You are on page 1of 11

Measurement 143 (2019) 58–68

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

A review of Fault Tolerant Control Systems: Advancements and


applications
Arslan Ahmed Amin a,b,⇑, Khalid Mahmood Hasan a
a
Department of Electrical Engineering, University of Engineering and Technology, Lahore, G.T. Road, Lahore 54890, Punjab, Pakistan
b
Department of Electrical Engineering, FAST National University of Computer and Emerging Sciences Chiniot Faisalabad Campus, Chiniot, Punjab, Pakistan

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, a comprehensive state-of-the-art review of Fault Tolerant Control Systems (FTCS) is pre-
Received 20 January 2019 sented with the latest advances and applications. The overall aim of an FTCS is to accommodate faults
Received in revised form 1 March 2019 in the system components during operation and maintain stability with little or acceptable degradation
Accepted 28 April 2019
in the performance levels. Such systems are of great importance for safety and mission-critical applica-
Available online 8 May 2019
tions where human lives are involved and service continuity is of extreme importance. In this paper, var-
ious types of faults in the sensors and actuators have been described. Types and classifications of Fault
Keywords:
Tolerant Control (FTC) including active, passive and hybrid approaches with the latest design techniques
Active Fault Tolerant Control
FTC advanced techniques
are presented. Finally, a review of the stability and reliability analysis of FTCS is presented with research
Controller reconfiguration gaps. The contribution of the study is to benefit new and existing researchers in FTCS domain with the
Passive Fault Tolerant Control latest trends and applications for future research directions.
Lyapunov stability Ó 2019 Elsevier Ltd. All rights reserved.
FTC reliability

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
1.1. What is FTC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
1.2. Types of FTC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2. Active Fault Tolerant Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.1. Architecture of AFTCS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.2. Mathematical analysis of AFTCS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3. Latest techniques used for AFTCS design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3. Passive fault Tolerant control system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.1. Architecture of PFTCS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.2. Mathematical analysis of PFTCS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.3. Latest techniques used for PFTCS design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4. Hybrid fault Tolerant control system (HFTCS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5. Hardware redundancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6. Stability and reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
7. Latest applications of fault Tolerant control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
8. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Funding. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Declaration of Competing Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

⇑ Corresponding author at: Department of Electrical Engineering, University of


Engineering and Technology, Lahore, G.T. Road, Lahore 54890, Punjab, Pakistan.
E-mail address: arslan_engineer61@yahoo.com (A.A. Amin).

https://doi.org/10.1016/j.measurement.2019.04.083
0263-2241/Ó 2019 Elsevier Ltd. All rights reserved.
A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68 59

1. Introduction controller about the fault and AFTCS reacts by reconfiguring the con-
troller based on the information. FDI plays a vital role in active FTC to
1.1. What is FTC detect and isolate the faulty components after which controller
reconfiguration is performed for adaptability to new conditions.
Fault in a system is defined as the deviation of a parameter from Due to the immediate actions taken by the controller for reconfigu-
the acceptable value and failure is defined as the inability of a sys- ration, it is termed as ‘active’. The active FTC (AFTC) has the ability to
tem to carry out its intended operation under specified conditions. deal with many types of faults and results in optimal performance,
Clearly, failure is much more drastic condition than just a mere however, it is sensitive to the result obtained from FDI i.e., it can give
fault. Fault tolerance is defined as the ability of a system to continue a wrong decision with excessive noise. Design becomes more diffi-
its operation regardless of faults. Faults are inevitable in every real cult for nonlinear systems with uncertainties [13,14].
system and affect the system’s stability and performance level. FTC In comparison with AFTCS, PFTCS does not require an FDI unit
can be described as an integral part of every safety critical system and is less computationally complex. It is fast enough to act
such as airplanes and nuclear facilities. Extensive details and appli- promptly to any abnormal condition but is able to accommodate
cations of FTC are presented in the references [1–3]. only a limited number of faults defined at the design stage and
It is observed that operation during faulty conditions will works offline [15]. Hybrid of both of these techniques is also being
degrade the performance of the overall system but degradation implemented and discussed in detail in Section 4.
up to a certain level can be acceptable as long as reliability is guar- Our contribution in this paper is an organized discussion on FTC
anteed. Such systems are termed as Fault Tolerant Control Systems and it’s all kinds with the latest applications and developments.
and are getting intensive research attention for improved perfor- The significance of the paper is the presentation of the thorough
mance and stability. FTC strategies are now being implemented and organized state of the art of FTCS for the benefit of new and
in production critical systems such as oil and gas and fertilizers existing researchers for future research directions. Further con-
where production loss cannot be accepted and system continual tents of the paper are organized as follows: Section 2 describes
performance is mandatory. The study on the distributed control the architecture, mathematical analysis and latest techniques of
systems is presented in this regard as being the base of many AFTCS, Section 3 describes details of PFTCS and Section 4 explains
industrial manufacturing systems [4]. hybrid of both of these schemes. Hardware redundancy is dis-
Most of the industrial processes are inherent of the nonlinear cussed in Section 5. Stability and reliability analysis techniques
type with time delays. Many of such systems are described as non- are mentioned in Section 6. The conclusion is presented in the last
linear Multi-Input Multi-Output (MIMO) systems. Such systems section with future directions.
are addressed from many years and various control structures have
been designed for such systems having capabilities to handle
2. Active Fault Tolerant Control System
uncertainties, disturbances, and delays. FTC for such nonlinear sys-
tems is addressed in [5–7]. Process control domain is also mostly of
2.1. Architecture of AFTCS
nonlinear nature [8] and FTC for these are discussed in [9–11] more
specifically related to safety issues of modern systems.
The architecture of ATFC is shown in Fig. 1. The main function is
performed by the FDI unit which is termed as Fault Detection and
1.2. Types of FTC Isolation Unit. It is generally implemented as an observer-based
mechanism to generate its own internal output which will then
Fault-Tolerant Control Systems are classified into two main cat- be compared with the actual plant values. Any deviation in sensor
egories: active FTCS (AFTCS) and passive FTCS (PFTCS) [12]. The or actuator out of some specified bound will be declared as a faulty
AFTCS consists of the following subsystems: Fault Detection and Iso- condition. Technically, if the error between actual and estimated
lation (FDI) module, a reconfiguration mechanism and a reconfig- values goes to zero asymptotically or exponentially, the control
urable controller. FDI module provides information online to the declares no fault. If the error signal goes out of bound, an error is

Fig. 1. Structure of Active Fault Tolerant Control.


