1.2
Computer Configurations of Supervisory Units J. W. LANE, C. S. LIN, T. H. TSAI
(1995)
S. RENGANATHAN
Types of Software Packages:
A. Control system design B. Process monitoring and control C. Process optimization D. Statistical process control
Partial List of Software Suppliers:
Allen-Bradley (B) (www.ab.com) DMC (C) (www.dmc.com) Gensym Corp. (D) (www.gensym.com) Icom (A) (www.icom.net.au) Iconics (B) (www.iconics.com) Setpoint (C) (www.setpointusa.com) T/A Engineering (B) (www.ta-eng.com) U.S. Data (B) (www.usdata.com) Wonderware (B) (www.wonderware.com)
INTRODUCTION The first use of digital computers was for acquiring data from the plant and logging that data without influencing plant operations. These systems were called data acquisition systems (DAS). Supervisory control evolved, as the information collected by DAS was processed in order to generate optimal set points for the individual control loops. These set points were first sent to existing analog controllers and later to the digital control algorithms within DCS systems. This method of operation is called supervisory control, which these days often also incorporates optimization functions. The discussion in this section will first cover the design when supervisory computers are operating with controllers, followed by a description of supervisory computers in combination with DCS systems.
HISTORY OF SUPERVISORY CONTROL In 1957, the Ramo–Woodridge Corporation introduced the RW 300, the first solid-state digital computer designed for process control, at the National Meeting of the American Institute of Chemical Engineers in Los Angeles. The computer was designed based upon a magnetic drum containing 8064 words of 20-bit length. The external input/output supported 540 one-bit digital inputs, 1024 analog inputs, 540 digital outputs, and 36 analog outputs.
(2005)
The RW 300 was designed to interface directly with the existing analog control system of a process. The role of this system was an enhancement to the existing process control instrumentation, and it functioned as a supervisory control device. The software program consisted of models of the process in varying degrees of sophistication, which were updated based upon real-time data and then perturbed using optimization or maximization routines. These were usually simple linear programming algorithm perturbation routines. The models provided new sets of operating conditions (set points), which would improve the process performance according to some performance criteria. These new conditions were then outputted to the process. Essentially, the process was moved from one set of steady-state conditions to another in response to changes in such uncontrolled disturbances as feed characteristics, ambient conditions, heat transfer coefficients, and catalyst activity. In spite of the seemingly small memory capacity of the RW 300 as compared to today’s standards, the RW 300 was successfully installed in a number of sophisticated chemical and refinery processes. The very first process control system to be successfully installed was on the Texaco phosphoric acid polymerization unit at Port Arthur, Texas. Within the next few years, several companies introduced their versions of the process control computer. These companies included IBM, General Electric, Foxboro, Control Data Corporation, and many others. The first process control computer to perform dynamic regulatory control of a process was installed 15
© 2006 by Béla Lipták
16
General
at the Riverside Cement Company’s Victorville plant in the United States and controlled two cement kilns and raw feed blending. The first improvement in the field of the process control computer was an increase in magnetic drum capacity. Memory size first doubled to 16K, then 32K, and then kept rising. The next improvement in the process control computer design was the introduction of high-speed core memory, which increased the computational speed from milliseconds to microseconds. A giant leap in technology came with the replacement of solid-state circuitry (diodes, resistors, and transistors) with silicon integrated circuits. The new technology offered greatly increased speed, capacity, and most of all, reliability. Another chapter in the evolution of the process control computer came with the development of direct digital control (DDC), wherein the function of the analog instrumentation was incorporated in the computer, and the analog controllers were thereby eliminated. The motivation was to reduce the control system cost and increase design flexibility for new processes. This new design concept initially met with mixed reviews, primarily because the computers — especially the software, which had become complicated — were not sufficiently reliable, and in most cases, when the computer went down, so did the process. However, as minicomputers began to replace the old process control computers, it became economically feasible to provide redundancy by installing dual computers. This was frequently done. Then a major advancement came in 1976 when Honeywell announced the first distributed digital control system (DCS), named the TDC 2000. The hallmark of the system was reliability based upon redundancy in microprocessor-based controllers, redundancy in communications, and redundancy in operator interface.
