This action might not be possible to undo. Are you sure you want to continue?

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 48, NO. 5, MAY 2003

**Reconfigurable Fault-Tolerant Control Using GIMC Structure
**

Daniel U. Campos-Delgado and Kemin Zhou

Abstract—In this note, a fault-tolerant control strategy is proposed from a robust control perspective by applying the recently introduced generalized internal mode control architecture. This fault-tolerant control design consists of two parts: a nominal performance controller and a robustness controller, and works in such way that when a sensor failure is detected, the controller structure is reconfigured by adding a robustness loop to compensate the fault. This note shows how to design such controllers for a single-input two-output gyroscope system so that the performance with the nominal controller may be maintained in the case of sensor failure and/or model uncertainties. Index Terms— trol. control, fault-tolerant control, gyroscope, robust con-

Fig. 1.

Generalized internal model (GIMC) control structure.

I. INTRODUCTION Keeping good performance in the event of sensor/actuator failures is crucial in many applications, especially if costly equipments are controlled. Consequently, fault detection and fault-tolerant control have become very active research topics in recent years. The main challenge in the fault detection field is to develop detection algorithms that could distinguish between faults and disturbances into the system or model uncertainties [2], [3]. Thus, filters are designed such that the effect of faults is maximized at the outputs while the effect of disturbances is minimized. Several approaches have been suggested: robust detection and isolation based on eigenstructure assignment [14], estimation based on 2 and optimization [4], [12], detection and isolation by frequency domain optimization [7], [13], detection based on model-based probabilistic approaches [6], etc. Most of the existing research is focused on linear systems, but extensions to fault detection and isolation of nonlinear systems have also been proposed in [5] and [8]. Furthermore, the applications of fuzzy logic [1] and wavelet transforms [17] to fault detection have been recently introduced. One way of synthesizing fault-tolerant controllers is by appealing to robust design techniques [2], [9], [19]. Unfortunately, the standard -based robust control design is based on the worst case scenario which may never occur and it is not surprising to see that such a control system does not perform very well even though it is robust to model uncertainties and sensor faults. On the other hand, reconfigurable faulttolerant structures have also been studied where the model-matching strategy is used to design linear [16] and nonlinear [10] controllers. In [11], a fault-tolerant strategy was suggested based on actuator and sensor fault compensations for a winding machine. The problem of adaptive compensation for actuator failures was recently addressed in [15].

K

H

H1

Motivated from the design limitations of techniques, one of the authors has recently introduced a new controller architecture in [18], called generalized internal model control (GIMC), which can overcome the classical conflict between performance and robustness in the traditional feedback framework. This controller architecture uses the well-known Youla controller parameterization in a completely different way. First of all, a high-performance controller, say 0 , can be designed using any standard method, and then a robustification controller, say Q, can be designed to guarantee robust stability and performance using any robust control techniques. The feedback control system will be solely controlled by the high performance controller will only be 0 in the normal case and the robustification controller active when there is a sensor fault or external disturbances. In this note, we shall test this GIMC control strategy on a gyroscope experiment in our lab. The note is organized as follows. Section II describes the philosophy behind the GIMC architecture. The details about the fault compensation strategy are shown in Section III. Section IV gives a description of the setup of the gyroscope experiment. In Section V, the GIMC controller is designed and implemented for the gyroscope experiment. Both numerical simulation and experimental results are reported. Section VI gives some concluding remarks.