60 A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68

generated. After detecting an error, the next step is to isolate the Float failure: This fault means that the actuator has become free
faulty component and estimate the possible value. This is termed and has no control over system performance.
as control law reconfiguration in which erratic value will be Runaway or Handover: This fault means that the actuator has
replaced by estimated value to maintain the stability of the system reached its maximum or minimum limit.
and avoid a shutdown. An alarm may be annunciated to warn the Loss of Effectiveness: Actuator becomes ineffective due to struc-
operator about the system faulty state to carry out necessary tural damage to the control surface.
actions to repair or replace the faulty components[16–18]. Broadly, actuator faults can be categorized as the additive and
Fault detection, isolation, controller reconfiguration stapes can multiplicative type and can be represented mathematically as:
be based on excessive computations and thus may take lots of Additive Fault:
time, thereby, making the response of the system quite slow.
uf ¼ u þ f u ð4Þ
Therefore, a major disadvantage of AFTCS could be its slow
response. But since it can cover a wide range of faults and maintain f
where u denotes actuator output, u denotes actuator input and f u is
stability, it is a very advantages control system [19]. actuator fault value.
Multiplicative Fault:
2.2. Mathematical analysis of AFTCS
uf ¼ ð1  qu Þu ð5Þ
Mathematical analysis of control system provides true insight where qy denotes fault gain and 0  qy  1.
into the function of a control system and its limitations about sta- Additive and Multiplicative Fault Combined:
bility. Therefore, it is very important to be able to mathematically
analyze faults and its various types both in sensors and actuators uf ¼ ð1  qu Þu þ f u ð6Þ
as these two components are the most vulnerable to faults. Math-
Consider the following non-linear system:
ematical analysis of the AFTCS is briefly discussed to give the read-
ers a basic idea of how advanced controllers work. Since the main x_ ðt Þ ¼ Axðt Þ þ Buðt Þ þ g ðx; u; t Þ ð7Þ
function of the FTC is to deal with faults, it is important to under-
stand the types and nature of faults in sensors and actuators. Any yðtÞ ¼ Cxðt Þ ð8Þ
fault in the sensor or actuator will create disturbance in the system x 2 Rn is state v ecotr; u 2 Rp is input v ector and y 2 Rm is output
and control objective will not be achieved unless some action is v ector: A, B and C are known system matrices of appropriate
taken to cope with the faults. dimensions. ‘n’ is the number of states, ‘p’ is the number of inputs
Mathematical modeling of faults and observer design for sensor and ‘m’ is the number of outputs.
and actuator faults are studied in ([20–22]) and briefly described
g ðx; u; t Þ is a nonlinear function vector on Rn . Also,
below.
Sensor faults can be classified as follows: zðt Þ ¼ Cxðt Þ ð9Þ
Bias: bias can be described as an offset in the sensor output. where zðt Þ is a clean output fictitious variable.
Mathematically, it can be written as Y ¼ X þ b where X is the true Faults in the sensor can be combined to give a single equation as
output and b is offset in the sensor value. It can be caused by follows:
improper calibration or physical deterioration in the sensor.
Drift: It is also a statistical deviation in the sensor output, linear yðtÞ ¼ ð1  qðt ÞÞzðtÞ þ f ðt Þ ð10Þ
as well as non-linear.
yi ðt Þ ¼ zi ðtÞ þ f i ðx; t Þ ¼ C i xi ðt Þ þ f i ðx; tÞ ð11Þ
Scaling (or Gain): A multiplicative type error in the scaling of
the sensor value and represented as slope as YðtÞ ¼ aðtÞX here g ðx; u; t Þ is assumed to be globally Lipschitz i.e.,
aðtÞ is the time varying scaling factor and its range is: 0  aðtÞ  1 k g ðx1 ; u; t Þ  g ðx2 ; u; t Þ k 6 k kx1  x2 k8u; t ð12Þ
Noise: Noise is random variations in the sensor values possibly
due to ambient, hardware or wire conditions. where k is Lipschitz constant.
Hard Fault: It is defined as a stuck value from a sensor. Mathe- Fig. 3 shows the basic scheme of the observer. The basic func-
matically, YðtÞ ¼ C where Cis a constant value. C = 0 means complete tion of the observer is to produce an estimated value of the plant
loss of sensor and any other value shows the sensor output is stuck. output as per the defined model of the system. Observer design
More broadly, sensor faults are classified as multiplicative and is the most vital part of the FDI scheme. Plant actual output from
additive types. Detailed mathematical fault equations of these the sensor is compared with this estimated values and residual is
are described below: calculated. The controller is designed such that error must
Additive Fault: approach zero asymptotically or exponentially for the stability of
the control system like Lyapunov second method. If the error goes
yf ¼ y þ f y ð1Þ
abruptly out of bound, error is declared in the system. This is math-
f
where y denotes actual faulty output, y denotes system output and ematically shown in the next section.
f y is sensor fault value. A general form of the observer is designed as follows: [23]
Multiplicative Fault:
  x_ ¼ Ax þ Bu ð13Þ
yf ¼ 1  qy y ð2Þ
y ¼ Cx ð14Þ
where qy denotes fault gain and 0  qy  1.
_
 
The combined equation for additive and multiplicative faults x ¼ A x þBu ð15Þ
would be as follows:  
  y¼Cx ð16Þ
y ¼ 1  qy yðt Þ þ f y
f
ð3Þ 
where x is the estimated state value. Subtracting (12) from (14)
and (13) from (15), we get
Fault in the actuator are of following types:
   
Lock failure: This fault means the actuator is locked or jammed _
x  x_ ¼ A x x ð17Þ
or stuck.
A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68 61

          
y y ¼ C x x ð18Þ V_ ¼ eTx RA þ R L C þ AT R þ C T LT R ex þ 2eTx R g x; u; t  g ð x; u;t Þ
  

x x ¼ ex is defined as the error between the estimated and 6 eTx RA þ R L C þ AT R þ C T LT R ex þ 1=l eTx R2 ex
actual state of the system. From fig (2), the observer equation   2
þl k g x; u;t  g ð x; u; tÞ k
can be written as follows:
    
_
 
6 eTx RA þ R L C þ AT R þ C T LT R ex þ 1=l eTx R2 ex þ l k2 k ex k2
x ¼ A x þBu þ L y y ð19Þ
  
where L is state feedback gain matrix. Subtracting (12) from (16), ¼ eTx RA þ R L C þ AT R þ C T LT R þ l k2 I þ 1=l R2 ex
we get
      ð30Þ
_
x  x_ ¼ A x x þ L y y ð20Þ Substituting the Eq. (28) into the above equation to get:
Putting (13) and (15) into (17), we get  
    V_ 6 eTx RA þ XC þ AT R þ C T X T þ l k2 I þ 1=l R2 ex ð31Þ
_
x  x_ ¼ ð A  LC Þ x x ð21Þ
If the following inequality holds, ex converges asymptotically to
e_ x ¼ ð A  LC Þex ð22Þ zero.

  RA þ XC þ AT R þ C T X T þ l k2 I þ 1=lR2 < 0 ð32Þ


y y ¼ Cex ð23Þ By applying the Schur complement, the Eq. (32) becomes equiv-
alent to (27) which completes the proof.
For a stable system, the error vector ex will decay to zero and we Similarly, the following theorem can be obtained.
can get the value of L for the observer.
As described in [20], the equation for the observer design for a Theorem 1:. The error ex ðt Þ approaches to zero exponentially with
nonlinear system would be as follows: rate K=2 if there exist matrices R, X and scalars l, ksuch that
     R ¼ RT > 0andl; k > 0 to satisfy following:
_
 
xðt Þ ¼ A x ðtÞ þ Bu þ g x; u; t þ L C x y ð24Þ " #
RA þ AT R þ XC þ C T X T þ lk2 I þ kR R
Let ex ðtÞ be the error vector <0 ð33Þ
R lI

ex ðtÞ ¼
b x ðt Þ  xðt Þ ð25Þ where

Using the approach for linear system observer design as for Eq. L ¼ R1 X ð34Þ
(20), we can obtain the error equation for a nonlinear system
Similar to the proof of Lemma 1, from (33) and (31), it can be
observer from Eqs. (7) and (24):
 obtained that
     
e_ x ¼ A  L C ex ðtÞ þ g x; u; t  g ðx; u; tÞ ð26Þ V_ ðtÞ  keTx Rex ¼ kV ðtÞ ð35Þ
For the system to be stable, the error e_ x must approach to zero Hence, we get
asymptotically. The second lyapunov stability criterion for nonlin-
ear systems can be used to determine this condition in which a lya- V ðtÞ  eTx V ð0Þ ð36Þ
punov function is defined and its derivative is taken. If the From (29) we get
derivative is proved to be negative, the system would be declared
as stable showing the energy dissipation [24]. Moreover, for sim- kmin ðRÞk ex ðt Þ k2 6 ekt kmax ðRÞk ex ð0Þ k2 ð37Þ
plification of various complex matrix operations, the Schur com- where kmin andkmax are minimum and maximum Eigen values of
plement will also be used [25]. The important steps have been matrix R respectively. Hence, we obtain the following norm of the
summarized below [20]. error vector:
The following lemma 1 gives sufficient condition to determine pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
stability for (26). k e x ðt Þ k 6 kmax ðRÞ=kmin ðRÞ ex ð0Þ ekt=2 ð38Þ
The residual equation can be obtained from Eq. (24) and (7):
Lemma 1:. The error ex ðtÞ approaches to zero asymptotically if there

exists a matrix R, X and scalar l such that R ¼ RT > 0andl > 0 to rðt Þ¼k
^ C x ðt Þ  yðt Þ k
satisfy following linear matrix inequality (LMI):
From equation (25) and (38), we obtain
" # pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
RA þ AT R þ XC þ C T X T þ lk2 I R r ðt Þ 6 kmax ðRÞ=kmin ðRÞ k C kex ð0Þ ekt=2 ð39Þ
<0 ð27Þ
R lI
Note that k ex ð0Þ k is unknown generally, so the threshold in
And the observer gain matrix can be selected as follows: (39) is unavailable. However, since it can be shown that:

L ¼ R1 X ð28Þ k C kex ð0Þk  krð0Þ

To prove it, consider the following Lyapunov function to prove Hence, the following criteria would be obtained for fault detec-
its derivative to be zero as explained earlier: tion for sensor fault
( pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi kt
V ðt Þ ¼ eTx Rex ðtÞ ð29Þ 6 kmax ðRÞ=kmin ðRÞ k r ð0Þ ke 2 ; there is no fault
r ðt Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi kt
ð40Þ
> kmax ðRÞ=kmin ðRÞ k rð0Þ k e 2 ; there is a fault
Now we will check V_ ðxÞ < 08x 2 D  f0g condition along the
trajectory for asymptotic stability. ex is a column vector, therefore, Same process technique can be used to detect and isolate faults
the overall multiplication of matrices eTx Rex ðtÞ will result in a scalar in actuators. After fault detection, isolation, and estimation, the
after performing all multiplication operations as described below: control law is designed to perform controller reconfiguration step.
62 A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68

2.3. Latest techniques used for AFTCS design is obtained through ANNs. In [41] a backstepping approach based
on adaptive neural network-based is presented for underwater
Latest approaches for the FDI design in AFTCS are discussed in vehicles with thruster fault. In [42] backstepping FTC design is pro-
this section for both linear and nonlinear systems. In fact, most posed for MIMO Nonlinear Systems with adaptive neural network
of the practical and industrial processes are of a nonlinear nature. control. In [43] an adaptive NN-based approach is proposed for
One popular approach for fault estimation is the use of Kalman nonlinear time-varying systems with time delay and Unmodeled
Filter(KF). Preliminary knowledge of the working of KF is described dynamics. In [44] NN-based AFTCS is described using the average
in great detail in [26–28]. KF provides state estimation of the sys- dwell time method for a class of switched nonlinear systems with
tem in two steps of time update and measurement update, which its application to RCL circuit.
is then used for comparison or in a residual generation to decide Fuzzy logic (FL) is a qualitative model based technique and is
the faulty state of the system. KF designed for nonlinear applica- also being used for fault diagnosis. FL controllers can handle com-
tions is known as the extended KF (EKF). EKFs are extensively used plex non-linear, multi-variable and time-varying system dynamics.
for FDI applications such as in [29] Extended Multiple Model Adap- Instead of 0 and 1, FLs employ linguistic terms to apply to real-
tive Estimator (EMMAE) is built for the multi–inputmultioutput world scenarios. Thus, FLs have the capability to deal with systems
(MIMO) system for sensor faults. The technique is built for use in having no precise mathematical representation and can approxi-
the industrial applications and demonstrated by an application mate nonlinear functions. The adaptive fuzzy control system is
on the boiler system. In [30] adaptive form of EKF is built for flight implemented in various ways for nonlinear systems. In [45] it is
parameter sensors of an aircraft. In [31] it is used in FDI design of designed with actuator faults considering bias and gain faults. In
stochastic nonlinear and non-stationary dynamic systems and [46] observer-based adaptive fuzzy FTC is designed for stochastic
applied to a perfectly stirred chemical reactor. nonstrict-feedback systems. In [47] it is designed for uncertain
Further, in [32] generalized observer and a dedicated observer switched stochastic systems with time-varying output constraints.
schemes are implemented using a Kalman filter bank. In [33] FDI In [48] Fuzzy-approximation method is proposed for pure-
is implemented for unreliable measurements and correlated noise feedback systems with unknown control directions and sensor fail-
using unscented KF (UKF) for nonlinear systems. In [34] a fault ures. In [49] Takagi-Sugeno (T-S) fuzzy system is used for single-
detection and location (FDL) system is built using a matrix of parameter-learning (SPL) approach for identification and approxi-
hybrid Kalman filters for the coexistence of faults and health mation of the uncertain nonlinear dynamics.
degradation for the nonlinear aero-engine model. In [35] an effi-
cient multiple-model (MM) approach with hybrid Kalman filters 3. Passive fault Tolerant control system
(MHKFs) is presented for a nonlinear and piecewise linear (PWL)
models of the system using a Bayesian approach. 3.1. Architecture of PFTCS
Linear parameter varying method is being extensively utilized
in FTC. In [36] a nonlinear wind turbine system is modeled with As compared to AFTCS, the architecture of PFTCS is quite simple
LPV and fault reconstruction approach is proposed. In [37] a robust and is shown in Fig. 2. There is no FDI unit and no controller recon-
adaptive polytopic observer system is proposed for LPV systems figuration is performed. Rather, the controller works in offline
with time delay. Linear matrix inequalities (LMI) solution approach mode in both normal and abnormal conditions with predefined
is shown in papers [38–40]. In [38] observer is designed for dis- parameters that mask the faulty readings from the components.
crete time LPV systems for actuator fault estimation. In [39] FDI The main advantages of passive FTC techniques can be described
unit is built for asynchronous machine using convex polytopic rep- as follows: Fault detection and controller reconfiguration process
resentation. In [40] robust FDI is built for uncertain LPV systems is not implemented and the system works with the same parame-
solving with LMIs. ters in normal and fault conditions. Due to the elimination of FDI
Artificial neural networks (ANN) are also used in fault diagnosis and controller reconfiguration process, the time response of the
especially for nonlinear systems due to their useful features of self- system becomes very fast as compared to AFTCS. However, the
learning, self-organization and nonlinear systems modeling capa- detection of various types of faults becomes very difficult for one
bilities. Logic can be defined in ANNs for inputs and outputs and passive FTC system since it is designed to work in the offline mode.
mapping can be performed to retrieve outputs for unseen inputs. Hence, its functionality would be limited and can be jeopardized
Due to parallel calculations, fast dynamic response of the system with complex and various kinds of simultaneous faults [3,15,50].

Fig. 2. Basic observer design philosophy.


A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68 63

3.2. Mathematical analysis of PFTCS tains the constant (MsgnðsÞ) plus proportional (ks) rate reaching
law as reported in [55].
Sliding Mode Control (SMC) is one of the most popular PFTCS 1
technique due to its advantages of robustness to external distur- u ¼ g ðx; x_ Þ ½ks  MsgnðsÞ ð49Þ
bances, system parameter variations and model uncertainties Where k and M are positive numbers and sgn is the signum
[51–53]. SMC controller design is mathematically described in function. The function g ðx; x_ Þ must be invertible for Eqs. (48) and
[54] and is mentioned below: (49).
Consider second order single input single output (SISO) nonlin- The overall control input will become:
ear affine system: h i
1
u ¼ g ðx; x_ Þ €x1  f ðx; x_ Þ  ke_  ks  MsgnðsÞ
d
ð50Þ
€xðtÞ ¼ f ðx; x_ Þ þ g ðx; x_ Þu þ df ð41Þ
x 2 Rn is the state vector, u 2 Rp is the control input vector and Substituting this input equation to (46), we get
y 2 Rm is the output vector. A, B and C are known system matrices s_ ¼ ks  MsgnðsÞ þ df ð51Þ
of appropriate dimensions. g ðx; x_ Þ is a nonlinear function vector on
Rn and df represents uncertainties and disturbances within For conventional SMC, the necessary condition would be
bounds jdf j < D. The system is continuous time but time index is 1 d 2
omitted for simplicity. Let the state variables be x1 ¼ x and s < 0; orss_ < 0 ð52Þ
2 dt
x2 ¼ x_ 1 , we have
To reduce the chattering phenomenon, we introduce a bound-

x_ 1 ¼ x2 ary layer and expression becomes
ð42Þ
x_ 2 ¼ f ðx; x_ Þ þ g ðx; x_ Þu þ df ss_ < gjsj ð53Þ
Let the desired trajectory be xd1 ,
then the error between actual Multiplying (51) with s, we get
x1 and the desired trajectory xd1 be written as:
ss_ ¼ ks2  MsgnðsÞs þ df s ¼ ks2  Mjsj þ df s ð54Þ
e ¼ x1  xd1 ð43Þ
With the proper choice of k and M, (53) can be satisfied.
Taking time derivative The chattering effect can be eliminated by using a discontinu-
ous saturation function sat ðsÞ instead of sgn function. The satura-
e_ ¼ x_ 1  x_ 1 ¼ x2  xd2
d
ð44Þ tion function satðsÞ is defined as follows:
The switching surface ‘‘s” is conventionally defined for the sec- 
sgnðsÞ if jsj > us
ond order systems as a combination of the error variables e and e_ sat ðsÞ ¼ ð55Þ
s=us if jsj < us
s ¼ e_ þ ke ð45Þ
whereus is the boundary layer around sliding surface s.
where k sets the dynamics in the switching plane (s ¼ 0Þ.
The control input u should be chosen so that trajectories
approach the switching surface and then stay on it for all future 3.3. Latest techniques used for PFTCS design
time instants.
The time derivative of s would become Robust control is an important characteristic of PFTCS. A litera-
ture review of the robust control is being described for more elab-
s_ ¼ f ðx; x_ Þ þ g ðx; x_ Þu þ df  €x1 þ ke_
d
ð46Þ oration. In [56] engineering applications with robust control are
The control input is expressed as the sum of two terms, as described; in [57] design and implementation of robust control
reported in [55]. laws are discussed and robust control for nonlinear applications
is discussed in survey paper [58].
u ¼ ueq þ u ð47Þ In [59] PFTCS is designed for actuator failures with polytope for-
The first one ueq is called as the equivalent control and is chosen mation of the system by calculating maximum physical values and
using the nominal plant parameters (df ¼ 0), so as to make s_ ¼ 0 peak gains. In [60] it is designed for a class of MIMO nonlinear sys-
when s ¼ 0. It is written as: tems with uncertainties. In [61] PFTCS deals with the outage, loss
  of effectiveness and stuck faults due to parameter uncertainty, dis-
1
ueq ¼ g ðx; x_ Þ €xd1  f ðx; x_ Þ  ke_ ð48Þ turbance and actuator faults. In [62] it is designed for an induction
motor with a nonlinear block control (NBC) transformation. [63]
The second term u is selected to deal with the uncertainties in Robust Adaptive PFTCS based on backstepping approach is
the system and to introduce a reaching law. The second term con- designed for strict-feedback nonlinear systems subjected to

