servo.pdf

Design, Analysis and Real-time Implementation of

Networked Predictive Control SystemsLIU Guo-Ping1, 2 SUN Jian3 ZHAO Yun-Bo2, 4

Abstract This paper is concerned with the design, analysis and real-time implementation of networked control systems using thepredictive control strategy. The analysis of the characteristics of networked control systems is detailed, which shows that a networkedcontrol system is much dierent from conventional control systems. To achieve the desired performance of closed-loop networkedcontrol systems, the networked predictive control scheme is introduced. The design, stability analysis and real-time implementationof networked predictive control systems are studied. It is illustrated by simulations and practical experiments that the networkedpredictive control scheme can compensate for random network communication delay and data dropout, achieve desired controlperformance and have good closed-loop stability.

Key words Networked control systems, networked predictive control, real-time implementation, stability analysis, network delay

Citation Liu Guo-Ping, Sun Jian, Zhao Yun-Bo. Design, analysis and real-time implementation of networked predictive controlsystems. Acta Automatica Sinica, 2013, 39(11): 17691777

Networked control systems (NCSs) are such control sys-tems where the control loop is closed via communicationnetworks, as illustrated in Fig. 1. In NCSs, perfect dataexchanges among the control components are not availablewhich however is a fundamental basis of conventional con-trol systems (CCSs). These communication constraints inNCSs greatly degrade the performance of the control sys-tem or even destabilize the system at certain conditionsand therefore are the main concerns of the research onNCSs[15].

Fig. 1 The networked control system

The research on NCSs has been done mainly withinthe control theory community. Researchers are concernedwith the theoretical analysis on the performance of NCSswhere the communication network in NCSs is modelled bypredetermined parameters to the control system. In thiskind of research, the communication constraints, mostlythe network-induced delay, may be carefully described andformulated, and incorporated into the description of thesystem to form a special class of CCSs, and then a CCSinstead of an NCS is considered. This kind of research sim-plies the modelling and analysis of NCSs and, more impor-tantly, all the previous results in CCSs can now be readilyapplied to NCSs, thus enabling it to be the mainstream fora signicant period[34]. Conventional control approachesapplied in this way include, for example, time delay system

Manuscript received July 8, 2013; accepted August 31, 2013Supported by National Natural Science Foundation of China

(61273104, 61021002, 61104097), and Projects of Major Interna-tional (Regional) Joint Research Program National Natural ScienceFoundation of China (61120106010)Invited Articles for the Special Issue for the 50th Anniversary of

Acta Automatica Sinica1. School of Engineering, University of South Wales, Pontypridd

CF37 1DL, UK 2. Center for Control Theory and Guidance Tech-nology Center, Harbin Institute of Technology, Harbin 150001, China3. School of Automation, Beijing Institute of Technology, Beijing100083, China 4. Department of Chemical Engineering, ImperialCollege, London SW7 2AZ, UK

theory[68], stochastic control theory[911], optimal controltheory[1214], switched system theory[1517], and so on.

In parallel with the theoretical analysis of NCSs, thesynthesis issue is also a main concern within the controltheory community. However, despite considerable work todate, the limitations are obvious: most work concentrateson the extension of existing control approaches to NCSswithout a full use of the characteristics of NCSs. Bear-ing in mind that it is the communication network whichreplaces the direct connections among the control compo-nents in CCSs that makes NCSs distinct from CCSs, it istherefore natural and necessary to highlight the eects ofthe communication network when investigating NCSs. Thisapproach, referred to as the co-design approach to NCSs,has been an emerging trend in recent years. In this kindof research, the communication constraints are no longerassumed to be predetermined parameters but act as des-ignable factors, and by their ecient use a better systemperformance can be expected[1827].

Within the co-design framework, a networked predictivecontrol approach is introduced in this paper. By makingeective use of the packet-based transmission, the commu-nication constraints can be actively compensated for withthis approach. Both theoretical design and analysis, andpractical implementation are presented to show the eec-tiveness of the networked predictive control approach.

1 Characteristics of networked controlsystems

The distinct characteristics of NCSs are briey discussedfrom the control theory perspective, focusing on the eectsbrought by the communication network to the control sys-tem. These eects are caused, roughly speaking, by threedistinct features of the communication network, namely,the network topology, the packet-based transmission andthe limited network resources.

1) Network topologya) Time-synchronization: Time-synchronization, or

clock synchronization, is a fundamental basis of imple-menting distributed communication networks[28]. Time-synchronization among the control components in NCSsmay not be a necessary condition if the network-induceddelay in the backward channel (backward channel delay)

1770 ACTA AUTOMATICA SINICA Vol. 39

is not required for the calculation of the control signalsand/or the network-induced delay in the forward channel(forward channel delay) is not required for the imple-mentation of the control actions, which is typically thecase in CCSs. However, as discussed in [29 30], time-synchronization together with the use of time stamps inNCSs can oer an advantage over CCSs in that the back-ward channel delay is known by the controller and the for-ward channel delay (round trip delay as well) is known bythe actuator. This advantage can then be used to derive abetter control structure for NCSs as discussed in this paper.

b) Drive mechanism: Two drive mechanisms for the con-trol components in NCSs are usually used, that is, time-driven and event-driven. The dierence between the twodrive mechanisms lies in the trigger method that initi-ates the control components. For the time-driven mecha-nism, the control components are trigged to work at regulartime intervals, while for event-driven mechanism the con-trol components are only trigged by a predened event.Generally speaking, the sensor is always time-driven, whilethe controller and the actuator can either be time-drivenor event-driven. For more information on the drive mech-anism for the control components, the reader is referredto [31] and the references therein. It is worth mention-ing though, with dierent drive mechanisms dierent sys-tem models for NCSs are obtained and event-driven controlcomponents generally lead to a better system performance.

2) Packet-based transmissionPacket-based transmission is one of the most important

characteristics of NCSs that are distinct from CCSs[3, 32].This characteristic can mean that the perfect data trans-

mission as assumed in CCSs is absent in NCSs, thus pro-ducing the greatest challenge in NCSs. The communica-tion constraints caused by the packet-based transmissionin NCSs include the network-induced delay, data packetdropout, data packet disorder, etc., which are detailed asfollows.

a) Network-induced delay: Due to the network insertedinto the control loop in NCSs, network-induced delays areintroduced in both the forward and backward channels,which are well known to degrade the performance of thecontrol systems. The delays in the two channels mayhave dierent characteristics[33, 34]. In most cases, how-ever, these delays are not treated separately and only theround trip delay is of interest[35].

According to the types of the communication networksbeing used in NCSs, the characteristics of the network-induced delay vary as follows[1, 36]:

i) Cyclic service networks (e.g., Toking-Ring, Toking-Bus): Bounded delays which can be regarded as constantfor some occasions;

ii) Random access networks (e.g., Ethernet, CAN): Ran-dom and unbounded delays;

iii) Priority order networks (e.g., DeviceNet): Boundeddelays for the data packets with higher priority and un-bounded delays for those with lower priority.

Network-induced delay has widely been addressed in theliterature to date, see, e.g., in [5, 10, 19, 37].

b) Data packet dropout: Data packet dropout can occureither in the backward or forward channel, and it makeseither the sensing data or the control signals unavailableto NCSs, thus signicantly degrading the performance ofNCSs.

In communication networks, two dierent strategies areapplied when a data packet is lost, that is, either to sendthe packet again or simply discard it. Using the terms from

communication networks, these two strategies are referredto as transmission control protocol (TCP) and user data-

gram protocol (UDP) respectively[28]. It is readily seenthat with TCP, all the data packets will be received suc-cessfully, although it may take a considerably long timefor some data packets; while with UDP, some data packetsmay be lost forever. As far as NCSs is concerned, UDP isused in most applications due to the real-time requirementand the robustness of control systems. As a result, the ef-fect of data packet dropout in NCSs has to be explicitlyconsidered, as done in, e.g., [10, 12, 3841].

c) Data packet disorder: In most communication net-works, dierent data packets suer dierent delays. Ittherefore produces a situation where a data packet sentearlier may arrive at the destination later or vice versa,that is, data packet disorder. This characteristic in NCSscan mean that a newly arrived control signal in NCSs maynot be the latest, which never occurs in CCSs. Therefore,the eect of data packet disorder has to be specially dealtwith[42].

3) Limited network resourcesThe limitation of the network resources in NCSs is pri-

marily caused by the limited bandwidth of the communica-tion network, which results in the following three situationsin NCSs that are distinct from CCSs.

a) Sampling period, network loads and system perfor-mance: NCSs are a special class of sampled data systemsdue to the digital transmission of the data in communica-tion networks. However, in NCSs, the limited bandwidthof the network produces a situation, where a smaller sam-pling period may not result in a better system performancewhich, however, is normally true for sampled data systems.

This situation happens because, with too small a sam-pling period, too much sensing data will be produced; thusoverloading the network and causing congestion, which willresult in more data packet dropouts and longer delays, andthus degrade the system performance. The relationshipbetween the sampling period, network loads and systemperformance in NCSs is illustrated in Fig. 2. For example,when the sampling period decreases from the value cor-responding to point a to b, the system performanceis getting better as in conventional sampled data systemssince the network congestion does not appear until pointb; However, the system performance is likely to dete-riorate due to the network congestion when the samplingperiod is getting even smaller from the value correspondingto point b to c. Therefore, the relationship shown inFig. 2 implies that there is a trade-o between the periodof sampling the plant data and the system performance inNCSs, that is, in NCSs an optimal sampling period ex-ists which oers the best system performance (point b inFig. 2).

b) Quantization: Due to the use of data networks withlimited bandwidth, signal quantization is inevitable inNCSs, which has a signicant impact on the system perfor-mance.

Quantization in the meantime is also a potential methodto reduce the bandwidth usage which enables it to be aneective tool to avoid the network congestion in NCSs andthus improve the system performance of NCSs. For moreinformation on the quantization eects in NCSs, the readeris referred to [44] and the references therein.

c) Network access constraint and scheduling: It is oftenthe case that the communication network is shared by sev-eral control applications in NCSs, either used to transmitthe sensing data, the control signals or both. In such a

No. 11 LIU Guo-Ping et al.: Design, Analysis and Real-time Implementation of 1771

case, the limited bandwidth of the network produces a sit-uation where all the subsystems cannot access the networkresource at the same time. A scheduling algorithm is there-fore needed to schedule the timeline of when and how longa specic subsystem can occupy the network resource[30].The system performance can be severely aected by thisscheduling algorithm, for leaving the control system open-loop for a long time can readily destabilize the system atcertain conditions.

Fig. 2 The relationship between the sampling period, networkloads and system performance in NCSs (reproduced from [43])

2 Design of networked predictive con-trol systems

The design of networked control systems to be studied inthis paper is to compensate for random network communi-cation delay and data dropout, achieve desired control per-formance and have a good closed-loop stability. Recently, anetworked control strategy, named as networked predictivecontrol, has been proposed by Liu et al.[45]. The networkedpredictive control strategy, as shown in Fig. 3, actively com-pensates for the random network delay and data dropoutsso that the closed-loop networked control performance is al-most the same as the one without the network, which canhardly be achieved by other existing methods. It mainlyconsists of two parts: the control prediction generator andnetwork delay compensator. The control prediction gener-ator generates a set of control predictions which achieve therequired control performance. The network delay compen-sator compensates for the unknown random communicationdelay and data dropout. A novel design of networked pre-dictive control systems is discussed in details in this section.

For the sake of simplicity, the following assumptions aremade.

Assumption 1. The network delay from the controllerto the actuator in the forward channel ca is bounded byn0 ca nf .

Assumption 2. The network delay from the sensor tothe controller in the feedback channel sc is bounded byn1 sc nb.

Assumption 3. The number of consecutive data pack-age drops in both the feedback and forward channels isbounded by nd.

Assumption 4. The data transmitted through a net-work are with a time stamp.

In a practical NCS, there exists data loss. For instance,if the data packet does not arrive at a destination in acertain transmission time (i.e., the upper bound of the net-work delay), it means this data packet is lost, based oncommonly used network protocols. From the physical pointof view, it is natural to assume that only a nite numberof consecutive data dropouts can be tolerated in order toavoid the NCS becoming open-loop. The time stamp ofthe data transmitted through a network is very importantfor networked control systems. This is because a controlsequence of a control system is based on time. In addition,the synchronisation is also an issue in networked controlsystems. There exist various ways to synchronise the timeclocks in digital components (or computers). In this paper,it is assumed that the components in the system have beensynchronised.

Consider a linear discrete-time plant to be controlled:

xt+1 = Axt +But

yt = Cxt (1)

where xt Rn, yt Rl and ut Rm are the state, out-put and control input vectors of the system, respectively,and A Rnn, B Rnm and C Rln are the systemmatrices.

Let j and k denote the time-varying delays of real-timedata in the forward channel and feedback channel, respec-tively. Also, let d1 = nf +nd, d2 = nb +nd , nm = n0 +n1and d = d1 + d2. From Assumptions 13, it is clear thatj {n0, n0 + 1, , d1} and k {n1, n1 + 1, , d2}.

It is assumed that the states of the plant are not mea-surable but the system output is available. To obtain thestate vector of the plant on the controller side, an observeris designed as

xtk+1|tk = Axtk|tk1 +Butk|tk1+L(ytk ytk|tk1)

ytk|tk1 = Cxtk|tk1 (2)

Fig. 3 The networked predictive control system


where xti|tk Rn (i < k) denotes the state predictionfor time t i on the basis of the information to time t kand ytk|tk1 Rl is the output vector of the observer,and the gain matrix L Rnl , which can be designedusing standard observer design approaches.

Since there exists a network delay, the output signal y attime t is delayed for k steps when it arrives on the controllerside. Although the observer provides a one-step ahead pre-diction of the plant states using the output at time t k,the state predictions from time t k + 2 to t+ d1 are stillnot known. Based on the information available on the con-troller side at time t k, the other state predictions up totime t+ d1 can be constructed by

xtk+i|tk = Axtk+i1|tk +Butk+i1|tk (3)

for i = 2, 3, , k + d1. When the states of the plant areestimated, there are many control methods available forthe system. To illustrate the networked predictive controlstrategy, the observer based state-feedback control methodis employed. So, the control predictions to be generated onthe controller side are

ut+ik|tk = Kxt+ik|tk (4)

for i = 1, 2, , k + d1, where K Rmn is the controllergain matrix. So, the control prediction generator on thecontroller side is to produce a set of control predictionscalculated by (4).

To deal with the control input data dropout in networkedcontrol systems, there are three main methods. Method 1is that if the control input data are dropped, the controlinput is set to zero. Method 2 is that if the control inputdata are dropped, the control input keeps the previous con-trol input until the new control input data arrive. Method3 is that if the control input data are dropped, the controlinput uses the control prediction. These methods have ad-vantages and disadvantages. Method 1 is simple but thecontrol input causes an unsmooth switching, which maynot be allowed in some control systems. Method 2 has asmooth switching control input but it is hard to achievethe desired control performance. Method 3 provides thedesired control performance but it costs a little communi-cation eciency.

At time t, the future network delay and data dropout areunknown. According to the networked predictive controlstrategy and Assumptions 1 and 3, the control predictions[

ut|tk ut+1|tk ut+d1|tk]at time t, which are

calculated by (4), are packed together and are sent fromthe controller side to the actuator side. If the minimumnetwork delay in the forward channel is known, the controlpredictions to be sent from the controller side at time t willbe[

ut+n0|tk ut+n0+1|tk ut+d1|tk]. From As-

sumption 3, the output data set[

yt yt1 ytnd]

at time t is transmitted from the sensor side to the con-troller side to avoid the output data loss in the feedbackchannel. In this way, the needed control prediction dataon the actuator side and the needed output data on thecontroller side are always available at time t.

To avoid the data packet disorder, two data buers areneeded to reorder the received data and keep the latestdata. One is for the control input data on the actuatorside and the other for the output data on the controllerside. So, under Assumption 4, the latest output data onthe controller side and the latest control input data on theactuator side are available for use.

On the actuator side, the control input will be taken as

ut = ut|tkj = Kxt|tkj (5)

which implies that the network delay compensator selectsut|tkj from the latest control predictions in the databuer on the actuator side. It has already been shownthat the control performance of the closed-loop networkedpredictive control system is similar to the one without thenetwork in [45]. Thus, the gain K can be designed like nor-mal state-feedback control systems without the network sothat the desired control performance of the closed-loop sys-tem is satised. Therefore, the network delay can activelybe compensated and desired control performance can beachieved by the above control strategy.

3 Stability analysis of networked pre-dictive control systems

Stability analysis is a very important part of the controlsystem design. Normally, the analysis of networked controlsystems is much harder than that of conventional controlsystems (i.e., without network). The stability analysis ofclosed-loop networked predictive control systems is simpli-ed in this section.

Let = k + j, which represents the round trip delay ofthe real-time data in the system. Combining (2) and (4)leads to

xt+1|t = Axt |t1 +BKxt |t1+L(Cxt Cxt |t1) =(A+BK LC)xt |t1 + LCxt

which is equivalent to

xt+1|t = (A+BK LC)xt|t1 + LCxt (6)Using (4), the state predictions at time t in (3) can becalculated by

xt+i|t =Axt+i1|t +BKxt+i1|t =(A+BK)xt+i1|t =

(A+BK)i1(A+BK LC)xt |t1+(A+BK)i1LCxt

for i 2, 3, , d. So, letting i = yieldsxt|t = (A+BK)

1(A+BK LC)xt |t1+(A+BK)1LCxt

So, with controller (5), the state of plant (1) can be givenby

xt+1 = Axt +BKxt|t =

Axt +BK(A+BK)1LCxt+

BK(A+BK)1(A+BK LC)xt |t1 (7)It is clear from (6) and (7) that the closed-loop networkedpredictive control system can be described as:

Zt+1 = A1Zt +BZt (8)

where

Zt =

[xt

xt|t1

], A1 =

[A 0LC A+BK LC

]

B =[BK(A+BK)1LC BK(A+BK)1(A+BKLC)

0 0

]


Actually, system (8) can be seen as a followingswitched system with a mode-dependent delay {nm, nm + 1, , d} . In recent years, switched systemswith mode-dependent delay have been well investigatedand many results of analysing their stability have beenobtained[1517]. Those results can be employed to analysethe stability of the closed-loop networked predictive controlsystem.

4 Implementation of networked predic-tive control systems

To implement networked predictive control systems, theNetCon (networked control) platform has been developed,as shown in Fig. 4. The platform is based on the visual con-guration technology for implementation of real-time con-trol systems through Intranet/Internet. It provides an idealintegrated solution for control system development and im-plementation, which can be used for research, development,experiment and assessment of control theory and controltechnology. This platform oers a tool chain for develop-ment, implement and evaluation of new control algorithmsand can also be used for rapid control prototyping. For thesystems based on network (Ethernet/Internet/wireless net-work), the user can construct networked control test rigs,carry out remote control, monitoring and supervision, andstudy networked control systems.

Fig. 4 NetCon platform for networked predictive controlsystems

The NetCon platform mainly consists of four parts: Net-Controller, NetConLink, NetConTop and NCS toolbox.NetController, which has a microprocessor for fast real-time computation of users control algorithms, and can con-struct both a networked control system and local controlsystem with the plant. Using the NetConLink, the usercan carry out block-diagram based simulation designed us-ing the NCS toolbox, then generate executable codes au-tomatically and upload them to the remote NetControllerthrough networks. Remote monitoring and on-line parame-ters tuning can also be implemented using the NetConTop.The NetController, control conguration workstation (run-ning the NetConLink) with the NCS toolbox and super-visory conguration workstation (running the NetConTop)can be employed to construct various structures of closed-loop control systems, such as networked control system overEthernet, networked control system over wireless local areanetwork, networked control system over Internet.

The NetController is the hardware part of the NetConplatform. Various real-time control algorithms can run onthe NetController. It is an ideal solution for Ethernet-

based systems and industrial modular design. The NetCon-troller is based on a cost-eective high-performance 32-bitRISC microprocessor, equipped with an embedded real-time Linux operating system and multi-channel standardI/O interface, including A/D, D/A, PWM, etc.

The NetConLink is the visual control conguration soft-ware of the NetCon platform and provides the user with aconvenient way to implement control strategies of real-timecontrol systems in a visual graphic form through the In-ternet. It is seamlessly integrated with Matlab/Simulink/Real-time workshop and various hardware drivers, and canautomatically and rapidly generate executable codes fromblock diagrams in Simulink.

The NetConTop is the visual supervisory congurationsoftware of the NetCon system. It provides the user varioustools to congure a visual diagram to monitor the operationconditions of real-time control systems running on the Net-Controller through a network. It has real-time data acqui-sition and management functions, is equipped with variousvisual congurable virtual graphs and virtual instruments,supports the client/server structure, and implements real-time remote monitoring, tuning and supervision

The NCS toolbox (networked control system toolbox)is developed in the Matlab/Simulink-based environment.With the sole NCS toolbox, the user can build his/hersimulations and real-time experiments of networked con-trol systems. This toolbox is composed of four parts: net-work simulation, network interface, plant interface and con-trol schemes. The network simulation part has the Ran-dom delay and Markov delay blocks to simulate com-munication networks as a random transmission delay unitand a Markov-chain transmission delay unit, respectively.The network interface part provides two categories of datasending/receiving blocks for simulation and real-time ex-periments, respectively, where the UDP network protocolis adopted for real-time data transmission. The plant in-terface part oers various input-output data acquisitionblocks, for example, A/D, D/A, DI, DO, PWM and PIOblocks, to exchange the information between the plant andthe NetController. The control scheme part gives many dif-ferent control algorithm blocks, for example, NPC scheme(networked predictive control), GPC scheme (generalisedpredictive control), MPC (model predictive control), andRLS estimator (recursive least squares algorithm).

Therefore, simulation and experiments of networked pre-dictive control systems can easily be implemented in theNetCon platform.

5 Simulations of networked predictivecontrol systems

To illustrate the performance of networked predic-tive control systems, a DC servo control system isconsidered[11]. The discrete-time model of the servo sys-tem is given by

A =

1.2998 0.4341 0.13431 0 0

0 1 0

, B =

10

0

C =[

3.5629 2.7739 1.0121]

The sampling period is 0.04 s. Let the desired poles ofthe closed-loop state feedback control system be [0.1,0.6+0.3i, 0.60.3i] and the desired poles of the observerbe [0, 0, 0]. Using the pole assignment method, the control


gain and observer gain are designed to be

K =[ 0.1998 0.1041 0.1793 ]

L =[

0.1739 0.1918 0.1463]T

In order to show the eectiveness of the proposed method,some simulations have been carried out using Mat-lab/Simulink. Two types of network delay are consideredin the simulations. One is the constant network delay, andthe other is the random network delay.

1) Constant network delayThe network delay, either the constant or random delay,

is often one of the causes of the poor performance and evenmakes the system unstable. The step responses of the sys-tem with 3-and 8-step constant round-trip network delayare given in Figs. 5 and 6. From these two gures, it canbe seen that the control performance deteriorates with theincrease of the network delay. Especially, when the networkdelay is 8-step, the closed-loop system is unstable. For thiscase, using the networked predictive control method pro-posed in this paper, the control performance is illustratedin Fig. 7. It can be seen from Fig. 7 that the eect of thenetwork delay is actively compensated and the performanceis the same as the case of no network delay if the model ofthe plant is perfectly accurate.

Fig. 5 Step responses with 3-step network delay

Fig. 6 Step responses with 8-step network delay

2) Random network delayIt is assumed that the round-trip network delay belongs

to an interval [3, 8] as shown in Fig. 8. The step responsesof the system without delay compensation or with delaycompensation are given in Fig. 9. It can be seen that thenetworked predictive control method can greatly improvethe performance of the networked control system in the caseof the random delay. It indicates that the proposed methodis capable of compensating for the network delay and datapacket dropout for the Internet based control system.

Fig 7 Step responses with 8-step network delay and delaycompensation

Fig. 8 Random network delay

Fig. 9 Step responses with random network delay and with orwithout delay compensation

6 Experiments of networked predictivecontrol systems

To test the networked predictive control method, a testrig has been established. This test rig consists of twoNetControllers and one DC servo system (Fig. 10). OneNetController was placed at the controller side and im-plemented the proposed networked predictive control algo-rithm and generated the future control input signals. Theother NetController was placed at the plant side and actedas the network delay compensator. These two NetCon-trollers were connected via Internet and UDP was adoptedas the communication protocol between them. The con-trolled plant is a servo system. This system consists of DCmotor, magnetic plate load and angle sensor. The controlsystem is designed to achieve a pre-set angle against the


magnetic plate load. The motor is driven by a servo ampli-er and the position angle of the servo system is measuredby a position sensor. This servo system and its model wereused for the simulation and practical experiment of the net-worked predictive control method through Internet.

Fig 10 The DC servo system

In order to illustrate the operation of the networked pre-dictive control method, experimental results presented inthis part. The implementation diagram is shown in Fig. 11.In the experiment, two NetControllers are placed at Univer-sity of South Wales in the UK and Tsinghua University inChina, respectively. The IP addresses of these two NetCon-troller are 193.63.131.219 and 159.226.20.109, respectively.

The round-trip network delay in the system varies from

0.12 s to 0.32 s according to the experimental results. Thesampling period is chosen as 0.04 s, so the round-trip net-work delay varies from 3 to 8 steps. The same controllerand observer as those used in the simulations are adoptedin the practical experiment. If this controller and observerare applied to control the servo system over Internet with-out any delay compensations, the closed-loop system willbe unstable, which is shown in Fig. 12. Using the proposednetworked predictive control method, the step response ofthe closed-loop system is shown in Fig. 13 where the localcontrol response is also given for the sake of comparison.From Fig. 13, it can be seen that the proposed method hascompensated for the eect of the network delay eectivelyand yields a similar control performance as the local con-trol.

7 Conclusions

The design, analysis and real-time implementation ofnetworked predictive control systems have been investi-gated in this paper. The control approaches and theoriesthat have been applied to NCSs have been summarized. Ananalysis on the characteristics of networked control systems(NCSs), which are much dierent from the conventionalcontrol systems, has been detailed. In order to overcomethe eect of network delay and data dropout, the networkedpredictive control method has been introduced. The sta-bility of the networked predictive control system can be

Fig 11 Implementation diagram of the networked predictive control system

Fig. 12 Step response without compensation Fig. 13 Step responses of local control and networkedpredictive control


studied using switched system approaches or switched de-lay system approaches. Simulations and practical experi-ments have illustrated that the proposed networked predic-tive control scheme has compensated for random networkcommunication delay and data dropout, and achieved de-sired control performance which is similar to the local con-trol.

References

1 Tipsuwan Y, Chow M Y. Control methodologies in net-worked control systems. Control Engineering Practice, 2003,11(10): 10991111

2 Antsaklis P, Baillieul J. Special issue on technology of net-worked control systems. Proceedings of the IEEE, 2007,95(1): 58

3 Baillieul J, Antsaklis P J. Control and communcation chal-lenges in networked real-time systems. Proceedings of theIEEE, 2007, 95(1): 928

4 Gupta R, Chow M Y. Networked control system: overviewand research trends. IEEE Transactions on Industrial Elec-tronics, 2010, 57(7): 25272535

5 Zhang L X, Gao H J, Kaynak O. Network-induced con-straints in networked control systems: a survey. IEEE Trans-actions on Industrial Informatics, 2013, 9(1): 403416

6 Gao H J, Chen T W, Lam J. A new delay system approachto network-based control. Automatica, 2008, 44(1): 3952

7 Lin C, Wang Z D, Yang F W. Observer-based networked con-trol for continuous-time systems with random sensor delays.Automatica, 2009, 45(2): 578584

8 Shi Y, Yu B. Robust mixed H2/H1 control of networked con-trol systems with random time delays in both forward andbackward communication links. Automatica, 2011, 47(4):754760

9 Zhang L Q, Shi Y, Chen T W, Huang B. A new method forstabilization of networked control systems with random de-lays. IEEE Transactions on Automatic Control, 2005, 50(8):11771181

10 Xu H, Jagannathan S, Lewis F. Stochastic optimal control ofunknown linear networked control system in the presence ofrandom delays and packet losses. Automatica, 2012, 48(6):10171030

11 Donkers M C F, Heemels W P M H, Bernardini D, Bempo-rad A, Shneer V. Stability analysis of stochastic networkedcontrol systems. Automatica, 2012, 48(5): 917925

12 Imer O C, Yuksel S, Basar T. Optimal control of LTIsystems over unreliable communication links. Automatica,2006, 42(9): 14291439

13 Sahebsara M, Chen T W, Shah S L. Optimal H1 ltering innetworked control systems with multiple packet dropouts.Systems and Control Letters, 2008, 57(9): 696702

14 Zeng W T, Chow M Y. Optimal tradeo between perfor-mance and security in networked control systems based oncoevolutionary algorithms. IEEE Transactions on IndustrialElectronics, 2012, 59(7): 30163025

15 Wang R, Liu G P, Wang W, Rees D, Zhao Y B. H1 controlfor networked predictive control systems based on switchedLyapunov function method. IEEE Transactions on IndustrialElectronics, 2010, 57(10): 35653571

16 Xia Y Q, Liu G P, Shi P, Rees D, Thomas E J C. Newstability and stabilization conditions for systems with time-delay. International Journal of Systems Science, 2007, 38(1):1724

17 Kruszewski A, Jiang W, Fridman E, Richard J, ToguyeniA. A switched system approach to exponential stabiliza-tion through communication network. IEEE Transactions onControl Systems Technology, 2012, 20(4): 887 900

18 Liu G P, Xia Y Q, Rees D, Hu W S. Networked predic-tive control of systems with random network delays in bothforward and feedback channels. IEEE Transactions on In-dustrial Electronics, 2007, 54(3): 12821297

19 Schenato L, Sinopoli B, Franceschetti M, Poolla K, Sastry SS. Foundations of control and estimation over lossy networks.Proceedings of the IEEE, 2007, 95(1): 163187

20 Zhao Y B, Liu G P, Rees D. Modeling and stabilizationof continuous-time packet-based networked control systems.IEEE Transactions on Systems, Man, and Cybernetics, PartB: Cybernetics, 2009, 39(6): 16461652

21 Longo S, Herrmann G, Barber P. Robust scheduling of sam-pleddata networked control systems. IEEE Transactions onControl Systems Technology, 2012, 20(6): 16131621

22 Liu K, Fridman E, Hetel L. Stability and L2-gain analysisof networked control systems under round-robin scheduling:a time-delay approach. Systems & Control Letters, 2012,61(5): 666675

23 Greco L, Chaillet A, Bicchi A. Exploiting packet size in un-certain nonlinear networked control systems. Automatica,2012, 48(11): 28012811

24 Katewa V, Gupta V. On a rate control protocol for net-worked estimation. Automatica, 2013, 49(5): 13101317

25 Guan Z H, Chen C Y, Feng G, Li T. Optimal tracking per-formance limitation of networked control systems with lim-ited bandwidth and additive colored white Gaussian noise.IEEE Transactions on Circuits and Systems I: Regular Pa-pers, 2013, 60(1): 189198

26 Chen W, Qiu L. Stabilization of networked control sys-tems with multirate sampling. Automatica, 2013, 49(6):15281537

27 Cea M G, Goodwin G C. Stabilization of systems overbit-rateconstrained networked control architectures. IEEETransactions on Automatic Control, 2013, 9(1): 357364

28 Stallings W. Data and Computer Communications (Sixthedition). Englewood Clis, NJ: Prentice Hall, 2000

29 Zhang W, Branicky M S, Phillips S M. Stability of networkedcontrol systems. IEEE Control Systems & Magazine, 2001,21(1): 8499

30 Zhao Y B, Liu G P, Rees D. Integrated predictive controland scheduling co-design for networked control systems. IETControl Theory & Application, 2008, 2(1): 715

31 Hu W S, Liu G P, Rees D. Event-driven networked predictivecontrol. IEEE Transactions on Industrial Electronics, 2007,54(3): 16031613

32 Murray R M, Astrom K J, Boyd S P, Brockett R W, SteinG. Future directions in control in an information-rich world.IEEE Control Systems Magazine, 2003, 23(2): 2033


33 Nilsson J, Bernhardsson B, Wittenmark B. Stochastic analy-sis and control of real-time systems with random time delays.Automatica, 1998, 34(1): 5764

34 Zhao Y B, Kim J, Sun X M, Liu G P. On the delay ef-fects of dierent channels in Internet-based networked con-trol systems. International Journal of Systems Science, 2012,44(11): 21192129

35 Hespanha J P, Naghshtabrizi P, Xu Y G. A survey of re-cent results in networked control systems. Proceedings ofthe IEEE, 2007, 95(1): 138162

36 Lian F L, Moyne J R, Tilbury D M. Performance evaluationof control networks: Ethernet, ControlNet, and DeviceNet.IEEE Control Systems Magazine, 2001, 21(1): 6683

37 Gao H J, Meng X Y, Chen T M. Stabilization of net-worked control systems with a new delay characterization.IEEE Transactions on Automatic Control, 2008, 53(9):21422148

38 Hu S, Yan W Y. Stability robustness of networked con-trol systems with respect to packet loss. Automatica, 2007,43(7): 12431248

39 Xiong J L, Lam J. Stabilization of linear systems over net-works with bounded packet loss. Automatica, 2007, 43(1):8087

40 Xue B Q, Li S Y, Zhu Q M. Moving horizon state esti-mation for networked control systems with multiple packetdropouts. IEEE Transactions on Automatic Control, 2012,57(9): 23602366

41 Wu D, Sun X M, Zhao Y B, Wang W. Stability analysis ofnonlinear switched networked control systems with periodi-cal packet dropouts. Circuits, Systems, and Signal Process-ing, 2012, 32(4): 19311947

42 Zhao Y B, Kim J, Liu G P, Rees D. Compensation andstochastic modeling of discrete-time networked control sys-tems with data packet disorder. International Journal ofControl, Automation and Systems, 2012, 10(5): 10551063

43 Zhao Y B. Packet-Based Control for Networked Control Sys-tems [Ph.D. dissertation], University of Glamorgan, UK,2008

44 Liberzon D. Quantization, time delays, and nonlinear sta-bilization. IEEE Transactions on Automatic Control, 2006,51(7): 11901195

45 Liu G P, Mu J X, Rees D, Chai S C. Design and stabilityanalysis of networked control systems with random commu-nication time delay using the modied MPC. InternationalJournal of Control, 2006, 79(4): 287296

LIU Guo-Ping Professor at the Univer-sity of South Wales, UK. He received hisB. Eng. and M.Eng. degrees in automa-tion from the Central South University ofTechnology (now Central South Univer-sity) in 1982 and 1985, respectively, and hisPh.D. degree in control engineering fromUMIST (now University of Manchester) in1992. His main research interest covers net-worked control systems and advanced con-trol of industrial systems. Corresponding

author of this paper. E-mail: [email protected]

SUN Jian Professor at Beijing Insti-tute of Technology. He received his Ph.D.degree from the Institute of Automation,Chinese Academy of Sciences in 2007. Hisresearch interest covers networked controlsystems and time-delay systems.E-mail: [email protected]

ZHAO Yun-Bo Research associate atImperial College London, UK. He receivedthe B. Sc. degree in mathematics fromShandong University, Shandong, China in2003, the M. Sc. degree in systems theoryfrom the Institute of Systems Science, Chi-nese Academy of Sciences, Beijing, Chinain 2007, and the Ph.D. degree from theUniversity of Glamorgan, Pontypridd, UKin 2008. His research interests include sys-tems biology, networked control systems,

and Boolean networks. E-mail: [email protected]

Documents

servo.pdf