H1

K

Q

H1 H1

II. GIMC FOR FAULT TOLERANT CONTROL Let 0 be a nominal model of the linear plant and 0 be a linear stabilizing controller for 0 . Suppose that 0 and 0 have the following coprime factorizations 0 = ~ 01 ~ and 0 = ~ 01 ~ . Then every controller 0 that internally stabilizes 0 can be written in the ~ )01 ( ~ + ~ ) for some following form = (~ 0 2 1 ~ (1) 0 (1) ~ (1)) 6= 0, see [19]. such that det( A new implementation of the controller parameterization called GIMC is proposed in [18] as shown in Fig. 1. Note that ( ) = ~ ( ) ( ) 0 ~ ( ) ( ). Hence, is the filtered error between the estimated output and the true output of the system (residual signal) [2]. Thus, = 0 if there is no model uncertainties, external disturbances or faults and then the control system will be solely controlled by the high performance controller 0 = ~ 01 ~ . On the other hand, the robustification controller will only be active when 6= 0, i.e., there are either model uncertainties or external disturbances or sensor/actuator faults. Since the signal represents the estimated output error (residual), this signal contains valuable information in case of a sensor failure. Consequently, in order to have the exact nominal performance in case of no-failure, could be monitored to detect a sensor or actuator failure and then activate the robustness loop, i.e., switch on the signal . Thus, the GIMC structure can also be implemented as shown in Fig. 2. This will improve the performance of the overall GIMC structure, since

P

V

K K

K V QN Q N

P

P K K P V U P M N P U QM Q H r

rs

N sus M sys r

Manuscript received November 13, 2001; revised June 14, 2002 and December 18, 2002. Recommended by Associate Editor L. Pandolfi. This work was supported in part by the Air Force Office of Scientific Research under Grant F49620-99-1-0179, by the National Aeronautics and Space Administration under Grant NCC5-573, by LEQSF under Grant NASA/LEQSF(2001-04)-01, by Consejo Nacional de Ciencia y Tecnologia (CONACYT) under Grant C01-FRC-12.22, and by Fondo de Apoyo a la Investigación (FAI) under Grant C01-FAI-11-8.91. D. U. Campos-Delgado is with the Facultad de Ciencias (UASLP), Zona Universitaria, C.P. 78290, S.L.P., Mexico (ducd@galia.fc.uaslp.mx). K. Zhou is with the Department of Electrical and Computer Engineering, Louisiana State University, Baton Rouge, LA 70803 USA (kemin@ece.lsu.edu). Digital Object Identifier 10.1109/TAC.2003.811263

r

Q

K

V U

r

r

q

0018-9286/03$17.00 © 2003 IEEE

VOL. MAY 2003 833 Fig. u s where s . the sensor or actuator fault can be modeled as model uncertainties. 2. 1 and 2 are uncertainty weights such that and 0 have the same number of unstable poles for all allowable 1. Fortunately. For example. Theorem 1: Suppose W1 . P0 2 H . [13]. the closed-loop stability. and K0 01 U is a stabilizing controller of P0 that satisfies the closed-loop V performance requirements. Note that with this formulation. Consequently. 48. However. Gyroscope system. 3. this is feasible if some knowledge about the frequency information of the faults and external disturbances is previously known [7]. GIMC structure with failure detector.e. the closed-loop transfer function from ref (reference) to u (control signal) is given by = 0 ~ ~ 01 UM =0 min T uref = (I + K0 P0 )01 K0 : (5) Proof: From Fig. NO. then the faulted input–output mapping P s can be represented as 1 1 1 ^( ) =[ + 1s ] ( ) ^( ) =[ + 1a ] ( ) I y s I u s () and actual faults. then ~ ~ 1 1 1 = (4) P may represent actuator failures.e. it can be seen that (with u s ( ) = ~ 01 0 ~ f V U W1 w s ( ) + 0 ( )g + ~ ( )0 ~ f P u s Q Nu s M ref = 0) ( )+ P0 u s W1 w s ( )g .. 5. 3.. and = 0 ( + 1) with 1 = 0 and 2 = P P W W P H W W P P I P W I W P I P P I W P W I Q is the optimal solution to the optimization problem Q kTzw k1 . this problem has an obvious answer if P0 2 H . disturbances could also be considered as faults. We shall consider a basic robustness requirement in this note. III. 3. Fig. if Q is chosen in this way. 4. In fact. see Fig. Fig. a 2 H represent the sensor and actuator perturbations due to the faults. we consider the following general class of uncertain models where Tzw is the closed-loop transfer function from signals w to z . if these terms are appended to the nominal plant P0 . Moreover. the sensor and actuator faults can be modeled in a multiplicative form y s Fig. As a result. there is no degradation of nominal performance in order to improve robustness. 5. FAULT COMPENSATION In general. 1 2 1 is an uncertainty block such that k1k1 1. GIMC architecture with uncertain plant. it is assumed that the fault detection scheme can distinguish between disturbances = 0 + 11 2 (2) where 0 is the nominal plant. kTzw k1 . W2 2 H . i. Hence. Generalized linear fractional transformation. i. our objective is to design Q to maximize the failure tolerance in the closed-loop system.IEEE TRANSACTIONS ON AUTOMATIC CONTROL. = ( + 1) 0 with 1 = and 2 = 0 may represent the sensor failures with 1 = 0 characterizing the lack of all sensor measurements. P s P s ( ) = 0 [ + 1a ] (actuator fault) ( ) =[ + 1s ] 0 (sensor fault) P I I P : min k zw k1 Q T (3) (1) Thus.

