Journal of Sound and Vibration 387 (2017) 115

Contents lists available at ScienceDirect

Journal of Sound and Vibration

journal homepage: www.elsevier.com/locate/jsvi

Non-probabilistic stability reliability measure for active

vibration control system with interval parameters
Yunlong Li, Xiaojun Wang n, Lei Wang, Weichao Fan, Zhiping Qiu
Institute of Solid Mechanics, Beihang University, Beijing 100191, China

a r t i c l e i n f o abstract

Article history: A systematic non-probabilistic reliability analysis procedure for structural vibration active
Received 29 October 2015 control system with unknown-but-bounded parameters is proposed. The state-space re-
Received in revised form presentation of active vibration control system with uncertain parameters is presented.
18 September 2016
Compared with the robust control theory, which is always over-conservative, the relia-
Accepted 9 October 2016
bility-based analysis method is more suitable to deal with uncertain problem. Stability is
Handling Editor: L.G. Tham
Available online 21 October 2016 the core of the closed-loop feedback control system design, so stability criterion is
adopted to act as the limited state function for reliability analysis. The uncertain para-
Keywords: meters without enough samples are modeled as interval variables. Interval perturbation
Non-probabilistic reliability
method is employed to estimate the interval bounds of eigenvalues, which can be used to
Active control of vibration
characterize the stability of the closed-loop active control system. Formulation of defining
Interval perturbation
Closed-loop system the reliability of active control system based on stability is discussed. A novel non-prob-
Stability analysis abilistic reliability measurement index is discussed and used to determine the probability
of the stability based on the area ratio. The feasibility and efficiency of the proposed
method are demonstrated by two numerical examples.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Active control of vibration is an important concern in the design of aerospace, as the demands for reduction of structural
vibrations are increasing and conventional passive vibration control is difficult to meet the design requirements [1,2]. For
high frequency vibration, passive control is effective and efficient. However, for the low frequency vibration, passive control
is inefficiency. Meanwhile, conventional passive vibration control need vibration isolation material, which is bulky and will
significantly increase the weight of the aircraft [3]. As the design of structure must meet the precise and tolerant of vi-
bration, active vibration control is becoming increasingly important and meaningful. The active control system is a com-
bination of sensors to measure the responses of the structure, actuators to apply secondary forces onto the structure and
control law to determine how the control force is applied. That is, the closed-loop active control system refers to a complex
multi-disciplinary problem, including material mechanic, structural dynamics, signal processing and control theory [4]. For
complex system, uncertainties are inherent such as material properties, geometric error and measurement error [5]. It is
well known that uncertainties in the active control system can degrade the control performance and even lead to system
instabilities. Thus, it is essential to explore the influence of uncertainties in the active vibration control system.
Traditional techniques dealing with the uncertainties in the field of control theory is using the worse-case measure of the

n
Corresponding author.
E-mail address: xjwang@buaa.edu.cn (X. Wang).

http://dx.doi.org/10.1016/j.jsv.2016.10.010
0022-460X/& 2016 Elsevier Ltd. All rights reserved.
2 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115

system, which is called robust analysis and robust design [6]. Zames firstly proposed the H control theory in 1981 [7]. Then,
the robust control theory was widely studied, and the active control methods were proposed to suppress the structural
vibration in 1980s [811]. Some representative research has emerged since then. Kar et al. proposed a static state feedback
controller to control the bending and torsional vibration of a flexible plate by using H-based control theory [12]. Li and Yam
proposed a model-based fuzzy controller to achieve the robust vibration suppress by updating the coefficient matrix online
[13]. Hu and Ng used integrated piezoelectric actuators to control the vibration of the flexible structure, the combination of
standard LQR (linear quadratic regulator) and nonlinear control signal are adopted to deal with the uncertainties in the
closed-loop control system [14]. A robust control strategy was proposed by Zuo and Slotine based on frequency-shaped
sliding control theory [15]. For some actual problems of active vibration control of engineering structures, the closed-loop
control system obtained based on robust control theory is always over-conservative [16]. That is, robust control theory can
achieve the requirement, but the active control system may consume more energy or need greater active control force
(secondary force). Generally speaking, robustness is a much stricter constraint than reliability in the field of active vibration
control. In order to reduce the conservation, reliability-based control theory is more suitable for dealing with the un-
certainties in the active vibration control system [16,17]. To design a reliable controller, reliability analysis of the active
control system is essential. Spencer et al. employed first and second order reliability methods (FORM/SORM) to calculate the
probabilistic reliability measures of a controlled structure [18], and the stability of the closed-loop control system was
chosen as the state limited function [19]. Breitung et al. proposed a reliability analysis method for nonlinear control system,
and similarly FORM/SORM methods based on system stability are employed [20]. Taflanidis et al. designed a reliable con-
troller to minimize the probability of structural failure [21]. The most work was done in the probabilistic frame, and gen-
erally the probability density function (PDF) requires more information of the uncertain parameters. Interval method as an
effective means of dealing with uncertainty can overcome the shortcoming of probabilistic method. Guo employed the
infinity norm of the uncertain variables vector to define the reliability, and based on this, a robust reliability method was
proposed as a measure of stability of controlled structure [22]. This definition of the reliability proposed by Guo only can
measure the case that the shortest distance from the origin to the failure surface is greater than 1. In view of this, Wang et al.
firstly proposed the non-probabilistic reliability based on the ratio of the volume of the safe region to the total volume [23].
The definition of non-probabilistic reliability proposed by Wang can measure the case that the reliability is smaller than 1.
Then, the non-probabilistic reliability concept was introduced into the reliability analysis of wing flutter [24]. A 15-degree
sweptback wing numerical example illustrate the validity of the non-probabilistic reliability index. Non-probabilistic re-
liability has received considerable attention in the structural design, but in the field of the active vibration control have not
been given sufficient attention. Certainly, the final goal is to design a reliable controller. However, the premise of design is
the reliability analysis. Thus, only the stability reliability analysis of the active control system is implemented in this paper.
In this paper, an approach for non-probabilistic reliability analysis of active control system is proposed by combining
interval perturbation method and non-probabilistic reliability measurement index. The composition of the paper is as
follows. In Section 2, there will be a brief review of vibration problem with interval uncertainties. In order to estimate the
interval bounds of eigenvalues of closed-loop systems with interval parameters, the interval perturbation method is pre-
sented in Section 3. In Section 4 reliability measurement for active vibration control system is developed based on non-
probabilistic reliability theory. Numerical examples are provided in Section 5 to show the accuracy and efficiency of the
proposed reliability analysis method, followed by conclusions in Section 6. The matrix notation is used as much as possible
to simplify the presentation.