Fig. 3. Structure of Passive Fault Tolerant Control.


64 A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68

Table 1
Comparison between AFTCS and PFTCS.

System’s Property AFTCS PFTCS


Architecture Complex Simple
Time Response Slow Fast
Fault Detection Online/Real Time Offline
Computations Large Relatively Small
Fault Detection and Isolation (FDI) Essential Not Required
Controller Reconfiguration Required Not Required
Noise Effect Can be corrupted by noise and the wrong decision can be made Robust to Noise
Time delay Possible due to noise No Time Delay
Faults nature Various Fixed predefined faults are accommodated
Control Structure Variable Fixed

actuator faults and external disturbances without utilizing neural passive decentralized unconditionally stabilizing (DUS) control
networks. for the sensor faults. In [75] it is proposed for a flight control sys-
As discussed earlier, Sliding Mode Control (SMC) is the most tem for partial actuator failures. In [76] Neural Network-Based
widely used approach for PFTCS design. It is a type of Variable hybrid FTC Scheme is proposed for a class of nonlinear systems
Structure Control (VSC) which basically combines the individual with unknown actuator faults with simulation results. In [77] it
features of different control structures and provides superior per- is proposed for uncertain nonlinear networked control systems
formance as compared to individual control structures. Chattering (NCS) under discrete event-triggered communication scheme
is caused by SMC due to switching, therefore, to remove the chat- (DETCS) with numerical simulation results. In [70] it is proposed
tering problem Super twisting algorithm is used. In [64] SMC based considering actuator saturation and L1 disturbances to compen-
FTC is proposed for quadrotor UAV in which two controllers in cas- sate for the loss of actuators’ effectiveness in pre and post-fault
cade form are designed, first one to deal with the uncertainties and diagnosis intervals.
disturbances to tolerate actuator faults and the second cascaded
SMC to control inner and outer velocities of the quadrotor. In
[65] higher order sliding mode based control (HOSM) scheme is 5. Hardware redundancy
proposed for the air path of a diesel engine. In [66] air-to-fuel ratio
(AFR) control based on Smooth Super-Twisting Algorithm (SSTA) Hardware redundancy is an important aspect in the design of
FTCS as described in [78]. Redundancy was originally developed
for a gasoline engine is presented. In [67], PFTCS based on SMC is
designed for sensor faults for a multi-rotor aircraft without fault to cover hardware components faults. Generally, sensors and actu-
ators are concerned with field devices while the control system is
detection and controller reconfiguration Fig. 3.
A brief comparison of AFTC and PFTC is mentioned in Table 1 placed in a separate environment. So, hardware redundancy con-
cept would deal with the control system side apart from field
regarding important system properties and their response against
each property. devices. Early computing systems used redundant structures with
the voting of parallel independent channels to abolish the effects of

4. Hybrid fault Tolerant control system (HFTCS)

Having studied both the pros and cons of both AFTCS and PFTCS,
a hybrid of both can be designed having advantageous properties
of both systems. For example, the time delay in AFTCS can be a
serious issue in safety-critical systems where minute time delay
can cause disasters. One solution is to use PFTCS with AFTCS, PFTCS
to respond quickly to faults while AFTC to optimize system
performance. Hence, hybrid FTCS can provide both desirable prop-
erties of stability to unexpected fast disturbances and optimal per-
formance. A hybrid approach is presented in [68] using multiple
controllers in which the proposed system does not suffer from
the incorrect decision effect from FDI and ensures stability. Param-
eters uncertainties are also considered and graceful performance Fig. 4. Concept of dual redundancy in a component.
degradation is established. The hybrid approach proposed in [69]
hybrid of FDI and DSU (decentralized stabilization unit) is pro-
posed to guarantee stability in the event of sensor faults and vir-
tual sensor outputs are reconstructed. In [70] unified approach is
presented to cope loss of effectiveness of actuator. Pre and post-
fault diagnosis are handled by PFTCS and AFTCS respectively. FDI
unit does not affect the stability and optimization of system gains
are carried out to handle actuator saturation and disturbance
effects.
The hybrid scheme has been found in various papers. In [71] a
survey of active, passive and hybrid FTC approached is presented.
In [72] it is proposed for aerospace vehicles which are based on
the multiple models. In [73] it is proposed and demonstrated by
controlling a McDonnell F-4C airplane through simulation. In
[74] it is proposed by combining FDI with the newly developed Fig. 5. Structure of triple modular redundancy.
A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68 65

faults or failed components. These also employed information code In [79] TMR scheme has been presented for VLSI voter system in
error detection and diagnostic techniques for the location of faulty which a mid-value or median is detected among four sensors. In
components and replace these through automatic changeovers. In [80] FTC is implemented on digital systems using TMR. In [81]
a modern complex system, fault tolerance is needed in components FTC voter circuit is implemented with TMR on CMOS circuit at
other than hardware such as software and embedded components. the Nanoelectronics level with simulation results. In [82] Imple-
Redundancy can be categorized into two types: direct and ana- mentation of TMR is demonstrated on operating systems. How-
lytical. In the direct redundancy, actual physical hardware redun- ever, there is a limitation of TMR that the reliability of each
dancy is carried out. For example, the use of two or three sensors channel must be greater than 0.5 and faulty modules must be
instead of one to measure the same quantity is called double and replaced immediately after the occurrence of a fault in the channel.
triple redundancy whereas in normal operation, only one sensor Moreover, if two channels become faulty simultaneously, the sys-
is sufficient for control purposes. These extra hardware compo- tem shuts down and cannot produce output [78,83].
nents make the system reliable in case of failure of the main ele- In the analytical redundancy, an observer based on the mathe-
ment. A voting system decides which channels are working matical model of the system is used to provide an estimate of the
correctly and which are faulty. This hardware redundancy concept signals of interest instead of having multiple sensors that measure
is also extended to the actuators. the same signal. Sensor value is compared with the estimated
The concept of dual redundancy in components is shown in value to generate residual. Any abnormal increase in residual
Fig. 4. The reliability of the overall system increases with the dual would indicate a fault in the sensor. This algorithm runs in the con-
redundancy because the system will fail only when both of the trol computer and no actual additional hardware is used. This fea-
components fail. The system will continue its performance in case ture is desirable in many systems especially in unmanned air
of a fault in any one component at a time (Fig. 5). vehicles (UAVs) and aircraft to reduce extra hardware, cost, and
Let reliabilities of individual components are denoted weight [84]. In paper [85] robust analytical redundancy for nonlin-
byR1 andR2 and overall reliability of the dual redundant system is ear systems is described that is insensitive to state and unknown
denoted by RDR . Mathematically, we can write parameters. In [86] Analytic redundancy-based fault diagnosis
technique (ARFDT) is implemented for maintenance systems and
RDR ¼ 1  ðbothcompoentsfailÞ ð56Þ
sequential probability ratio test is used in fault isolation. The
If R1 ¼ R2 ¼ R method compensates sensor outputs in case of outages and is
implemented on aircraft system sensors.
RDR ¼ 1  ð1  RÞ2 ð57Þ