the switching is performed only once during the system operation.834 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. then a weighted H1 approximation has to be solved. if there exists a 2 H1 such that Tyr 2 H1 . note that once a fault is detected. 6. Q is chosen according to where S = (I + K0 P0 ) = min kFl (G. since M 01 we have ~ ~ ~ ~~ ~ ( ) = V 01U ref(s) 0 V 01U M 01 N u(s) ~ ~ Therefore. the closed-loop system is not linear time-invariant anymore.. Moreover. it is natural to ask about the stability of the overall closed-loop with this scheme. We want to point out that it may not always be desirable to have the compensation signal q active all the time if P0 is stable and the controller Q in Theorem 1 is used. 4. 5. The generalized plant G will now be given by 1 1 = ( + 1) G Consequently. Simulation with sensor fault at 8 s. as it is seen from (6). Therefore. 48. NO. For P0 2 H1 . as seen in Fig. in this case. the signal q is not continuously switched on and off. Q)k1 Q G (7) (8) outage) and the closed-loop will become unstable. this problem can be put in an LFT framework. This is because. if an output uncertainty (sensor fault) is considered for the unstable plant P0 . after the fault is detected and q is activated. and z s W2 u s . the closed-loop = will be stable again if P0 2 H1 and Q is selected according to (4). Hence. On the other hand. Turef = [I + K0 P0 ]01 K0 since K0 = V 01 U . 2. Consequently. the control system will operate in an open-loop fashion with an equivalent open-loop controller I K0 P0 01 K0 and the system performance has no tolerance to disturbances and model mismatches. Thus. ~ ~ 01 ~0 = 0SPMK0 SP00V (9) 0~ ~ where S = (I + P0 K0 )01 . Consequently.e. this will not be the case if additional performance criteria are included in the H1 design. since P0 simplification ()= () ~ ~ = M 01 N and after some (6) Tzw ~ ~ ~ = 0W2 [I + K0 P0 ]01 V 01 [U + QM ]W1 : min kTzw k1 = 0 Q Q and internal stability is guaranteed if k k1 < = . the switch is activated and the compensation signal q enters into the system and remains active until the proper sensor replacement is done. i. the optimal solution to the robust stability problem is ~~ = 0U M 01 ~ 2 H1 . the control system is equivalent to running in an open loop with a reference and an open-loop controller K I K0 P0 01 K0 . 2 where the compensation signal q is introduced only when a sensor failure is detected. if the plant P0 is not stable. So. the uncertainty could take the value 1 = 0I (sensors [ + ] 1 1 1 . Next. However. the closed-loop stability is again recovered if k k1 < = = and Q satisfies (8). Of course. VOL. MAY 2003 Fig. us Thus. Thus. It is important to mention that with the configuration of Fig. it is recommended to adopt the controller structure in Fig. Therefore. with the generalized plant G given by =[ + ] ref ~ 01 SK = 0W2M W01W1 W2 S0V ~ 0 01 . P I P0 . note that ~ ~ ~ ~ u(s) = V 01 U (ref(s) 0 P u(s))+ Q N u(s) 0 M P u(s) ~~ with Q = 0U M 01 and after some calculations and simplifications. Otherwise. Alternatively. Note that in this case 1 since this will represent that the maximum tolerable uncertainty is always k1k1 < 1. it is assumed that the fault is not intermittent. if the q signal is introduced into the closed-loop.