2. Problem statement

Consider the following controlled structural system with n degrees of freedom modeled by finite element method

( t ) + Pw ( t ) + Kw ( t ) = f f ( t ) + Bc fc ( t )
Mw (1)

where M n n , P n n and K n n are the mass, damping and stiffness matrices, respectively. Bc n s is the location
matrix of the control input, f s 1 is control input and s is the number of actuators. The vectors
w n 1, w n 1, w n 1 are the displacement, velocity, and acceleration vectors, respectively. Here, the vector
ff s 1 represents external disturbance forces, which can be taken to zero in this paper (the external disturbance cannot
affect the stability of the system). Ignoring the external disturbance vector f f , one can transform the vibration Eq. (1) into a
state-space equation by using the state vector x ( t ) = wT ( t ), w T ( t ) as follows
T

x = Ax + Bu
y = Cx + Du (2)

where
 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115 3

0 I
A=
M1K M1P
0
B = 1
M Bc
i1
C=

0, ... , 0 1 0, ... , 0
D=0 (3)

and y represents the output (displacement) of the ith node.

The essential reason for using active vibration control system is to stabilize the responses of systems. In general, most of
the structures are stable, the introduction of active control system is to improve the performances. However, the active
control system can terminate the symmetry of the structural system, bringing the possibility of instability. Furthermore,
because of various errors in design, manufacture and measure, uncertainties in controlled structure are inevitable, which
may exacerbate the possibility of instability of the system. Thus, the possibility of instability of structures with active control
system is explored in this paper.
Considering the uncertainties of the structures with active controller, the global matrices M, P and K of the structure in
Eq. (1) are not deterministic and depend on a p-dimensional uncertain parameter vector b = b1, b2 , . . . bp . Since there are
T

not sufficient data to get information on the probability of uncertain variables, we assume that the uncertain parameters bj
(component of the uncertain vector) fluctuate within the corresponding interval denoted by b Ij . That is

bj b Ij = b j , bj = bcj bj , bcj + bj

b b I = b , b = b c b, b c + b (4)

where b = ( b j ) and b = bj are the lower and upper bound of the uncertain parameter vector b = b1, b2 , . . . bp , re-
( ) T

( )
spectively. bc = bcj and b = ( bj ) are the nominal value and the radius of interval vector b I , respectively. The system
matrices A and B are functions of the uncertain parameters and can be denoted as A ( b) and B ( b), respectively. The interval
system matrices A and B can be obtained by using the interval algorithm or other optimization algorithm.
Then, the state-space Eq. (2) can be rewritten as follows

x ( t ) = ( Ac + Ae) x ( t ) + ( Bc + Be) u ( t )
y ( t ) = Cx ( t ) (5)

where Ac and Bc are the nominal values, A and B are the uncertain radii, and e = 1, 1, respectively.
It is assumed that the state variables of structures can be measured and the system is controllable, and the following
state feedback control law is adopted.

u ( t ) = Gx ( t ) (6)

where G is the gain matrix. Substituting Eq. (6) into Eq. (5) yielding the following closed-loop system

x ( t ) = ( Ac + Ae Bc G BGe) x ( t )
y ( t ) = Cx ( t ) (7)

As mentioned above, the vital motivation for using active vibration control system is to stabilize the responses of sys-
tems. However, due to the existence of uncertainties, the gain matrix G obtained by using the nominal value of system
matrices may not meet the requirements. That is to say, when the gain matrix G obtained by using the nominal value is
applied to the vibration system, there may be the possibility of instability in the closed-loop control system. In this paper,
the possibility of instability will be discussed in the reliability field. In the common reliability theory, limit state function is
the essential concept which plays an important role. Thus, we will find out a limit state function which can character the
instability of the closed-loop control system in order to construct the reliability analysis framework.