6. Stability and reliability


RDR ¼ 2R  R2 ð58Þ
If R ¼0.9, RDR ¼ 0:99 Stability is the most important aspect of FTC especially in air-
In redundant modular systems, faulty modules are discon- craft where human lives are involved and in critical production
nected and isolated from the system and new modules are inserted facilities such as nuclear power plants. The Lyapunov method is
in a highly tight control environment. If a fault arises in a model, it the most popular method in determining the stability of control
is detected instantly and replaced with a healthy module. systems. For linear time-invariant systems, Lyapunov first method
Triple Modular Redundancy (TMR) design is one of the high- is used which states that the Eigen values must lie in a left half
reliability hardware architecture in hardware redundancy domain. complex plane. For nonlinear systems, the Lyapunov second
Three parallel channels work to produce results and voting among method is mostly used in which a Lyapunov function is defined.
them is carried out to generate the final output. 2 out of 3 voting The Lyapunov function must be positive definite and satisfy the
process results free of error performance and no single point of fail- Lyapunov equation. Its derivate must also be negative definite
ure. The three channels are identical and independent each per- showing energy dissipation with the passage of time indicating
forming its own action parallel to one another. If a fault arises in asymptotic stability of the systems [24]. Lyapunov Second criterion
any channel, it is detected by the voter mechanism and the final can be written mathematically as:
output is generated from the rest two channels. Meanwhile, the
X
n X
n
operator is alerted of the faulty situation by annunciations and V ðxÞ ¼ xT Px ¼ pij xi xj ; P ¼ PT ð63Þ
alarms. Limitation of the TMR system is that it can only mask i¼1 j¼1
one module fault. A failure in either of the remaining modules
would cause the voter to produce an erroneous result. However, V ðxÞ is the Lyapunov function and P is the system matrix:
it has been determined that the dependability of a TMR system
P > 0ðpositive definiteÞ if and only if ki ðP Þ > 0 8i
can be improved by removing failed modules from the system.
Let reliabilities of individual components are denoted Q is a positive definite matrix.
byR1 ; R2 andR3 and overall reliability of TMR is denoted by RTMR . Math-
ematically, reliability function of TMR can be determined as follows: if P > 0; Q > 0 then the system x_

RTMR ¼ R1 R2 R3 þ ð1  R1 ÞR2 R3 þ R1 ð1  R2 ÞR3 þ R1 R2 ð1  R3 Þ ð59Þ ¼ Ax is ðglobally asymptoticallyÞ stable; i:e:; Rki < 0

¼ R1 R2 R3 þ R2 R3  R1 R2 R3 þ R1 R3 R1 R2 R3 þ R1 R2 R1 R2 R3 kmin ðQ Þ T


V_ ðxÞ ¼ xT Qx  kmin ðQ ÞxT x   x Px ¼ aV ðxÞ ð64Þ
kmax ðPÞ
¼ R1 R2 þ R2 R3 þ R1 R3  2R1 R2 R3 ð60Þ
kmin ðQ Þ
If R1 ¼ R2 ¼ R3 ¼ R a¼ >0 ð65Þ
kmax ðPÞ
2 2 2 3
RTMR ¼ R þ R þ R  2R ð61Þ
V ðxÞ > 0; V_ ðxÞ < 08x 2 D  f0g ð66Þ
RTMR ¼ 3R2  2R3 ð62Þ
Above Eq. (51) is the main condition in the controller design
If R ¼0.9, RTMR ¼ 0:972 which establishes the stability criterion of the system.
66 A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68

MTTF is calculated for the unrepairable systems such that com-


ponents can’t be repaired [89–91].
Mean Time to Repair (MTTR) is a basic measure to determine
the maintainability of repairable items. It represents the average
time required to repair a failed component or device. These param-
eters are explained well covering the state of the art with limita-
tions and application example to the nuclear power plant is
mentioned in [92].
In [93] reliability modeling of FTC is described using semi Mar-
kov process model and reliability function computation. In [94]
reliability enhancement of multilevel modular converters is men-
tioned by using FTC techniques in bulk power transmission exam-
ple and increased reliability is shown both in normal as well as in
faulty conditions. Markov chains have been used to investigate
Fig. 6. Bathtub curve for failure rate of a component. reliability by solving differential equations. In [95] reliability study
is carried out by dividing the system lifetime into different phases
and identifying key components that can cause system failure.
In [87] Lyapunov stability and asymptotic stability for second Importance analysis of these key components and subcomponents
order systems is explained. In [88] concepts and connections are carried out to design automatic control schemes and optimiza-
between Lyapunov stability and passivity for nonlinear descriptor tion of these schemes are carried out to obtain overall greater sys-
systems are described. In [24] controllability and stability of nonlin- tem reliability and lifetime.
ear systems are explained with Lyapunov functions and feedback.
Reliability can be defined in terms of dependability and avail-
ability of the system under specified time and conditions. Reliabil- 7. Latest applications of fault Tolerant control
ity depends on the failure rate of the system components,
operating time, and environmental conditions in which the system The FTC has become a vital control design element in non-
is being used. Reliability of a system or component is the probabil- process as well as for process conditions. FTCS has become an
ity of performing its required function under the stated operating essential part of unmanned air vehicles (UAVs) to maintain system
environment for a specified duration of time. An important param- stability with acceptable performance in case of actuator and sen-
eter in the reliability study is a failure rate. The failure rate can be sor faults [96,97]. In [98] multilevel reconfiguration for actuator
defined as the expected number of failures per unit time. It is gen- and sensors is described based on LPV framework for NASA aircraft.
erally represented as k. For example, if a component fails after NASA has used FTC in modern drone control system design [99].
1000 h, its failure rate would be k ¼ 10001
or 1  103 failures/h. Triconix has used TMR in safety instrumented or emergency shut-
Failure rates of components usually follow a bathtub curve dis- down (ESD) systems due to its high integrity in critical process
tribution as shown in Fig. 6. This is a widely used tool in the relia- applications that require a significant degree of safety and avail-
bility study and is named according to its shape. In this curve, the ability [100]. The system also contains intrinsically safe barriers
lifetime of a component is shown in three phases: early life, con- as per requirement of hazardous environment safety that contains
stant failure and wear out period. In early life, the component a low level of energy to prevent spark in case of short-circuiting in
experiences a greater number of faults due to imperfections and the field. Moreover, an additional feature to communicate with a
anomalies in the manufacturing process. After continual learning, Distributed Control System (DCS) has also been incorporated. Gen-
constant failure rate period starts in which the component seldom eral Electric (GE) has designed a MarkVIeS system for SIL-3 safety
experiences failure. After this useful period, wear out period starts system applications based on TMR [101,102]. These systems are
in which component experiences several failures due to wear and utilized in the turbine protection and control as well as process
tear by excessive usage. safety applications. These systems provide high performance in
Failure rate (k) will be constant during the useful period or mid- extreme harsh systems. Rockwell Automation has designed Pro-
section of the curve and its reliability function would be exponen- cess Safety & Critical Control system named as GuardLogix 1756
tial as mentioned below: utilizing TMR for Safety Integrity Level (SIL)-3 applications [103].
Siemens has designed S7-400F/FH for fail-safe automation plants
Rðt Þ ¼ ekt ð67Þ with increased safety features [104]. The system can be used for
It is also known as the exponential failure law for calculating SIL-3 applications and has a modular design. As TMR has the lim-
time-varying reliability function. itation of working with two healthy channels, further research
One of the main quantitative measure for determining the reli- direction for more reliability enhancement of TMR would be to
ability of the system is the mean time to failure (MTTF). MTTF enable the system working on a single channel tolerating simulta-
shows the mean lifetime of an item. It is the average time during neous faults in the other two channels.
which item will be expected to last in operation. There is a model-
ing assumption of MTTF that the failed item has infinite repair time
or simply stating the item cannot be repaired. MTTF is computed 8. Conclusions
by the following formula:
Z In this paper, a comprehensive state of the art review of fault
1
MTTF ¼ Rðt Þdt ð68Þ tolerant control systems (FTCS) is presented consisting of the latest
0 advances and applications. An FTCS aims to accommodate faults in
MTTF can also be calculated by taking the reciprocal of failure the system components during operation and maintain stability
rate (k) as follows: with little or acceptable degradation in the performance levels.
Mathematical modeling of various types of faults in the sensors
1 and actuators has been described with respect to FTC strategies.
MTTF ¼ ð69Þ
k Types and classifications of FTC are described with the latest
A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68 67