Hence. q4 and !4 . q2 . consists of a high inertia brass rotor suspended in an assembly with four angular degrees of freedom (q1 . For comparison. 037). In this way. Jn . the system is put in the standard LFT framework by using (9). we shall consider a special case where the gimbal axis 3 is locked so that bodies A and B become one rigid body. the robustness controller Q is designed to tolerate sensor faults as presented in (1). first a nominal controller K0 is obtained using (model) as a plant model. In this note. The resulting plant is useful for demonstrations of gyroscopic torque action where the position and rate. 2. However. and D) are the scalar moments of inertia about the ith (i = 1. q4 ). Expiremental responses without fault. respectively. u is the control input that represents the torque applied by motor #2 (T2 in Fig. 48. q2 . EXPERIMENT DESCRIPTION The fault-tolerant structure will now be tested in a multiple-input–multiple-output (MIMO) gyroscope experiment. in bodies A. Consequently. As a result. a nonlinear model can be obtained. the following optimization problem is formulated: 0 G= 0 P0 : I P0 . The rotor spin torque (T1 ) is provided by a dc motor (motor #1). Next. through motor #2 while the rotor is spinning. Q)k1 . This uncertainty representation can also be viewed as output multiplicative uncertainty. and D. sensor faults are modeled by uncertainty in the output signal. Kn (n = A.B. The first transverse gimbal assembly (body C) is driven by another dc motor (motor #2) that applies a torque to axis 2 (T2 ). 5. and In . Both motors have restrictions on rotation rates to commands change. the nominal plant P0 is given by P0 10:33s = s(s2 +1389:2) 038:69 (11) _ x= 0 0 0 q4 I q y= 2 = 1 0 +K +K +K + 0 0 I +K0J K +K 0J 0 0 I J 1 0 +I x+ x I 1 +I 0 0 u 0 (10) where is the spin rate of the rotor disk (body D). 030) and for the observer (035. P0 2 H1 . VOL. = A high-gain observer-based controller is designed such that the closed-loop poles are located at: for the state feedback (010 6 40j . the optimization scheme gives a controller Q with = 1:0. a small penalty " is imposed on the control signal. Taking the coordinate frame definition from Fig. IV. Note that this model is valid only for small displacements around the operating point. P0 has all its poles on the imaginary axis.IEEE TRANSACTIONS ON AUTOMATIC CONTROL. In the next step of the GIMC controller design. 5. this value is expected since this represents that the complete sensors outage cannot be tolerated. 5. NO. MAY 2003 835 Fig. 036.B. Q is chosen according with the following optimization criterion (robust-stabilization) = minQ kFl (G. CONTROLLER DESIGN Following the procedure outlined previously.729 j . Consequently. The plant. an H1 controller is designed based also on measurement uncertainty. which are represented by saturations of 610 V in their input voltages. in order to make the H1 formulation well posed. This controller results in a fairly fast response with a small overshoot while keeping the controlling voltage between the limits of the saturation. may be controlled by rotating angle 2. 7. Thus. 3) direction.C. A linearized version around the origin can be derived as follows: V.C. 5). Hence. where the open-loop poles are located at 0 and 619. q3 . shown in Fig.