3. Interval perturbation method for estimating the interval bounds of eigenvalues of closed-loop systems with in-
terval parameters

The stability of the closed-loop feedback system is the core of the control system design. The so-called stability is that if
the input of the system is within a certain bounded interval, the responses (output) of the active control system stay within
another bounded interval for all time. This physical definition can also be interpreted as a mathematical definition, which is
a formal definition of stability proposed by Lyapunov. This definition of stability is not an effective method to check the
stability of a given closed-loop feedback system. Hence, many methods had been proposed to check the stability of a given
4 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115

closed-loop feedback system, which require less effort than the Lyapunov definition method. Among them, eigenvalue-
based methods are the more effective methods. The stability of the closed-loop active control system is directly related to
the root of the characteristic equation and the eigenvalues of the system matrix for a system in state space format. In this
paper, the eigenvalue based method is adopted to check the stability. Thus, the interval perturbation method to estimate the
interval bounds of eigenvalues of closed-loop systems with interval parameters is proposed in this section.
Consider the closed-loop system as follows

x ( t ) = A x ( t ) (8)

where A = ( Ac + Ae BcG BGe) is an interval matrix. In practice, Lyapunov stability condition is equivalent to the
eigenvalues criterion, which is that a feedback closed-loop control system is stable if and only if there are no eigenvalues
with a positive real part of the state matrix. Now, the stability problem of feedback closed-loop system is transformed into
an eigenvalues problem of interval matrix. The basic problem is formulated as follows:

A = (9)

where is the eigenvalue of A and is the associated eigenvector. As mentioned above, the state matrix A is an interval
matrix, which is the function of the uncertain parameter vector b = b1, b2 , . . . bp . Thus,
T

A = A ( b) (10)

The interval bounds of the eigenvalues and the corresponding eigenvectors, subject to the constraint conditions Eq. (4),
are

= { : , A ( b) = , b b } I
(11)

The bound of the set defined by Eq. (11) is generally very complicated. Obviously, it is difficult to find the exact interval
bounds of the eigenvalues. For simplicity, but without loss of accuracy, we can depend on the first-order Taylor expansion to
get the estimates of the eigenvalues. In other word, we can find out a closed interval for each eigenvalue i , i = 1, 2, . . . , 2n,
such that

I = , = iI , iI = i , i , i = 1, 2, ... , 2n
( ) (12)

where i = min i and i = max i .

bbI bbI
In order to get the interval bounds of the eigenvalues, first-order interval perturbation algorithm is introduced in this
section. We first review the matrix perturbation theory which will be used in the interval perturbation method for esti-
mating the interval bounds of eigenvalues of closed-loop systems. Consider the following standard eigenvalue problem
c
A ic = ic ic , ( ic )T A c = ic ( ic )T i = 1, 2, ... , 2n (13)
c c
where A is the unperturbed matrix, ci is ith right eigenvector of A and ci is ith left eigenvector corresponding to the ith
eigenvalue ic , respectively. Generally, the right eigenvectors ci , i = 1, 2, ... , 2n can be normalized as follows

( ic )T ic = 1, i = 1, 2, ... , 2n (14)
c
Now we give a small change A to the unperturbed matrix A , and the perturbed eigenvalue problem is obtained in the
following form
c
( A )
+ A ic = i ic i = 1, 2, ... , 2n (15)
c
where i is the ith perturbed eigenvalues. Assumed that A , A are known in this problem, and the perturbation item of i is
i . Substituting i = ic + i into Eq. (15) and expanding the equations yield
c
A ic + A ic = ic ic + i ic i = 1, 2, ... , 2n (16)

Considering the first equation of Eq. (13), one can rewrite Eq. (16) as follows

A ic = i ic i = 1, 2, ... , 2n (17)
T
Premultiplying Eq. (17) by ( ci
) yields
T
i = ( ic ) A ic (18)

That is
T
i = ic + i = ic + ( ic ) A ic (19)

Due to the active control system destroying the symmetry of the structural system, the eigenvalues of the state matrix A
 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115 5

will be complex interval numbers. Thus, Eq. (19) can be rewritten in real and imaginary part forms as follows

(
i = ir + j iy = irc + ir + j iyc + iy ) (20)

where ir and iy are the real part and imaginary parts of the eigenvalues, respectively. ir and iy are the corresponding
perturbation item. j = 1 . Comparing Eq. (20) and Eq. (19) yields the following equations

( ic )T A ic = ir + jiy (21)

Considering the real part and the imaginary part of the eigenvectors ci , we can obtain

( irc + jiyc )T A ( irc + jiyc ) = ir + jiy (22)

Expanding Eq. (22) and respectively considering the real part and the imaginary part yields
T
( )T
irc = ( irc ) A irc iy
c
A iy
c

T c T c
iyc = ( irc ) A iy
c
+ ( iy ) Air (23)

Firstly, for the real part of the complex eigenvalues i , i = 1, 2, . . . , 2n, based on the interval extension principle, the
following interval expression of the real part can be obtained by using interval algorithm

irI = irc + irI (24)

where irI = ir , ir = ( cir ) A cir ciy A ciy , A = A , A and A is the radius of the interval state matrix A .
T I T I I
( )
Using the interval operations algorithm and letting ir be the radius of real part of the complex eigenvalues i yields

irI = ir , ir , i = 1, 2, ... , 2n (25)

T
where ir = irc
ir , ir = irc
+ ir and ir = + cir T A
. cir ciy A ciy
Similarly, the imaginary part of the complex eigenvalues i , i = 1, 2, . . . , 2n can be obtained by using the above op-
erations and Eq. (23) as follows

iyI = iy, iy , i = 1, 2, ... , 2n (26)

