Transactions of the Institute of Measurement and Control 27, 2 (2005) pp. 103 /118
Time delay and data loss compensation for Internet-based process control systems S.H. Yang1, X. Chen1, L.S. Tan2 and L. Yang1 1
Computer Science Department, Loughborough University, Loughborough, LE11 3TU UK 2
Department of Computer Science, Central China Normal University, Wuhan, PR China
Introducing the Internet into process control systems has brought a lot of benefits and some difficulties to control engineers as well. The major challenge of Internet-based process control systems is how to deal with Internet time delay and data loss. This paper analyses Internet communication features and proposes a control structure with a variable sampling time to overcome the Internet transmission delay. In order to take advantage of the structure fully and attack the major difficulties, two compensation elements have been designed and implemented in the system. These two compensators are located in the feedback and feedforward channels in the architecture respectively. Our simulation and experimental results illustrate that the control structure and two compensation elements can efficiently deal with the Internet time delay and data loss. Our simulation results also show that the time delay and data loss in the feedforward channel seem to cause more serious influence on the control performance and is more difficult to be compensated for. Key words: compensation; control system design; data loss; Internet; Internet-based process control; time delay.
1.
Introduction
In past years, the success in adopting the Internet to deliver business services has demonstrated a lot of advantages, such as cost reduction and flexibility. In the control
Address for correspondence: S. H. Yang, Senior Lecturer in Computer Science, Loughborough University, Loughborough LE11 3TU, UK. E-mail:
[email protected]
–
2005 The Institute of Measurement and Control
10.1191/0142331205tm133oa
104
Internet-based process control systems
theory and application areas, researchers began to exploit the advantages of the Internet for control systems, namely Internet-based control systems, and its controllability and stability. These new types of control systems are characterized as globally remote monitoring and adjustment of plants over the Internet. With the prevalence of the Internet, plants stand to benefit from the ways of retrieving data and reacting to plant fluctuations from anywhere around the world at any time. Most researchers in this area from higher education institutions focus on developing web-based virtual control laboratories for distance learning purposes (Shaheen et al., 1998; Overstreet and Tzes, 1999; Yang and Alty, 2002; Srinivasagupta and Joseph, 2003). Recently, some small-scale demonstrators of the Internet-based control have shown some promising results (Halley and Gauld, 1999; Yang et al., 2002). Meanwhile, a few companies are more likely to produce Internet-based control systems as a control device (Cushing, 2000). Furthermore, the design issues for the development of Internet-based process control systems have been identified in our recent work (Yang et al., 2003a). Overall, in most cases, Internet-based control systems have been developed by means of extending discrete control systems, which do not explicitly consider Internet transmission features. For example, the laboratory control system developed by Overstreet and Tzes (1999) achieves the Internet-based control by adding the Internet communication elements between the remote controller and the sampling switches, which maintain the discrete control structure. This structure implies that the Internet communication is no more than a boundary timedelay element to the control system. This assumption is not true, since Internet transmission is characterized by unpredictable time delay and data loss. Therefore, a new control structure and relative elements are required for Internet-based control systems. This paper aims to analyse the features of Internet transmission and build proper control architecture with time delay compensations efficiently to overcome the Internet time delay and data loss. This study is organized as follows: section 2 describes the features of Internet transmission. Section 3 provides a control structure with a tolerant period of time for sampling. Two compensators located at the feedback and feedforward channels are designed in section 4 fully to exploit the benefit of the control architecture. A simulation study is used to assess the performance of the compensators in section 5. Section 6 presents the experimental results and section 7 contains the conclusions.
2.
Internet communication features
Differently from other private transmission media, the Internet is a public transmission media, which could be used by many end-users for different purposes. The major obstacle of Internet-based control systems is how to overcome uncertain transmitting time delay and data loss problems. According to a study of Internet transmission (Luo and Chen, 2000), the performance associated with time delay and data loss shows large temporal and spatial variation. As the results show, the performance decreases with distance as well as the number of nodes traversed. The performance also depends on the processing speed and the load on the network
Yang et al.
105
nodes. Uncertain transmission time delay and data loss problems are not avoidable for any Internet-based application. The reasons why the variable time delay occurs are as follows (Magyar et al., 2001): / /
/ /
network traffic changes all the time because multiple users share the same computer network; routes or paths of data transmission decided by Internet Protocol (IP) are not certain. Data is delivered through different paths, gateways, and networks whose distances vary; large data is separated into smaller units such as packets. Moreover, data may also be compressed and extracted before sending and after receiving; using TCP/IP protocols, when error in data transmission occurs, data will be retransmitted until the correct data is received. Therefore, data loss and time delay in data transmission can be treated identically.