design techniques. Finally, a review of the stability and reliability regression-based observer model, Meas. Control (2019), https://doi.org/
10.1177/0020294018823031.
analysis of FTCS is presented.
[19] J. Lunze, J.H. Richter, Reconfigurable fault-tolerant control: a tutorial
Future research areas in FTC are mentioned below for further introduction, Eur. J. Control. 14 (2008) 359–386, https://doi.org/10.3166/
exploration by researchers. ejc.14.359-386.
[20] Youqing Wang, S. Donghua Zhou, Joe Qin, Hong Wang, Active fault-tolerant
control for a class of nonlinear systems with sensor faults, Int. J. Control
 FTC scheme implementation for process plants Autom. Syst. 6 (2008) 339–350.
 Fault isolation for multiple actuator faults [21] F. Yang, H. Zhang, B. Jiang, X. Liu, Adaptive reconfigurable control of systems
 Computational time reduction for AFTCS with time-varying delay against unknown actuator faults, Int. J. Adapt Control
Signal Process. 28 (2014) 1206–1226, https://doi.org/10.1002/acs.2439.
 Improving PFTCS capabilities for optimum performance [22] H. Alwi, C. Edwards, C.P. Tan, Fault Tolerant Control and Fault Detection and
 Integration of AFTCS and PFTCS Isolation. Fault Detect. Fault-Toler. Control Using Sliding Modes, Springer-
 Integration of FTC and sophisticated hardware redundancy Verlag, London, 2011, pp. 7–27.
[23] Norman S. Nise, Design via State Space, in: Control Syst. Eng., 6th ed., Wiley-
algorithms like TMR Blackwell, 2012, pp. 663–710.
 Cost optimization due to complexity and hardware additions [24] F. Clarke, Lyapunov functions and feedback in nonlinear control, in: Optim.
 Physical size reduction Control Stab. Nonsmooth Anal., Springer, Berlin, Heidelberg, 2004, pp. 267–
282, doi:10.1007/978-3-540-39983-4_17.
 Graceful performance degradation [25] T. Ando, Schur complements and matrix inequalities: operator-theoretic
 Industrial applications critical to safety and production approach, in: F. Zhang (Ed.), Schur Complement Its Appl., Springer US, Boston,
MA, 2005, pp. 137–162, doi:10.1007/0-387-24273-2_6.
[26] Y. Pei, S. Biswas, D.S. Fussell, K. Pingali, An elementary introduction to Kalman
Funding filtering. ArXiv171004055 Cs, 2017.
[27] G. Welch, G. Bishop, An Introduction to the Kalman Filter, 2006.
The author(s) received no financial support for the research, [28] X. Liu, A. Goldsmith, Kalman filtering with partial observation losses. 2004
43rd IEEE Conf. Decis. Control CDC IEEE Cat No04CH37601, vol. 4, 2004, p.
authorship, and/or publication of this article. 4180-4186 Vol.4. doi:10.1109/CDC.2004.1429408.
[29] A.M. Bardawily, M. Abdel-Geliel, M. Tamazin, A.A.A. Nasser, Sensors fault
estimation, isolation and detection using MIMO extended Kalman filter for
Declaration of Competing Interest
industrial applications. 2017 10th Int. Conf. Electr. Electron. Eng. ELECO,
2017, p. 944–8.
The author(s) declare no conflict of interest in preparing this [30] L. Van Eykeren, Q.P. Chu, J.A. Mulder, Sensor fault detection and isolation
paper. using adaptive extended kalman filter*, IFAC Proc. 45 (2012) 1155–1160,
https://doi.org/10.3182/20120829-3-MX-2028.00195.
[31] Y. Chetouani, Using the kalman filtering for the fault detection and isolation
References (FDI) in the nonlinear dynamic processes, Int. J. Chem. React. Eng. 6 (2008),
https://doi.org/10.2202/1542-6580.1411.
[32] D.H. Trinh, H. Chafouk, Fault detection and isolation using Kalman filter bank
[1] J. Jiang, Fault-tolerant Control Systems—An Introductory Overview, Acta
for a wind turbine generator. 2011 19th Mediterr. Conf. Control Autom. MED,
Autom. Sin. 31 (2005) 161–174.
2011, p. 144–9. doi:10.1109/MED.2011.5983155.
[2] M.S. Mahmoud, Y. Xia, Fault-tolerant control systems, in: Anal. Synth. Fault-
[33] H. Chen, L. Yan, Y. Xia, M. Fu, B. Xiao, G. Hu, et al. UKF based fault detection
Toler. Control Syst., Wiley-Blackwell, 2013, pp. 135–173, doi:10.1002/
and state estimation for nonlinear systems with correlated noise. 2017 36th
9781118700365.ch4.
Chin. Control Conf. CCC, 2017, p. 5294–9. doi:10.23919/ChiCC.2017.8028193.
[3] P.M. Frank, Trends in fault-tolerant control of engineering systems, IFAC Proc.
[34] Y. Yuan, X. Liu, S. Ding, B. Pan, Fault detection and location system for
37 (2004) 377–384, https://doi.org/10.1016/S1474-6670(17)31053-4.
diagnosis of multiple faults in aeroengines, IEEE Access 5 (2017) 17671–
[4] M.S. Mahmoud, Y. Xia, Industrial fault-tolerant architectures, in: Anal. Synth.
17677, https://doi.org/10.1109/ACCESS.2017.2744639.
Fault-Toler. Control Syst, Wiley-Blackwell, 2013, pp. 307–319.
[35] B. Pourbabaee, N. Meskin, K. Khorasani, Sensor fault detection, isolation, and
[5] M. Benosman, A survey of some recent results on nonlinear fault tolerant
identification using multiple-model-based hybrid kalman filter for gas
control, Math. Probl. Eng. (2010), https://doi.org/10.1155/2010/586169.
turbine engines, IEEE Trans. Control Syst. Technol. 24 (2016) 1184–1200,
[6] M.S. Mahmoud, Y. Xia, Fault-tolerant nonlinear control systems, in: Anal.
https://doi.org/10.1109/TCST.2015.2480003.
Synth. Fault-Toler. Control Syst., Wiley-Blackwell, 2013, pp. 175–218, https://
[36] H. Shao, Z. Gao, X. Liu, K. Busawon, Parameter-varying modelling and fault
doi.org/10.1002/9781118700365.ch5.
reconstruction for wind turbine systems, Renew Energy 116 (2018) 145–152,
[7] J.-Y. Shin, C.M. Belcastro, Performance analysis on fault tolerant control
https://doi.org/10.1016/j.renene.2017.08.083.
system, IEEE Trans. Control Syst. Technol. 14 (2006) 920–925, https://doi.org/
[37] H. Hamdi, M. Rodrigues, C. Mechmeche, Braiek N. Benhadj, Observer-based
10.1109/TCST.2006.876911.
fault diagnosis for time-delay LPV descriptor systems, IFAC-Pap 51 (2018)
[8] T.A. Hughes, Introduction to Process Control. Meas. Control Basics Fifth Ed. 5
1179–1184, https://doi.org/10.1016/j.ifacol.2018.09.702.
edition, Research Triangle Park, NC: International Society of Automation; 2015.
[38] Z. Wang, M. Rodrigues, D. Theilliol, Y. Shen, Actuator fault estimation
[9] A. Fekih, Fault tolerant control design for complex systems: Current advances
observer design for discrete-time linear parameter-varying descriptor
and open research problems. 2015 IEEE Int. Conf. Ind. Technol. ICIT, 2015, p.
systems, Int. J. Adapt Control Signal Process. 29 (2015) 242–258, https://
1007–12. doi:10.1109/ICIT.2015.7125229.
doi.org/10.1002/acs.2469.
[10] Blanke M, Staroswiecki M, Wu NE. Concepts and methods in fault-tolerant
[39] C. Mechmeche, S. Rihane, H. Dallagi, S. Nejim, Fault detection and isolation for
control. Proc. 2001 Am. Control Conf. Cat No01CH37148, vol. 4, 2001, p.
linear parameter varying system application to asynchronous machine, in:
2606–20 vol.4. doi:10.1109/ACC.2001.946264.
2018 Int. Conf. Adv. Syst. Electr. Technol. ICASET, Hammamet, IEEE, 2018, pp.
[11] X. Yu, J. Jiang, A survey of fault-tolerant controllers based on safety-related
327–334, doi:10.1109/ASET.2018.8379877.
issues, Annu. Rev. Control 39 (2015) 46–57, https://doi.org/10.1016/j.
[40] D. Henry, A. Zolghadri. Robust fault diagnosis in uncertain linear parameter-
arcontrol.2015.03.004.
varying systems. 2004 IEEE Int. Conf. Syst. Man Cybern. IEEE Cat No04CH37583,
[12] J. Jiang, X. Yu, Fault-tolerant control systems: A comparative study between
vol. 6, 2004, p. 5165–70 vol.6. doi:10.1109/ICSMC.2004.1401014.
active and passive approaches, Annu. Rev. Control 36 (2012) 60–72, https://
[41] Y. Wang, M. Zhang, P.A. Wilson, X. Liu, Adaptive neural network-based
doi.org/10.1016/j.arcontrol.2012.03.005.
backstepping fault tolerant control for underwater vehicles with thruster
[13] Y. Zhang, J. Jiang, Bibliographical review on reconfigurable fault-tolerant
fault, Ocean Eng. 110 (2015) 15–24, https://doi.org/10.1016/j.
control systems, Annu. Rev. Control 32 (2008) 229–252, https://doi.org/
oceaneng.2015.09.035.
10.1016/j.arcontrol.2008.03.008.
[42] H. Gao, Y. Song, C. Wen, Backstepping design of adaptive neural fault-tolerant
[14] I. Hwang, S. Kim, Y. Kim, C.E. Seah, A survey of fault detection, isolation, and
control for MIMO nonlinear systems, IEEE Trans. Neural Netw. Learn Syst. 28
reconfiguration methods, IEEE Trans. Control Syst. Technol. 18 (2010) 636–
(2017) 2605–2613, https://doi.org/10.1109/TNNLS.2016.2599009.
653, https://doi.org/10.1109/TCST.2009.2026285.
[43] S. Yin, H. Yang, H. Gao, J. Qiu, O. Kaynak, An adaptive NN-based approach for
[15] M. Benosman, Passive fault tolerant control, Robust Control Theory Appl
fault-tolerant control of nonlinear time-varying delay systems with
(2011), https://doi.org/10.5772/14334.
unmodeled dynamics, IEEE Trans. Neural Netw. Learn Syst. 28 (2017)
[16] Z. Gao, C. Cecati, S.X. Ding, A survey of fault diagnosis and fault-tolerant
1902–1913, https://doi.org/10.1109/TNNLS.2016.2558195.
techniques—part I: fault diagnosis with model-based and signal-based
[44] L. Tang, D. Ma, J. Zhao, Neural networks-based active fault-tolerant control for
approaches, IEEE Trans. Ind. Electron. 62 (2015) 3757–3767, https://doi.org/
a class of switched nonlinear systems with its application to RCL circuit, IEEE
10.1109/TIE.2015.2417501.
Trans. Syst. Man. Cybern. Syst. (2018) 1–13, https://doi.org/10.1109/
[17] M. Pawlak, Active fault tolerant control system for the measurement circuit in
TSMC.2018.2847283.
a drum boiler feed-water control system, Meas Control 51 (2018) 4–15,
[45] Y. Li, S. Tong, Fault diagnosis and fuzzy fault-tolerant control design of
https://doi.org/10.1177/0020294018754714.
nonlinear systems with actuator faults. 2016 35th Chin. Control Conf. CCC,
[18] A.A. Amin, K. Mahmood-ul-Hasan, Robust active fault-tolerant control for
2016, p. 6512–6514. doi:10.1109/ChiCC.2016.7554381.
internal combustion gas engine for air–fuel ratio control with statistical
68 A.A. Amin, K.M. Hasan / Measurement 143 (2019) 58–68

[46] H. Ma, Q. Zhou, L. Bai, H. Liang, Observer-based adaptive fuzzy fault-tolerant [74] S.W. Su, J. Bao, P.L. Lee, A hybrid active–passive fault-tolerant control
control for stochastic nonstrict-feedback nonlinear systems with input approach, Asia-Pac. J. Chem. Eng. (2006).
quantization, IEEE Trans. Syst. Man. Cybern. Syst. (2018) 1–12, https://doi. [75] X. Yu, J. Jiang, Hybrid fault-tolerant flight control system design against
org/10.1109/TSMC.2018.2833872. partial actuator failures, IEEE Trans. Control Syst. Technol. 20 (2012) 871–
[47] Y. Liu, H. Ma, H. Ma, Adaptive fuzzy fault-tolerant control for uncertain 886, https://doi.org/10.1109/TCST.2011.2159606.
nonlinear switched stochastic systems with time-varying output constraints, [76] Q. Shen, B. Jiang, P. Shi, Neural Network-Based Fault Tolerant Control
IEEE Trans. Fuzzy Syst. 26 (2018) 2487–2498, https://doi.org/10.1109/ Scheme Against Un-modeled Fault, in: Q. Shen, B. Jiang, P. Shi (Eds.), Fault
TFUZZ.2018.2814596. Diagn. Fault-Toler. Control Based Adapt. Control Approach, Cham, Springer
[48] X. Liu, D. Zhai, T. Li, Q. Zhang, Fuzzy-approximation adaptive fault-tolerant International Publishing, 2017, pp. 163–190, doi:10.1007/978-3-319-52530-
control for nonlinear pure-feedback systems with unknown control 3_7.
directions and sensor failures, Fuzzy Sets Syst. 356 (2019) 28–43, https:// [77] J. Wang, X. Yao, W. Li, Hybrid Active-Passive Robust Fault-Tolerant Control of
doi.org/10.1016/j.fss.2018.03.017. Event-Triggered Nonlinear NCS, Open Electr. Electron Eng. J. 11 (2017),
[49] S. Gao, H. Dong, B. Ning, X. Yao, Single-parameter-learning-based fuzzy fault- https://doi.org/10.2174/1874129001711010068.
tolerant output feedback dynamic surface control of constrained-input [78] E. Dubrova, Hardware Redundancy. Fault-Toler. Des., Springer-Verlag, New
nonlinear systems, Inf. Sci. 385–386 (2017) 378–394, https://doi.org/ York, 2013, pp. 47–69.
10.1016/j.ins.2017.01.014. [79] M.D. Krstic, M.K. Stojcev, G.L. Djordjevic, I.D. Andrejic, A mid-value select
[50] J. Eich, B. Sattler, Fault Tolerant Control System Design Using Robust Control voter, Microelectron Reliab. 45 (2005) 733–738, https://doi.org/10.1016/j.
Techniques, IFAC Proc. 13 (1997) 1237–1242, https://doi.org/10.1016/S1474- microrel.2004.07.006.
6670(17)42566-3. [80] R. Sß inca, C. Szász, Fault-tolerant digital systems development using triple
[51] S.K. Spurgeon, Sliding mode observers: a survey, Int. J. Syst. Sci. 39 (2008) modular redundancy, Int. Rev. Appl. Sci. Eng. 8 (2017) 3–7, https://doi.org/
751–764, https://doi.org/10.1080/00207720701847638. 10.1556/1848.2017.8.1.2.
[52] Merheb AR, Noura H, Bateman F. Passive fault tolerant control of quadrotor [81] M.H. Rahman, S. Rafique, M.S. Alam, A fault tolerant voter circuit for triple
UAV using regular and cascaded Sliding Mode Control. 2013 Conf. Control modular redundant system, J. Electr. Electron Eng. 5 (2017) 156–166, https://
Fault-Toler. Syst. SysTol, 2013, p. 330–335. doi:10.1109/ doi.org/10.11648/j.jeee.20170505.11.
SysTol.2013.6693910. [82] S. Hudson, R.S. Shyama Sundar, S. Koppu, Fault control using triple modular
[53] P.K. Nandam, P.C. Sen, Industrial applications of sliding mode control, in: redundancy (TMR), in: S.S. Rautaray, H. Das, J. Nayak, P.K. Pattnaik (Eds.),
Proc. IEEEIAS Int. Conf. Ind. Autom. Control, 1995, p. 275–280. doi:10.1109/ Prog. Comput. Anal. Netw., Springer Singapore, 2018, pp. 471–480.
IACC.1995.465829. [83] A.A. Amin, K. Mahmood-Ul-Hasan, Advanced fault tolerant air-fuel ratio control
[54] H. Noura, D. Theilliol, J.-C. Ponsart, A. Chamseddine, Actuator and sensor of internal combustion gas engine for sensor and actuator faults, IEEE Access 7
fault-tolerant control design, in: Fault-Toler. Control Syst. Des. Pract. Appl., (2019) 17634–17643, https://doi.org/10.1109/ACCESS.2019.2894796.
Springer-Verlag, London, 2009, pp. 7–40. [84] A.D. Pouliezos, G.S. Stavrakakis, Analytical redundancy methods, in: Real
[55] G.K. Singh, K.E. Hole, Guaranteed performance in reaching mode of sliding Time Fault Monit. Ind. Process., Springer Netherlands, Dordrecht, 1994, pp.
mode controlled systems, Sadhana 29 (2004) 129–141. 93–178, doi:10.1007/978-94-015-8300-8_2.
[56] T. Li, B. Zhang, Z. Feng, B. Zheng, Robust control with engineering [85] A. Shumsky, Robust Analytical Redundancy Relations for Fault Diagnosis In
applications, Math Probl. Eng. (2014), https://doi.org/10.1155/2014/567672. Nonlinear Systems. Asian J Control n.d.;4:159–70. doi:10.1111/j.1934-
[57] Petkov P, Kralev J, Slavov T. Design And Implementation Of Robust Control 6093.2002.tb00342.x.
Laws. Proc - 29th Eur Conf Model Simul ECMS 2015 2015:6–18. doi:10.7148/ [86] C. Chi, W. Zhang, X. Liu, Application of analytic redundancy-based fault
2015-0006. diagnosis of sensors to onboard maintenance system, Chin. J. Aeronaut. 25
[58] J. Wang, Robust and nonlinear control literature survey (No. 19), Int. J. Robust (2012) 236–242, https://doi.org/10.1016/S1000-9361(11)60383-X.
Nonlinear Control (2010), https://doi.org/10.1002/rnc.1619. [87] D.S. Bernstein, S.P. Bhat, Lyapunov stability, semistability, and asymptotic
[59] X. Yu, Y. Zhang, Design of passive fault-tolerant flight controller against stability of matrix second-order systems, J. Mech. Des. 117 (1995) 145–153,
actuator failures, Chin. J. Aeronaut. 28 (2015) 180–190, https://doi.org/ https://doi.org/10.1115/1.2836448.
10.1016/j.cja.2014.12.006. [88] C. Yang, J. Sun, Q. Zhang, X. Ma, Lyapunov stability and strong passivity