T
where iy = iyc iy , iy = iyc + iy and ir = cir T A ciy + ciy A cir .
The interval eigenvalues of the interval state matrix A can be obtained by combining Eq. (25) and Eq. (26) as follows

iI = irI + j iyI
irI = ir , ir
iyI = iy, iy (27)

where
T c T c T c T c
ir = irc irc A irc iy A iy , ir = irc + irc A irc + iy A iy
T c T c T c T c
iy = iyc irc A iy
c
iy A ir , iy = iyc + irc A iy
c
+ iy A ir (28)

Generally, the uncertainties of the active control system are small, which satisfies the conditions of using the pertur-
bation [25,26]. The presented method is practical and effective to solve the standard eigenvalues problem of interval matrix.
So far, based on the interval mathematics and orthogonality of the right eigenvectors and the left eigenvectors of the interval
matrix, the closed interval iI , which satisfy Eq. (12), for each eigenvalue i is found. Then, how to use the interval eigen-
values to measure the non-probabilistic reliability of the active vibration control system with interval parameters will be
introduced in next section.

4. Non-probabilistic reliability measurement for active vibration control system

The closed-loop feedback active control system is designed originally for achieving certain performance requirements, as
the structure design needs more accurate and less tolerant of steady or transient vibration. However, due to the un-
certainties of the closed-loop control system, the performance will de degraded and furthermore the uncertainties can
destroy the stability leading the closed-loop system to instability. Thus, the reliability issues of closed-loop system are more
prominent than the original system without control input. The traditional techniques for dealing the uncertainties are
seeking the worst cases to design the controllers, which is the so-called robust control and a conservative control algorithm.
6 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115

Fig. 1. Stable region of closed-loop system with two uncertain parameters.

Reliability based technique is an effective manner of dealing uncertain problem. Similarly, in order to design a robust control
system, robust analysis will be achievement firstly. In this section, a novel non-probabilistic reliability measurement for
active vibration control system with interval uncertain parameters, which will be useful and meaningful for the reliability
control design in the future, will be proposed.

4.1. Reliability problem formulation of active control system based on stability

A stable system is a dynamic system with a bounded response to a bounded input [27]. Based on the modern control
theory, we know that the sufficient and necessary condition for a closed-loop feedback control system to be stable is that
each eigenvalue of the state matrix A has a negative part. The reliability of a closed-loop control system is defined as the
probability of the system is stable. If the system is instable, will be considered as a failure active control system design. Thus,
the reliability of the closed-loop system is the probability Sstable of the closed-loop control system being stable. Then, the
reliability of the closed-loop control system can be defined as follows

Sstable = Pos { 2n
Re i ( b) < 0
i=1
( )} (29)

where Re [ ] denotes the real part of the complex number. Pos denotes the possibility.
As shown in Fig. 1, the stable region is complicated, so the probabilistic methods are inconvenient to us to obtain the
reliability. Moreover, the probability density function cannot be accurately estimated without sufficient data. In order to
achieve the reliability analysis of the closed-loop system, we should get the limit state function of this problem. Considering
the Eq. (29), we can adopt the limit state function for single component as follows

M = M (Re cr , Re ib ) = Re cr Re ib = 0, i = 1, 2, ...2n (30)

where

M>0 Safe
M=0 Limit State Surface
M<0 Failure (31)

Because of the influences of the uncertain factors in the closed-loop active control system, Re cr and Re ib will be
uncertain varies. As mentioned above, Re cr and Re ib will vary within the interval obtained by Eq. (28)

Re cr Re cr Re cr
Re ib Re ib Re ib (32)

where Re cr and Re cr are the lower bound and upper bound of the critical value, respectively. Re ib and Re ib are
the lower bound and upper bound of the each eigenvalue of the closed-loop control system. Based on the interval
mathematic, the critical value interval and eigenvalue interval can be

( Re cr )I = Re cr , Re cr
( Re ib )I = Re ib , Re ib (33)
 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115 7

Fig. 2. Interference model for comparing interval number and deterministic number.

I I
where Re ib can be obtained by Eq. (28) and Re cr can be given by the designer of the closed-loop active control
( ) ( )
system.
I
For the stability based reliability analysis, Re cr is generally equal to deterministic number 0. Then, the interval
( )
descriptions of the eigenvalues and the critical value can be placed on a number axis as shown in Fig. 2.
During the design process of closed-loop active control system, the real parts of the eigenvalues of the state matrix must
be smaller than zero, indicating that the closed-loop active control system with respect to the nominal values must be
stable. However, due to the uncertainties, the real parts of the eigenvalues may be larger than zero, as shown in Fig. 2 (the
shaded region). Just like the terminology in probabilistic reliability theory, Fig. 2 can be called non-probabilistic interval
interference model.
Generally, the eigenvalues of the closed-loop active control system can be placed in any positions of the s-plane. In some
special cases, the real part of the eigenvalue will be required to within a certain interval. Similarly, if the real part of the
eigenvalue exceeds the given interval, we also consider that the closed-loop system is failure. In those cases, the following
non-probabilistic interval interference model will be useful as shown in Fig. 3. During the design process of the controller of
closed-loop system, the real parts of the eigenvalues of the state matrix must be smaller than a certain interval, indicating
that the closed-loop active control system with respect to the nominal values must satisfy the performance requirement.
However, due to the uncertainties, the real parts of the eigenvalues may be larger than the lower bound of the given critical
interval, as shown in Fig. 3.
For the designer, in order to meet the reliability requirements whether performance reliability or stability reliability, the
algorithm of the reliability should be obtained firstly, which is the non-probabilistic reliability measurement index discussed
in the next section.

4.2. Non-probabilistic reliability measurement index

In this section, the stable reliability of closed-loop control system will be calculated based on the non-probabilistic
interval reliability model proposed in the above section. For the first case where the critical value is a deterministic number
0, the possibility that the real part of the eigenvalue is smaller than 0 will be

Re ib
Sstable ( M > 0) =
Re ib Re ib (34)

which can be defined as the ration of the length of shaded region in Fig. 2 to the total length of the interval. For the other
case where the critical value is an interval number, we can extend the idea to the comparing of two intervals.
Representing the critical interval and interval of the eigenvalue in a plane yields the solid rectangle which is crossed by
the failure plane Re cr = Re ib as shown in Fig. 4. The safe region in the figure represent that the real part of the
eigenvalue is smaller than the critical value, that is M > 0. The possibility that the real part of the eigenvalue is smaller than
the critical value or M > 0 can be defined as the ration of the area of the safe region to the total area of the solid rectangle, as
follows
Asafe
Sstable = S ( M > 0) =
Atotal (35)

The possibility that the real part of the eigenvalue is larger than the critical value can be defined as the ratio of the area of
the failure region to the total area of the solid rectangle
Afailure
Sinstable = S ( M < 0) =
Atotal (36)

Fig. 3. Interference model for comparing two interval numbers.

8 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115

Fig. 4. Scheme for the space of variables.

The possibility of stability can be referred as the non-probabilistic reliability measurement index. We can easily find that
Sstable + Sinstable = 1 (37)

which is similar to the probabilistic method for reliability analysis.

Certainly, the designer generally wishes that the closed-loop active control system can be 100 percent stable for a certain
period of time. This goal can be reached by using the robust control theory, which is a worst-case measure of the stability of
the active control system. However, this control theory using the bounds of the uncertain parameters is more conservative,
which will make the system redundant. Conservative can be avoided effectively, if the reliability based design method of
active control system is adopted in the process of controller design. When the upper bound of the real part of the eigenvalue
of the state matrix is equal to the lower bound of the critical value, this state can be defined as critical state as shown in
Fig. 5, which means that the reliability of the active control system is 100%. Compared with robust control theory, the
reliability based control theory is more appropriate from the perspective of dealing with uncertainties.
In order to be more intuitive to show the conservatism of robust control, the reliability of the robust control system can
be defined as the sum of the shortest distance from the safe region to the failure plane and 1 as shown in Fig. 6. At this
point, the reliability of the system is greater than 100%. The larger reliability index means that the control system is more
conservative. The greater the reliability index is; the more conservative system will be. In this way, we can use the reliability
index of the controlled system as a measure of the conservatism of the system. The greater the conservatism of the system
is, the greater active control force will be needed.

Sstable = S ( M > 0) = 1 + (38)

In the field of structure reliability analysis, the failure plane mostly is linear, we can easily calculate the areas of the safe
region and failure region. However, in other research fields, there may be nonlinear failure plane. For the general nonlinear
limit state function (shown in Fig. 7), the above concept of the non-probabilistic safety measure can still be applied, as a

Fig. 5. Scheme for the critical state.

 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115 9

Fig. 6. No-probabilistic reliability measurement of robust control system.

Fig. 7. Scheme for the nonlinear limit state function.

ratio of the appropriate areas. Certainly, we can use calculus to obtain the area ration. However, the limit state function is
generally unknown, so the Monte Carlo simulation method will be popular in the calculation of the area ration.
Generally, the maximum eigenvalue can be used to measure the reliability of the active control system with interval
parameters. However, due to the uncertainties, the real part of other eigenvalue whose interval radius is much larger will be
likely greater than zero. Therefore, every eigenvalue of the state matrix has to be considered. The failure mode of the system
can be represented in Fig. 8.
Here, the possibility that the real parts of all the eigenvalues are smaller than the critical value or M > 0 can be defined as
the product of every reliability as follows
2n
Sstable = S1S2, ... , S2n = Si
i (39)

where Si is the single reliability obtained by the ith eigenvalue.

Now, a novel non-probabilistic reliability concept and measurement index for active vibration control system with in-
terval uncertain parameters has been proposed. In order to clarify the above development, two numerical examples are
implemented in the next section.

5. Numerical examples

Example 1. Consider the active control problem of F4E fighter aircraft presented by Ngamsom and Hoberock in 2003 as
follows [28]. State variable x ( 1), x ( 2) and x ( 3) are normal acceleration, angular velocity of pitch and drift angle of lift

Fig. 8. Illustration of series system.

10 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115

rudder, respectively.

x ( t ) = ( Ac + A) x ( t ) + ( Bc + B) u ( t )

where
0.8251 17.76 90.245 91.44

A = 0.1734 0.7549 11.1 , B = 0
0 0 250 250
0.1645 0.35 5.905 6.34

A = 0.0914 0.0963 0.29 0.1, B = 0 0.1
0 0 0 0

To illustrate the conservatism of the robust control, we should measure the reliability of the robust control system firstly.
For this problem, we obtain the robust controller K = 58.2757, 44.1966, 19.4212 based on robust control theory [29]. With
the obtained robust controller, the interval bounds of eigenvalues of the closed-loop controlled system can be obtained
based on Eq. (28)

1I = 411.03, 385.97
2I = 411.03, 385.97
3I = 56.60, 3.4022

Then, the reliability index of the controlled system can be obtained as follows based on above intervals of eigenvalues
Sstable = 385.97 + 1 = 386.97 > > 1
Sstable = 385.97 + 1 = 386.97 > > 1
Sstable = 3.402 + 1 = 4.402 > > 1

Based on the definition of reliability index which is greater than 100%, we conclude that the controlled system with
robust controller K = 58.2757, 44.1966, 19.4212 is more conservative. That is, robust control can achieve the goal of de-
sign, but there is greater safety margin.
Given the controller of the closed-loop control system obtained by reliability-based method K = 8.8, 71, 1.02,
which can be obtained by pole assignment method (the desired real-part of eigenvalues are
399.11,
399.11,
3.025), we
also can give the reliability of the controlled system [30]. Based on the standard eigenvalue solving method, the eigenvalues
and the corresponding eigenvectors of the nominal system can be obtained as shown in the Tables 1 and 2. In order to show
the conservatism of the robust control, the active control forces of robust control and reliable control system are shown in
Fig. 9. As shown in Fig. 9, in order to keep the system robust, the system need greater force, which means the complex
actuator is needed in the robust system. Under normal condition, the weight of the complex actuator is bigger than the
simple one. For aerospace design, we need reduce the weight as much as possible. That is, we need to reduce the con-
servatism of the traditional robust control.
From Table 1, we can obviously realize that all the real parts of the eigenvalues of the state matrix are smaller than zero
(is negative). That is, the nominal closed-loop active control system is stable. However, due to the uncertainties in the
system, the closed-loop control system may be instable. The right eigenvectors of the nominal system are orthonormalized,
which satisfies the condition Eq. (14). So, the interval analysis method proposed in the previous section can be used to
obtain the lower and upper bounds of the eigenvalues of the system with interval parameters.
The bounds of the real and imaginary parts of the interval eigenvalues of the closed-loop control system with bounded-
but-unknown uncertainties obtained by interval perturbation (Elapsed time is 0.007274 s, Intel s Core i7-2600 CPU @
3.40 GHz, RAM16 GB) are listed in the Table 3, and as a comparison the results derived by Monte Carlo method (100000
times simulation, Elapsed time is 11.083179 s, Intel s Core i7-2600 CPU @ 3.40 GHz, RAM16 GB) are also presented.
While, the stable reliability is also summarized in the Table 3. Under this condition (or controller), the nominal closed-loop
control system is stable, but there will be the possibility of instability. In order to intuitively give the instability, the re-
sponses of the closed-loop control system are shown in the following Fig. 10. The initial control is x = 2, 0, 1 and there is
no external excitation.

Table 1
The eigenvalues of the nominal system.

1cr 1cy 2cr 2cy 3cr 3cy

399.11 199.17
399.11
199.17
3.025 0

 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115 11

Table 2
The right eigenvectors of the nominal system.

1c r 1c y 2c r 2c y c3r c3y

0.3424 0.915 104
0.3424 0.915 104 0.9924 0

0.021 0.0105 0.021
0.0105
0.123 0
0.939 0 0.939 0 0.9656 102 0

Fig. 9. Comparison of active control force of robust and reliable control system.

Table 3
The bounds of the eigenvalues.

iI ir ir iy iy Component reliability Total reliability

Interval perturbation 1I
400.493
397.734 199.011 199.337 1 63.72%

2I
400.494
397.734
199.337
199.011 1
3I
14.047 7.997 0 0 63.72%

Monte Carlo 1I
401.907
395.946 173.396 221.106 1 65.86%

2I
401.907
395.946
221.106
173.396 1
3I
14.865 7.704 0 0 65.86%

Fig. 10. The state variables of the nominal system with controller.
12 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115

Fig. 11. An instable case of closed-loop control system (reliability 63.72%).

According to the nominal system matrix, the closed-loop active control system can be designed to make the system
stable as presented in Fig. 10.However, as shown in Fig. 11, even though the nominal control system is stable, due to the
existence of uncertainties, there will be the possibility of instable. This numerical example shows that the reliability analysis
of active control system is very necessary, and it can provide another way for the controller design, which can improve the
reliability and reduce the conservation of the traditional robust control. This problem is a stable reliability problem of active
control system based on the theory we proposed in the previous section, which is an interference model for comparing
interval number and deterministic number. In next example, the reliability analysis of eigenstructure assignment active
control system is discussed.

Example 2.

Plate structure is an important and common component in the field of aerospace. The vibration of the plate seriously
affects the performance of aircraft. The active control of a cantilever plate is discussed in this example. Young's modulus
E = 70GPa , Poisson's ratio = 0.3, thickness t = 2mm and the mass density = 2700kg /m3. The damping matrix is pro-
portional to the stiffness matrix and mass matrix and P = M + K is used in this example, in which = 5 and
= 1.8 105. The thickness of elements and the Young's modulus exhibit some uncertainties and are considered to be
interval numbers which are taken as t I = 1.8mm , 2.2mm and EI = 68GPa, 72GPa.
As shown in Fig. 12, in order to validate the MATLAB codes, the finite element model of the cantilever plate is built in the
ANSYS. As the characterization of the dynamic responses, the natural frequencies of the cantilever plate obtained by using
the ANSYS and MATLAB are listed in the Table 4. As shown in Table 4, the MATLAB codes used to design the closed-loop
control system is effective and creditable. Due to the operability of the MATLAB codes, the following results obtained by
using the MATLAB. The commercial finite element software ANSYS is just a method to validate.

Fig. 12. The finite element model of cantilever plate in ANSYS.

 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115 13

Table 4
Natural frequencies of the cantilever plate.

Natural frequency ANSYS (Hz) MATLAB (Hz)

1 123.71 123.70
2 483.79 483.70
3 655.15 655.15
4 859.84 859.84
5 1494.1 1494.1
6 1562.5 1562.5
7 2506.5 2505.6
8 2596.1 2596.1
9 2650.2 2650.2
10 3531.4 3531.4

Fig. 13. The illusion of closed-loop control system.

Fig. 14. Comparison of the uncontrolled and controlled system in Case 1 (impulse excitation).

The critical interval values of the eigenvalues are 1Icri = 8.22, 7.3 108, 2Icri = 7.5, 7.3 108,
= 7.2, 7.0 108, 4I cri = 7.0, 6.8 108 and 5Icri = 6.7, 6.5 108, respectively. That is, we want to set
3I cri
the eigenvalues in these positions. Due to the uncertainties, we assume that the design is effective as long as the eigenvalue
falls into each interval. For convenient, only two cases are discussed as shown in Fig. 13. The controller obtained by the
designer is G = 10000 for the case with one actuator, and G = [7500, 2330, 1280] for the case with three actuators. And,
negative feedback control strategy is adopted.
The effect of closed-loop control system is illustrated in Fig. 14. From Fig. 14, we can discovery that for nominal system
the controller can meet the requirements. However due to the uncertainties, there may be possibility that the eigenvalues
may be greater than the critical interval. Using the reliability index proposed in this paper, we can obtain the reliability of
the active control system with the current controller as listed in Table 5 and Table 6. Certainly, in order to demonstrate the
accuracy, effective and efficiency of the reliability proposed in this paper, Monte Carlo simulation method (100,000 samples)
is also adopted. From Table 5 and Table 6, we can see that the reliability obtained by using the interval method is smaller
than the accurate results by using the Monte Carlo. It is the interval natural expansion that caused this phenomenon. This is
a compromise between accuracy and efficiency. For large-scale practical engineering problems, the method proposed in this
paper is a better choice.
14 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115

Table 5
The bounds of the eigenvalues and reliability (Case1).

iI Component reliability Total reliability Time

(
ir 108 ) (
ir 108 )
Interval perturbation 1I
8.504
7.958 93.19% 93.19% 0.845 s
2I
8.099
7.573 1
3I
7.780
7.266 1
4I
7.509
7.004 1
5I
7.273
6.766 1

Monte Carlo 1I
8.260
8.204 99.5% 99.5% 27125 s

2I
7.862
7.814 1
3I
7.544
7.504 1
4I
7.275
7.239 1
5I
7.034
7.002 1

Table 6
The bounds of the eigenvalues and reliability (Case 2).

iI Component reliability Total reliability Time

(
ir 108) (
ir 108 )
Interval perturbation 1I
8.552
8.007 95.45% 95.45% 1.033 s
2I
8.130
7.605 1
3I
7.807
7.294 1
4I
7.537
7.033 1
5I
7.286
6.781 1

Monte Carlo 1I
8.264
8.205 99.8% 99.8% 33238 s

2I
7.870
7.823 1
3I
7.551
7.510 1
4I
7.282
7.245 1
5I
7.042
7.009 1

6. Conclusions

An approximate methodology for the reliability problem of active control system with uncertain parameters has been
formulated. Reliability concept is the best way to solve the uncertain problem, so the non-probabilistic reliability analysis
method has been introduced into the vibration active control system. Eigenvalue criterion as a simple and effective strategy
for stability analysis has been employed in this paper. In order to get the interval bounds of the eigenvalues, interval
perturbation method is proposed, which can effectively reduce the amount of computation and shorten the time of cal-
culation. Based on the interval bounds of the eigenvalues, a novel non-probabilistic reliability index has been put forward.
Compared with the Monte Carlo simulation method, the proposed method is more efficient. However, though the two
numerical examples, it can be find that reliability obtained by using the proposed method is smaller than the accurate
solutions. Thus, the controller designer has to make a tradeoff between accuracy and efficiency. Certainly, the reliability
analysis of the active control system with uncertain parameters is the first step. The future work of the researcher will focus
on how to design a controller (reliable controller) to satisfy the requirements of reliability.