In detail, the Internet time delay is characterized by the processing speed of nodes, the load of nodes, the connection bandwidth, the amount of data, the transmission speed, etc. The Internet time delay Td (k) at the instant k can be described as follows (Han et al., 2001): Td (k)
n X li i 0
M tRi tLi (k) bi C
n X li i 0
M tRi bi C
n X
tLi (k) dN dL (k)
(1)
i 0
where li is the ith length of the network link, C the speed of light, tRi the routing speed of the ith node, tLi (k) the delay caused by the ith node’s load, M the amount of data, and bi the bandwidth of the ith link. dN is a term, which is independent of time, and dL (k) is a time-dependent term. Because of the time-dependent term dL (k), it is somewhat unreasonable to model the Internet time delay for accurate prediction at every instant. Internet time delay must be explicitly handled through control system architecture and/or control algorithms.
3.
Internet-based control architecture
It might be arguable that the conventional discrete control structure, which uses a fixed sampling interval, is not suitable for Internet-based control systems. The discrete control structure requires a predictable execution time for closed control loops. Conversely, Internet transmission time delay is unpredictable, which breaks the foundation of the conventional discrete control structure. For example, an actuator may receive a control signal from a controller after the sampling interval passes. As the result, the control system may lose a number of control signals because of the fixed sampling interval. Therefore, a suitable control structure is required to deal with the uncertain execution time, time delay and data loss for Internet-based control systems. Overstreet and Tzes (1999) added the Internet between the remote controller located at the remote site and two sampling switches located at the local site for both the feedforward and feedback channels. Principally, the traditional discrete feedback control structure is still maintained in their control system. The transmission delays at the feedback and feedforward channels will affect the control system
106
Internet-based process control systems
performance. Since the Internet transmission delay is unpredictable, the sampling interval in the discrete control structure should be not fixed in order to acquire the latest information. For example, the latest control information might arrive the local site just after the sampling time passes. As the consequence, the control system with a fixed sampling interval potentially increases the system time delay. Based on Overstreet and Tzes’s (1999) work, a new Internet-based control system structure is proposed here, in which a tolerant time Dt and two compensators are introduced to handle the unpredictable information transmission delay. The new structure is shown in Figure 1. The functions of the two compensators will be discussed in the next section. The value of the tolerant time Dt is chosen as less than the sampling interval, so that the control law could still be satisfied. Instead of acquiring control signals at a single instant, the new structure allows the data acquisition occur from every instant Ts Dt to every instant Ts Dt, which potentially maximizes the possibility of succeeding the transmission on time. As the result of introducing the tolerant time, a sampling switch pair located on both the remote and local sites replaces a single switch in Overstreet and Tzes’s (1999) work to facilitate the transmission action. The sampling switches are triggered and synchronized by the timers provided by the network services. /
4.
/
Predictive control with time delay compensation
Even though any type of controller, including PID controllers, can be implemented in the new control structure proposed above, the Dynamic Matrix Controller (DMC) (Cutler and Ramaker, 1980) is chosen in the design of the two time delay compensators shown in Figure 1 because DMC is widely accepted in industry. The compensator in the feedback channel with a data buffer is designed to overcome the time delay occurring in the transmission from the local site to a remote site. The compensator in the feedforward channel is designed to overcome the time delay occurring in the transmission from a remote site to the local site. All the data were sent over the Internet with a time stamp generated by a global timer in the Internet-based control system. The receiving time will be compared with the time stamp for each data to see if a delay has occurred or the transmission is normal. Supposed the process is described in the step response model as follows: y(t)
X
gi Du(t i)
i 1
Figure 1 Control structure with time delay compensation
(2)
Yang et al.