Generic Features of DCS Currently, most large-scale process plants such as oil refineries, petrochemical complexes, and various other processing plants are controlled by microcomputer-based DCS (distributed control systems). As is discussed in detail in Chapter 4, these systems generally include the following features: 1. Cathode ray tube (CRT)-based operator consoles and keyboards, which are used by plant operators or engineers to monitor and control the process 2. Controllers, multifunction control modules, and programmable logic controllers (PLCs), which provide the basic control computation or operation 3. A communication network, which is used to transfer the information between control modules and operator consoles across the node on the network 4. I/O (Input/Output) modules, which are used to convert the field instrumentation signals from analog to digital and digital to analog form for controller modules and console displays 5. Fieldbus communication links, which are used for communication between remote I/O devices and control modules 6. Historical module, which is used for data storage for pertinent process data or control data and for online data retrieval or archiving 7. Computer interface, which is used for communication between the nodes on the DCS network and the supervisory computer 8. Software packages for real-time monitoring, control, reporting, graphics, and trending The generic arrangement for these components is shown in Figure 1.2a. This arrangement is a natural outcome for
Operator console
Operator console
Operator console Printer
Communication network Historical data
Control room Controller
Controller
Supervisory computer
Controller
Field bus Field
Process
FIG. 1.2a Generic features of DCS.
© 2006 by Béla Lipták
I/O Module
I/O Module
I/O Module
Robot interface
1.2 Computer Configurations of Supervisory Units
large-scale process control because the parallel operations of the control tasks are required for both continuous and batch process control. Therefore, the distributed control is a natural way of satisfying the parallel nature of the process from a geographical point of view. DCS replaces conventional control systems to perform basic control. The supervisory control functions that in the past were performed by the computer systems described above, plus additional functions such as online information evaluation, are usually still performed by a supervisory computer, which is linked to a DCS.
•
•
COMPUTER-SUPERVISED ANALOG CONTROLLERS The computer’s capability to perform complex mathematical calculations and to make logical decisions provide a unique opportunity for improving the performance of any process through the application of supervisory control. All processes are affected by disturbances; some can be measured or calculated, others are not. Examples of disturbances are feed composition and concentration variations, ambient conditions, drifts in catalyst activities, heater and heat exchanger fouling, and economics. Functions Supervisory control has three fundamental functions: 1. Calculation of the present state of the process in terms of key process parameters such as yield, severity, and efficiency. 2. Calculation of critical process constraints, such as compressor capacity, column flooding limits, explosion limits, vacuum condenser capacity, and raw material. 3. Calculation of new process conditions, in terms of set points, which will meet the requirements of the objective function. The objective function can be one of a number of goals such as maximizing throughput, yield, and profit or minimizing deviation. Tasks The supervisory control computer requires: 1. 2. 3. 4.
A process model Cost functions Optimization algorithms Constraints and limits
The supervisory computer typically performs the following types of tasks:
•
17
in creating the required process models. Because of the complex nature of the process itself, the computation of the model is a difficult task and might require iterative computation to satisfy convergent criteria. Determines the present operating state of the process based on the online, real-time information based on temperatures, pressures, and feed characteristics to obtain reactor yields or determines the desired state according to the constraints and optimization criteria. Determines the optimum control strategy based on the online, real-time information to achieve the control command by adjusting the manipulated variables or set points at the DCS level. Once such an algorithm is used in the supervisory computer, the response of the process to control strategy commands will gradually move this process from an initial state to a final desired state, avoiding the process constraints and following the optimum path such that one of the following objective functions is obtained: minimum cost, minimum energy, maximum yield, or maximum profit. Predicts impending alarms based on rigorous mathematical models using present and past history, process data, and control commands. Anticipating alarm conditions in advance of the process reaching these conditions is a vital function of supervisory control. Proper grouping of the pertinent process variables in a preclassified manner allows the severity of the alarms to be identified and allows actions to be pinpointed that can then be initiated quickly by operators.
