06356720

Embed Size (px)

Citation preview

  • 7/28/2019 06356720

    1/6

    A Robust Approach to Fault Tolerant Controller Design Based on

    GIMC structure for Non-Minimum Phase systems

    Abstract-This paper represents a robust scheme for

    designing fault tolerant controller (FTC) based on

    Generalized Internal Model Control structure, and extends

    it to non-minimum phase LTI systems. In this scheme, first

    the nominal controller is designed to have a closed-loop

    stability and acceptable performance, and then by

    correcting the nominal controller based on the idea of

    GIMC, a robustification signal is imposed to compensate the

    actuator faults. The proposed scheme, deals with non-

    minimum phase systems by utilizing the idea of

    transmission zero-assignment. It will be shown that the

    problem of zero-assignment problem for state-accessible

    systems is equivalent to a pole-placement problem with state

    feedback on a reduced-order subspace, which always has asolution. The advantages of the proposed control scheme

    have been verified by the simulation on the quadruple-tank

    process subjected to the actuator fault.

    Keywords: Fault Tolerant Controller (FTC), Non-

    Minimum Phase System, Transmission Zero-Assignment,

    Generalized Internal Model Control (GIMC).

    1. Introduction

    In the early stages of control applications, closed-loopperformance was the main objective for the control

    engineer. To achieve this goal, the implementations of

    these feedback configurations involved sensors, actuators,electronic instrumentation, and signal processors.

    However, during a normal operation, these parts could

    fail in some degree, and the resulting performance of the

    closed-loop system will be largely deteriorated, or even

    instability can be observed. So, it is necessary to design

    controllers which are capable of not only diagnosing

    faults in closed-loop systems, but also maintaining

    closed-loop stability and performance in the presence offaults. A closed-loop control system which can tolerate

    component malfunctions, while maintaining desirable

    performance and stability properties is said to be a Fault-

    Tolerant Control system (FTCs) (see, [1]-[2]).

    FTC can be approached from two perspectives:passive and active [3]. Passive approaches make use ofrobust control techniques to ensure that a closed-loop

    system remains insensitive to certain faults using constant

    controller parameters and without use of on-line fault

    information [4]-[5]. Active approaches react to the

    system component faults actively by reconfiguring

    control actions so that the stability and acceptableperformance of the entire system can be maintained. In

    such control systems, the controller compensates for the

    impacts of the faults either by selecting a pre-computed

    control law or by synthesizing a new one on-line [6]-[7].

    The survey papers [8] and [9] give the state of the art in

    the field of FTC.

    One of the approaches in FTC design is control

    systems based on Generalized Internal Model Control.

    In traditional feedback control systems, it is well

    recognized that a robust controller design is usuallyachieved at the expense of performance. This is not hard

    to understand since most robust control design techniques

    are based on the worst possible scenario which may never

    occur in a particular control system. The GIMC structure

    was proposed for this problem [10]. This structure is IMC

    generalized by introducing outer feedback controller. The

    distinguished feature of GIMC is that, this controller

    architecture has the potential to overcome the conflict

    between performance and robustness in the traditional

    feedback framework by showing structurally how the

    controller design for performance and robustness may bedone separately. This GIMC structure was used as FTC in

    [11]-[12] and [13]. In [11] it is applied to gyroscope

    system so that to keep stability and performance of

    perturbed plant when one of the sensors breaks down. In

    [12], it is applied to dc-motor as speed regulation

    problem and experimentally shows good performance ofthis structure in the case of sensor failure. Moreover,

    GIMC has been applied to Magnetic Suspension System

    in [13] which is unstable system and experiment was

    carried out in the case of sensor failure. These research

    works have shown that the GIMC is very useful for

    system perturbation. However, all those previous resultsare restricted to only minimum phase systems. In other

    words, application of GIMC approaches has been limited

    to the minimum phase systems. This prohibits applying

    the powerful GIMC structure technique into the plants

    which are non-minimum phase.

    The objective of this paper is to extend the use of

    powerful GIMC structure into the case of non-minimum

    phase systems. In particular, the paper extends concepts

    of GIMC presented in [10]-[11] and [12] into the area ofnon-minimum phase systems with eliminating this

    limitation in controller design. In this paper, we applyGIMC structure to quadruple-tank process which has

    unstable zeros. Then we utilize the idea of zero-

    assignment to place the unstable zeros of the plant in the

    stable region [14]. Thus, it will be possible to implement

    the GIMC structure on this plant.