[60] A. Nasiri, S.K. Nguang, A. Swain, D. Almakhles, Passive actuator fault tolerant analysis for nonlinear descriptor systems, IEEE Trans. Circuits Syst. Regul.
control for a class of MIMO nonlinear systems with uncertainties, Int. J. Pap. 60 (2013) 1003–1012, https://doi.org/10.1109/TCSI.2012.2215396.
Control (2017) 1–12, https://doi.org/10.1080/00207179.2017.1367102. [89] E. Dubrova, Dependability evaluation techniques, in: Fault-Toler. Des.,
[61] X.J. Li, G.H. Yang, Robust adaptive fault-tolerant control for uncertain linear Springer-Verlag, New York, 2013, pp. 19–42.
systems with actuator failures, IET Control Theory Appl. 6 (2012) 1544–1551, [90] Peebles PZ. Probability. Probab. Random Var. Random Signal Princ. 4th Ed. 4th
https://doi.org/10.1049/iet-cta.2011.0599. Edition edition, TMH; 2002.
[62] H. Tohidi, K. Erenturk, S. Shoja-Majidabad, Passive fault tolerant control of [91] V. Kumar, L. Singh, A.K. Tripathi, Reliability analysis of safety-critical and
induction motors using NBC, J. Control Eng. Appl. Inform. 19 (2017), 49–58– control systems: a state-of-the-art review, IET Softw. 12 (2017) 1–18, https://
58. doi.org/10.1049/iet-sen.2017.0053.
[63] J.X. Zhang, G.H. Yang, Robust adaptive fault-tolerant control for a class of [92] V. Kumar, L. Singh, A.K. Tripathi, Reliability analysis of safety-critical and
unknown nonlinear systems, IEEE Trans. Ind. Electron. 64 (2017) 585–594, control systems: a state-of-the-art review, IET Softw. 12 (2018) 1–18, https://
https://doi.org/10.1109/TIE.2016.2595481. doi.org/10.1049/iet-sen.2017.0053.
[64] Merheb AR, Noura H, Bateman F. Passive fault tolerant control of quadrotor [93] H. Li, Q. Zhao, Reliability Modeling of Fault Tolerant Control Systems. Proc.
UAV using regular and cascaded Sliding Mode Control. 2013 Conf. Control 44th IEEE Conf. Decis. Control, 2005, p. 2397–402. doi:10.1109/
Fault-Toler. Syst. SysTol, 2013, p. 330–5. doi:10.1109/SysTol.2013.6693910. CDC.2005.1582521.
[65] Murtaza G, Butt YA, Bhatti AI. Higher order sliding mode based control [94] M. Abdelsalam, H. Diab, S. Tennakoon, A. Griffiths, Reliability enhancement of
scheme for air path of diesel engine. 2016 Int. Conf. Emerg. Technol. ICET, modular multilevel converter by applying fault tolerant control. 2016 51st
2016, p. 1–6. doi:10.1109/ICET.2016.7813217. Int. Univ. Power Eng. Conf. UPEC, 2016, p. 1–5. doi:10.1109/
[66] R. Anjum, I. Khan, A. Yar, A.I. Bhatti, Air-to-fuel ratio control of gasoline UPEC.2016.8114008.
engines using smooth sliding mode algorithm, in: 2017 13th Int. Conf. Emerg. [95] D.U.I. Hongyan, Reliability optimization of automatic control systems based
Technol. ICET, 2017, p. 1–6. doi:10.1109/ICET.2017.8281731. on importance measures: a framework, Int. J. Perform. Eng. 12 (2015) 297–
[67] Fu C, Tian Y, Peng C, Gong X, Zhang L, Guo X. Sensor faults tolerance control 300.
for a novel multi-rotor aircraft based on sliding mode control. Proc Inst Mech [96] Unmanned Aircraft. NASA n.d. https://www.nasa.gov/subject/9566/
Eng Part G J Aerosp Eng 2017:0954410017731590. doi:10.1177/ unmanned-aircraft/.
0954410017731590. [97] Gipson L. Unmanned Aircraft Systems Integration in the National Airspace
[68] M. Maki, J. Jiang, K. Hagino, A stability guaranteed active fault-tolerant System (UAS in the NAS). NASA n.d. http://www.nasa.gov/aeroresearch/
control system against actuator failures, Int. J. Robust Nonlinear Control 14 programs/iasp/uas.
(12) (2004) 1061–1077, https://doi.org/10.1002/rnc.932. [98] Péni T, Vanek B, Szabó Z, Bokor J. Supervisory fault tolerant control of the
[69] S.W. Su, J. Bao, P.L. Lee, A hybrid active–passive fault-tolerant control NASA AirStar aircraft. 2014 Am. Control Conf., 2014, p. 666–71. doi:10.1109/
approach. Asia-Pac J Chem Eng n.d.;1:54–62. doi:10.1002/apj.7. ACC.2014.6859264.
[70] M. Khatibi, M. Haeri, A unified framework for passive–active fault-tolerant [99] Stepanyan V. Identification and Reconfigurable Control of Impaired Multi-
control systems considering actuator saturation and L1 disturbances, Int. J. Rotor Drones, San Diego, CA, United States: 2016
Control (2017) 1–11, https://doi.org/10.1080/00207179.2017.1365172. [100] Triconex Safety System | Schneider Electric Solutions n.d. https://schneider-
[71] Z. Gao, C. Cecati, S.X. Ding, A survey of fault diagnosis and fault-tolerant electric.pepperl-fuchs.com/se/en/733.htm.
techniques—part II: Fault diagnosis with knowledge-based and hybrid/active [101] MarkVIeS. GE Autom 2014. http://www.geautomation.com/products/mark-
approaches, IEEE Trans. Ind. Electron. 62 (2015) 3768–3774, https://doi.org/ vies-safety-management-system.
10.1109/TIE.2015.2419013. [102] GE Mark VIeS Safety Management Solution - ISA n.d. https://www.isa.org/
[72] J.D. Boskovic, R.K. Mehra, Hybrid fault-tolerant control of aerospace vehicles, templates/news-detail.aspx?id=136202.
in: Proc. 2001 IEEE Int. Conf. Control Appl. CCA01 Cat No01CH37204, 2001, p. [103] Trusted Triple Modular Redundant (TMR) Controller | Rockwell Automation
441–6. doi:10.1109/CCA.2001.973905. n.d. https://www.rockwellautomation.com/global/products/safety-
[73] M. Maki, J. Jiang, K. Hagino, A stability guaranteed active fault-tolerant instrumented-systems/overview.page?
control system against actuator failures, Int. J. Robust Nonlinear Control 14 [104] S7-400/S7-400H/S7-400F/FH n.d. https://mall.industry.siemens.com/mall/
(2004) 1061–1077, https://doi.org/10.1002/rnc.932. en/WW/Catalog/Products/5000014 (accessed December 8, 2018).

You might also like