Supervisory Computer for Analog Loops The functions of a supervisory computer control system are shown in Figure 1.2b. It controls four process loops, but can be extended to as many loops as needed. Typically, 50 loops are assigned to one computer. In this configuration, if analog controllers are being supervised, the proportional, integral and derivative (PID) settings of the analog controllers are kept intact. They are not changed. The process variables are measured, and their transmitted measurements are sent not only to the analog controller but also to the analog to digital (A/D) converter of the multiplexer. These signals are processed by the computer, which takes into consideration the optimal control algorithm, the external constraints, and the management decisions, while generating the set points for the corresponding analog controllers. These set points are sent back to their corresponding controllers through the D/A converter of the demultiplexer. This operation is carried out for all the loops in a sequential, cyclic fashion on time-sharing basis. Process Models
•
Determines the process operating constraints, such as a column flooding condition in distillation or a surge condition of a compressor. Basic material balance, energy balance, or heat transfer calculations are utilized
© 2006 by Béla Lipták
The model of a process can be a simple first-order linear one or a more complicated multivariable nonlinear model. It is essential to obtain a fairly accurate process model, and for
18
General
CV
CV
C
D/A DEMUX
CS
P
P
T
T
Computer
C
CS EOC SOC
A/D MUX
SP
SP C
T
P
T
C
P CV
CV Legend P –Process C –Analog controller CV –Control value T –Transducer/Transmitter
CS SOC EOC A/D D/A
– Channel select – Start of conversion – End of conversion – Analog to Digital – Digital to Analog
FIG. 1.2b Block schematic of supervisory computer control for four loops.
many chemical processes such as distillation or boiler operation, obtaining an accurate process model is a difficult task. In some cases, particularly in old, existing plants, obtaining an accurate and complete model is not possible. When conventional methods of obtaining a model fail, heuristic models are advised. Some of the heuristic models include the fuzzy logic–based model, the neural network–based model (see Section 2.18) and the genetic algorithm–based neural models. These models can be obtained by mapping a set of inputs and their corresponding outputs from the process. There are also neural or fuzzy controllers, which can generate their own optimal control algorithms. The behavior of these nonconventional models and controllers is similar to that of humans; these controllers are well suited for controlling complicated and nonlinear processes. Conventional Models Process models can be obtained by conducting a process reaction test. Many higher-order processes have over-damped responses. Such a response is similar to the response of a first-order with dead time process. In other words, the higher-order process can often be approximated by models that are first order plus dead. If there are oscillations in the system, then it can be approximated by an under-damped second-order + dead time model. The parameters of the dead time and the first- or
© 2006 by Béla Lipták
second-order approximation can be experimentally deter1 mined. The different methods available are impulse response methods, step response methods, and frequency response 1 methods. For multivariable systems, if one can identify all the inputs and outputs and if they are measurable, then the least square method can be used to obtain a state model of the process. This will be an offline model or post-flight model. A number of experiments can be conducted for a sufficiently long time to obtain the input and output data, and those values 2,3 are used in the least squares (LS) algorithm. In systems where the parameters are likely to change with time, it is preferred to have online estimation of the system parameters. The recursive least square algorithm, discussed in References 2 and 3, can be used. Optimization The main purpose of supervisory control is to obtain optimal performance of the overall process. When the supervisory computer works out and generates the set points for individual local controllers, coordinating the operation of the individual loops is an important task. The overall process performance will only be optimal if all the equipment constraints are taken into consideration and raw material availability, market needs, and management requirements are all satisfied. When the supervisory computer generates the set points for individual analog controllers, it is important that these controllers be properly tuned. The tuning of PID and other controllers for various criteria are discussed in detail in Chapter 2 and are summarized in Table 1.2c. In 1/4 decay ratio response, the oscillations of the controller output are such that for a small disturbance the second positive peak is one-fourth of the amplitude of the first positive peak. The PID controller settings are tuned to provide this output response. In the integral tuning criteria listed in Table 1.2c, the integral values are minimized. COMPUTER-SUPERVISED DCS SYSTEMS Most of the DCS software has evolved from the classical analog control. Therefore, many sophisticated supervisory control functions, such as calculating operating constraints for process units or determining heat transfer coefficient trends for maintaining heat exchangers, cannot be easily performed at the DCS level due to hardware or software restrictions. A review of supervisory control techniques and strategies is provided later in this section. Production Monitoring and Control Production monitoring and control for a given operation may include any or all of the following functions: • • •
Order entry/assignment Scheduling Production reporting
1.2 Computer Configurations of Supervisory Units
TABLE 1.2c Criteria for Controller Tuning 1. Specified Decay Ratio, 1 Usually /4
Decay ratio = ISE =
3. Minimum Integral of Absolute Error (IAE)
IAE =
4. Minimum Integral of Time and Absolute Error (ITAE)
ITAE =
• • • •
∞
∫
2. Minimum Integral of Square Error (ISE)
second peak overshoot first peak overshoot
[e(t )]2 dt ,
0
where e(t) = (set point process output)
∫
∞
|e(t )| dt
0
∫
∞
part of a standard DCS function. In most cases, such systems do not have sufficient capacity to meet users’ needs, and a computer-based data archiving system is necessary. Similarly, some DCS packages do not have sufficient storage capacity nor processing speed for processing a large quantity of data, and they normally do not support the desired database formats required by many users. In addition, many computer systems are supported by huge quantities of standard software packages, making the implementation of a realtime information system affordable. Many of the standard packages are not available in the DCS environments.