Acknowledgment

The authors would like to thank the National Nature Science Foundation of the P. R. China (No. 11372025, No. 11572024
and No. 11432002), the Defense Industrial Technology Development Program (No. JCKY2016601B001 and No.
JCKY2016205C001).
 Y. Li et al. / Journal of Sound and Vibration 387 (2017) 115 15

References

[1] P. Gardonio, Review of active techniques for aerospace vibro-acoustic control, J. Aircr. 39 (2) (2002) 206214.
[2] D.J. Inman, Vibration with Control, John Wiley & Sons, 2006.
[3] A. Benjeddou, Advances in piezoelectric finite element modeling of adaptive structural elements: a survey, Comput. Struct. 76 (1) (2000) 347363.
[4] Y. Li, X. Wang, R. Huang, Z. Qiu, Active vibration and noise control of vibro-acoustic system by using PID controller, J. Sound Vib. 348 (2015) 5770.
[5] Z. Qiu, X. Wang, Comparison of dynamic response of structures with uncertain-but-bounded parameters using non-probabilistic interval analysis
method and probabilistic approach, Int. J. Solids Struct. 40 (20) (2003) 54235439.
[6] B.F. Spencer Jr, M.K. Sain, J.C. Kantor, C. Montemagno, Probabilistic stability measures for controlled structures subject to real parameter uncertainties,
Smart Mater. Struct. 1 (4) (1992) 294.
[7] G. Zames, Feedback and optimal sensitivity: model reference transformations, multiplicative seminorms, and approximate inverses, Autom. Control
IEEE Trans. 26 (2) (1981) 301320.
[8] T. Bailey, J.E. Ubbard, Distributed piezoelectric-polymer active vibration control of a cantilever beam, J. Guid. Control Dyn. 8 (5) (1985) 605611.
[9] J.L. Fanson, T.K. Caughey, Positive position feedback control for large space structures, AIAA J. 28 (4) (1990) 717724.
[10] N.W. Hagood, W.H. Chung, A. Von Flotow, Modelling of piezoelectric actuator dynamics for active structural control, J. Intell. Mater. Syst. Struct. 1 (3)
(1990) 327354.
[11] E.F. Crawley, J. De Luis, Use of piezoelectric actuators as elements of intelligent structures, AIAA J. 25 (10) (1987) 13731385.
[12] I.N. Kar, T. Miyakura, K. Seto, Bending and torsional vibration control of a flexible plate structure using H1-based robust control law, Control Syst.
Technol. IEEE Trans. 8 (3) (2000) 545553.
[13] Y.Y. Li, L.H. Yam, Robust vibration control of uncertain systems using variable parameter feedback and model-based fuzzy strategies, Comput. Struct. 79
(11) (2001) 11091119.
[14] Y. Hu, A. Ng, Active robust vibration control of flexible structures, J. Sound Vib. 288 (1) (2005) 4356.
[15] L. Zuo, J.E. Slotine, Robust vibration isolation via frequency-shaped sliding control and modal decomposition, J. Sound Vib. 285 (4) (2005) 11231149.
[16] S. Guo, Y. Li, Non-probabilistic reliability method and reliability-based optimal LQR design for vibration control of structures with uncertain-but-
bounded parameters, Acta Mech. Sin. 29 (6) (2013) 864874.
[17] S. Wang, H.Y. Yeh, P.N. Roschke, Robust control for structural systems with parametric and unstructured uncertainties, J. Vib. Control 7 (5) (2001)
753772.
[18] B.F. Spencer, M.K. Sain, C. Won, D.C. Kaspari, P.M. Sain, Reliability-based measures of structural control robustness, Struct. Saf. 15 (1) (1994) 111129.
[19] B.F. Spencer Jr, M.K. Sain, J.C. Kantor, C. Montemagno, Probabilistic stability measures for controlled structures subject to real parameter uncertainties,
Smart Mater. Struct. 1 (4) (1992) 294.
[20] K. Breitung, F. Casciati, L. Faravelli, Reliability based stability analysis for actively controlled structures, Eng. Struct. 20 (3) (1998) 211215.
[21] A.A. Taflanidis, J.T. Scruggs, J.L. Beck, Reliability-based performance objectives and probabilistic robustness in structural control applications, J. Eng.
Mech. 134 (4) (2008) 291301.
[22] S. Guo, Robust reliability as a measure of stability of controlled dynamic systems with bounded uncertain parameters, J. Vib. Control (2010).
[23] X. Wang, Z. Qiu, I. Elishakoff, Non-probabilistic set-theoretic model for structural safety measure, Acta Mech. 198 (12) (2008) 5164.
[24] X. Wang, Z. Qiu, Nonprobabilistic interval reliability analysis of wing flutter, AIAA J. 47 (3) (2009) 743748.
[25] Z.P. Qiu, P.C. Mller, A. Frommer, An approximate method for the standard interval eigenvalue problem of real non-symmetric interval matrices,
Commun. Numer. Methods Eng. 17 (4) (2001) 239251.
[26] Z.P. Qiu, S.H. Chen, I. Elishakoff, Bounds of eigenvalues for structures with an interval description of uncertain-but-non-random parameters, Chaos
Solitons Fractals 7 (3) (1996) 425434.
[27] W.L. Brogan, Modern Control Theory, Pearson Education, India, 1974.
[28] P. Ngamsom, L.L. Hoberock, Using robust stability analysis theorems for robust controller design, J. Dyn. Syst. Meas. Control 125 (3) (2003) 669671.
[29] K. Zhou, J.C. Doyle, K. Glover, Robust and Optimal Control, Prentice hall, New Jersey, 1996.
[30] Y.L. Li, Research on Reliable Control Theories and Methods of Active Control of Structural Vibration and Acoustics (Ph.D. Thesis), Beihang University, Beijing,
2016.