107
where y is the process output variable, Du is the increment of the control action, gi is the coefficient of the step response, t is the current time instant. The DMC general control law can be given as (Camacho and Bordons, 1999): u (GT GlI) 1 GT (w f)
(3)
in which l is the penalization factor for the control costs, I is the unit matrix, the superscript T means the transfer of the vector. The system dynamic matrix G is defined as Equation (4), in which the elements are taken from the coefficients of the step response model of the process, g1,. . .,gp shown in Equation (2): 3 2 0 ... 0 g1 6 g2 g1 0 0 7 7 6 :: 6n n n 7 : 7 6 (4) G 6 7 g g g g m m 1 2 1 7 6 : :: 4n n n n 5 gp gp 1 gp m2 gp m1 p m where m is the control horizon, p is the prediction horizon. The control action u is defined as: u [Du(t)
Du(t1) Du(tm 1)]T
(5)
The reference trajectory vector w is defined as: w [w(t1) w(t2) w(tp)]T
(6)
The free response vector f is defined as: f (tp)]T
f [f (t1) f (t2)
(7)
The free response f(tk ) is computed as: /
N X (gki gi )Du(t i) f (tk) ym (t)
(8)
i 1
where ym (t) is the measurement value of the process output variable, N is the process horizon. The reference trajectory w(tk) is computed as: /
w(t) ym (t)
w(tk) aw(tk 1)(1 a)r(tk)
k 1; :::; N
(9)
where a is a parameter between 0 and 1. r(tk) is the setpoint of the controller. /
4.1
Compensation at the feedback channel
The time delay occurring in the transmission from the local site to a remote site causes the controller in the remote site to be unable to receive the feedback signal ym (t) from the local site on time. Once the time delay occurs, ym (t) in Equations (8) and (9) will be
108
Internet-based process control systems
replaced with the predictive value y(t½t); ˆ which is calculated from the step response model in the Equation (2) based on the available measures at the instant t. When the transmission recoveries, the accumulated error e(t) during the period of the time delay is used to compensate the effect. Assuming the period of the time delay is D, consequently, the accumulated error e(t) during the period of the delay is expressed as: e(t)
D X (ym (t i) y(t i½t i)) ˆ
(10)
i 1
The compensation item chosen in this study is the accumulated error multiplied by an adjustable parameter b between 0 and 1. This compensation item is added into the free response f(tk) shown in Equation (8). The history data is stored in the data buffer. Overall, the computation of the free response and the reference trajectory can be summarized as: /
if the time delay occurs 8 N > X > > > f (tk) y(t½t) ˆ (gki gi )Du(t i) > < i 1 w(t) y(t½t) ˆ > > > > w(tk) aw(tk 1)(1 a)r(tk) > : k 1; . . . ; N
(11)
otherwise, if the transmission is normal 8 D N > X X > > >f (tk) ym (t)b (y (t i) y(t i½t i)) ˆ (gki gi )Du(t i) > m < i 1 i 1 w(t) ym (t) > > > > w(tk) aw(tk 1)(1 a)r(tk) > : k 1; . . . ; N
(12)
Equations (11) and (12) form the compensator at the feedback channel.
4.2
Compensation at the feedforward channel
The objective of the compensation at the feedback channel is to reduce the effect of the control signal blanks caused by the transmission delay. In Equation (5) u is a vector composed of the m 1 future control increments. Normally, only the one at the current instant, Du(t), is taken into action, the future control increments from Du(t1) to Du(tm 1) are simply not in use. Therefore, it is possible to use these available future control increments in the situation where the time delay occurs. When the time delay occurs in the feedforward channel, the elements in the control vector u in Equation (5) are shifted one step forward at every sampling interval. Equation (5) becomes to the equation (13), in which the first element is Du(t1), and /
/
/
/
/
Yang et al.
109
the last element is replaced with zero. If the time delay is longer than m 1, u will be a zero vector after delaying m 1 steps, and the system will be in an open control mode. /
/
u [Du(t1) Du(t2) Du(tm 1) 0]T
(13)
Using an out-of-date control signal when a feedback delay occurs is only an empirical solution to the feedback time delay. Ideally, the control action shown in Equation (13) pushes the process in the right direction so that the effect of the control signal blank can be reduced. However, it could lead to serious problems, such as the process becoming unstable because of the mismatch between the predictive model and the actual process, or a great potential process disturbance. These influences can be partly overcome through adjusting the control horizon m. In the worse case, the control horizon canbe set as 1, so that the control signal is maintained at a fixed value, which means that the control system is operated in an open control mode. The following experimental work and simulation will show some proofs for these compensation methods.
5.
Simulation results
The objectives of the simulation study are to investigate the effect of the Internet time delay and data loss on the control system and to evaluate the performance of the two compensators at the feedback and feedforward channels. The major benefit of this simulation study includes: 1) isolating the control issues from the Internet communication; 2) amplifying the frequency of the Internet time delay and data loss; and 3) providing an identical circumstance for the evaluation. The setpoint step change and the step disturbance have been introduced in the simulation in order to assess the control system performance.