|e(t )| t dt
0
Quality measurement and control by statistical process control or statistical quality control (SPC/SQC) Warehouse and shipping management Inventory management Customer service (order status, etc.)
Traditionally, these functions were either performed by manual operation or by the application of management information systems (without the availability of online data). As computers became more powerful and less costly, it became technically feasible to include many or all of the above functions in an integrated online information and control system that includes several computers linked by one or a number of sophisticated communication networks. Because of the increasing pressure from worldwide competitions, these integrated systems, sometimes called CIM (Computer Integrated Manufacturing), are rapidly becoming a necessary tool for many major operations. One example is for a mature company to allocate each order to one of its three refineries in an optimum manner. In order to do this task correctly, the most current information must be available from all three refineries, including all processing unit loads and efficiencies, maintenance schedules, and inventories. CIM systems are complicated and will not be discussed in detail in this section. However, it is important to note that a supervisory control computer may be only a small part of a network system, and “system integration” can be an important added task, demanding both technical and managerial attention. Online Information System An online information system may include any or all of the following functions:
SUPERVISORY CONTROL TECHNIQUES One of the most important jobs for an engineer responsible for implementing supervisory control is to make sure that the computer functions and DCS functions support each other.
Supervisory Control Algorithms As was discussed earlier, before the DCS became popular in process industries, supervisory control was used to command single-loop analog controllers. These controllers served to achieve certain limited goals, such as to obtain a uniform temperature profile for a given multi-pass furnace or to determine the optimum blending for gasoline products. The functions that the analog controllers could not accomplish were more-or-less delegated to the digital computer to perform. These included the necessary logic, sequence, or analytic computations to improve the process control operation. From this experience, the practical value of writing the control algorithm in velocity form becomes clearer. Coordinated supervisory control with integrated feedback or feedforward combined in a velocity control algorithm can be activated or deactivated without creating perturbations to the process. The use of velocity algorithms in a computer-supervised DCS system is illustrated in Figure 1.2d and is explained below. Advantage of Velocity Algorithm The velocity algorithm implemented in the supervisory computer can be converted to the position algorithm, which is usually applied at the DCS level. The control command, u, in the supervisory computer at the current sampling time, n, can be expressed as follows: u(n) = Sat [u(n − 1) + ∆]
• • •
Data collection, checking, and verification Data reconciliation Data storage and retrieval
Many DCS systems include only limited online information functions, perhaps providing, for example, a history module as
© 2006 by Béla Lipták
19
1.2(1)
where ∆ = incremental output from computer command; u(n) = current computer position command at sampling instant n; u(n – 1) = past sampled position command; Sat(x) = upper limit if x ≥ upper limit, = lower limit if x ≤ lower limit, = x if lower limit < x < upper limit.
20
General
Upper bound flag at valve
Supervisory computer +
DCS
Gateway
SP +
SP + −
Data table
SP (N-1) Incremental output from supervisory control
Ethernet
Control network
SP(N) PV
∆>0
AND
∆≤0
AND
OR
PID
Control module
Flow loop
0 OR 1 +
0 1 Switch
0 OR 1 Max +
Min
Preassigned value From other sources
Selection logic
The incremental value can be the combination of feedforward or feedback as follows:
∑ k ∆ f b = ∑ k ∆ ff i
i
i
j
1.2(2)
j
Setpoint Output 0 1 Switch
0 1 Switch 0 OR 1
∆=
Selection Upper logic Loop
Lower bound flag at valve
∆PID 0
FIG. 1.2d Velocity algorithm for supervisory control.
ESD flag
Selection logic z−1
Switch 0 OR 1
Initialized value
FIG. 1.2e Supervisory control with antiwindup.
j
where ∆f bi = incremental value from the i feedback loop; th ∆ ffj = incremental value from j feedforward loop; ki = assignth th able constant for i loop; kj = assignable constant for j loop. The advantage of this algorithm for initialization of the control output command from the supervisory computer is that initialization can be obtained simply by setting ∆ equal to 0 and making u (n − 1) equal the value obtained from DCS (for example, a set point value for a slave loop). th
This antiwindup algorithm can be written by modifying Equation 1.2(1) as follows: u(n) = Sat [u(n − 1) + (1 − sw) * ∆]
1.2(3)
where sw = 1 if antiwindup logic is true and sw = 0 if antiwindup logic is false. The switching logic is defined as: I + = N + (∆ > 0)
Antiwindup Control Strategy For supervisory control, the command from the supervisory computer to the set point for the control loop at the DCS level is essentially a form of cascade control. Caution is needed when control loops are in cascade. The upper or cascade master control command (supervisory computer command) requires information from the lower or secondary loop at the DCS level in order to command correctly. This information includes valve saturation status and lower loop current set point value at the DCS. This information, which resides at the DCS level, requires constant monitoring by the supervisory computer of the realtime information communicated through interface modules on the DCS network. It is very important to prevent the computer command from causing set point windup at the DCS level. The block diagram for antiwindup is shown in Figure 1.2e. This algorithm protects the upper loop command from windup. This protection is accomplished by freezing the supervisory control command output to the DCS set point at its last value if the DCS output to the valve is saturated and if the direction of the upper loop output increment would cause further valve saturation at the DCS level. Otherwise the upper loop within the supervisory computer will not be frozen at its last value.
© 2006 by Béla Lipták
I − = N − (∆ ≤ 0) sw = (I + .AND. Iu ) .OR. (I − .AND. I1) where N + (∆ > 0) = 1
if
=0 N − (∆ ≤ 0) = 1
if ≤ 0 if ≤ 0
>0
= 0 if > 0 Iu = 1 if valve8 position reaches upper bound = 0 if valve position at upper bound is not reached I1 = 1 if valve position reaches lower bound = 0 if valve position at lower bound is not reached Combined Discrete and Continuous Algorithm For industrial process control, the control algorithm, if arranged in a structured manner, can be implemented by using a user-friendly configuration arrangement instead of writing program code for each special algorithm. Such an arrangement can be extremely useful to facilitate the implementation of supervisory control by control engineers. Through the experience gained using several DCS systems implemented at the supervisory computer level, the generalized
1.2 Computer Configurations of Supervisory Units
Selection logic
Selection logic Mental value 0 OR 1
0 OR 1 +
Switch
+
Output
Min
Preassigned value From other source
Switch
Switch 0 OR 1
Selection logic
ESD flag
Max
Switch 0 OR 1
Selection logic
There are typically hundreds of display page screens provided in both DCS and the supervisory computer. These operational display pages can be easily accessed through the keyboard, touch screen, mouse, or trackball to arrive at the destination screen for display. The display access method and the display architecture should be similar for both the supervisory computer and the DCS screen displays. The supervisory control monitor and the operator console at the DCS level should have consistent operating procedures and should be easy for the operators to use.
z−1 Initialized value
FIG. 1.2f Generalized output control algorithm.
output control algorithm shown in Figure 1.2f was found to be useful and practical. It is designed to satisfy the application of control output for batch or continuous process control and is combined with emergency shutdown logic. Algorithm Initialization For supervisory control, the upperlevel (supervisory) control algorithm must be initialized when the output from the supervisory computer commanding the lower loops in the DCS is initiated or when an emergency process condition requires that the supervisory control command be prevented from acting in an unrealistic manner. Because the control algorithm is structured in the velocity algorithm form, initialization is simple and universal. That is, ∆ = 0, and u(n − 1) = initial value from destination loop at the DCS level. System Failure When the supervisory computer fails, the control commands generated at the computer level cannot be transferred to the DCS level. The designer of the integrated control system should utilize a watchdog timer or communication link message to flag the lower DCS loops as a precautionary step, to prevent unnecessary disturbance to the process due to supervisory computer failure. The intent of the supervisory control command should be designed so that fault-tolerant control can be achieved in the event that the computer fails. One approach, which can be utilized to achieve fault tolerance, is to implement the generalized output control algorithm at the DCS level, as shown in Figure 1.2f. Proper manipulation of the switch blocks and proper use of the “computer failed” flag at the DCS level ensures that the process control actions at the DCS level will not be perturbed by a computer failure.
Display Access Method The access methods should be designed such that a minimum number of keystrokes are required to arrive at the destination display, where the controller of interest resides. One convenient access method is to use a tree-structured, multilevel display architecture. This is shown in Figure 1.2g. For example, if a two-level structure is used for 16 elements (pages) in each level, then the total number of pages of screen will be 16 × 16 = 256 pages. In such case, each screen page at the second level can cover 16 elements, and each element can be any digital or analog input/output combination. Therefore, a total of 4096 combined I/O points can be displayed at the second level. If the correct search path is selected, only two keystrokes are required to arrive at the destination controller for operation. Display Feature for Supervisory Control Traditional analog controllers on the panel board are arranged in cluster of rows or columns. In the DCS, controllers are usually arranged in groups of 8, 12, 16, or 32 for screen displays in a onedimensional horizontal arrangement or a two-dimensional matrix arrangement (Figures 1.2i and 1.2j). However, if the same type of the group display for DCS is used for a supervisory control application, the information that resulted from the supervisory computer calculations cannot be properly displayed to inform the operators. A typical example is the multi-input/multi-output control commands from the supervisory computer to the lower loops in DCS. This type of control algorithm does not exactly fit
Level 1 1
DCS and Supervisory Computer Displays In order to judge whether or not the supervisory computer or DCS is performing properly, the displays on the CRT screen become the only man-machine interface that can be utilized for performance evaluation.
2
16
3
Level 2 1
© 2006 by Béla Lipták
21
2
16 Level 3 1
2
FIG. 1.2g Multilevel display architecture.
n–1
n
22
General
Input sources TAG. 1 TAG. 2 TAG. 3
TAG. N TAG. X
TAG.OP1 TAG.OP2 TAG.OP3
Multi-input/multi-output control strategy and live data
Operator console input
TAG. Y TAG. Z Output destination
Output TAG. A TAG. B TAG. C destination
FIG. 1.2h Multi-input/multi-output strategy display.
the conventional single-input, single-output structure. For this reason a new type of display for the multi-input/multi-output control display structure should be created. This display would indicate the source and destination of information for different locations, magnitude, type of process signals, etc. One of the simplest ways to represent the information for supervisory strategies is to create a multi-input/multi-output control strategy display. This display is shown in Figure 1.2h. Alarm Access Architecture Several types of alarms exist for each process variable from the DCS, including high, low, high-high, low-low, and high rate of change. These alarms will appear on the alarm screen at the DCS operator’s console in chronological order with tag names, time of occurrence, magnitude of alarmed variable, alarm limit, and so on. The alarm buffer can be created to accommodate the burst of alarms occurring in a short period of time. For supervisory computers, the alarm management program should be created differently in order to speed the alarm access and to assist the alarm diagnostic operation (to help the operators pinpoint alarm cause and location). To do so, the priority classification or grouping of the pertinent process alarm information is required. Tag names belonging to an important process unit or equipment should be grouped to give a group alarm flag so that any individual alarm or alarms within that group can be identified logically. This will help speed the decision making of the operator when time is critical. The alarm access architecture can also be grouped according to geographical regions of the process plant or unit so that field operators can be informed (by the control room console operator) to take precautionary measures when necessary. Voice Input Man-Machine Interface In process control, voice input and output can be an important method for conveying pertinent information about real-time process conditions to the operator. The cause or location of alarms can be announced without viewing the screen display, which clearly can be an
© 2006 by Béla Lipták
advantage when the operator is away from the console. When process variables are approaching an alarm limit, voice output can be generated to alert the operator of the impending alarm condition. Because of the parallel nature of process control activities, the input methods for man-machine interface should be improved to take advantage of both hands and feet to speed up the operational maneuver. Advanced Control Strategies Advanced control, as it is defined here, refers to the multiinput/multi-output control algorithm. The computations usually involve a combination of continuous and discrete algorithms. These algorithms can be implemented either at the DCS or at the supervisory level or at a combination of both of these levels. Simple PID control algorithms are usually implemented at the DCS level. Those algorithms are structured with singleinput/single-output characteristics. Robust multi-input/multioutput algorithms are often implemented at the supervisory computer level and should be carefully structured to satisfy multi-periodic sampled data control requirements. Advanced Control at the DCS Level Advanced control strategies may fit into the software architecture of DCS systems with sophisticated controllers. Such control modules usually include multi-input and multi-output capabilities inside the control module. Therefore, if the advanced control algorithm is programmed with multi-input/multi-output that have assignable sequence computation steps, and the periodic computation is synchronous with the periodic execution of the control module (such as less than a 30-s range), then the advanced control should be implemented at the DCS level. Examples include ratio controls, feedforward controls, logical sequencing, and start-up and shutdown ramping operations during the transition period. A typical application of the multi-input/multi-output algorithm is heater pass temperature balancing control combined with the total feed-flow control. The multi-input/multi-output algorithm in this case is structured to fit the single-input/single-output type, suitable for implementation at the DCS level with single-input/multi-output algorithm structure combined to form the multi-input/multi-output algorithm. The execution of the multi-input/multi-output control period in this case is identical with that of the DCS controller. The control period is usually a multiple basic scan rate of the given controller module in the DCS. The size of the program is usually small. Consequently, the CPU time limit can be satisfied easily. Advanced Control in the Supervisory Computer For control strategies requiring substantial computation time to complete the program execution, the regular DCS controller is not suitable. This function may involve a process model or iterative
1.2 Computer Configurations of Supervisory Units
computations to obtain the convergent solution. In this case a dedicated computer is required. To accomplish this computation task, the periodic execution of the program is typically greater than a 30-s period. This period is a function of such process response dynamic parameters as the time constant, pure transportation delay, program execution elapse time, interaction with the lower loop control execution frequencies, DCS network speed, update frequency of the gateway data table, and external periodic interrupt. Because of the physical distances between the input sources at the DCS and the supervisory computer and because of the multiple scan frequencies involved, the precise timing for the arrival for the commands at the DCS loop set points cannot easily be predicted. It is up to the designer of the advanced control strategy to minimize the timing problem. An example of advanced control being implemented in supervisory computers and using the multi-input/multi-output concept is gasoline blending with optimal product cost control. The criteria for optimal cost can be minimum product quality, maximum profit subject to the availability of the blend component tank inventories, or delivery time constraints. An online optimization algorithm, such as a linear programming method or gradient method, can be utilized to obtain the solution for commands that can be manipulated. The commands to the set points of the flow ratio parameters can be so directed by the supervisory computer that optimum volume ratios are maintained for the components. Mixed Implementation For certain situations, the DCS control module cannot facilitate the advanced control algorithm because of the infrequent execution or the excessive size of the program. In such a situation it is necessary to partition the multi-input/multi-output structure to fit the actual hardware constraint. An example of such a partitioned structure is batch control, where the sequence control can be handled by a supervisory computer, while regulatory and ramping controls can be handled by the DCS control.
COMPUTER INTERFACE WITH DCS
Supervisory computer Link/LAN
Gateway Power supply
LAN/Modem adapter Internal bus DCS communication network controller
Memory
CPU
DCS net
FIG. 1.2i Generic gateway.
gateway device collects the listed data from the DCS communication network or data highway and transfers them to the supervisory computer through a different communication link or network. At the same time, the gateway also receives the data from the supervisory computer or nodes on the computer local area network (LAN) and transfers these to the nodes of the DCS where the information was requested. The generic hardware arrangement of the gateway with respect to the supervisory computer and the DCS network is shown in Figure 1.2i. Typical data transmission speeds for the DCS network are from 1 to 10 million bits per second (MBS). Two types of computer connections are explained here using Honeywell’s configuration of the gateway to illustrate the functionality of each type. Type one, called a computer gateway, communicates with a computer via a dedicated link, as shown in Figure 1.2j. Type two is the plant network module, which communicates with computers through LAN. The computer LAN usually differs from the proprietary DCS network in communication speed and protocol. This is shown in Figure 1.2k. The physical connection of the LAN is typically a bus-type topology; however, the logical connection of the
Hardware Supervisory computers are interfaced to the DCS through a gateway, which is constructed to receive the data or messages from an originating source (see Chapter 4). The gateway resides on nodes of the local control network or data communication network of the DCS. Usually, the communication network topology is either a ring or bus type. The nodes on the communication network consist of the operator console, engineering console, process control unit, controller subsystem, and gateway. The gateways are usually configured through the DCS from the engineering console to establish a list of variables associated with the tag names of interest in the DCS node on the gateway data table. The
© 2006 by Béla Lipták
23
Computer Computer gateway
Local link
Application program link driver
DCS communication network Control system
Control system
Operator station
FIG. 1.2j Gateway for supervisory computer and DCS.