the system with the GIMC controller keeps the performance of the nominal controller in the case of no fault. producing a useless controller. but in the case of fault. it is seen that the system with the nominal controller becomes unstable in the event of sensor fault. Thus. VOL. the control signal applies an input voltage to motor #2 in order to rotate this gimbal. Now. the simulation of the response to a square wave reference signal is computed with sensor fault in Axis 2 (q2 ) at t0 = 8 seconds as shown in Fig. 8. which produces a corresponding angular velocity in the gimbal #4 (q4 ) direction. specially if external disturbances are considered. Finally. However. The failure condition is then detected by monitoring the signal r (residual). Other approaches like monitoring the statistics of r could lead to a more efficient detection. On the other hand. 7. Now. After the sensor fault. the experimental response with fault in Axis 2 (q2 ) sensor is tested. the experimental response of the nominal controller K0 is not much different from the simulation due to the high gain of this controller. q2 (t) the true reading and t0 ^ the failure time. 1 J = kr(t)k2 ^ t . The following simple approach is chosen since external disturbances are not considered. As expected. the real advantage of the GIMC architecture is obvious: even in the case of sensor fault (outage) the tracking capabilities are not significantly affected. instead it is related to interactions among gyroscope components.. the system with this controller still maintains good tracking. Finally. it is needed to choose a strategy to detect a sensor failure from the residual r . i. the H controller response is drastically affected. there are two sensor measurements q2 and q4 (Axes 2 and 4). Because. q tries to compensate the nominal control signal u to preserve stability and to ^ keep the tracking response. 48. the resulting H controller K (s) gives again = 1:0. The approach taken is by no means the most efficient way to detect the sensor failure. if external disturbances are considered to influence the system. Like in the GIMC controller. the response with the GIMC controller is affected since there is large error between the model and true system. However. i. the reading from that sensor is switched off after 2. these controllers are also tested experimentally. the robustness loop is switched on only in the case of a sensor failure. The responses to a square wave command signal (without fault) are shown in Fig. check the standard behavior of this signal for no-failure case and set a threshold value such that if r goes above it the signal q is activated. 5.e. Nevertheless.2 s. MAY 2003 Fig.K and GIMC are investigated. The plot shows how the nominal controller K0 immediately produces an unstable response.. only the overall application is intended to be illustrated here. In this feedback configuration. the systems with H and GIMC controllers remain stable. since almost no movement in the Axis 4 direction is detected. Nevertheless. it is not coming from the dc-motor characteristics. the idea proposed in Fig. Consequently. 1 1 Meanwhile. No external disturbances are considered during testing. it still maintains stability with reasonable tracking. On the other hand. but the drawback is its slow response.836 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. the performance of the three controllers K0 .2 s. small displacements of gimbal #2 (q2 ) do not produce any motion in Axis 4.. 8. The fault is programmed at 2. 1 1 1 q2 (t) = ^ q2 (t) t < t0 0 t t0 where q2 (t) represents the measurement. From these plots.e. The results present a peculiar behavior of the experimental configuration. as predicted by simulation. Thus. Experimental responses with sensor fault at 2.T = t 0T rT ( )^( )d > Jth ^ r (12) . But due to friction mainly. The results are shown in Fig. 2 is followed. It can also be seen that the system response with the H controller is not clearly affected by the sensor fault.e.t. We consider the case when there is a sensor outage of Axis 2. 6. This behavior could be characterized by a dead-zone nonlinearity. but it is more complicated than that. NO.2 s. While the rotor is spinning. a detection strategy based on the norm of a weighted residual r over a finite ^ period of time T [17] can be adopted. i. In addition.

Fig. Experimental response: switching on the robustness signal after detecting sensor failure (6 s). 48. 5. 10.IEEE TRANSACTIONS ON AUTOMATIC CONTROL.35 s in detection. Simulation with sensor fault at 8 s and delay of 0. . 9. MAY 2003 837 Fig. NO. VOL.

