TECH NOLOGY
Direct Digital Control Takes Shape How it will function ultimately is still in question, but there are some indications of the direction in which hardware is evolving How will direct digital control function? This question has set process computer and instrument manufacturers to a good deal of soul searching. A definitive answer still eludes them. Nonetheless, some of the current thinking and some experimental results have surfaced at the Instrument Society of America's 19th annual conference and exhibit in New York and may provide an idea of the direction in which DDC is evolving. Here's how the current situation shapes up for manufacturers: • Users and manufacturers alike accept D D C and see a big potential market for it. Yet so far there is only a smattering of experimental data to act as a guide in developing hardware. • Guidelines for manufacturers, hammered out in two workshops involving representatives from better than two dozen potential user companies (C&EN, May 18, page 21) are available and represent what users feel they will need. But the guidelines are in large part general statements and pose many questions that manufacturers must answer as they get down to specifics. • Manufacturers are working under a time element that has provided little opportunity for DDC to undergo a hardware evolution. It only came to light as a concept in 1961-62. As Fischer & Porter's K. Russel Knoblauch states: "DDC has gone from concept to the brink of hardware in an amazingly short time. In my 40 years in the automation business, I've seen nothing like it." The concept of DDC, if not the way of implementing it, is pretty well defined. Basically, it involves using a computer to provide, on a time-shared basis, the control functions for the many control loops of a process. The computer thus supplants the individual loop controllers conventionally used and is the link between sensing elements and final actuators. The computer, however, has no optimizing or 66
C&EN
OCT. 26,
1964
supervisory function: It operates to maintain a process at predetermined set points, doesn't determine them on the basis of over-all plant goals. Although some D D C installations are in the planning stage in the U.S.—• at Monstanto, Esso, D u Pont, and Dow —they are experimental. D D C is thus still largely just a concept, and translating the concept into hardware is now the job of the equipment manufacturers. So far, two are making public efforts in this direction—Westinghouse with its Prodac 50 computer, and Foxboro with its M/97400 system based on Digital Equipment Corp.'s PDP-5. Westinghouse is working with Monsanto, and Foxboro with Esso. Perplexing. The most important, and the most perplexing, problems facing manufacturers of D D C equipment are the questions of purpose, reliability, and economic justification. Purpose, for example, boils down to two main arguments—the first for direct replacement at lower cost, and the second for greater control sophistication. It is for this reason that the Users' Workshop last May set up general specifications for two types of D D C computers, Type I and Type II. Type I would be primarily for direct replacement. It would be a specialpurpose, wired-program machine but with some minimum programing flexibility. It would be designed to handle up to about 100 control loops and cost about $600 to $700 per loop in the 50-loop range. Type II would be a general-purpose, stored-program machine that would require logging, data reduction, and special control functions and capabilities in addition to the Type I functions. It would typically handle 14-bit words and have 16,000 words of memory. It would cost $50,000 to $100,000, or $100 to $1000 per loop. Most manufacturers feel that D D C will go the Type II route primarily, and that Type I would find only very limited use. (This is the approach
being taken by Westinghouse and Foxboro.) Bill Ware, Honeywell's market manager for digital systems points out two reasons for this, without taking equipment cost into account: First, very complex programs can easily be entered into a storedprogram machine, and, second, the inevitable changes to the program can be easily made in this machine. At the May workshop, users also defined a Type III computer, today's conventional optimizing computer, saying that because of its greater complexity than Type II it would not find application for DDC. The Type HI would handle 18- to 24-bit words, have core and drum memories of up to 32,000 words, and cost upward of $150,000. In comparing Type II with Type III, Mr. Ware points out that the shorter word length of the Type II is permitted primarily to lower the cost of the computer. But, he says, it isn't long enough to address directly the amount of memory stipulated. Thus, the program for the Type II could be 30 to 7 0 % longer than that for the Type III and more than offset the cost advantage. The minimum word length of 18 bits indicated for the Type III would seem to be the lowest possible cost compromise for the Type II, he says. Furthermore, the same type of computer logic design and software system is required to solve control algorithms in the Type II as is needed to solve plant optimization formulas in the Type III. The main difference between the two computers thus becomes the question of reliability—99.95% availability specified by the users for DDC, compared to the 99.5% currently achieved by Type Ill's. Reliability is a sticky problem. In a conventional control system, an analog controller can fail, generally without too noticeable an effect. But with DDC, if the computer fails, it could result in complete plant shutdown, depending on the process. Thus such
reliability terms as per cent availability or mean-time-before-failure (MTBF) leave much to be desired. A company, as Mr. Ware says, would probably not look too kindly on the failure of its computer, even though the lack of failure of two dozen computers installed by competitors meant that the MTBF was higher than specified. Means of achieving complete reliability (such as a redundant computer) are much too costly. Most manufacturers feel that the answer lies in providing various forms of backup. For example, Mr. Ware says, some installations will warrant including manual control stations on some loops and complete conventional control on others. But the question remains to plague manufacturers: How much cost above that for conventional control can be borne by users for the improved control with DDC and still leave economic justification for DDC? Specifics. Meanwhile, some work has been continuing on specific technical implementation of DDC. Foxboro has made studies including analog and hybrid computer simulation and actual digital computer control of pilot processes. It has also made detailed engineering studies of several industrial processes and has constructed experimental equipment assemblies to study problem areas that arise with DDC. John W. Bernard and Joseph F . Cashen, both of Foxhoro, point out that technical requirements of D D C pose some unique questions. These include type of valve actuation, form
(rate). Many forms of equivalent digital control equations can be used, but for use with full-speed-actuation valves, all that's required is to calculate a change in position. In this case, reset and rate terms become dimensionless constants. This means that the wide range of mode adjustments needed in analog controllers won't be needed with DDC, if sampling time is chosen in relation to process dynamics. Furthermore, in general, use of the derivative term gives no practical improvement, and a modified proportional-integral equation gives the best results. • Sampling time—the frequency at which measurements are made of a variable—is a function of the process dynamics and the control equation used. At a practical balance of equipment size against system performance, experience to date shows that the fastest sampling time needed for control would be from 1.0 to 1.5 sec. for liquid flow control. Other types of loops would vary widely, but sampling time would not be greater than 10 to 20 sec. for a reasonable response to set point changes and for alarm requirements. • The input quantization level must be good enough to preserve the basic sensitivity of the measurement transducer, whereas the best output quantization is only a function of control performance. So far, both input and output quantization of one part in a thousand have been used, since other levels produced no significant control improvement or savings in equipment complexity.
of feedback law, sampling time, and level of quantization needed when putting process variables into digital form. In a conventional analog system, Mr. Bernard and Mr. Cashen explain, a continuous measurement is sent to the controller, which sends a continuous signal to the control valve. Set point and mode adjustments (proportional, reset, and rate) are manually made in the controller. In a digital system, however, a sampled quantized measurement signal is serially detected by the computer, which sends a multiplexed, quantized signal to the control valve. The same set point and mode adjustments are required as in the analog system, but, in addition, sampling time and quantization must be specified. Some of the conclusions Foxboro has reached on the basis of its experimental work so far: • From a number of practical considerations, transmitting a change of valve position from the computer, rather than absolute valve position, is preferred for DDC. Of the two main ways of doing this—continuous actuation from one signal to the next as the valve changes position, or full speed actuation to a new position where the valve holds until the next signal—full speed actuation with the fastest practical speed gives the best performance. • The position control equation used today in conventional analog controllers contains terms for error, gain (proportional), integral time constant (reset), and derivative time constant
. . . will be replaced by one timeshared computer system such as this
In direct digital control, many analog control loops such as this . . .
Analog signals
Control signals
from sensing elements
to valves
Process
Control signal to valve
Single analog signal
Analog/digital i^
Direct digital —
converter
*
•
control computer
Analog controller
i—r
Set point
— •
Digital/analog converter
1
Set points, nnodes djustments i
1
Mode adjustments
. . . but specific means of implementing DDC awaits development of hardware by manufacturers OCT.
26, 1 9 6 4 C & E N
67