    The paper is organized as follows. Section 2 describesthe problem formulation. Section 3 presents the structure

    of the FTC scheme. First the general methodology is

    introduced and the extension to non-minimum phase

    system is explored. At last, the structured algorithm for

    designing FTC is presented. Simulation results on aquadruple-tank process subjected to the actuator faults aregiven in Section 4. Finally, Section 5 makes some

    concluding remarks.

    M. Pezeshkian*, H. Ozma**, and M. J. Khosrowjerdi***

    *M.S. Graduated from Sahand University of Technology of Tabriz,[email protected]

    **M.S. Graduated from Sahand University of Technology ofTabriz, [email protected]

    *** Assistant Professor of Sahand University of Technology of

    Tabriz, [email protected]

    2011 2nd International Conference on Control, Instrumentation and Automation (ICCIA)

    978-1-4673-1690-3/12/$31.002011 IEEE 564

  • 7/28/2019 06356720

    2/6

    2. Problem Formulation

    The objective of this paper is to design FTC to

    compensate the faults occurred in the actuators of non-

    minimum phase LTI systems. The faults addressed in this

    paper are modeled as additive faults. But, in problem

    formulation we consider the general system with possible

    faults in both actuators and sensors. Consider a system

    ( )P s affected by possible faults lf R described by

    (1)1

    2

    ( )x Ax Bu F f

    P sy Cx Du F f

    -

    where nx R represents the vector of states, mu R the

    vector of inputs, andp

    y R the vector of outputs. Thus

    matrix 1n p

    F Ru

    stands for the distribution matrix of the

    actuator or system faults, and 2p l

    F Ru

    for sensor faults.

    The nominal system ( , , , )A B C D is consideredcontrollable and observable. On the other hand, the

    system response y can be analyzed in a transfer matrix

    form (frequency domain) as follows:

    (2))()()()()( sfsPsusPsy fu

    where

    (3)

    1

    1 2

    ( )

    ( )

    u

    f

    P C sI A B D

    P C sI A F F

    A left coprime factorization for each transfer matrix can

    be derived as follow [15]

    (4)

    1

    1

    u

    f f

    P M N

    P M N

    where , , fM N N RHf and can be obtained as follow

    fM N N

    (5)1 2

    2

    .A LC L B LD F LF

    C I D F

    where matrix n pL Ru

    is such that A LC be Hurwitz.Now, it is assumed that a nominal controller K stabilizes

    the nominal plant uP , and it provides a desired closed-

    loop performance. The controller K is considered

    observable, and consequently, it can also be expressed bya left coprime factorization as

    (6)1K V U

    where ,U V RH f and

    K K

    K K

    A BK

    C D

    (7)> @K K K K K K K

    K K

    A L C B L D LU V

    C D I

    The nominal controller can be synthesized following

    classical techniques or optimal control: PI , PID ,

    2/LQG H , Hf loop shaping design, and so on. The

    proposed scheme for designing fault tolerant controller is

    depicted in Figure 1. In this structure, the feedback

    system is controlled only with nominal controller

    1K V U while there are no faults in the systems. Butwhen fault occur in any component of the system, the

    compensator signal q is activated to attenuate or

    compensate the effect of the fault in the closed-loop

    system. In next section, the way to how to produce the

    compensator signal q based on GIMC structure is

    introduced.

    Figure 1: The proposed FTC scheme

    3. FTC Structure

    In previous section, it has been seen that the proposed

    FTC structure is based on the production of the

    compensator signal q , see Figure 1. In this section, we

    introduce the way to construct the compensator signal q

    based on GIMC structure.

    3.1 Overall FTC structure

    Consider the nominal plant uP and let K be a linear

    stabilizing controller for uP . Suppose that K and uP have

    the coprime factorizations as Equations (4) and (6). Then

    every controller K that internally stabilizes uP can be

    written in the following form:

    1( ) ( )K V QN U QM

    where Q RHf and det( ( ) ( ) ( )) 0V Q Vf f f z [15]. A

    new implementation of the controller parameterization

    for producing the compensator signal q is shown in

    Figure 2.

    Figure 2: FTC structure for producing signal q

    565

  • 7/28/2019 06356720

    3/6

    As can be seen from Figure 2

    (8)( ) ( ) ( ) ( ) ( )s N s u s M s y sX

    In fault detections, this signal is called residual signal,

    the difference between the estimated output and the true

    output of the system [1]. If system is working in fault free

    condition, then 0X and the control system will be

    solely controlled by the nominal controller 1K V U . In

    the other hand, when fault occur in the system, then

    0X z and the compensator signal q will be active to

    compensate or attenuate the effect of the occurred fault.

    Since the signal X represents the estimated output error

    (residual), this signal contains valuable information in

    case of a component failure.

    Define the following nominal closed-loop transfermatrices:

    i. input sensitivity: 1i uS I KP { ii. output sensitivity: 1o uS I P K {

    iii. complementary output sensitivity: 1o u uT I P K P K

    {

    Lemma 1. In the FTC configuration of Figure 2

    considering additive faults, the resulting closed-loop

    characteristics for the control signal u and output y are

    given by

    1 1( )( )i i fu S Kr S V UM Q N f

    (9)1 1

    ( )( )o o fy T r S M I NV Q N f

    and the closed-loop system is stable, provided that

    Q RHf .

    Proof. From the Figure 1, for signal u we have:

    > @ > @1 1( ) ( )u V q U r y V Q U r yX

    > @1 ( ) ( )V Q Nu My U r y

    by substitution of Equation (4) in above equality:

    1f u fu V QN f Ur UP u UP f

    by some arithmetic operation we can deduce following

    equation

    1 1i i fu S Kr S V UM Q N f

    by substitution ofu from above equation in Equation (2),

    signaly can be deduced.

    By considering Equation (9), variety of scenarios for

    designing the compensator signal q can be introduced.

    For instance, for minimizing the fault effect at the sensorsignal, the problem 1 can be expressed:

    Problem 1: Consider Figure 2, given 0J ! find the

    compensator Qsuch that fyT Jf .

    For solving Problem 1, according to Equation (9),

    following optimization strategy is suggested:

    1 1min ( )( )o f

    Q RH

    S M I NV Q N f

    f f

    (10)min ( , )L Q

    Q RH

    F G Qf

    f

    where QG represents the generalized plant given by (see

    Figure 3)

    (11)

    1

    0

    o f o u

    Q

    f

    S P S P V G

    N

    In Equation (10), LF represents lower Linear Fractional

    Transformation (lower LFT). Thus, the problem ofdesigning fault tolerant controller leads to the problem of

    solving a robust control problem that can be handled

    easily using standard softwares such as Matlab [17].

    Figure 3: LFT framework for compensatorQ

    In a special case, if the nominal plant uP has a stable

    inverse, i.e. be a minimum phase system, a complete

    output decoupling for faults can be achieved. Thefollowing lemma suggests the best optimal solution to

    problem 1, when the nominal plant has a stable inverse.

    Lemma 2. If the nominal plant satisfies uP,1

    uP RH

    f ,

    then1

    N RH

    f , and by considering Equation (9) and

    taking1

    Q VN RH

    f , the resulting output signal is

    completely decoupled from the faults, that is

    1i fu S Kr N N f

    (12)oy T r

    Proof. By considering1

    Q VN

    and substituting it in

    Equation (9), the Equation (12) can be easily deduced.

    Note that the compensation proposed in Lemma 2 is

    particularly useful for an actuator or system faults, since

    the output is perfectly decoupled from faults. In addition,

    the above Lemma is useful when the nominal plant uP be

    the minimum-phase systems.

    In the following sub-section, we introduce a method

    to overcome to this weakness and expand the above

    lemma to non-minimum phase systems.

    3.2 Expansion of FTC to non-minimum phasesystems

    566

  • 7/28/2019 06356720

    4/6

    In this sub-section, our goal is to introduce the

    method to assign the unstable zeros of the non-minimumphase system to the stable region by using the idea of

    zero-assignment to overcome the weakness of the Lemma

    1 [14]. In this method, we will show that the problem of

    zero-assignment problem for state-accessible systems is

    equivalent to a pole-placement problem with statefeedback on a reduced-order subspace, which always hasa solution.

    Consider the nominal system described by Equation

    (1). In this paper, we deal with the actuator or system

    faults, so 2 0F . And, without loss of generality, assume

    also that 0D and pair ( , )A B is controllable. Assume

    further that matrix CB is non-singular, i.e. ( )rank CB m .

    This system is written here for simplicity

    (13)1x Ax Bu F f

    y Cx

    It can be shown that there exist an orthogonal matrixn n

    T Ru such that

    2[0 ]T

    TB B

    where 2m m

    B Ru and is non-singular. The matrix T can

    be computed via QR decomposition on the matrix B .

    By using the coordinate transformation x Txl , the

    distribution matrix of the actuator faults 1F from

    Equation (13) has the following structure

    (14)11

    112

    c

    FF TF

    F

    In addition, the nominal system described by Equation

    (13), when 0f , has the following representation

    (15)11 12 11

    21 22 2 22

    0

    ccBA

    A A xxu

    A A x Bx

    11 22

    xy C C

    x

    (16)

    where 1n m

    Rx , 2m

    Rx and other matrices have

    appropriate dimension. In addition, the matrix 2m m

    RC u

    is non-singular. Transmission zeros of this system is

    given by [14]

    111 12 2 1 0n mI A A C CO

    .

    Define new output for system (15) and (16) as

    (17)1y y Mx

    where( )n n m

    M Ru is a design matrix. Substituting for

    1x and y from Equations (15) and (16) in Equation (17)

    yields:

    11 22

    S

    xy S

    xS

    (18)

    where

    (19)

    1 1 11

    2 2 12

    S C MA

    S C MA

    Now, transmission zeros of the transformed system (15)

    with a new output (18) is given by:

    (20)111 12 2 1 0n mI A A S SO

    Equation (20) is equivalent to the pole-placement for

    the pair 11 12( , )A A with the state feedback matrix

    12 1zK S S . It can be easily shown that this pole-

    placement problem, and in fact, this transmission zero-

    assignment problem have always a solution. For this,

    consider the following lemma.

    Lemma 3. The matrix pair 11 12( , )A A is controllable if

    and only if the pair ( , )A B is controllable.

    Proof. Because of the special form of the plant (15) and

    using the fact that 2det( ) 0B z , it follows that

    11 12

    21 22 2

    [ ]0

    rank sI A B rank sI A A

    A sI A B

    > @11 12rank sI A A m for all s

    This implies

    11 12[ ] [ ]rank sI A B n rank sI A n m $

    and from the Popov-Belevitch-Hautus (PBH) rank test it

    follows that ( , )A B is controllable if and only if the pair

    11 12( , )A A is controllable.

    By assumption, the pair ( , )A B is controllable. Thus

    according to Lemma 3 and appropriate selection of the

    gain matrix1

    2 1zK S S , there is always solution to

    transmission zero-assignment problem. Therefore, we can

    place right half-plane zeros of the plant, if there was any,

    in the left half-plane and convert non-minimum phase

    system to minimum phase system.By substituting for 1S and 2S from Equation (19) in

    the gain matrix zK , we have

    12 12 1 11( ) ( ) zC MA C MA K

    or

    11 12 2 1( )z zM A A K C K C

    by appropriate selection of zK , the matrix 11 12( )zA A K is

    non-singular and we have

    (21)1

    2 1 11 12( )( )z zM C K C A A K

    567

  • 7/28/2019 06356720

    5/6

    By finding M from Equation (21) and substituting it in

    Equation (17), the compensated plant with y as a new

    output has transmission zeros placed in an appropriate

    place.

    Remark. The remarkable notice in this method is that it

    is clear that 1lim ( ) 0x t

    as to f

    . Thus, according toEquation (17)

    lim ( ) ( )t

    y t y t o f

    which shows that this method has no impact on closed-

    loop system tracking problem.

    FTC design algorithm:

    Data: given plant with Equation (1)

    Assumption: ( , )A B is controllable and ( , )A C is detectable

    2 0D F

    ( )rank CB m

    x Step1) For nominal plant in Equation (1), designnominal controller to achieve desired performance

    x Step2) find left coprime factorization of nominalplant and nominal controller as Equations (4) and

    (5)

    x Step3) assume the overall structure of the FTC asdepicted in Figure 2

    x Step4) solve Problem 1 and find compensator Q If the nominal plant was stable and also was minimum

    phase, then go to step 10. But if the plant was stable and

    non-minimum phase, continue

    x Step5) using coordinate transformation T ,construct cF as Equation (14). Furthermore,

    transform the nominal plant uP to new coordinate

    system as Equations (15) and (16)

    x Step6) find zK such that the poles of the matrix11 12( )zA A K is placed in appropriate place

    x Step7) find M according to Equation (21)x Step8) construct the compensated plant (15) and

    (18)

    x Step9) replace 1( , , , )A B F C in Equation (1) with( , , ),c c cA B F S and construct new uP and fP in

    Equation (3). Then go to step 2

    x Step10) Find the compensator Q as Lemma 2.The above algorithm is constructive and can be

    implemented using standard softwares such as Matlab

    [16].

    4. Numerical example: the quadruple-tank process

    To demonstrate the effectiveness of the proposed FTC

    scheme, we consider the quadruple-tank process [17]. A

    linearized dynamic model of the quadruple-tank system is

    given in the state space formulation as [17]:

    ( )

    0.50 00

    0 0.500

    0.0159 0 0.0419 0 0.035700 0.0111 0 0.0333 0 0.01050 0 0.0419 0 0 0.10770 0 0 0.0333 0.07280

    x x u f

    y x

    where 4x R is the level of water in each tank,

    1 2( )T

    u u u is the voltage applied to pumps 1 and 2,

    1 2( )T

    f f f is the actuator faults associated with pump 1

    and 2. The linearized system is stable and transmission

    zeros of the system is located at 1 0.1336z and

    2 0.2088z . So the system has one unstable zero that

    made it non-minimum phase system. According to

    method described in section 3.2, first by coordinate

    transformation x Txl , we transfer the system into new

    coordinate as Equations (14) and (15). Then, we find gain

    matrix zK so that to locate the transmission zero of the

    system in 1 0.1336z , and 2 0.2088z . For this, the

    calculated M according to Equation (20) is

    6.8943 3.21674.0387 1.2166

    M

    . Now, for the compensated plant

    with new output as in Equation (17), implement the

    algorithm. For designing fault tolerant controller, we take

    the nominal controller K as a Hf -controller, which

    calculated in [18]. In order to analyze the effectiveness of

    the proposed FTC, the actuator fault is applied as

    depicted in Figure 4, and also assumed that only the

    actuator 1 is faulty. The output of the system with

    nominal Hf

    -controller is shown in Figure 5. As shown

    in this figure, the output can perfectly track the reference

    input. The output of the system with nominal controller

    when fault occur in the actuator is shown in Figure 6. As

    can be seen, the output cannot track the reference inputproperly with the occurrence of the actuator fault. In

    Figure 7, the output y by using the proposed FTC

    structure has been shown. The simulation was carried out

    by assuming fault detections delay of about 2 second. As

    shown in this figure, although the open-loop system is

    subjected to actuator fault, one can see that the closed-loop system is internally stable and tracking performance

    is accurate. The same results can be achieved with

    constant faults (bias) and other kind of time-varyingactuator faults that show the effectiveness and good

    performance of the proposed FTC scheme.

    Figure 4: Occurred fault in actuator 1

    568

  • 7/28/2019 06356720

    6/6

    Figure 5: Outputs of the system with nominal controller without

    actuator fault

    Figure 6: Outputs of the system with nominal controller when actuator

    fault occur at t=300s

    Figure7: Output signals with FTC, when actuator fault occur in t=300s

    5. Conclusion

    In this paper a robust scheme for designing fault

    tolerant controller (FTC) based on Generalized Internal

    Model Control structure, and its extension to non-

    minimum phase LTI systems is presented. In this scheme,

    first the nominal controller is designed to achieve closed-loop stability and acceptable performance, and then by

    correcting the nominal controller based on the idea ofGIMC, a robustification signal is imposed to compensate

    the actuator faults. In particular, we extend the concepts

    of GIMC into the area of non-minimum phase systems.

    The proposed scheme, deals with non-minimum phasesystems by utilizing the idea of transmission zero-

    assignment. It has been shown that the problem of zero-

    assignment problem for state-accessible systems is

    equivalent to a pole-placement problem with state

    feedback on a reduced-order subspace, which always has

    a solution. At last, a constructive algorithm for FTC

    design has been proposed and simulation results on the

    quadruple-tank process, which has unstable zeros,

    subjected to the actuator fault have shown satisfactorylevel of control performance.

    References

    [1] J. Chen and R. J. Patton,Robust Model-Based Fault Diagnosis forDynamic Systems. Norwell, MA: Kluwer, 1999.[2] M. Blanke, M. Kinnaert, J. Lunze, and M. Staroswiecki,Diagnosis and Fault Tolerant Control, Berlin, Germany: Springer-

    Verlog, 2003.[3] R. J. Patton, "Fault Tolerant Control: The 1997 situation," In theProceeding of the 3

    rd

    IFAC symposium on fault detection, supervisionand safety for technical processes, pp. 1033-1055, 1997.[4] Sh. Suryanarayanam, M. Tomizuka, and T. Suzuki, " Design ofSimultaneously Stabilizing Controllers and Its Application to Fault-

    Tolerant Lane-Keeping Controller Design for Automated Vehicles,"IEEE Transaction on Control Systems Technology, Vol. 12, NO. 3, pp.

    329-339, 2004.

    [5] H. Niemen, and J. Stoustrup, "Passive Fault Tolerant Control ofDouble Inverted Pendulum a case study," Control Engineering

    Practice, Vol. 13, pp. 1047-1059, 2005.

    [6] B. Jiang, M. Staroswiecki, and V. Cocquempot, "FaultAccommodation for Nonlinear Dynamic Systems," IEEE Transaction

    on Automatic Control, Vol. 51, pp. 1578-1583, 2006.

    [7] D. Ye and G. H. Yang, "Adaptive Fault-Tolerant TrackingControl Against Actuator Faults with Application to Flight Control,"

    IEEE Transaction on Automatic Control, Vol. 14, pp. 1068-1096, 2006.

    [8]

    M. Blanke, M. Staroswieki, and N. E. Wu, "Concepts andMethods in Fault Tolerant Control," In the Proceeding of American

    Control Conference, Arlington, VA, pp. 2602-2620, 2001.

    [9] Y. Zhang and J. Jiang, "Bibliographical review on reconfigurablefault-tolerant control systems," Annual reviews in control, Vol 32, pp.

    229-252, 2008.

    [10] K. Zhou and Z. Ren, "A new controller architecture for highperformance, robust, and fault-tolerant control," IEEE Transaction on

    Automatic Control, Vol. 46, NO. 10, pp. 1613-1618, 2001.

    [11] D. U. Campos-Delgado and K. Zhou, "Reconfigurable faulttolerant control using GIMC structure,"IEEE Transaction on AutomaticControl, Vol. 48, NO. 5, pp. 832-838, 2003.[12] D. U. Campos-Delgado, S. M. Martinez, and K. Zhou, "IntegratedFault Tolerant Scheme with Disturbance Feedforward," Proceeding ofthe 2004 American Control Conference. Boston, Massachusetts, pp.1799-1804, 2004.[13] Y. Nakaso, and T. Namerikawa, "GIMC-based Fault Detectionand Its Application to Magnetic Suspension System,"Proceeding of the17th World Congress, the International Federation of AutomaticControl. Seoul, Korea, pp. 7363-7368, 2008.[14] A. K. Sedigh, Analysis of Multivariable Control System, K. N.Toosi University of Technology, 1996.[15] K. Zhou, and J. C. Doyle, Essential of Robust Control, UpperSaddle River, NJ: Prentic-Hall, 1998.[16] G. Balas, R. Chiang, A. Packard, and M. Safonov,Robust ControlToolbox 3: User Guides. The Math Works, Inc, 2007.[17] K. H. Johansson, The Quadruple-Tank Process: A MultivariableLaboratory Process with an Adjustable Zero, IEEE Transaction onAutomatic Control, Vol, 8, NO. 3, pp. 456-465, 2000.[18] M. Pezeshkian, and M. J. Khosrowjerdi, Active Fault TolerantController Implemented on The Quadruple-Tank Process, 17thInternational Conference on Electrical and Engineering, Tehran, Iran,May 2009.

    569