Networked Control Systems Overview

7/26/2019 Networked Control Systems Overview

http://slidepdf.com/reader/full/networked-control-systems-overview 1/16

47Networked ControlSystems Overview

47.1 Introduction

47.2 Fundamental Issues in Networked Control Systems The Idea of Feedback for Control Systems An Overview toComputer-Based Control • Architectures for Control Systems

47.3 Timing Analysis of Control Systems Timing Assumptions in the Closed-Loop Operation • TimingAnalysis of Different Architectures that can Support theClosed-Loop Operation

47.4 Effect of Time Delays in the Performance of Control SystemsPerformance of Control Systems • Constant Communication-Induced Time Delays vs. System Performance • VaryingCommunication-Induced Time Delays vs. System Performance

• Network Type and Message Scheduling vs. SystemPerformance

47.5 Conclusions

47.1 Introduction

Networked Control Systems (NCS) can contain a large number of interconnected control devices thatexchange data through communication networks; examples of application areas include industrial andbuilding automation, office and home automation, intelligent vehicle and transportation systems, andadvanced aircraft and spacecraft, among other automated systems. NCS provide several advantages such

as modular and flexible system design (e.g., distributed processing and interoperability), simple and fastimplementation (e.g., small volume of wiring and powerful easy-to-use configuration tools), and power-ful system diagnosis and maintenance utilities (e.g., alarm handling and supervisory packets). However,the combination of sensors, controllers, and actuators with communication networks also makes theanalysis and design of NCS a complex task [WIT95] because it requires the integration and good under-standing of several disciplines including control systems, communications systems, and real-time systems.

First of all, by distributing the three main functions of the closed-loop operation (sampling, controlcomputation, and actuation) across a communication network, the conventional control theory thatsupports the control strategy must be reconsidered. Traditionally, the control community has seen thecomputing platform as providing the determinism that discrete-time control theory requires [AST97].

Control theory has considered implementation other than dedicated processors systems only to a very small extent. Consequently, when computing resources (processor time and communication bandwidth)are limited, control theory rarely advises on how to design controllers to take these limitations into

Pau Martí, Ricard Villà andJosep M. Fuertes

Technical University of Catalonia

Gerhard FohlerMälardalen University

© 2005 by CRC Press LLC



account [ARZ00]. In order to ensure system performance, communication-induced time delays have tobe considered at the controller design stage [MAR01].

In addition, the choice of the communication network will be subject to the timing requirementsimposed by the closed-loop control application. The use of communication networks such as field buses[LIA01] and more recently TCP/IP-based networks, typically using Ethernet [DEC01], is an increasing

tendency in all automation fields. However, each type of network has properties that may precludeits utilization in an NCS or may affect the performance of the closed-loop operation if not properly ana-lyzed.

Moreover, the performance of NCS depends not only on the control strategy and reliable communi-cation network but also on the message scheduling algorithm [WAL01]. When the communicationresources are limited, sensors, controllers, and actuators, along with other nodes, have to compete foraccessing the network. The criterion to assign the network to nodes (or messages), that is, message sched-uling, is aimed to solve this problem. However, different scheduling strategies will derive in shortening orincreasing the message latencies, thus influencing system performance [NIL98].

The contribution of this paper is to review fundamental aspects of control systems, communications

systems, and real-time systems in order to make NCS analysis and design easier. In Section 47.2, we review the basics of computer-based control systems and characterize NCS with other computer-based controlsystems. In Section 47.3, we review control aspects of NCS, focusing on the timing required by well-known control strategies that are typically used when designing discrete-time controllers. In Section 47.4,we analyze communication networks with respect to NCS as well as message scheduling strategies.Finally, Section 47.5 concludes and points out open research topics.

47.2 Fundamental Issues in Networked Control Systems

The objective of computer-based control is to use computers (and associated technologies, e.g., networks)

to manipulate the available inputs of a dynamic system in order to cause this system to behave in a man-ner more desirable than it otherwise would [LUE79]. The general functionality of a computer-basedclosed-loop control system can be split into three main activities: sampling , control algorithm computation

(calculation for short), and actuation.The basic concepts of control systems, like the idea of feedback, are reviewed in the next section. In the

subsequent section, we describe the closed-loop operation of a computer-based control system and fur-ther we analyze different types of architectures (based on processing on devices and networks) in orderto characterize the role that a communication network may play in control applications.

The Idea of Feedback for Control Systems

A control system is an interconnection of components forming a system configuration that will provide adesired system response [DOR95]. The basis for analysis of a system is the foundation provided by linearsystem theory, which assumes a cause–effect relationship for the components of the systems. A compo-nent or process to be controlled (also called a physical system or plant) can be represented by a block (asshown in Figure 47.1, top) where the input–output relationship represents the cause–effect relationship.

In control terms, the controlled variable is the quantity or condition that is measured and controlled(i.e., process output). The manipulated variable is the quantity or condition (i.e., process input) that isvaried by the controller so as to affect the value of the controlled variable. Depending on the type of infor-mation that the controller uses to vary the manipulated variable, we distinguish between open-loop con-

trol and closed-loop control. The defining feature of an open-loop control is that the controller functionthat varies the manipulated variable is determined completely by an external process that it aloneaccounts only for the desired output response (Figure 47.1, middle).

A closed-loop control system (also called a closed-loop system) utilizes an additional measure of thecontrolled variable (the actual outputs or other significant variables) to compare it with the desired out-put. Consequently, the controller function is determined on a continuing basis by the behavior of the




system itself (as expressed by the behavior of the outputs). The measure of the output is called the feed-

back signal , because it is fed back (possibly in a modified form due to the controller action) to the process.In summary, in closed-loop systems, to control means measuring the value of the controlled variable of the system and applying the manipulated variable to the system to correct or limit deviation of the meas-ured value from a desired value. A simple closed-loop control system is shown in Figure 47.1, bottom.

There are many reasons why closed-loop control is often preferable to open-loop control. Feedback isoften superior to open-loop from a performance standpoint and it can automatically adjust to unfore-seen system changes or to unanticipated disturbance inputs. A disturbance is a signal that tends to

adversely affect the value of the output of the system.The feedback concept has been the foundation for control systems analysis and design, which enables

us to control the desired output and improve accuracy while maintaining stability . Traditionally, closed-loop systems were analog-control systems (also called continuous-time systems); practically all the controlsystems implemented today are based on computer control, that is, they are computer-controlled systems

(also called discrete-time systems).

An Overview to Computer-Based Control

A computer-based control system (see, e.g., [AST97]), containing both continuous-time signals and dis-

crete-time signals, can be described schematically as in Figure 47.2.The functionality of such a system can be described as follows:Sampling : The controlled variable, y (t ) (the quantity or condition that is measured and controlled, i.e.,

process output) is usually a continuous-time signal. It is converted to digital form by the Analog-to-Digital(A/D) converter that provides a sequence { y (t k)} at the sampling instants, t k. The elapsed time between suc-cessive sampling instants is called the sampling period and is denoted by h. The A/D converter acts as asampler that returns a digital snapshot value of a continuous-time signal in the form of an n-bit word.

Calculation: The computer interprets the converted signal, { y (t k)}, as a sequence of numbers, processesthe input sequence using the control algorithm (implementing a previously designed control law), andgives a new sequence of numbers, {u(t k)}.

Actuation: This sequence is converted to an analog signal, u(t ), by a Digital-to-Analog (D/A) converterin order to be applied to the process. The D/A converter acts as a hold circuit that takes a discrete-timesignal and converts it into a continuous-time signal. Usually, a zero-order hold is used, in which case theresulting continuous-time signal is piecewise constant between successive D/A conversions. The manip-

ulated variable u(t ) is the quantity or condition (i.e., process input) that is varied by the controller so asto affect the value of the controlled variable.

Input

Input

Input

Process to be controlled

Open-loop (feedforward) control

Closed-loop (feedback) control

Output

Desired output

Desired output

Output

Output

Plant

Plant

Plant

Controller

ControllerComparison

FIGURE 47.1 Feedback in control systems.




Note that a real-time clock in the computer-based system synchronizes the events.In summary, we can describe the functionality of a computer-based closed-loop system as the sequence

of the three main operations (sampling, calculation, and actuation) that have to be repeatedly executedat every sampling period h, as illustrated in Figure 47.3. The sampling period h, beyond conforming tothe Shannon sampling theorem [SHA48], can be chosen following one of various rules-of-thumb,depending on the desired performance of the closed-loop system and the dynamics of the plant to be con-

trolled. An accepted rule-of-thumb is that the sampling frequency should be 4 to 20 times the system’scut-off frequency (which is usually approximated by the system natural frequency ω n). See [AST97] forfurther discussion on the selection of the sampling period.

Architectures for Control Systems

Computer-based control is used in many application areas,such as factory automation, process control, robot-ics, automotive systems,and others.In such applications, different types of computer-based platforms are usedto control processes, and are expected to react within precise time constraints to external events according tothe application requirements. They are expected to behave correctly both in the value and timing domains:

inputs are processed and the adequate outputs are provided with an accurate timing. Therefore, the computer-based architecture that supports the control application, and that will provide the different timings, is of mainconcern in terms of system performance. Basic architectures for computer-control are discussed next.

The traditional communication architecture for control systems in all fields of automation has been point-to-point (see [BEN00] for a short review of the history of digital computers for control systems). In such anarchitecture, systems components (sensors and actuators) that interact with the physical process are directly

Computer-basedsystem

Controlalgorithm

A/D D/A

Clock

Plant

{y (t k )} y (t ) {u (t k )} u (t )

FIGURE 47.2 Diagram of a computer-based control system.

Actuation

Calculation

Sampling

Time

u (t k −1) u (t k +1)u (t k )

y (t k −1) y(t k +1)y (t k )

t k −1 t k +1t k

h h

FIGURE 47.3 Diagram of the three main operations in a closed-loop system.




wired to controllers using the traditional 4 to 20 mA current loop. Systems of this type are also referred to asdirect digital control systems or local closed-loop systems (see Figure 47.4, top). In point-to-point architec-tures, each sensor or actuator exchanges data with the controller using a dedicated communication link.Therefore, control data (sensor data and command data) do not suffer delays other than the one expectedwhen transmitting in a point-to-point link. Nowadays, for digital data, this architecture is widely used for

Programmable Logic Controller (PLC)-based control architectures. For analog data,point-to-point architec-tures based on Proportional, Integral and Derivative (PID) controllers are commonly used in the automationfield (as stressed in [AST00], PID controller is the first solution that should be tried when feedback is used).

However, such isolated point-to-point architectures do not suit modern automation systems [SCH97]where the goal is to design flexible systems that can accomplish various tasks with small reconfiguration cost.The reconfiguration of a point-to-point architecture to expand physical setups and functionality is not an easy task and it is often easier to build a completely new system than to rewire all the existing system components.In addition, for such systems, diagnosis and maintenance have to be performed locally as no remote opera-tion (control and supervision) is allowed. In the end, this results in inefficient systems and increased costs.

The common bus network architecture can solve the problems of the point-to-point architecture for

control systems [LEN93]. Depending on how the control tasks are distributed across the network, we candistinguish two architectures: Distributed Control Systems (DCS) (Figure 47.4, middle) and NCS (Figure47.4, bottom). Such architectures, main goal is to reduce maintenance, lower cost, reduce weight andpower, simplify installation, and improve reliability.

In DCS, most of the real-time control tasks (sensing, calculation, and actuation) are carried outwithin the individual control nodes. The interconnected nodes mainly use the network to transmitalarm signals, monitoring information and local control actions. The network is also used for systemconfiguration and setup. Therefore, a DCS can be viewed as a networked collection of point-to-point con-trol systems. Note that for such architectures, the control data suffer the same delays as for point-to-pointarchitectures, because it does not travel in the network. DCS are widely spread in all fields of automation.

Actuator

ActuatorActuator

Actuator Actuator

Point-to-point control

Distributed control system

Networked control system

Controller

Controller

Controller

Controller

Controller

Sensor

SensorSensor

Sensor Sensor

Communication network


Process

Process

Process

Process Process

FIGURE 47.4 Architectures for control systems.




One step further in distributing the processing intelligence are NCS. NCS are the application of theintelligent network concept (where all nodes have processing power, allowing for the modularization of the functionality and standard interfaces for interchangeability and interoperability) to control systems[RAY89]. An NCS is a completely distributed and networked feedback control system, where plant sen-sor–controller–actuator nodes in closed-loop operation drive the network principal traffic. In such archi-

tectures, all processing control nodes (sensors, controllers, and actuators) are interconnected through thecommunication network and, apart from the data typically exchanged in DCS (mainly supervisory data),they transmit sensory and command messages. Note that sensory and command messages (which we alsocall control data) carry the information required to accomplish the closed-loop operation: sensory mes-sages carry the sampled data, which are used by the controller node to compute the control signal, andcommand messages carry the command signals that are applied to the process by the actuator node.Therefore, for such architectures, control data are subject to delays caused by the communication net-work (apart from necessary transmission and propagation delays).1 Depending on the network and themessage scheduling policy, the induced delays will take different values, thus affecting the quality of theclosed-loop control operation at different levels.

47.3 Timing Analysis of Control Systems

In this section, we summarize the timing assumptions considered by discrete-time control theory in theclosed-loop operation and compare them with the timing analysis that different architectures for com-puter control provide for closed-loop systems.

Timing Assumptions in the Closed-Loop Operation

From classical discrete-time control operation (which we briefly described in the section An overview to

computer-based control), we can derive the following timing assumptions about the three main parts of a closed loop (recall Figure 47.2):

● The input data collection or sampling is performed by the sampler at equidistant time instants givenby the sampling period, h.

● The time delay , τ, that is, time elapsed between related2 sampling and actuation instants (whichshould include the execution time spent by the control algorithm calculation) is constant (eitherinstantaneous3 or not, depending on how the controller was designed).

● The output data transmission or actuation is performed at the actuation instants, which occurs atthe time delay completion.

Although these assumptions are the basis for discrete-time control theory, in practice, it is not possi-ble to keep all of them in a computer-based implementation (as we further discuss in the next section).For example, it is clear that calculations take time, therefore the time delay, which includes the calcula-tion of the control law, cannot be zero. The implementation approach underlying the assumption of an

1The transmission and propagation delays are constant parameters that characterize any transmission of datathrough a communication media [STA03]. The transmission delay is the time taken by each device — sensor or con-troller — to send each message given a transmission rate and the propagation delay is the time taken by the commu-nication link to transport each message.

2Related sampling and actuation instants refer to the sampling and actuation actions performed at each execution

of a closed-loop algorithm for each sampling period.3Recall that the idea of using digital computers as components in control systems started around 1950 [AST97].

The notion of instantaneousness comes from the theory of linear time-invariant continuous-time systems, fromwhich discrete-time systems theory emerged. In continuous-time systems, analog controllers compute almost instan-taneously. As a consequence, some of the models and methods used in discrete-time theory, coming from continu-ous-time theory, maintain the instantaneous assumption.




instantaneous time delay is to build the system in such a way that the time delay is minimized in order toignore it in the controller design. However, this does not mean that the control computation executiontime is zero. It means that the control computation execution time is not relevant to the controlling pur-poses if compared to the closed-loop system dynamics.

Note that sampling period and time delay are the timing parameters of interest in the analysis and

design of discrete-time controllers. In fact, the well-known control models and methods used in discrete-time control theory were developed for systems characterized with a constant sampling period ( h) and aconstant time delay (τ).

Timing Analysis of Different Architectures that can Support the Closed-Loop

Operation

In the previous section, we have explained the timing assumptions that discrete-time control theory expects in the closed-loop operation. In this section, we give an overview of the real expected timingbehavior of closed-loop systems that are either implemented using the point-to-point architecture or the

networked architecture. Note that we do not discuss distributed control systems architectures because aswe argued in the section Architectures for control systems, in this architecture, control data (sensor andcommand messages) suffer the same delays as for point-to-point architectures.

Looking at the closed-loop operation for any of the two architectures (Figure 47.5), the following func-tionality is expected: the sensor samples the process to be controlled with a given sampling period h. Eachtime the data have been collected, the sensor forwards it to the controller using the corresponding com-munication link, introducing a communication delay τsc (sampler to controller delay). The kth controlcomputation execution start time is given by the sampling time t k plus the τsc delay. The control compu-tation introduces a delay (τc ) used to calculate the command signal(s). The controller forwards the com-mand signal(s) to the actuator, introducing another communication delay τca (controller to actuator

delay). Finally, the actuator performs the actuation at the time given by t k +τ

sc +τ

c +τ

ca.In both architectures, the time delay τ (time elapsed between related sampling and actuation instants)is given by τ = τsc + τc + τca. However, depending on the architecture, this delay will be either constant ortime varying. Note that in most of the actual closed-loop implementations, controllers are implementedin dedicated processors. Therefore, we will also consider the time delay τc introduced by the controller to

Actuator

Actuator

Sensor

SensorController c

Controller c

ca

ca sc sc

Actuation

Calculation

Sampling

Timeu (t k −1) u (t k +1)u (t k )

y (t k −1) y (t k +1)y (t k )

t k −1 t k +1t k h h h

sc

ca ca ca

sc sc

c c c

Actuation

Calculation

Sampling

Timeu (t k −1) u (t k +1)u (t k )

y (t k −1) y (t k +1)y (t k )

t k −1 t k +1t k h h h

sc

ca ca ca

sc sc

τc τc τc


Process

Process

h

h

Point-to-point control Networked control system

FIGURE 47.5 Timing analysis of the closed-loop operation: point-to-point (left) vs. networked (right) architecture.




be almost constant. We regard the conversion operation delay for A/D conversion in the sensor and D/Aconversion in the actuator to be negligible. If these operations take too long a time, they could be includedfor the timing analysis in the time delay τc . For implementations other than dedicated processors, the τc

delay will depend on the concurrency of the other executing tasks (see [MAR02] for further reading).Note also that the sampling period is constant for both architectures (Figure 47.5). Samples, y (t k), are

taken at equidistant times, given by h. Therefore, the sampling period does not raise challenges other thanthe usual ones that are found in the analysis and design of traditional computer-based closed-loop con-trol systems. Consequently, from now on, we will focus on the time delay as a critical timing parameterfor ensuring system performance. Specifically, we are interested in the communication-induced timedelays, τsc and τca.

If we focus on the point-to-point architecture (Figure 47.5, left), we can model the time delays (τ) intro-duced by the point-to-point communication links as constant delays. Both τsc and τca will depend on con-stant parameters such as the transmission time and the propagation delay of the link. Therefore, the controlsignal u(t k) is applied to the process at equidistant times, given by the sampling time plus the time delay.

However, looking at the networked architecture (Figure 47.5, right), communication-induced time

delays may vary at each closed-loop instantiation. Such delays, apart from being characterized by constantparameters such as the transmission time and propagation delay (as for the previous point-to-pointarchitecture), will depend on new parameters because control nodes exchange data (sampling and con-trolling messages) through a shared communication link. Issues like the network medium access proto-col and the message scheduling will also determine τsc and τca (see the section Network type and messagescheduling vs. system performance for further discussion). Therefore, the actuation signals u(t k) are sentat nonequidistant time instants, given by the sampling time plus a time delay that may vary at each exe-cution, as illustrated in Figure 47.5, right.

After this timing analysis, we can conclude that the theoretical expected timing, constant time delays,assumed by discrete-time control theory is not realistic if compared to the timing derived from net-

worked architectures, varying time delays (see [TÖR98] for further reading in timing analysis). This mis-match will degrade system performance, as we illustrate in the next section.

47.4 Effect of Time Delays in the Performance of ControlSystems

In classic control theory, several properties are used to evaluate the performance of closed-loop systems.After discussing them, we will evaluate the effect of communication-induced time delays in system per-formance. For illustrative purposes, we will use a specific example: an inverted pendulum controlled by aset of control nodes that exchange control data through a communication network.

Performance of Control Systems

The primary evaluation is concerned with meeting the closed-loop response performance specifications4

and stability 5 (see, e.g., [DOR95]). Beyond these requirements, looking at the closed-loop response, sincecontroller designs attempt to minimize the system error to certain anticipated inputs or perturbations,

4The desired controlled system characteristics, in a formal specification of the problem, are given through thedesign parameters, which can be met by specifying the closed-loop poles location. However, rather than specifyingdesign parameters, usually it is more meaningful to specify quantities, such as at which time the controlled system

recovers from a perturbation (settling time), or the allowed error of the controlled system response to certain antici-pated inputs.

5The concept of stability is very important when analyzing dynamic systems. From a conceptual point of view, acontrolled system is said to be in an equilibrium state if in the absence of any perturbation, the system output remainsin the same state. And stability is related to the ability to return to an equilibrium state from a perturbation. For morereading on stability, see [AST97].




traditional performance criteria focus on the system error. The system error is defined as the differencebetween the desired response of the system and the actual response of the system. The smaller the differ-ence, the better the performance. Figure 47.6 illustrates these concepts. In Figure 47.6, right, we representthe response of a process that is being controlled. Before the perturbation arrival, the system is in equi-

librium. Upon perturbation arrival, the system loses its equilibrium state. Therefore, the actual controlledsystem response is different from the desired system response, causing the system error. From thismoment on, the controller action tries to correct the deviation in order to bring the controlled systemback to the equilibrium state again (this is illustrated by the oscillatory curve with decreasing amplitude).The time it takes to reach again the equilibrium state and the maximum amplitude of the deviation deter-mine the performance of the controller.

Constant Communication-Induced Time Delays vs. System Performance

In the section Timing assumptions in the closed-loop operation, we have discussed that controllers

designed using classical discrete time control theory are dependent on the sampling period and time delay and that for NCS, as we saw in the section Timing analysis of different architectures that can support theclosed-loop operation, time delays are the timing parameters of main concern. Therefore, it is interestingto outline which are the effects of communication-induced time delays on the performance of closed-loop control systems. Recall that we are evaluating time delays in the sense of the time elapsed from sam-pling to actuation. To do so, in this section, we will focus on constant time delays, because such type of

Closed-loop system

Error+

−

Desired

response

Actualresponse

Controller Plant

Desired system response

Actual system response

System error

System error

Error No errorNo error

Perturbation arrival

Equilibrium

FIGURE 47.6 Performance of closed-loop control systems.




delays can easily be accounted for in the controller design and included in the closed-loop operation (see[MAR01] or [YEP02] for more details). In the next section, we will focus on varying time delays.

Consequently, for illustrative purposes, if we experiment with constant sampling-to-actuation delays (mul-tiples of a nominal delay τn) and include them in the controller algorithm that is executing in a controller nodeof an NCS, we can summarize the effect they have in the closed-loop operation as shown in Figure 47.7. Note

that the system we consider is an NCS that consist of a set of processing nodes (sensor, controller and actua-tor) that communicate data across a communication network, which introduces a constant time delay for thesampling message (form sensor to controller) and for the command message (from controller to actuator).

In Figure 47.7, we show four responses ( y (t )) of a generic controlled system affected by a perturbation(of magnitude 1 in the error axis) for four different constant sampling-to-actuation delays (τn, 2τn, 3τn ,

and 4τn). Although the controller has been designed to account for the delays, there will always be degrad-ing effects on the controlled system response. The degrading effects on system performance can be sum-marized as follows:

Delayed response: The first effect is that a sampling-to-actuation delay delays the application of each

control action to the plant. Therefore, as it can be seen from Figure 47.7, left, if the time elapsed fromsampling to actuation is longer at each one of the four system responses, the more delayed each closed-loop system has its response (in terms of recovering from the perturbation). Note that the response of thesystem labeled with a nominal delay τn (the shortest one) always reacts before (its curve precedes) thanall the others, which have longer delays.

Increased error : The second effect is that the later the control action is applied to the process (the longer thedelay), the larger the error the closed-loop system response has. As it can also be seen in Figure 47.7, left, theinitial error is 1. However, looking at the detailed view (Figure 47.7, right), the error that has to be correctedis not 1 but 1+E i, where E i is the added error due to each delay. Note that the added error of the system labeledwith a delay of 4τn is larger than the others, as it suffers the longest delay.

These two effects can be clearly seen in Figure 47.7 right, where we show the detailed initial response of each system. In each system, the first sampling that detects a perturbation (error equal to 1) on the systemis performed at time 0. While the sample travels across the network in the sensory message to the con-troller, the controller calculates the control action, and sends it to the actuator, the error of the controlledsystem increases. It is when the actuator applies the command action to the system (at the delay comple-tion) that the error starts being corrected. The later this action is performed, the larger the system errorthat must be corrected. Compare, for example, E 4 with E 2. E 4 is the error of the fourth system (with a 4τn

delay) at the first actuation instant. E 2 is the error of the second system (with a 2τn delay) at the first

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

2

1.5

1

0.5

0

−0.5

−1.5

−1

−2

0 0.02 0.04 0.06 0.08 0.1 0.12

1.08

1.06

1.04

1.02

1

0.98

0.96

Time Time

E r r o r

E r r o r

System response with a nominal delay 1

System responsewith a delay 41

Desired system response

41

31

21

1E 1

E 2

E 3

E 4

Delay effects (complete view) Delay effects (detailed view)

FIGURE 47.7 Performance of a generic controlled system for different time delays.




actuation instant. As it can be seen, the error E 4 is more than twice E 2. However, 4τn is exactly 2×2τn. Theerror increases faster than the delay increases.For further reading on this topic, see, for example, [YOO00].

From this generic study, it has been shown that the system error generated by a perturbation (which hasto be corrected by a controller) may dramatically increase if, from the corresponding sampling, the actuationoccurs too late. And this will depend on the network-induced time delays. However, it has to be pointed out

that time delays can be assumed in closed-loop systems as far as they are constant,known at design stage, andthat the degradation they introduce can be made to conform to the performance specifications. Alternatively,if the communication delays are not constant but exhibit fluctuations, care must be taken because the systemperformance may drastically decrease as we show in the next section. As we will discuss in the subsequent sec-tion, the type of network that supports the networked application and the message scheduling policy willdetermine the delay pattern that should be accounted for in the analysis and design of networked controllers.

Varying Communication-Induced Time Delays vs. System Performance

In this section, we illustrate, using an example, the problems that varying communication time delays can

introduce in control loops closed over communication networks. First of all, we introduce the setup.Afterwards, we analyze the effect of varying time delays in the system performance.

The system we consider is an NCS that consist of a set of control and noncontrol nodes that commu-nicate data across a communication network. The control nodes (sensor, controller, and actuator) are incharge of controlling an inverted pendulum mounted on a motor-driven cart (see Figure 47.8, left). Eachcontrol node executes the corresponding control law code shown in Figure 47.8, right. The noncontrolnodes do not participate in the closed-loop operation but use the network whenever needed, increasingthe traffic load and introducing unexpected longer message latencies. Therefore, the network introducesvarying time delays for the sampling message (from sensor to controller) and for the command message(from controller to actuator) due to the interferences of the other noncontrol nodes.

The inverted pendulum control problem can be stated as follows: the inverted pendulum (of length l and mass m) can only swing in a vertical plane parallel to the direction of the cart (of mass M ). In the

Algorithm- SensorNode

Algorithm- ControllerNode

Algorithm- ActuatorNode

x

m

M

g

l

θ

u

0

0

(M + m ) . g

m . g

M . l M . l

M M −

−

1

1

11

1

10 0 0

0

0 0 0

0 0

0 00

0 0 0000

==

=

.

. + u (t)

y (t)

{ wait(h)

read_angle ( ),

send_sampling_message ( );

}

{ = receive_sampling_message ();

u= calculate_output(Controller(h, ));

send_controller_message (u);

}

{ u= receive_controller_message(); write_output(u);

}

Control nodes codesSketch and mathematical model

Algorithm-SensorNode

Algorithm-ControllerNode

Algorithm-ActuatorNode

FIGURE 47.8 Inverted pendulum model (left) and algorithms for the distributed operation (right).




presence of a perturbation, to balance the pendulum, the cart is pushed back and forth on a track of lim-ited length. Balancing fails when the inclination of the pendulum exceeds preset limits, or when the carthits the stops at the end of the track. The aim is to find a controller to balance the inverted pendulum,preventing it from failing, and to bring the cart to the center of the track.

The state of the inverted pendulum on a cart is described by the cart position ( x ), its velocity (v ), the

pendulum angle (θ), and its angular velocity (ω ). The force applied to the cart, (u), is the manipulatedcontrolling command calculated by the controller according to the actual angle and position (controlledvariables). A linear time-invariant state-space model of the inverted pendulum, used for the controldesign can be viewed in Figure 47.8 (where, e.g., we take M =2 kg, m=0.1 kg, l =0.5 m, and g =9.81 m/sec2).

For the sake of simplicity, we will focus only on the angle (θ). The goal of our controller is to maintainthe desired vertical position of the inverted pendulum at all times. Specifically, the performance require-ment is to recover from a perturbation in less than 1 sec. To do so, we close the networked loop by usinga state-feedback controller designed using classical methods, with strictly periodic sampling h, andaccounting for a constant time delay τ.

The inverted pendulum response obtained by the distributed controller if it would execute in isolation

(on a networked architecture but with no competitors, thus, with a constant time delay) in the presenceof a perturbation (modeled as an initial condition) can be seen in Figure 47.9, left. Since it is beyond thescope of this work to specifically discuss state-feedback controller design, we assume that our design isgood enough for illustrative purposes (we call it ideal). Note that the performance requirement is met;that is, the pendulum recovers from the perturbation in less than 1 sec.

The inverted pendulum response obtained by the distributed controller executing in the networkedarchitecture and suffering varying time delays due to the competitor’s interferences is quite different. Asit can be seen in Figure 47.9, right, the inverted pendulum response suffers different degrees of degrada-tion depending on the randomly induced communication delays of different magnitudes that we havegenerated (simulating the interference produced by the messaging from the noncontrol nodes).

Concretely, in Figure 47.9, right, we show four curves: the dotted line corresponds to the ideal responseof the inverted pendulum, with a constant time delay of τ. The other three curves correspond to threeresponses obtained by three controllers, each one suffering random delays of different magnitudes (delayslonger than the nominal τ but bounded by 10, 25 or 50% of τ). As it can be seen, the longer the randomgenerated delay, the larger the degradation. Note that for the case with longest delays (delays bounded by 50% of τ), the system becomes unstable (the inverted pendulum falls).

The degradation caused by varying time delays can be explained as follows. From a control perspec-tive, the control system with varying delays is no longer time-invariant. That is, the controller designedto account for a constant time delay (time invariant system) is not effective when delays vary (time

0 0.5 1 1.5 2 0 0.5 1 1.5 2

3

2.5

2

1.5

1

0.5

−3−2.5

−2

−1.5

−1

−0.5

0

3

2

1

−3

−2

−1

0 A n g l e

A n g l e

Time Time

Ideal ()

Ideal ()

+rand (50% )

+rand (25% )

+rand (10% )

Ideal response Degraded response

FIGURE 47.9 Inverted pendulum system response: ideal (left) vs. degradation (right) due to varying time delays.




varying system). Therefore, the standard computer control theory (targeted for time-invariant systems)cannot be used in the analysis and design of NCS with varying time delays. The problem of analysis anddesign of control systems when the communication delays are varying in a random manner is complex and it is still an open research field (see [CHO01] for an introductory tutorial).

Following the seminal work on communication and control [HAL90], an approach to overcome these

problems is to model the communication delays as probabilistic distributions. The design of the con-troller has to account for those delays or alternatively has to view the temporal nondeterminism of net-work-induced delays as an uncertainty, similar to a plant uncertainty, or a disturbance, and then designthe control systems to be robust against such uncertainty.

For example, [NIL98] presents various probabilistic communication delays models and accordingly,solves an LQG optimal control problem for them. Using similar techniques, in [XIA01] communicationresources allocation and linear systems design are jointly optimized.

Another approach is to assume a constant sampling-actuation delay, forcing a synchronous actuation(synchronous with respect to the sampling) [WIT98]. This can be achieved by forcing the actuation tooccur at equidistant times, given by the worst-case-message-latency. However, this may unnecessarily

impose a longer time delay than the actual ones that appear during system runtime, thus forcing a pes-simistic timing behavior in the closed-loop system that may result in a graceful but unnecessarily systemperformance degradation [YEP02].

Up to now, we have not discussed stability issues. As can be seen in Figure 47.9, right, the degradationthat the controlled system suffers drives the system to almost instability. It is well known that delays cancause instability. Therefore, in the analysis of control loops closed over communication networks, delayshave to be included for the stability analysis. See [ZHA01] for a starting reading point.

Note, however, that the randomness of the communication-induced delays will depend on the type of communication network and message scheduling that supports the closed-loop operation, as introducedin [LIA02]. This is discussed next.

Network Type and Message Scheduling vs. System Performance

Varying communication time delays can seriously degrade the performance of the controlled system, asoutlined in the previous section (see Figure 47.9, right). Moreover, when delays appear in a random man-ner, the lack of well-established control techniques for time-varying systems makes the analysis anddesign of NCS a complex task.Taking into account that the type of delays determine system performance,the type of network and more concretely, the rules that allow the exchange of messages, that is, messagescheduling policy, is a key issue in terms of system performance. By choosing the adequate message sched-uling (or protocol), the impact of communication-induced time delays can be drastically reduced because

delays can be somehow controlled.It is well known that random-based scheduling policies (e.g., Ethernet [DEC01]) conflict with the

determinism that control theory imposes on an implementation. Standard Ethernet communicationintroduces random delays and no guarantee can be given on the maximum message latency. Therefore,to design NCS with Ethernet as a communication backbone is not appropriate from the control tractabil-ity point of view. However, there is a strong interest in using the cheap and simple Ethernet technology for industrial and embedded systems. This far, however, the lack of real-time services has prevented thischange in the used network technology. Nevertheless, in recent literature, works can be found addressingthe lack of determinism by using switched Ethernet techniques (see [LOB01] for further reading onEthernet and control issues in the automation field).

To overcome this problem of the lack of determinism in Ethernet, we can think of well-known com-munication networks with deterministic scheduling policies such as table-driven (e.g., WorldFIP[ALM02]), master/slave (e.g., Profibus [TOV99]), or nonpreemptive fixed priority-based (e.g., ControllerArea Network, CAN [TIN94]) scheduling policies. Such scheduling policies limit message latencies, thusfacilitating the analysis and design of closed loops over communication networks and also ensuring aver-age or priority-based levels of performance.




For example, table-driven or master/slave scheduling policies can guarantee constant message latencies.Moreover, latencies (delays) are known before runtime. For example, in a master/slave configuration, theorder of the polling determines the order in which nodes have access to the network. And when a node isaccessing the network, it uses all the available bandwidth. Therefore, the network-induced delays depend onknown parameters (transmission time and propagation time). If delays are known before run time, the

analysis and design of the distributed controller taking into account the delay induced by the networkedarchitecture are straightforward [AST97]. Note that such protocols, although providing constant time delays(which makes the control design job easy), may not provide the optimum performance. This is because they may be enforcing delays longer than the ones that could be obtained using other protocols based on moreflexible priority-based mechanisms such as CAN. Recall that a delay is always a source of degradation, as weoutlined in the section Constant communication-induced time delays vs. system performance.

For nonpreemptive fixed priority-based scheduling policies such as CAN, although message latenciesare bounded, they are not known before system runtime. This is because messages may suffer collisions.In this case, the deterministic collision resolution mechanism that CAN incorporate ensures that the mes-sage with highest priority will continue being transmitted. Moreover, using [TIN94], the maximum laten-

cies for all messages can be precalculated. Therefore, although message delays are not exactly known, theknown upper bounds can be used for the analysis and design of the networked controller. Moreover, suchscheduling techniques can be used to shorten the message latencies by adequately assigning the priorities.

47.5 Conclusions

NCS constitute a particular type of applications over communication networks where the principal traf-fic is driven by plant sensor–controller–actuator nodes in closed-loop systems. As we have reviewed inthis chapter, the analysis and design of such systems are complex tasks involving the integration and goodunderstanding of several disciplines: control systems, communication systems, and real-time systems. In

particular, when a network is used dedicated just for one control loop, without any external or other nodeinterferences, the resulting networked system becomes trivial. But when the network is shared among dif-ferent nodes and there is even just one sensor–controller–actuator closed-loop control subnetwork, caremust be taken on how the expected or unexpected delays can affect the controlled system performance.This chapter has shown the way to characterize the different architecture issues in a networked controlsystem, mainly in what refers to the different delays that occur in the messaging related to the controlactions. In general, if those delays are negligible with respect to the dominant dynamics of the controlledplant, they can be overcomed; but when those time delays are significant, a more precise analysis has tobe conducted, both in the design or redesign of the control algorithm, on the network characteristics, andon the system’s scheduling. These last topics are still an open issue that is currently being investigated by

several research groups.

References

[ALM02] Almeida, L., Tovar, E., Fonseca, J.A., and Vasques, F., Schedulability analysis of real-time trafficin WorldFIP networks: an integrated approach, IEEE Transactions on Industrial Electronics, 49,1165–1174, 2002.

[ÅRZ00] Årzen, K.-E., Cervin, A., Eker, J., and Sha, L., An Introduction to Control and Scheduling Co-Design, Proceedings of the 39th IEEE Conference on Decision and Control, Vol.5, Sydney,Australia, Dec. 2002, pp. 4865–4870.

[AST97] Åström, K.J. and Wittenmark, B., Computer-Controlled Systems. Theory and Design, 3rd ed.,Prentice-Hall, Englewood Cliffs, NJ, 1997.

[AST00] Åström, K.J. and Hägglund, T., The Future of PID Control, Preprints of IFAC Workshop onDigital Control. Past, Present and Future of PID Control, Terrassa, Spain, 2000, pp. 19–30.

[BEN00] Bennett, S., The Past of PID Controllers, Preprints of IFAC Workshop on Digital Control. Past,Present and Future of PID Control, Terrassa, Spain, 2000, p. 13.




[CHO01] Chow, M.-Y. and Tipsuwan, Y., Network-Based Control Systems: A Tutorial, Proceedings of the27th Annual Conference of the IEEE Industrial Electronics Society, Vol.3, Denver, CO., U.S.A.,2001, pp. 1593–1602.

[DEC01] Decotignie, J.-D., A Perspective on Ethernet as a Fieldbus,Proceedings of the 4th IFAC InternationalConference on Fieldbus Systems and their Applications, Nancy, France, November 2001.

[DOR95] Dorf, R.C. and Bishop, R.H. (1995). Modern Control Systems, Seventh Edition, Addison-Wesley

[HAL90] Halevi, Y. and Ray, A., Analysis of integrated communication and control system networks, ASME Journal of Dynamic Systems, Measurement and Control , 112, 365–372, 1990.

[LEN93] Lenhart, G., A fieldbus approach to local control networks, Advances in Instrumentation and

Control , 48, 357–365, 1993.[LIA01] Lian, F., Moyne, J. and Tillbury, D., Performance evaluation of control networks: ethernet, con-

trolnet, and devicenet, IEEE Control Systems Magazine, 21, 66–83, 2001.[LIA02] Lian, F., Moyne, J. and Tillbury, D., Networked design consideration for distributed control sys-

tems, IEEE Transactions on Control Systems Technology , 10, 297–307, 2002.

[LOB01] Lo Bello, L., and Mirabella, O., Design Issues for Ethernet in Automation, Proceedings of the 8thIEEE International Conference on Emerging Technologies and Factory Automation,Vol.1, AntibesJuan-les-pins, France, Oct. 2001, pp. 213–221.

[LUE79] Luenberger, D.G., Introduction to Dynamic Systems. Theory, Models and Applications, John Wiley & Sons, New York, 1979.

[MAR01] Martí, P., Fuertes, J.M., and Fohler, G., An Integrated Approach to Real-Time DistributedControl Systems Over Fieldbuses, Proceedings of the 8th IEEE International Conference onEmerging Technologies and Factory Automation, Vol.1, Antibes Juan-les-pins, France, Oct. 2001,pp. 23rd 177–182.

[MAR02] Martí, P., Fuertes, J.M., Fohler, G., and Ramamritham, K., Improving Quality-of-Control using

Flexible Timing Constraints: Metric and Scheduling Issues, Proceedings of the 23rd IEEE Real-Time Systems Symposium, Austin, TX, USA, Dec. 2002, pp. 91–100.

[NIL98] Nilsson, J., Bernhardsson, B., and Wittenmark, B., Stochastic analysis and control of real-timesystems with random time delays, Automatica, 34, 57–64, 1998.

[RAY89] Ray, A., Introduction to networking for integrated control systems, IEEE Control System

Magazine, 9, 76–79, 1989.[SCH97] Schickhuber, G. and McCarthy, O. Distributed fieldbus and control network systems,

Computing & Control Engineering Journal , 8, 21–32, 1997.[SHA48] Shannon, C.E., A mathematical theory of communication, Bell System Technical Journal, July

and October, 27, 379–423 and 623–656, 1948.

[STA03] Stallings, W., Data and Computer Communications, 7th ed., Prentice-Hall, Englewood Cliffs, NJ,2003.

[TIN94] Tindell, K. and Burns, A., Guaranteeing Message Latencies on Controller Area Network (CAN),Proceedings 1st International CAN Conference, 1994, pp. 1.2–1.11.

[TÖR98] Törngren, M., Fundamentals of implementing real-time control applications in distributedcomputer systems, Journal of Real-Time Systems, 14, 219–250, 1998.

[TOV99] Tovar, E. and Vasques, F., Real-time fieldbus communications using profibus networks, IEEE

Transactions on Industrial Electronics, 46, 1241–1251, 1999.[WAL01] Walsh, G.C. and Ye, H., Scheduling of networked control systems, IEEE Control Systems

Magazine, 21, 57–65, 2001.

[WIT95] Wittenmark, B., Nilsson, J., and Törngren, M., Timing Problems in Real-Time Control Systems:Problem Formulation, Proceedings of the American Control Conference, Vol.3, Seattle, WA, U.S.A.,1995, pp. 2000–2004.

[WIT98] Wittenmark, B., Bastian, B., and Nilsson, J., Analysis of Time Delays in Synchronous andAsynchronous Control Loops, Proceedings of the 37th IEEE Conference on Decision and Control,Vol.1, Tampa, FL, U.S.A., 1998, pp. 283–288.




[XIA01] Xiao, L., Johansson, M., Hindi, H., Boyd, S., and Goldsmith, A., Joint Optimization of Communication Rates and Linear Systems, Proceedings of the 40th IEEE Conference on Decisionand Control, Vol.3, December 2001, pp. 2321–2326.

[YEP02] Yepez, J., Marti, P., and Fuertes, J.M., Control Loop Performance Analysis over NetworkedControl Systems, Proceedings of the IEEE 28th Annual Conference of the Industrial Electronics

Society, Vol.4 , Sevilla, Spain, November 2002, pp. 2880–2885.[YOO00] Yook, J.K., Tilbury, D.M., and Soparkar, N.R., A Design Methodology for Distributed Control

Systems to Optimize Performance in the Presence of Time Delays, Proceedings of the AmericanControl Conference, Vol.3, June 2000, pp. 1959–1964.

[ZHA01] Zhang, W., Branicky, M.S., and Phillips, S.M., Stability of networked control systems, IEEE

Control Systems Magazine, 21, 84–89, 2001.

Networked Control Systems Overview

Documents

Transcript of Networked Control Systems Overview