5.1
Design of the simulation
The simulation study is conducted in the Matlab/Simulink environment. The system structure is shown in Figure 2, including two compensators, two delay elements, a DMC and a process model. The process model describes a water heater (Camacho and Bordons, 1999), where the cold water is heated by means of a gas burner. The output of the model is the outlet temperature of the water heater, the input of the model is the gas flow-rate, which adds the energy to the water through the gas burner. The sampling interval Ts is chosen as 1 s; the tolerant time Dt as 0.3 s. In order to provide an identical simulation circumstance for various simulation tasks, an identical time delay pattern was employed at both the feedback and feedforward channels, which was randomly generated. Figure 3(a) illustrates the feedback delay pattern; Figure 3(b) shows the feedforward delay pattern. The maximum time delay is set as 10 s. In the simulation study, the prediction horizon p is chosen as 10; the control horizon m is 5; the reference trajectory parameter, a is 0.7; the parameter b is 1. The setpoint step change is set as 1, and the step disturbance as 0.5. There are four proposed scenarios: there is no any time delay or data loss; the time delay and data loss only exist at the feedback channel; the time delay and data loss
110
Internet-based process control systems
Figure 2 Simulation structure
only exist at the feedforward channel; and the time delay and data loss exist at both channels. Corresponding to the real world, the first scenario refers to the ideal situation; the second refers Asymmetrical Digital Subscriber Lines (ASDL) communication and the feedforward channel obtains a high bandwidth; the third refers to ASDL communication and the feedback channel obtains a high bandwidth; and the last refers to the symmetrical communication.
5.2
Simulation result
Figures 4, 5 and 6 give the simulation results for the above four scenarios. The scenario with no delay and no data loss is used as a reference in the other three scenarios. Comparisons have been made between no time delay, time delay without compensation and time delay with compensation in each scenario. Figure 4(a) and (b) shows the process response and the control signal for the scenario, in which the time delay and data loss only exist at the feedback channel. The first disturbance is caused by the setpoint step change, the second by the step disturbance. Since the simulation circumstance is kept being identical in all the simulations, it is assumed that the differences in the responses are purely contributed by the feedback delay and data loss, and the employment of the compensator.
Figure 3 Input and output delay pattern; (a) feedback, (b) feedforward
Yang et al.
Figure 4 (a) Comparison of feedback delay effect; (b) comparison of feedback delay effect
111
112
Internet-based process control systems
Figure 5 (a) Comparison of feedforward delay effect; (b) comparison of feedforward delay effect
Yang et al.
Figure 6 (a) Comparison of feedback plus feedforward delay effects; (b) comparison of feedback plus feedforward delay effects
113
114
Internet-based process control systems
Figure 4(a) shows that the feedback compensator can significantly reduce the influence caused by the step change in the setpoint and the step disturbance even though there are still some difference between that with the compensator and that with no time delay. Figure 4(b) shows that the control signal is in a reasonable range. Similarly, Figure 5(a) and (b) gives the process response and control signal for the case, in which the time delay and data loss only exist at the feedforward channel. The time delay and data loss causes the process variable oscillation, which might potentially lead the process towards instability. The effect of the time delay in the response to the compensator has been dramatically reduced in the setpoint step change and the step disturbance. However, the effect has not been fully compensated for. Figure 6(a) and (b) represents the process response and control signal for the case, in which the time delay and data loss exists at both channels. As expected, outputs become worse (more oscillation and overshoot), although they finally reach a stable value. The response to the compensators maintains the performance at an accepted level. Comparing Figures 4(b), 5(b) and 6(b), it can be seen that more control energy is required in this case. Comparing Figures 4(a) and 5(a), the effect of the time delay at the feedforward channel is greater than that at the feedback channel, and is more difficult to be compensated for. Therefore the feedforward channel might need a high bandwidth. It matches the current situation in most computer networks.
6.
Experimental results
To evaluate further how well the proposed predictive control algorithm with time delay compensation and the control structure with a variable sampling interval cope with the Internet communication features, a water-tank teaching rig in our process control laboratory is used as a case study. Experiments with and without the time delay compensation have been carried out on this teaching rig.
6.1
Experiment design
The process to be controlled through the Internet is a water tank in the process control laboratory at Loughborough University. The control target is to maintain the liquid level of the water tank at a desired value. The experimental system layout is shown in Figure 7. The tank is filled by the inlet flow controlled by a hand valve and is emptied into a drainage tank through a connection pipe and a pump. In order to maintain the liquid level of the tank at a desired value, a local PID controller controls the outlet flow. The predictive controller is located at the remote control system, which is deployed on a laptop computer, to change the setpoint of the local PID controller. The DAQ instrument is used to gather the liquid level signal from and send a control command to the water tank. The local control system of the tank and the DAQ instrument are connected and wired through an RS-232c serial port. Through the serial cable, the real-time data is exchanged between the local control system and the instrument. The local control system acts as a web server and a video server as well. A web camera provides visual information to users. Because the web camera is independent
Yang et al.
115
Figure 7 Experimental system layout
from the DAQ, it is considered an extra sensor. The web server provides the Internet service (IIS 5.0) and establishes connections between the remote control system and the local control system. The remote control system is connected to the Internet through a telephone modem providing a 33.6K bandwidth. Because of the limited transmission capability of the modem, Internet congestion and time delay is often encountered, which is exactly the experimental circumstance we expected for the remote predictive controller with time delay compensation. The step response model of the tank is obtained by applying a step change in the setpoint of the local PID controller. In order to evaluate the entire condition of the controller performance, the setpoint of the remote predictive controller is driven by a square wave with the wave centre at 50%, which is the desired value of the liquid level of the water tank. The setpoint driven by the square wave is shown in Figure 8 in the solid line. For the control parameter, the prediction horizon p is 10; the control horizon m is 5; the reference trajectory a is 0.7. Three categories of the experiments have been carried out, one without involving the Internet (illustrated by the line with crosses in Figure 8), the second over the Internet but without employing time delay compensation (illustrated by the dashed line in Figure 8), the third over the Internet and with the implementation of time delay compensation (illustrated by a dotted line in Figure 8). The one not involving the Internet was carried out in the local network and is used as a standard reference for comparison, in which the network communication effect can be completely ignored. In order to reflect the controllers’ performance and the process reaction, process data is directly acquired from the DAQ before the Internet transmission.
116
Internet-based process control systems
Figure 8 Comparison of typical performances
6.2
Experimental results and analysis
The typical results of the three experiments for the step response are shown in Figure 8. It is obvious that the one with the compensation over the Internet has less overshoot and faster response to the change in the setpoint than that without the compensation over the Internet. The experimental statistical results are summarized in Table 1. Every catalogue of the experiments includes three elements: the average transmission time, the standard deviation and the data loss. The average transmission time indicates the sum of the execution times at the local control system and the remote control system, and the data transfer time from the local site to the remote site and vice versa. The standard deviation represents the dispersant degree. The data loss records the times of the transmission failure, which could be caused by the transmission data loss and/or the transmission timeout. Concerning the second and the third experiments, the high standard deviation values (327.17 ms and 377.92 ms respectively) for both Internet transmissions illustrate the existence of the unpredictability of the Internet transmission. The average transmission times (494.32 ms and 555.60 ms) give the order of the Internet time delay. The control transmissions in the two experiments over the Internet (the second and the third) have 13 and 20 times of data loss. The Internet circumstance for the one with the compensation (the third experiment) is worse than that without the compensation (the second experiment). Several criteria can be used to evaluate the controller performance such as IAE (Integral Absolute Error), ITAE (Integral Time and Absolute Error) and ISE (Integral Square Error). Only the ISE criterion is employed here. As shown in Table 1, the ISE value increases from 9561.9 to 10923 because of the Internet time delay if the
Yang et al.
117
Table 1 Comparison of the experimental results Experiments
Traditional controller/local network communication Non-compensation controller/ Internet communication Compensation controller/ Internet communication
Average transmission time (ms)
Standard deviation (ms)
Data loss (times)
Integral Square Error (ISE)
11.95
23.85
0
9561.9
494.32
327.17
13
555.60
377.92
20
10923 9957.7
compensation is not employed. However, the third experiment shows that even in a worse circumstance compared with the second one, the ISE value of the experiment with the compensation (9957.7) is still less than that of the experiment without the compensation. Therefore, it empirically shows the efficiency of the compensation. Unfortunately, the advantages of the variable sampling interval introduction are difficult to determine in this experiment.
7.