Thus. Hamelin.838 IEEE TRANSACTIONS ON AUTOMATIC CONTROL. Feb. 44. pp. Ind. 1710–1715. 1999. Dinca.” IEEE Trans. M. vol. Huang. and R. vol. 2002.” IEEE Trans. Russell. Fault Detection and Diagnosis in Industrial Systems. Mag.. 46. 23. Dyna. TX. Contr. robust and fault-tolerant control. “A geometric approach to nonlinear fault detection and isolation.” J. M. Guid. June 2001. Control.. Automat. S. VOL. 1. Moreover. 384–390. The response is presented in Fig. Chiang. pp. C. L. J. [17] H.. vol. Doyle. Decision Control. Yang. MA: Kluwer. F. L. Kim and Y. T. Contr. pp. Y. 1 1 REFERENCES [1] S. Chang. pp. Shen. Robust and Optimal Control. Ren. 2108–2114. vol. Jth is a threshold value ^ to prevent false triggering. Joshi.” J. VI. An important issue concerning the GIMC fault-tolerant structure of Fig. Contr.” in Proc. 1879–1884. [5] C. 1999. De Persis and A. Kinnaert. “Robust tion of uncertain dynamic systems. 1063–1079. S.. 2001. “Robust fault detection and isolation with unstructured uncertainty using eigenstructure assignment. [2] J. Oct. Theilliol. 10. Control. vol. Collins Jr. the performance of the GIMC architecture is drastically improved since there is no sacrifice of performance for robustness before the fault. “Fuzzy inference systems implemented on neural architectures for motor fault detection and diagnosis. vol. is the possible delay in the failure detection. “Integrated design for fault detection systems in time-frequency domain. 47. The fault-tolerant approach is tested in the experimental setup with failure of Axis 2 (q2 ). Sauter and F. TX. Trussell. 1999. [19] K. Control. MAY 2003 ~ ~ where r(s) = R(s)[N (s)u(s) 0 M (s)y (s)]. optimization.. Contr.” IEEE Trans. no. 5. 2000. Wang. and the speed of response of the nominal controller. U. sensor outage. 44. J.. 2000. 878–882. [14] L. and S. 23. “A frequency domain approach to fault detection of uncertain dynamic systems. Song. Tao. Ding. 3. Automat. . 1999. Auized tomat. E. “Robust variable structure controller design for fault-tolerant flight control. [7] X. Hammouri. no. there is a strong relation among the size of the maximum delay allowable. H. 50–57. and P. and H.” J.. Australia. the GIMC controller is presented as an extension of the H technique that overcome the conservativeness of the latter one. IEEE Conf. San Antonio. pp. pp. 1998. Guo. 21. 1993. Aldemir. 853–865. Nov. Yang and J. 506–511. further analysis must be performed to obtain a measure of this maximum delay. Apr.5 s. D.. G. Zhang.” IEEE Control Syst. Sydney. IEEE Conf.” IEEE Trans. 47. [13] D. pp.. Guid. J. Since the compensation signal q is added after the failure is detected. Dyna. and G. London.” IEEE Trans. G. In order to analyze this behavior. [11] H. no.35 seconds. pp. pp. Dec. Chen and R. Sauter. pp.: Springer-Verlag. Robust Model-Based Fault Diagnosis for Dynamic Systems. Rizzoni. 2001. vol. NO. simulation was carried out assuming a failure in Axis 2 (q2 ) sensor at 8 s and a detection delay of 0. vol. “A model-based probabilistic approach for fault detection and identification with application to the diagnosis of automotive engines. and T. M. the performance will be dramatically affected and even the system can become unstable. Contr.. pp. CONCLUSION The motivation of this work has been to present how the GIMC architecture can be adopted to the fault-tolerant control problem in MIMO. Mar. “A new controller architecture for high performance. the detection delay will have a big impact in the overall performance and stability. NJ: Prentice-Hall. The classical H -based controller has shown to be useless in our experiment due to unmodeled nonlinearities in the system. 46. 