closed-loop system after the occurrence of faults and disturbances.

In Section 5, simulation results are developed to show the effectiveness of fault estimation based sliding fault tolerant control design for the model of VTOL aircraft system.

#### 3. Simultaneous Actuator and Sensor Faults Estimation Design

Based on the polytopic LPV system representation, the objective is to design a robust simultaneous actuator and sensor faults estimation. Transformation coordinate is introduced to decouple faults, and two polytopic subsystems are constructed. The first one is subject actuator faults and the second subsystem is affected only by sensor faults.

The present subsection is given to determinate the stability of sliding mode observers based faults estimation.

##### 3.1. Transformation Coordinate System

Under Assumption 2, for the polytopic LPV system (4), there exists the following transformations : where , , , , , , and are invertible.

Based on and , system (4) is transformed into two subsystems [26]:where (9) is affected by actuator faults and (10) is subject to only sensor faults.

Define a new state and introduce a coordinate transformation , where with .

Therefore, systems (9) and (10) can be rewritten as the following form:The following subsection is dedicated to conceive two sliding mode observers in order to estimate states and simultaneous actuator and sensor faults.

##### 3.2. Sliding Mode Observers Design

For subsystems (12)-(13), we propose to conceive two polytopic sliding mode observers with the following structures: denote, respectively, the estimation of . is chosen to be a stable matrix; is the SMO observers gains.

The discontinuous output error injection terms and are defined byTo know the bound of the faults and , we use an adaptive algorithm described aswhere , are a symmetric definite Lyapunov matrix, and . With are the state estimation errors.

##### 3.3. Convergence Analysis

Consider Lyapunov candidate function:where , , and . We define and . The time derivative of can be obtained as From the fact that is bounded, it follows from (16) and (18) that It turns out thatLikewise, the derivatives and with respect to time can be derived asandSubstituting (23), (24), and (25) into (20), it results that To minimize the effect of disturbances, we use the performances. The performance of can be defined as where is a small positive constant; suppose that is the controlled output for the error system. is assumed to have full rank matrix, having the following structure: where .

In that order, small gain of the transfer function from the external disturbances to the estimation errors combination means a small influence of on .

To attain robustness to the disturbances in the sense, we impose the following constraint on our stability criteria:We conclude thatwith In this way, if , for , then along the system trajectories with attenuation level .

Integrating the inequality in (30) from 0 to, we have With zero initial condition, we have Thus, we obtain

So, if there exists the matrices , , , and the observer estimation errors are stable with attenuation level subject to .

The estimate actuator and sensor faults and will be obtained by scaling the so-called equivalent injection signals. Specifically where and are two small positive scalars.

Based on Assumptions 2–5, the challenge of this paper is to construct a robust controller. Hence, the closed-loop system response converges to zero even in the case that actuator fault, sensor fault, and disturbance simultaneously have effects on the system dynamics.

#### 4. Fault Tolerant Control Design

This section explores a fault tolerant control design for the polytopic LPV system (4) based on the provided information about simultaneous actuator and sensor faults estimation. In this way, we propose to conceive two FTC strategies, subject to provide a corrective action after the occurrence of faults and disturbances, as well as stabilize the LPV system, such that we have the following:(i)State feedback fault tolerant control.(ii)Sliding mode fault tolerant control.

##### 4.1. State Feedback Fault Tolerant Control

Based on both actuator fault and system states estimation, we propose to conceive a robust controller aswhere and represent, respectively, the state-feedback control and fault compensation gains. We assume that .

Substituting (38) into (4), it remains to develop the following model aswhere is the compensated output vector, represent the measurement output, , and .

The active fault tolerant design strategy based on state feedback control is shown in Figure 1 for the polytopic LPV system.

Hence, the stability of the closed-loop systems (39) is given in the next theorem.

Theorem 6. *The closed-loop LPV system (39) is stable and the performance is guaranteed with an attenuation level , if there exist symmetric positive definite matrices and matrices and scalar satisfying the following LMI optimization problem, ,**minimizes subject towhere*

*Proof. *In order to prove the stability of the closed-loop system, we consider the following Lyapunov function: The time derivative of can be obtained asThe objective is to guarantee the stability of the closed-loop system such thatTo attain the required performance (44), the following relation should hold:Inequality (45) implies thatwhere .

Before and after multiplying (46) by and its transpose, respectively, and using the Schur complement, inequality (46) is rewritten asUsing Lemma 1, we can express the following relation: or equivalently From this, inequality (47) directly leads to (40).

This completes the proof.

Sliding mode control has been widely studied in the literature based on its computational simplicity and in particular strong robustness against uncertainties, disturbances, and measurement noise. The proposed sliding mode controller with adaptive law is assigned to provide a corrective action in order to compensate simultaneous actuator and sensor faults effects.

##### 4.2. Sliding Mode Fault Tolerant Control Design

The objective is to force the controlled output for the closed-loop system to zero infinite time and induce a sliding motion on the surface, as We propose to conceive a sliding mode controller of the form is a linear part such that It is assumed that and .

The nonlinear component is assumed to have the form where . The sliding surface is designed as with such that and is an arbitrary matrix.

In addition, and are a small constant. is used to determine the unknown scalar that is defined by and is a positive constant.

To analyze the sliding motion corresponding to the sliding surface S, we consider the Lyapunov function aswith and .

Referring to the open-loop system (4), the time derivative of with respect to time givesIt follows from (56) and (57) thatIt results from (58) that in the subset if the parameter , such that .

In the following, we propose to analyze the system stability on the sliding mode; we give the equivalent control input asSubstituting (59) into (4), the closed-loop system state satisfies the differential equation as where , , and

The following closed-loop polytopic LPV system is rewritten as follows:The proposed design of sliding fault tolerant control based on output feedback is summarized in Figure 2.

A sufficient condition to obtain the stability of the closed-loop polytopic LPV system (61) on the sliding surface S despite the presence of uncertainties, actuator, and sensor faults is given in the next theorem.

Theorem 7. *Given positive scalar , the closed loop LPV system (61) is stable with performance, if there exist symmetric positive definite matrices and matrices satisfying the following LMI optimization problem, ,**minimize subject to where *

*Proof. *In order to ensure the closed-loop stability, we started by considering the following Lyapunov function: where . By taking into account the closed-loop system (61), is handled as The objective is to ensure the robustness against the additive term . For this, we define According to comparison principle and similarly to Theorem 6, it remains to prove thatwhere .

If relation (67) is satisfied, which implies that , it remains to conclude that the closed-loop polytopic LPV system is robustly stable with respecting the attenuation level .

#### 5. Illustrative Example

In this section, the best performances of the proposed fault tolerant control are verified by considering the following:(i)Simultaneous actuator and sensor faults estimation based state feedback controller.(ii)Simultaneous actuator and sensor faults estimation based sliding mode controller.

##### 5.1. Linear Parameter Varying Model

To demonstrate the effectiveness of the developed methods, we use a VTOL aircraft model taken from [27].

Consider the LPV system in the form of (2) and denote the state vector as where , , , and represent, horizontal velocities, vertical velocities, pitch rate, and pitch angle, respectively. where and represent, respectively, collective pitch control and longitudinal cyclic pitch control.

The polytopic LPV system matrices are given as follows:

The weighting functions are defined as where and .

The initial value of original and estimated states are set as and .

We assume also that , , ,