Conclusions
With the prevalence of the Internet, process managers and operators have being offered a virtual global control room for their process plants. This paper contributes to a way of overcoming the Internet time delay and data loss for Internet-based process control systems. Since the Internet time delay is affected by the number of nodes and the transmission loads, it is variable and unpredictable. In this paper, it has been found that the traditional discrete control architecture is not suitable for an Internet-based control system due to the data transmission delay. A control structure with a variable sampling interval has been proposed to offer a solution for the Internet-based control system from the architecture aspect. Taking advantage of the new control structure, two compensations for the DMC have been developed to compensate for the effect of the Internet transmission time delay at the feedback and feedforward channels. This new control strategy with the time delay compensation has shown a promising way to compensate efficiently for the effect of the Internet time delay on the control performance by reference to a simulation and experimental work. Another interesting finding from our simulation results is that the feedforward time delay and data loss seem to cause more serious influences on the control performance, and are more difficult to compensate for in the controller design. Therefore, the feedforward channel is desired to give a high bandwidth if possible. Internet-based process control systems have extremely large potential application areas, even though the current applications are mainly focused on virtual control laboratory and distance learning. Internet-based process control systems are suitable for highly geographically distributed devices or processes, such as windmill power stations, small-scale hydroelectric power stations, and food manufacturing and logistic warehouse energy control. In our recent work (Yang et al., 2003b) Internet-based
118
Internet-based process control systems
process control systems have been integrated into a remote support system for process plants over the Internet, in which Internet-based process control systems are used remotely to monitor, adjust and maintain real-time control systems located at a local site through the Internet. This virtually eliminates the need for any control software supplier’s expert to conduct onsite maintenance. Therefore time and money can both be saved. New developments in various Internet technologies, such as real-time communication protocols and data encryption algorithms for Internet security, will definitely bring Internet-based process control systems into process industries.
Acknowledgements The contribution is part of the work of the EPSRC (Grant No. GR/R13371/01) funded project ‘Design of Internet-based process control’. The authors are grateful to Drs David W. Drott and David W. Edwards for their collaboration in the experimental work.
References Camacho, E.F. and Bordons, C. 1999: Model predictive control . London: Springer-Verlag. Cushing, M. 2000: Process control across the Internet. Chemical Engineering May, 80/ 82. Cutler, C.R. and Ramaker, B.C. 1980: Dynamic matrix control / a computer control algorithm. Automatic Control Conference , San Franciso. Halley, A. and Gauld, P. 1999: Integration where Java meets process control. Control and Instrumentation 31, 57/58. Han, K.H., Kim, S. Kim, Y.J. and Kim, J.H. 2001: Internet control architecture for Internet-based personal robot. Autonomous Robots 10, 135/47. Luo, R.C. and Chen, T.M. 2000: Development of a multibehaviour-based mobile robot for remote supervisory control through the Internet. IEEE Transactions on Mechatronics 5, 376/85. Magyar, B., Forhecz, Z. and Korondi, P. 2001: Observation and measurement of destabilization caused by transmission time delay through the development of a basic telemanipulation simulation software. IEEE WISP, 125/30. Overstreet, J.W. and Tzes, A. 1999: An Internet-based real-time control engineering la-
boratory. IEEE Control Systems Magazine 19, 320/26. Shaheen, M., Loparo, K.A. and Buchner, M.R. 1998: Remote laboratory experimentation. American Control Conference , Philadelphia, PA, 1326/29. Srinivasagupta, D. and Joseph, B. 2003: An Internet-mediated process control laboratory. IEEE Control Systems Magazine 23, 11/18. Yang, S.H. and Alty, J.L. 2002: Development of a distributed simulator for control experiments through the Internet. Future Generation Computer Systems 18, 595/611. Yang, S.H., Chen, X., Edwards, D.W. and Alty, J.L. 2002: Development of an Internet-based process control system. In Grievink, J. and Van Schijndel, J., editors. Computer-aided chemical engineering. ESCAPE-12, The Hague: pp. 601/606. Yang, S.H., Chen, X. and Alty, J.L. 2003a: Design issues and implementation of internet-based process control systems. Control Engineering Practice 11, 709/20. Yang, S.H., Chen, X. and Yang, L. 2003b: Global support to process plants over the Internet. In Chen, B. and Westerberg, A.W., editors. Computer-aided chemical engineering , PSE03, Kunming, China, 15, 1399/ 404.