2000. “Fault-tolerant control in dynamic systems: application to a winding machine. Norwell. 5. El Yaagoubi. pp. Zhou and Z. Stoustrup. Ding. Altug.. Decision Control. the system cannot be stabilized anymore. Upper Saddle River. 46. and E. Automat. 2200–2205. [18] K. 44.” presented at the IEEE Conf. and P. Once the failure is detected the control signal is compensated but still some oscillation is observed. San Antonio. Nov. Chow. H. no. the size of the saturation in the control signal. 44. Automat. “Observer-based approach to fault detection and isolation for nonlinear systems. 1999. 9 for the no-delay case. Therefore. vol. From this plot. after 6 as shown in Fig. pp. 613–1618. vol. Contr.K. 1722–1727. 20. Electron. Zhou. Isidori. Obviously.” IEEE Trans. and K. 430–437. Dyna. X. from the analysis of the properties of the signals in the GIMC architecture. “An adaptive actuator failure compensation controller using output feedback. Gertler. 1996. J. C. [8] H. Hamelin. 6. [6] L. For a delay >0. 33–49. 1999. The GIMC architecture is motivated from H design and this particular configuration has appealing characteristics for fault-tolerant control. “Frequency-domain optimization for robust fault detection and isolation in dynamic systems. Hsu. [15] G. [3] L. Automat. H. vol. Thus. Contr. Oct. The advantage of this new architecture over standard control is successfully demonstrated on a single-input–two-output configuration of a gyroscope system. Frank. 1.” IEEE Trans. and G. it is possible to derive a reconfigurable control strategy that keeps the nominal performance for no-sensor failure and maintains stability and good tracking in case of a sensor fault in our system. “Fault-tolerant decentralcontrol for symmetric composite systems. If this delay is too large... Qiu and J.” IEEE Trans. Ye. 2. Glover. Automat. “Robust FDI systems and in Proc. and R(s) is a filter chosen to maximize the sensitivity to faults and minimize the effect of disturbances [7]. Hence. Guid. and D. Patton. D. Braatz. “Robust reconfigurable control for parametric and additive faults with FDI uncertainties. 1993. Automat.. it can be seen that the delay affects the output performance since the output became oscillatory after the fault is presented.” IEEE Trans. K. 2000. Contr. S. it is clear that the controller is able to stabilize the system and achieve good performance even after the failure. vol.. Lam. 48. [10] D. pp. M. Decision Control. and it can be compared with Fig. vol. [16] Z. and S. 857–864. L. [9] S. pp. Kim.. 1 estimation and fault detec[4] E. Noura. Chen. 2002.” [12] Z.

- IJCAS_v6_n3_pp.339-350
- A Review of Methods for Input-output Selection
- gas turbine Opportunities_and_challenges
- p805
- Troubleshooting Bizhub 421_501
- What Makes Some Control Problems Hard
- A Control Engineer's Guide to Sliding Mode Control
- Checker Product Guide
- State Space Approch LQG
- lqr feedforward
- Design and Development of Mission System for Micro Aerial Vehicle using AM3358ZCZ Processor
- ALCS Lect03fv
- Full Report(Tck Present
- Lecture 24
- Pendulo Invertido
- Pole Placement Design - MATLAB
- P.L.D. Peres; J.C. Geromel -- H2 Control for Discrete-time Systems Optimality and Robustness
- Sensor
- Assignment 3
- Adaptive Controller Designs for Robot Manipulator Systems Using Lyapunov Direct Method
- 2011 NOTIFIER List Price Only
- LTBT-VLD
- JFI.pdf
- Absorption Spectroscopy
- ALS Series Product Information Gv
- Dynamic Inversion With Zero-dynamics Stablization for Quadrotor Control
- Tank Level Control
- Regelungstechnik
- mechanical engineering syllabus kurukshetra university
- 1-s2.0-S0925400501008565-main.pdf

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd