Вы находитесь на странице: 1из 6

2015 IEEE Conference on Control Applications (CCA)

Part of 2015 IEEE Multi-Conference on Systems and Control


September 21-23, 2015. Sydney, Australia

Modelling and Control of Wireless Networked Control Systems: A


Fixed Structure Approach
Ahmad W. Al-Dabbagh and Tongwen Chen

Abstract— This paper proposes a system topology for poten- nodes is given in [5], [6], [7], [8], [9], [10], [11], [12].
tial use as a Wireless Networked Control System (WNCS). The The nodes are resource constrained, where they can perform
topology has a plant to be regulated, a dedicated controller computational tasks and can wirelessly communicate with
that makes control decisions, and an intermediate network
of nodes. The network assists the controller in stabilizing the each other and with the sensors and actuators of the plant.
closed-loop system and provides the means for connecting the This topology is referred to as the Wireless Control Network
sensors and actuators of the plant and the inputs and outputs (WCN).
of the controller. Further, the paper presents a framework Inspired by the concept of the WCN, we propose in
for modelling the closed-loop system such that controller and
network systems can be designed. The paper also provides an this paper an extended system topology for potential use
LMI-based algorithm for computing optimal controller and as a WNCS. The topology has a standard plant to be
network systems that achieve a stable closed-loop system while regulated, a dedicated controller that makes control decisions
meeting performance measures. The deigned systems have a and an intermediate network that assists the controller in
fixed structure and hence the associated design complexity stabilizing the closed-loop system and provides the means
is reduced. Finally, the paper also illustrates with examples
the effectiveness of the proposed design and computational for connecting the plant and the controller. Further, the plant
procedures. system is associated with sensor nodes and actuator nodes,
the controller system is associated with input nodes and
I. INTRODUCTION output nodes, and the network system is associated with its
The growing interests in distributed applications make the own nodes. The nodes of the systems can be geographically
deployment of wireless communication networks in control distributed and are connected using communication links.
systems a necessity, as they can reduce network cabling Also, the paper presents a framework for modelling the
and wiring, improve deployment location and manoeuvring closed-loop system such that controller and network systems
flexibility, and require less maintenance efforts. In this paper, can be designed. Further, it provides an LMI-based algorithm
the use of wireless communication in a closed-loop control for computing optimal controller and network systems to pro-
system is referred to as a Wireless Networked Control vide a stable closed-loop system and also satisfy performance
System (WNCS). In industrial applications, the communi- measures. In addition, two examples are used to demonstrate
cation usage in a WNCS can be categorized into six classes the effectiveness of the proposed procedures.
which belong to safety, control and monitoring functions [1]. The paper is organized as follows: Section II presents
The classes range from the least critical actions involving the details of the topology and its components; Section III
data logging and downloading/uploading related tasks to addresses the modelling and formulation of the closed-loop
the most critical actions involving emergency related tasks. system to facilitate the design of the controller and network
To facilitate the use of a WNCS for all classes, further systems; Section IV proposes the algorithm for computing
research is needed to address the shortcomings associated the controller and network systems; Section V provides
with deploying wireless communication in control systems. examples to demonstrate the effectiveness of the proposed
Indeed, researchers have been addressing several horizons design and computational procedures; and finally Section VI
of WNCSs. For example, modelling WNCSs with variable provides concluding remarks and discusses future research
sampling intervals, delays and possible packet dropouts directions.
using delay impulsive systems [2], using redundant pack-
ets transmissions in communication protocols to adjust the
probability of successful communication [3], and designing II. SYSTEM DEFINITION
optimal plant control and communication power management
policies [4]. In addition, an interesting approach that replaces
the use of the traditional dedicated controller with a set of The system topology is depicted in Fig. 1. It consists of
a plant, a controller and an intermediate network. The plant
*The authors would like to thank the Natural Sciences and Engineering has a set of sensor nodes denoted by S = {s1 , s2 , ..., sp }
Research Council of Canada (NSERC), the Alberta Innovates - Technology
Futures and Alberta Innovation & Advanced Education, and the University that sample and transmit the plant measurements, and a set
of Alberta for the generous financial support. of actuator nodes denoted by A = {a1 , a2 , ..., am } that
The authors are with the Department of Electrical and receive and apply corrective actions in the plant. The plant
Computer Engineering, University of Alberta, Edmonton, AB
T6G 2V4, Canada. (Emails: aaldabba@ualberta.ca, system denoted by G is given as a linear discrete-time system
tchen@ualberta.ca) defined as

978-1-4799-7787-1/15/$31.00 ©2015 IEEE 1051


Plant Network Controller
s1 γ1 
.. v1 v2 v3 v4 .. xNi (k + 1) = ωii xNi (k) + ωij xNj (k)
. . vj ∈Nvi
sp γd  
v5 v6 v7 v8 + λij yj (k) + ψij gj (k),
a1 .. θ1
. sj ∈Nvi θj ∈Nvi
.. ..  
. vN . ui (k) = υij xNj (k) + ξij yj (k)
am θt
vj ∈Nai sj ∈Nai

+ δij gj (k),
Fig. 1. The Proposed System Topology with a Plant to be Regulated, a θj ∈Nai
Dedicated Controller and an Intermediate Network  
fi (k) = πij xNj (k) + σij yj (k)
vj ∈Nγi sj ∈Nγi

+ φij gj (k).
xG (k + 1) = AxG (k) + B1 w(k) + B2 u(k), θj ∈Nγi
z(k) = C1 xG (k) + D11 w(k) + D12 u(k), (1) Aggregating the equations for all the nodes, the network
y(k) = C2 xG (k) + D21 w(k), system denoted by N is defined as
where xG (k) ∈ Rn is the plant state vector, w(k) ∈ Rs xN (k + 1) = ΩxN (k) + Λy(k) + Ψg(k),
is the plant external input vector, u(k) ∈ Rm is the plant
controlled input vector, z(k) ∈ Rq is the plant output vector u(k) = ΥxN (k) + Ξy(k) + Δg(k), (3)
to be controlled and y(k) ∈ Rp is the plant measured output f(k) = ΠxN (k) + Σy(k) + Φg(k),
vector, and all matrices have suitable dimensions.
where the matrices Ω ∈ RN ×N , Λ ∈ RN ×p , Ψ ∈ RN ×t , Υ ∈
The controller has a set of input nodes denoted by Γ =
Rm×N , Ξ ∈ Rm×p , Δ ∈ Rm×t , Π ∈ Rd×N , Σ ∈ Rd×p , and
{γ1 , γ2 , ..., γd } that receive feedback to be used in making
Φ ∈ Rd×t contain the weight of each communication link
control decisions, and a set of output nodes denoted by
between the nodes of the network, plant and the controller.
Θ = {θ1 , θ2 , ..., θt } that transmit the control decisions. The
Remark 1: The sensor nodes of the plant as well as the
controller system denoted by K is given as a linear discrete-
output nodes of the controller can communicate directly to
time system defined as
the actuators of the plant and the inputs of the controller, or
xK (k + 1) = AK xK (k) + BK f(k), indirectly through the nodes of the network. This specifica-
(2) tion is captured in (3).
g(k) = CK xK (k) + DK f(k),
where xK (k) ∈ Rr is the controller state vector, f(k) ∈ Rd A. Advantages
is the controller input vector and g(k) ∈ Rt is the controller The proposed system topology has the advantages of the
output vector, and all matrices have suitable dimensions. WCN: compositionality, simple scheduling, low overhead,
The network has a set of nodes denoted by V = and multiple sensing and actuation points [9]. In addition, it
{v1 , . . . , vN } that assist the controller in making control offers the following advantages:
decisions and provide the means to connect the sensors and 1) Generality: The modelling approach followed in this
actuators of the plant and the inputs and outputs of the paper uses a standard plant with external inputs and outputs,
controller. More specifically, the inputs of the network are a dynamic output feedback controller and an intermediate
the outputs of the plant and the controller and the outputs of network connecting the plant and the controller. This allows
the network are the inputs of the plant and the controller. for maintaining generality in the modelling as well as inte-
In addition, the system topology has communication links grating the approach with the results of the widely studied
denoted by E that allow for information exchange between system and control theory.
the nodes of the plant, controller and network systems. They 2) Practicality: The use of a dedicated controller satisfies
are categorized into two types: (i) directional links that allow the requirements needed by many existing applications for
for information flow in one direction, such as from the having centralized controllers in their closed-loop control
sensors to the nodes of the network, and (ii) bidirectional systems, such as in oil & gas, utilities and manufacturing
links that allow for information flow in both directions such applications, to name a few. In addition, the nodes of the
as between the nodes of the network. network can be distributed geographically which offers the
Further, similar to the WCN [9], the nodes of the network, flexibility in deploying the sensors and actuators as well
actuators of the plant and inputs of the controller update their as the controller. Also, the approach for allowing direct
values iteratively. Denoting the state of node vi by xNi and and indirect communication is practical as the need for
the nodes in the neighbourhood of vi by Nvi , the values indirect communication through nodes can be dependent on
of any node vi , actuator node ui , and input node of the the distance between, and the location of the nodes that need
controller fi are defined as to communicate.

1052
3) Availability: The nodes of the network offer the avail- where x = [xTG xTN xTK ]T is the closed-loop state vector.
ability of different routes for communication between nodes. The closed-loop system matrices A , B, C and D can be
If one route is unavailable, an alternative route can deliver the represented using affine matrices on each of the controller
information. Also, similar to the WCN [9], the nodes of the and network systems using the approach in [13] and [14].
network are capable of stabilizing the plant. This allows for
controlling the plant even when the controller is unavailable. A. Controller Definition

III. MODELLING AND FORMULATION Defining the matrices


The plant, controller and network systems described in (1)- 
(3) and combined in a standard feedback setup are depicted A + B2 ΞC2 B2 Υ
AP  ,
in Fig. 2. Defining the matrices ΛC2 Ω

B1 + B2 ΞD21
BP1  ,
A11  A + B2 (Ξ + Δ(It×t − DK Φ)−1 DK Σ)C2 , ΛD21

A12  B2 (Υ + Δ(It×t − DK Φ)−1 DK Π), B2 Δ
BP2  ,
Ψ
A13  B2 Δ(It×t − DK Φ)−1 CK ,
C P1  C1 + D12 ΞC2 D12 Υ ,
A21  (Λ + Ψ(It×t − DK Φ)−1 DK Σ)C2 ,
CP2  ΣC2 Π ,
A22  Ω + Ψ(It×t − DK Φ)−1 DK Π,
DP11  D11 + D12 ΞD21 ,
A23  Ψ(It×t − DK Φ)−1 CK ,
DP12  D12 Δ,
A31  BK (Id×d − ΦDK )−1 ΣC2 ,
DP21  ΣD21 ,
A32  BK (Id×d − ΦDK )−1 Π,
DP22  Φ,
A33  AK + BK (Id×d − ΦDK )−1 ΦCK ,
B1  B1 + B2 (Ξ + Δ(It×t − DK Φ)−1 DK Σ)D21 , results in the augmented plant and network system denoted
by P with inputs w and g and outputs z and f defined as
B2  (Λ + Ψ(It×t − DK Φ)−1 DK Σ)D21 ,
B3  BK (Id×d − ΦDK )−1 ΣD21 ,
xP (k + 1) = AP xP (k) + BP1 w(k) + BP2 g(k),
C1  C1 + D12 (Ξ + Δ(It×t − DK Φ)−1 DK Σ)C2 , z(k) = CP1 xP (k) + DP11 w(k) + DP12 g(k), (5)
C2  D12 (Υ + Δ(It×t − DK Φ)−1 DK Π), f (k) = CP2 xP (k) + DP21 w(k) + DP22 g(k),
C3  D12 Δ(It×t − DK Φ)−1 CK ,
where xP = [xTG xTN ]T is the augmented plant and network
D  D11 + D12 (Ξ + Δ(It×t − DK Φ)−1 DK Σ)D21 , system state vector. Letting Φ = 0 in (5) to simplify
results in the augmented closed-loop system with input w derivation and maintain linearity, and defining the matrices
and output z defined as

⎡ ⎤ ⎡ ⎤ AP 0(n+N )×r
A11 A12 A13 B1 AP  ,
0r×(n+N ) 0r×r
x(k + 1)  ⎣ A21 A22 A23 ⎦ x(k) + ⎣ B2 ⎦ w(k), 
BP1
A31 A32 A33 B3 B P1  ,
 
 
0r×s
A B
(4) 
BP2 0(n+N )×r
z(k)  C1 C2 C3 x(k) + Dw(k), B P2  ,
0r×t Ir×r
 

C C P1  C P1 0q×r ,

CP2 0d×r
C P2  ,
0r×(n+N ) Ir×r
z  w
G DP11  DP11 ,

y u DP12  DP12 0q×r ,
- 
DP21
- N DP21  ,
0r×s
g f with the controller parameter defined as
K  
D K CK
K ,
B K AK
Fig. 2. The Plant, Network and Controller Systems Representation in a results in the closed-loop matrices affine on the controller
Standard Feedback Setup parameter K defined as

1053
and network systems can be designed using the algorithm
A (K)  AP + BP2 KCP2 , presented in the following section.
B(K)  BP1 + BP2 KDP21 ,
(6) IV. CONTROLLER AND NETWORK SYNTHESIS
C (K)  CP1 + DP12 KCP2 ,
D(K)  DP11 + DP12 KDP21 . The controller and network systems are designed and
computed for the two following objectives:
B. Network Definition Objective 4.1: Given an unstable plant system G to be
Similarly, letting DK = 0 in (4) to simplify derivation and regulated and a predefined network system N , find a con-
maintain linearity, and defining the matrices troller system K that stabilizes the augmented plant and
network system P described in (5) assuming that P is both
⎡ ⎤ stabilizable and detectable.
A 0n×N 0n×r
A  ⎣ 0N ×n 0N ×N 0N ×r ⎦ , Objective 4.2: Given an unstable plant system G and a
0r×n 0r×N 0r×r predefined controller system K (not neccessarily stabilizing
⎡ ⎤ for G), find a network system N that can provide a stable
0n×n 0n×N 0n×r
closed-loop system.
AK  ⎣ 0N ×n 0N ×N 0N ×r ⎦ ,
In Objective 4.2, the design of the network system N
0r×n 0r×N AK
⎡ ⎤ involves the specification of the weights of all the com-
B1 munication links between the nodes of the network, plant
B1  ⎣ 0N ×s ⎦ , and the controller. Further, both Objective 4.1 and Objective
0r×s 4.2 are practical. The former addresses the scenario where
⎡ ⎤
B2 0n×N 0n×d the network, with its nodes and communication links, is
B2  ⎣ 0N ×m 0N ×N 0N ×d ⎦ , predefined and a stabilizing controller needs to be computed
0r×m 0r×N 0r×d for the plant and the network systems. The latter addresses
⎡ ⎤
0n×m 0n×N 0n×d the scenario where the controller is predefined and a network
BK  ⎣ 0N ×m IN ×N 0N ×d ⎦ , needs to be computed and inserted between the plant and the
0r×m 0r×N BK controller systems.
For both objectives, the computation of the controller and
C1  C1 0q×N 0q×r ,
⎡ ⎤ the network systems is performed by finding the K and N
C2 0p×N 0p×r parameters using a similar procedure to that in [12] and [14].
C2  ⎣ 0N ×n 0N ×N 0N ×r ⎦ , Denoting Q  {K, N }, the following lemma is given in [14].
0t×n 0t×N 0t×r Lemma 4.1: For an asymptotically stable closed-loop sys-
⎡ ⎤
0p×n 0p×N 0p×r tem, there exist matrices X, Y, Z, and Q such that the
CK  ⎣ 0N ×n IN ×N 0N ×r ⎦ , following holds:
0t×n 0t×N CK ⎡ ⎤
X Z A (Q) B(Q)
D11  D11 , ⎢ ∗ Y C (Q) D(Q) ⎥
⎢ ⎥ > 0. (8)
D12  D12 0q×N 0q×d , ⎣ ∗ ∗ X −1 0 ⎦
⎡ ⎤
D21 ∗ ∗ ∗ I
D21  ⎣ 0N ×s ⎦ , Then, consider the following definition in [13] and [14].
0t×s Definition 4.1: For a discrete-time system with input
w and output z, the energy-to-peak gain is defined as
with the network parameter defined as ∇ep  supwl 1  z l∞ , and the energy-to-energy
⎡ ⎤ 2
Ξ Υ Δ gain is ∇ee  supwl 1  z l2 . The l2 -norm and
N  ⎣ Λ Ω Ψ ⎦,
2
the l∞ -norm of a discrete-time signal are expressed as

Σ Π Φ  . l 2  k=0  .  and  . l∞  supk≥0  . ,
2

results in the closed-loop matrices affine on the network respectively, where the spatial norm  ·  is Euclidean based.
parameter N defined as
The following theorem is obtained from [14].
Theorem 4.2: For an asymptotically stable closed-loop
A (N )  (A + AK ) + (B2 + BK )N (C2 + CK ), system and a given positive scalar κ, the following statements
B(N )  B1 + (B2 + BK )N D21 , are true:
(7)
C (N )  C1 + D12 N (C2 + CK ), (i) ∇ep < κ iff there exist matrices Q, Z, X and Y such that
κI − Y > 0 and (8) holds.
D(N )  D11 + D12 N D21 .
(ii) ∇H2 = ||C (Q)(zI − A (Q))−1 B(Q) + D(Q)||2 <
With the closed-loop system representation in terms of K, κ iff there exist matrices Q, Z, X and Y such that κ2 −
and N defined in (6) and (7), respectively, the controller trace(Y ) > 0 and (8) holds.

1054
(iii) ∇ee = ||C (Q)(zI − A (Q))−1 B(Q) + D(Q)||∞ < κ initial values for X are found, a change in the configuration
iff there exist matrices Q, X and Y such that κ2 I − Y > 0 of the system topology, such as the connection of nodes,
and (8) holds when Z = 0. needs to be addressed. Further, when the algorithms con-
For the computation of Q, the non-convex term X −1 in verge, they provide a local optimal design for the controller
(8) is linearized by defining the linearization operator  of and network systems.
a matrix X −1 at a positive definite value Xk by [14], Thus, given an unstable plant system G, a predefined con-
troller system K (or network system N ), and initial values
(X −1 , Xk ) = Xk−1 − Xk−1 (X − Xk )Xk−1 for X computed using Algorithm 1, Algorithm 2 computes
(9)
= Xk−1 (2I − XXk−1 ). a local optimal system N (or K) satisfying the stability
Substituting (9) in (8), and using Theorem 4.2, the follow- and performance measures listed in Theorem 4.2. Further,
ing two algorithms are developed for computing Q [14]. as the computed network and controller systems are of fixed
structure, their design is associated with more flexibility and
Algorithm 1 Computation of Initial Values reduced complexity. In other words, the systems can be
designed to be of any order and do not need to match the
1. Start with setting the feasibility parameter η > 0 and
order of the controlled system.
k = 0. Specify X0 with an arbitrary symmetric matrix.
2. Compute X from the convex optimization problem, V. ILLUSTRATIVE EXAMPLES
[X] = arg min η In this section, two examples addressing Objective 4.1 and
Y,Q,X,Z
Objective 4.2 are used to illustrate the design and compu-
subject to tational procedures associated with computing the controller
⎡ ⎤
X Z A (Q) B(Q) and network systems to provide a stable closed-loop system
⎢ ∗ Y C (Q) D(Q) ⎥ and to satisfy the energy-to-peak gain performance measure.
⎢ ⎥>0
⎣ ∗ ∗ Xk−1 (2I − XXk−1 ) 0 ⎦ The algorithms are implemented using MATLAB’s Robust
∗ ∗ ∗ I Control Toolbox. Further, in Algorithm 1, Xk is initialized as
an arbitrary symmetric matrix such that X0 = RRT , where
3. End if η < 0. Else, set k = k + 1 and return to Step 2.
R denotes a random matrix with large entries.
In both examples, the plant to be regulated is specified as
Algorithm 1 provides the initial values for X. In addition, a 2nd order system, and is defined as
it provides values for Q; however, the computed Q simply
stabilizes the closed-loop system and does not account for  
1.1 0 0.1 0.2
any performance. xG (k + 1) = xG (k) + w(k)
0.3 0.2 2 1

Algorithm 2 Computation of Optimal Controller and Net- 1.8
+ u(k),
work Systems 0
 
1. Start with setting the convergence threshold  > 0 1 0.1 0.1 0.6
z(k) = xG (k) + w(k)
and k = 0. Use the initial values for X computed using 0.2 0 0.2 0.1

Algorithm 1. 0
2. Compute Y, Q and κ from the convex optimization + u(k),
0.2
problem,
y(k) = 0.8 1 xG (k) + 0 0.2 w(k).
[Y, Q, κ] = arg min κ
Y,κ,Q,X,Z
A. Example 1: Controller Computation for Objective 4.1
subject to
⎡ ⎤ Given the plant defined above and a 4th order network
X Z A (Q) B(Q) system defined as
⎢ ∗ Y C (Q) D(Q) ⎥
⎢ ⎥ > 0,
⎣ ∗ ∗ Xk−1 (2I − XXk−1 ) 0 ⎦ ⎡ ⎤
0.1 0.2 0.1 0.2
∗ ∗ ∗ I ⎢ 0.5
⎢ 0.1 0.5 0.1 ⎥ ⎥ x (k)
xN (k + 1) = ⎣
and 0.2 0.2 0.2 0.1 ⎦ N
κI − Y > 0 if solving for ∇ep , 0.1 0.1 0.1 0.1
⎡ ⎤ ⎡ ⎤
κ2 − trace(Y ) > 0 if solving for ∇H2 , or 0.1 0.2
⎢ 0.2 ⎥ ⎢ ⎥
κ2 I − Y > 0 (with Z = 0) if solving for ∇ee +⎢ ⎥ y(k) + ⎢ 0.2 ⎥ g(k),
⎣ 0.1 ⎦ ⎣ 0.2 ⎦
3. End if κ < . Else, set k = k + 1 and return to Step 2.
0.3 0.1

u(k) = 0.1 0.2 0.1 0.1 xN (k) + 0.2y(k)
Given that the initial values for X are found using Al-
gorithm 1, Algorithm 2 is able to compute an optimal Q + 0.5g(k),

parameter (namely, controller or network parameters). If no f(k) = 0.2 0.3 0.3 0.2 xN (k) + 0.5y(k),

1055
the resulting augmented system P is controllable and de- a stable closed-loop system and satisfy the energy-to-peak
tectable. To compute a 2nd order controller system to stabi- gain performance measure.
lize P , Algorithm 1 was first used to obtain the initial values
VI. DISCUSSION AND CONCLUSION
for X. Then, Algorithm 2 was used with setting  = 1.1
to compute the K parameter. This was used in extracting This paper presented a system topology for potential use as
the controller matrices. The algorithm converged after 28 a WNCS consisting of a plant, a controller and an intermedi-
iterations resulting in a controller system defined with the ate network. The closed-loop system was modelled such that
following matrices: controller and network systems can be designed. The paper
also presented an LMI-based algorithm for computing the
 controller and network systems to provide a stable closed-
−0.0152 0.0670
AK = , loop system and satisfy performance measures. Finally, two
0.0670 0.0793
 examples were used to illustrate the effectiveness of the
−0.3960 proposed procedures.
BK = ,
0.1398 Further, the algorithms presented in this paper can com-

CK = −0.3960 0.1398 , pute either a controller or a network system, and not both.
DK = −1.2512. In our future work, we will address the joint computation of
the controller and network systems for a given plant. Also,
B. Example 2: Network Computation for Objective 4.2 the proposed system topology assumes that the nodes can
Given the plant defined above and a 2nd order controller always communicate. In our future work, we will address a
system defined as more realistic system model that accounts for communication
  constraints in wireless networks.
0.5 0.4 1.2
xK (k + 1) = xK (k) + f(k), R EFERENCES
1 0.2 0.9
[1] “ISA-100.11a-2011,” Wireless Systems for Industrial Automation: Pro-
g(k) = 2 1.3 xK (k), cess Control and Related Applications, 2011.
[2] P. Naghshtabrizi and J. P. Hespanha, “Implementation considerations
and to compute a 4th order network system to stabilize the for wireless networked control systems,” in Wireless Networking Based
closed-loop system, Algorithm 1 was used to compute the Control. Springer, 2011, pp. 1–27.
[3] A. R. Mesquita, J. P. Hespanha, and G. Nair, “Redundant data trans-
initial values for X. Then, Algorithm 2 was used with setting mission in control/estimation over wireless networks,” in American
 = 1.1 to compute the N parameter. This was used in Control Conference, St. Louis, MO, USA, 2009, pp. 3378–3383.
extracting the network matrices. The algorithm converged [4] K. Gatsis, A. Ribeiro, and G. J. Pappas, “Optimal power management
in wireless control systems,” IEEE Transactions on Automatic Control,
after 42 iterations resulting in a network system defined with vol. 59, no. 6, pp. 1495–1510, 2014.
the following matrices: [5] M. Pajic, S. Sundaram, J. Le Ny, G. J. Pappas, and R. Mangharam,
“The wireless control network: synthesis and robustness,” in 49th IEEE
Conference on Decision and Control, Atlanta, GA, USA, 2010, pp.
Ξ = −0.1003, 7576–7581.
[6] S. Sundaram, M. Pajic, C. N. Hadjicostis, R. Mangharam, and G. J.
Υ = 0.1411 −0.0633 0.0003 0.0634 , Pappas, “The wireless control network: monitoring for malicious
Δ = −0.0462, behavior,” in 49th IEEE Conference on Decision and Control, Atlanta,
⎡ ⎤ GA, USA, 2010, pp. 5979–5984.
0.1411 [7] M. Pajic, S. Sundaram, G. J. Pappas, and R. Mangharam, “Network
⎢ −0.0633 ⎥ synthesis for dynamical system stabilization,” in 45th Asilomar Con-
Λ=⎢ ⎥
⎣ 0.0003 ⎦ , ference on Signals, Systems and Computers, Pacific Grove, CA, USA,
2011, pp. 821–825.
0.0634 [8] ——, “Topological conditions for wireless control networks,” in 50th
⎡ ⎤ IEEE Conference on Decision and Control and European Control
−0.0938 0.0565 0.0330 −0.0344 Conference, Orlando, FL, USA, 2011, pp. 2353–2360.
⎢ 0.0565 −0.1592 −0.0723 0.0886 ⎥
Ω=⎣ ⎢ ⎥, [9] ——, “The wireless control network: a new approach for control over
0.0330 −0.0723 −0.0246 0.0531 ⎦ networks,” IEEE Transactions on Automatic Control, vol. 56, no. 10,
pp. 2305–2318, 2011.
−0.0344 0.0886 0.0531 −0.0293
⎡ ⎤ [10] M. Pajic, S. Sundaram, J. Le Ny, G. J. Pappas, and R. Mangharam,
0.0574 “Closing the loop: a simple distributed method for control over
⎢ −0.0166 ⎥ wireless networks,” in 11th ACM/IEEE International Conference on
Ψ=⎢ ⎥
⎣ 0.0227 ⎦ ,
Information Processing in Sensor Networks, Beijing, China, 2012, pp.
25–36.
0.0349 [11] M. Pajic, R. Mangharam, G. J. Pappas, and S. Sundaram, “Topological
conditions for in-network stabilization of dynamical systems,” IEEE
Σ = −0.0462, Journal on Selected Areas in Communications, vol. 31, no. 4, pp.
794–807, 2013.
Π = 0.0574 −0.0166 0.0227 0.0349 , [12] R. Mangharam and M. Pajic, “Distributed control for cyber-physical
Φ = −0.2407, systems,” Journal of the Indian Institute of Science, vol. 93, no. 3, pp.
353–388, 2013.
where each entry of the network matrices corresponds to the [13] R. E. Skelton, T. Iwasaki, and K. M. Grigoriadis, A Unified Algebraic
weight of the respective communication link. Approach to Linear Control Design. Taylor & Francis, 1998.
[14] J. Han and R. E. Skelton, “An LMI optimization approach for
The computed controller and network systems are one structured linear controllers,” in 42nd IEEE Conference on Decision
solution from the set of possible designs that can provide and Control, vol. 5, Maui, Hawaii, USA, 2003, pp. 5143–5148.

1056

Вам также может понравиться