Historical module
24
General
Computer
Computer
Computer software
Computer
User application Local area network (LAN)
languages
Plant network module
Application executive Communication driver DCS communication network
Control system
Control system
Operator station
Computer gateway
Historical module
FIG. 1.2k Gateway for computer LAN and DCS.
DCS network
FIG. 1.2m Computer software structure.
computer-to-computer gateway or to plant network modules can be of the star-type topology. One approach for computer-integrated manufacturing is to design the hardware and software for the gateway such that integration of manufacturing or process data and plant management information can be achieved easily. Such an interface gateway or network module requires: 1. Highly efficient information communication between the DCS network data or messages and the data or messages in the computers on the LAN nodes 2. High security and reliability for both the DCS network access and the computer LAN access
in the data table at the interface module or at the nodes on the DCS. Usually, the data classified for reading are variables such as set points (SP), process variables (PV), control loop outputs (OP), digital inputs, digital outputs, control parameters, device status, controller status, or alarm status. However, the data to the DCS, which is classified for writing, usually are much more restrictive for security reasons. For a computer, the structure of the software architecture is are shown schematically in Figure 1.2m. The data representation in the interface module is converted into the computer format for transmission.
The actual architecture to this type of hardware connection is shown in Figure 1.2l.
CONCLUSIONS
Software The computer interface module or gateway functions such that the commands from the computer can be transmitted to accomplish read/write processing of data or messages
VAX/VMS User application network interface
VAX/VMS User application network interface
DECnet
Network interface data access
VAX/VMS User application network interface
Wide area network
DECnet
Plant network Network interface Module data access TDC 3000
LCN (Local Control Network)
FIG. 1.2l Honeywell TDC 3000 and VAX computers.
© 2006 by Béla Lipták
VAX/VMS User application network interface
At the present time, most DCS vendors are able to interface their proprietary DCS systems with computers. The supervisory computer can be used for data analysis and control commands. Most of the supervisory computers are handling the time-critical tasks. However, some personal computers are able to access directly from historical files for engineering analysis of the process using a database software package. The supervisory computer can assist the operator with the following tasks: performing the supervisory control; alarm classification and prediction; process performance evaluation; process diagnostics; communication routing to other host computers; and online optimization of production. When the supervisory computer is used with conventional analog controllers, it can be used for data analysis and generation of set point control commands.
References 1. 2.
Renganathan, S., Transducer Engineering, Chennai, India: Allied Publishers Ltd., 2003. Mendel, J. M., Discrete Techniques of Parameter Estimation, New York: Marcel Dekker, 1973.
1.2 Computer Configurations of Supervisory Units
3. 4.
Renganathan, S., “Adaptive Fixed Memory Indentification of Discrete Data Systems,” I.I.T., Madras, India. Johnson, C., Process Control Instrumentation Technology, 4th ed., New Dehli: Prentice Hall of India Pvt. Ltd., 1996.
Bibliography Adelman, M. L., “Programmable Controller/Personal Computer Automation Control System,” InTech, Vol. 34, July 1987, pp. 37–42. Berry, A. P., “DCS Enhancements for Advanced Simulation,” ISA/93 Technical Conference, Chicago, September 19–24, 1993.
© 2006 by Béla Lipták
25
Dartt, S. R., “Assess Today’s Option Is Distributed Process Control,” Chemical Engineering Progress, October 1990, pp. 61–70. Gallun, S. E., Matthews, C. W., Senyard, C. P., and Later, B., “Windup Protection and Initialization for Advanced Digital Control,” Hydrocarbon Processing, June 1985, pp. 63–68. Shaw, G., “What to Look for in an Industrial PC,” InTech, March 1993. Shaw, J. A., “Design Your DCS to Reduce Operator Error,” Chemical Engineering Progress, February 1991, pp. 61–65. Shook, P., “PCs Getting Plant Floor Respect,” Control, April 1993. Stafford, U., Serianni, A. S., and Varma, A., “Microcomputer Automated Reactor for Synthesis of Carbon 13 Labeled Monosaccharides,” AIChE Journal, Vol. 36, December 1990, pp. 1822–1828. Tsai, T. H., Lane, J. W., and Lin, C. S., Modern Control Techniques for the Processing Industries, New York: Marcel Dekker, 1986.
